Adopting Modern Fitness Sensors to Improve Patient Care [original]

Ulm University | 89069 Ulm | Germany Faculty of Engineering,

Computer Science

and Psychology

Institute of Databases and

Information Systems

Adopting Modern Fitness Sensors

to Improve Patient Care

Master’s Thesis at Ulm University

Submitted by:

Christoph Bachmaier

[email protected]

Reviewer:

Prof. Dr. Manfred Reichert

Prof. Dr. Martin Theobald

Advisor:

Dipl.-Inf. Marc Schickler

2015

Printed on August 31, 2015.

Some rights reserved.

This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0

International License. To view a copy of this license, visit:

http://creativecommons.org/licenses/by-nc-sa/4.0/

Abstract

Technology found in modern fitness sensor devices advances at a very fast pace and current

smartwatches are on the verge of closing the gap between being an everyday object and a

medically reliable monitoring device.

In this thesis, the possibility of adopting fitness sensor devices in medical environments

is explored and use cases in which sensor devices can be deployed are examined. Their

successful transfer from the area of sports to medical analyses and treatments may help

patients to deal with their illnesses and to improve the level of patient care found today.

Privacy and security issues as well as social concerns associated with such a disruptive

evolution are discussed and practical tests of a pulse oximeter in various activities of

daily living are conducted. The collected health data depicts a close representation of

the performed activities. Furthermore, three types of fitness sensor devices were used in

different real-life scenarios and the resulting data is compared. The results show that the

recorded vital signs may differ significantly, depending on the scenario.

iii

Contents

1 Introduction 1

2 Related Work 3

3 Vital Signs and Sensors 7

3.1 VitalSigns...................................... 7

3.1.1 BloodOxygen ............................... 8

3.1.2 Heartbeat.................................. 8

3.1.3 BloodPressure............................... 9

3.1.4 BloodGlucose ............................... 10

3.1.5 PhysicalActivity .............................. 10

3.1.6 SkinTemperature ............................. 11

3.2 Methods of Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.2.1 Blood Oxygen Saturation . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.2.2 Heart Rate and Pulse . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2.3 BloodPressure............................... 15

3.2.4 Blood Glucose Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.2.5 Acceleration ................................ 18

3.2.6 SkinTemperature ............................. 19

4 Fitness Sensor Devices 21

4.1 DeviceOverview .................................. 21

4.2 Wireless Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.2.1 Bluetooth .................................. 23

4.2.2 BluetoothSmart .............................. 24

4.2.3 Near Field Communication . . . . . . . . . . . . . . . . . . . . . . . . 26

4.3 Hands-onTests................................... 27

4.3.1 iHealth Wireless Pulse Oximeter PO3 . . . . . . . . . . . . . . . . . . 27

Contents

4.3.2 Measurements............................... 30

Basic Indoor Activities . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

OutdoorCycling .............................. 32

Outdoor Activities of Daily Living . . . . . . . . . . . . . . . . . . . . . 35

4.3.3 Comparing Sensor Devices . . . . . . . . . . . . . . . . . . . . . . . . 38

Sitting.................................... 39

Standing .................................. 42

Cycling ................................... 44

5 Medical Applications 47

5.1 Sensitivity vs. Specificity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

5.2 PatientCompliance................................. 50

5.3 Doctor’sAcceptance................................ 51

5.4 UseCases ..................................... 52

5.4.1 Automatic Oxygen Regulation . . . . . . . . . . . . . . . . . . . . . . . 52

5.4.2 Improved Fall Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.4.3 Dosage of Medication . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

5.5 Treat the Patient, Not the Disease . . . . . . . . . . . . . . . . . . . . . . . . 57

6 Concerns 61

6.1 Privacy and Security Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

6.1.1 DataGathering............................... 62

6.1.2 DataTransfer................................ 62

6.1.3 DataStorage................................ 63

6.1.4 Attackers and Countermeasures . . . . . . . . . . . . . . . . . . . . . 64

6.2 EthicalIssues.................................... 65

6.2.1 GeriatricCare ............................... 65

6.2.2 Quantified Self and Loss of Liberties . . . . . . . . . . . . . . . . . . . 66

7 Conclusion and Future Work 69

A SPSS Output 73

A.1 TestsofNormality ................................. 73

A.2 Q–Q Plots and Frequency Histograms . . . . . . . . . . . . . . . . . . . . . . 74

List of Figures 77

Contents

List of Tables 81

List of Abbreviations 83

References 85

vii

1 Introduction

In recent years, public awareness of the importance of fitness and health has increased.

People try to live healthier and exercise more. However, that often lasts only for a short

period of time and most people fall back into bad habits. As a way to stay motivated longer,

many people start keeping track of their sport activities and the resulting progress.

Fitness

sensor devices

can be a helpful tool in this difficult mission. They record

vital signs

and other

measurable characteristic values during the activities and provide ways for their users to later

analyze the data and share results with friends, which gives them additional motivation due

to peer recognition. Companies have recognized this demand and the need for technological

progress in this area to help their customers quantifying themselves.

As a result, the range of available fitness sensor devices has widened considerably. Ad-

vancements were made not only regarding the size of the devices and their connectivity

via wireless communication technologies such as

Bluetooth

Bluetooth Smart

and

Near

Field Communication

, but also regarding their measuring capabilities. What once were

single purpose sensor devices, has now advanced into devices carrying multiple sensors

combined. Current fitness sensor devices can track, among other data, the user’s

pulse rate

blood oxygen saturation

skin temperature

and his

physical activity

. Some of these features

can also be found in

smartwatches

which are another up and coming trend in mobile de-

vices. Nowadays, smartwatches mainly take over some of the displaying functionality from

smartphones but they also support measurement techniques such as

pulse oximetry

that

determines pulse rate and blood oxygen saturation.

Many of the vital signs and characteristic values that can be measured by fitness sensor

devices are equally relevant to the medical field, in particular to examinations, diagnostics

and treatments. When combining this functionality with the wearing comfort and discretion

of the form factor of a watch, it can be assumed that fitness sensors are well suited to

complement current medical equipment. The potentially resulting benefits regarding patient

1 Introduction

care would not only be an improvement for the patients themselves but also for the medical

personnel treating them. Well-known technology companies seem to have already identified

this gap in monitoring equipment, such as Google, who announced that it is planning to

release a sensor wristband meant to be used in professional medical contexts rather than

for simple fitness tracking [9].

Even though fitness sensor devices may be able to provide medical information that has

been missing thus far, one has to recognize that this type of information, namely health data,

is considered very private. If the devices are worn constantly over a prolonged period of

time, what they are meant to be, the analysis of the recorded data poses a substantial risk

to the users’

privacy

. As health is also a very personal topic, there are

social

and

ethical

issues that have to be considered very carefully.

This thesis aims to show that fitness sensor devices are suitable and advanced enough

to be adopted as medical equipment. It examines the usability and informative value of a

fitness sensor in a set of diverse real-life scenarios. It further analyzes the accuracy of the

recorded vital signs compared to fitness devices that deploy different methods of measuring.

Interviews with doctors and medical personnel were conducted to find fitting use cases in

medicine that can benefit from the introduction of such sensor devices. Technical, medical

and social aspects as well as limitations that may occur are discussed.

The remainder of this thesis is structured the following way. First, an overview of related

research is given in Chapter 2. Chapter 3 then lays out basic information about vital signs

and describes methods of measuring these different vital signs. Chapter 4 first gives a listing

of current fitness sensor devices. It goes on to present wireless communication technologies

commonly used in these devices and in smartphones. Following that, the fitness sensor

device is introduced which was used in real-life scenarios. Recorded health data is analyzed

in regard to its capability to reflect activities performed during the performance of tests.

Statistical tests are conducted to compare the resulting data of three fitness sensor devices

to determine their deviations. In Chapter 5, additional medical concepts relevant to this

thesis are explained, after which use cases are discussed in which sensor devices are

deployed. Chapter 6 discusses privacy and security issues as well as social challenges

that such application brings with it, while Chapter 7 summarizes the findings and discusses

future areas of research.

2 Related Work

Automating certain aspects of patient care will be one of the many advantages of the

adoption of sensor devices into health care and it will not only help reduce the work load of

medical personnel but also improve existing processes. In [

], Marschollek et al. identify

fall events as a major problem in regards to mortality, morbidity and financial costs, especially

among the elderly. As a result, it is important to make an accurate

risk assessment

about

which patients are most likely to fall in order to be able to take precautions.

In a study Marschollek et al. conducted, they examine the fall risk of

119

inpatients. They

use conventional assessment tests, assessments done by an interdisciplinary care team, a

newly developed logistic regression model and a newly developed sensor-based system.

The regression model is based on basic clinical data such as

sex

body mass index

age

and

also includes results from assessment tests such as the

Timed Up and Go test (TUG)

. The

sensor-based assessment is based on gait and motion parameters such as

kinetic energy

pelvic sway

step length

and others that were recorded during a TUG and a

meter walk.

The sensor equipment used is a wireless

triaxial accelerometer

that is attached to a belt

worn around the waist. After one year, they revisit and interview

patients (mean age

81.3

years) about possible falls they had experienced.

of the initial

119

patients could not be

included due to death, progressed illness, withdrawn consent or corrupted data. The results

show their two proposed methods of basing the decision on logistic regression models and

sensor data (accuracy of

72%

and

70%

) to be slightly better than the three existing methods

(accuracy of

48%

50%

and

55%

). However, they concede that the moderate performance of

the existing methods may be due to the small sample size.

In [

], Schobel et al. present

XFitXtreme

, a proof-of-concept implementation of a mobile

business application that integrates multiple sensor devices. Besides the fitness training

scenario, they suggest three other possible use cases: A mobile health care business

application may support physician in their daily

medical rounds

and supply additional health

2 Related Work

data in

unforeseen alarming

situations. It may enable rescue services to

faster determine

what kind of help a patient they encounter needs and it may also give

additional information

about a patient filling out a psychological questionnaire. After laying out the different ways

that can be used in development, they chose to make XFitXtreme a

native

application

over a

web-based

hybrid

application. Only that method provides them access to all the

needed components like the Bluetooth communication stack which the other types do not

expose or not to the extent necessary for the designed application. During the development

process, the authors use two sensor devices: a

heart rate monitor

and a

pulse oximeter

They encounter problems caused by

discrepancies

between these devices when it comes to

establishing a Bluetooth connection, by

lacking documentation

of Bluetooth packet formats

and by interference on the used frequencies.

Another approach to monitoring in health care similar to the one discussed in this work

is the use of

wireless sensor networks (WSNs)

. WSNs consist of a number of very small

autonomous sensor devices called sensor nodes or

motes

which can communicate wire-

lessly and usually relay their gathered information through a

gateway node

to an application.

CodeBlue

[

], developed in 2004 by Harvard researcher Malan et al., is the first WSN that

was designed for a medical environment. As this was an early design of a sensor network

for medical environments and mass casualty events, the authors concentrate mainly on

the basic challenges such as setting up an

ad hoc

network without a preexisting wireless

infrastructure and motes that have to be programmed via very low-level software.

In [

], Kargl et al. discuss security and privacy issues on the basis of the WSN-based

ReMoteCare

, a Pervasive eHealth Monitoring System (PEMS) they developed. They define

the three distinct environments in which WSNs can be deployed:

individual home monitoring

hospital monitoring

and

large-scale monitoring

. The authors further list different types of

attacks and derive corresponding types of attackers for them. An

external passive

attacker

is not part of the sensor system and thus can not control it but can gather medical data

by eavesdropping. Even with encryption in place he may still use the encrypted traffic to

track users’ locations or activities. An

external active

attacker can try to interact with the

sensor system by modifying or forging genuine traffic. He can also try to overload the

system using a denial-of-service (DoS) attack, temporarily rendering it unusable. Attackers

that have control over system components, like employees or users, are named internal

attackers. The

internal passive

attacker can also only listen to the wireless communication,

however, he might circumvent encryption mechanisms. The

internal active

attacker can do

the most damage as he controls part of the system and can easily modify or forge data. The

authors then specify different parts of the system that could be attacked and propose the

countermeasures

encryption

authentication

integrity checks

and prevention of DoS attacks

redundancy

. These security measures also take care of some of the privacy issues

because only authorized medical staff has access to the recorded files. Issues regarding

location privacy could be further lessened if location data is exclusively generated by the

local sensor device and only revealed in emergency situations.

Security and privacy issues of health care information technology in general are discussed

in [

]. Meingast et al. use the scenarios of

electronic patient records

and

in-home remote

patient monitoring

to raise the questions of who owns health data, what should be stored

and where and who has access to this data. They further discuss the possibilities of mining

such medical data. Their solution, once more, is the use of encryption, authentication and

access controls and also the development of

rules

and

policies

that regulate the access to

patient data.

Igual et al. [

] present a review of recent studies about fall detection systems, a topic

that is discussed in Section 5.4.2 of this thesis. They identify two categories in which

all existing studies can be divided:

context-aware systems

(

151

papers) and

wearable

devices

(

197

papers).

studies are included in both categories as they use a mixture of

these two methods. Context-aware systems use stationary sensors that were previously

distributed in the target area. Such sensors can be

video cameras

floor sensors

pressure

sensors

infrared sensors

and

microphones

. Then, relevant features are extracted in order

to distinguish falls from ordinary

activities of daily living (ADL)

. A big advantage of context-

aware systems is that there is no need for the user to constantly wear a sensory device

on his body. At the same time, this also poses a disadvantage as the fall detection system

is thereby limited to the area in which the sensors were placed beforehand. The opposite

holds true for systems that use wearable sensor devices. Falls can be detected

everywhere

inside and outside, but only if the users wears the sensor device. The wearable sensor

device used in this kind of systems can be a

stand-alone

accelerometer or gyroscope, or

they are sensors

integrated

into smartphones. Wearable sensor systems can use simple

threshold-based

methods for fall detection or

machine-based

methods. The latter are more

sophisticated but require a higher level of mathematical understanding. Many studies report

high

accuracy

sensitivity

and

specificity

levels between

90%

and

100%

, however a major

flaw in nearly all studies is the absence of elderly test subjects, the overall low number of

2 Related Work

test subjects, the lack of long-term observations and the lack of tests conducted in real-life

situations.

Important points of criticism regarding

single-purpose

fall detection systems are

cost

and

acceptance

of the application specific sensor devices and equipment they depend on. In

[

], Sposaro et al. propose the deployment of a fall detection system that uses built-in

sensors of smartphones for the detection. Due to the widespread use of smartphones, even

among the elderly, there often are no additional costs and the users are already familiar with

the interface to operate the system. They are also more discrete than dedicated devices.

The authors present

iFall

as their solution, a fall detection application for Android — not to be

mistaken for iFall developed by Salomon et al. [

], which uses a dedicated hardware device.

Besides the actual detection of falls using the internal sensors, they focus on preventing

situations in which a fall is

wrongly

detected and emergency services are requested by

mistake. They implement several

safeguards

to reduce the number of false positives. Two of

them are a screen shown to the user after a detected fall on which he can abort any further

actions and the fact that the starting position of the smartphone before an assumed fall is

taken into account.

A slightly different approach is taken by Majumder et al. [

], in which they use built-in

smartphone sensors and try to

prevent

falls before they happen. In their implementation

iPrevention

, an iOS application for iPhones, they record and analyze the

gait

of the user.

They use the gait and its patterns found in the way people normally walk to establish a

baseline. If certain

abnormalities

occur in the gait, a visible and audible alarm is triggered

that informs the user about his increased risk of falling so he can take appropriate measures

to prevent a fall from happening.

3 Vital Signs and Sensors

This chapter covers the medical basics that are needed throughout the remainder of this

work. It describes the vital signs that modern sensor devices are capable of recording and

gives details about different methods of how these vital signs can be measured manually

and digitally with sensors. Invasive methods are generally excluded because they can only

be performed in a medical environment and are not relevant to the everyday use of fitness

sensor devices by medical laymen. One exception is made for

blood glucose

in Section 3.1.4

as the accuracy of currently available non-invasive methods is too low.

Even though some of these characteristic signs such as

blood glucose

and

blood pressure

are not built into

multi-sensor

fitness devices because of the particular way they have to

be measured, there are

stand-alone

sensor devices that support them. If needed, these

stand-alone devices can be used in conjunction with other fitness sensor devices for the

time being, until new ways are found that allow them to be integrated with other sensors.

3.1 Vital Signs

To get a basic understanding of what each vital sign represents, this section gives a short

explanation for each of the vital signs relevant to this work. Although not vital signs in

a strict clinical sense, the characteristics

physical activity

and

skin temperature

are also

listed here. Especially physical activity is a feature that is included in a large number of

consumer sensory devices. Even if these two characteristics may not be significant enough

to draw conclusions about possible medical conditions of a patient, they may be used as

supplementary

source of information to interpret other recorded vital signs more reliably

and to make more accurate medical diagnoses.

3 Vital Signs and Sensors

3.1.1 Blood Oxygen

The human body needs

oxygen (O2)

in order for it to survive. Oxygen is inhaled with every

breath and is transported to the lungs where it is transferred to the blood by diffusion. A

very small amount of oxygen is dissolved in the blood (only about

) while most of it is

chemically bound to the so called

hemoglobin (Hb)

, a protein that is a component of red

blood cells. Each hemoglobin molecule can bind up to four O

-molecules. The

oxygenated

blood is used to distribute oxygen throughout the body where it gives off some of the oxygen.

The then

deoxygenated

blood takes on

carbon dioxide (CO2)

, an element that is produced

when nutrients and oxygen are converted into energy to sustain the human body. Carbon

dioxide has to be released from the body so some of it is dissolved in the blood and some of

it bound to hemoglobin and other molecules and then transported back to the lungs where

the carbon dioxide is diffused into the lungs and exhaled [

]. Section 3.2.1 describes ways

to measure the amount of oxygen contained in blood, called the oxygen saturation.

3.1.2 Heartbeat

In order for the steps mentioned in Section 3.1.1 to work, the blood somehow has to be

transported through the body. This function is carried out by the heart and the beating of

the heart. One

heartbeat

, also called one

cardiac cycle

, consists of the two major phases

systole

and

diastole

. In the systole, the heart is contracted by the heart muscle, builds up

pressure and pumps oxygenated blood out into the

aorta

, the main artery, and deoxygenated

blood out into the blood vessels in the lungs.

Artery

is the medical term for a blood vessel

that carries blood

away

from the heart. Arteries carry deoxygenated blood when leading

towards the lungs and oxygenated blood when going into the rest of the body. Blood vessels

that carry blood

towards

the heart are called

veins

which also carry both kinds of blood.

Oxygenated blood from the lungs to the heart and deoxygenated blood from the rest of the

body towards the heart (cf. Figure 3.1). In the second phase, the diastole, the heart muscle

relaxes and the heart is filled with blood again. Even without taking diseases into account,

there are many factors that can influence how frequent a heart has to beat:

age

sex

physical

fitness

psychological state

weather

and more. Since the heart is such a crucial organ (it

pumps around

8,600

liters of blood per day), it is essential that it functions correctly [

Different methods of measuring the heartbeat are discussed in Section 3.2.2.

3.1 Vital Signs

To Lungs

From Lungs

To Body

From Body

Figure 3.1:

Simplified blood flow through the human heart. Red arrows stand for oxygenated

blood, blue arrows represent deoxygenated blood [47].

3.1.3 Blood Pressure

Like every liquid that is pumped through pipes, the circulating blood exerts pressure on

the pipes it is pumped through. In case of the human body, those pipes are blood vessels

arteries

and

veins

. How high this

blood pressure

is depends on a variety of factors.

Some parameters such as

age

sex

genetic predisposition

and overall

state of health

have

a constant or only slowly changing effect. Others affect the blood pressure rapidly but just

for a short time span. These short term effects include

physical activities

psychologically

stressful

situations and

diseases

. Furthermore, blood pressure shows an

endogenic circa-

dian periodic

pattern which means that it adapts to the human 24-hours day and has its

minimal values at around 3:00 a.m.

and its maximal values at around 3:00 p.m.

local

time. Section 3.2.3 gives insights on how to measure the blood pressure level. Low blood

pressure, called

hypotension

, and its entailing poor blood flow can result in, for instance,

cold hands and feet, headaches and dizziness. If the blood pressure is too high for a longer

period of time, a condition called

hypertension

, it can strain blood vessels and cause heart

attacks, strokes and kidney damage [44, 24, 53, 1].

1Latin for ante meridiem; in the morning

2Latin for post meridiem; in the evening

3 Vital Signs and Sensors

3.1.4 Blood Glucose

Food supplies vital nutrients to the human body that are then converted to energy. A

particular kind of essential nutrients are

carbohydrates

which consist of sugar-molecules.

Monosaccharides are carbohydrates that consist of only one molecule or sugar. It is also

called

glucose

and sometimes just referred to as sugar. As one of the main sources of

energy, glucose, is transported through the circulatory system by the blood. The glucose

level discussed in Section 3.2.4 is a measurement for the amount of glucose in blood. This

level can fluctuate and if it is low, humans become hungry, the signal that new energy is

needed. If it sinks too low, it can cause seizures and reduced brain functions. A constantly

heightened glucose level in the blood can be the result of

missing insulin

or a built up

insulin

resistance as insulin normally helps to regulate the blood glucose level. The high level can

be a symptom of

diabetes mellitus

which is going to be an important issue in the future.

Its alarming development is visualized in Figure 3.2. The estimated number of

387

million

people (aged

years old) suffering from diabetes in the year 2014 is expected to

increase to 592 million by the year 2035 [30, 24, 53].

100

200

300

400

500

600

2000 2005 2010 2015 2020 2025 2030 2035

People with Diabetes [Million]

Time [Year]

Figure 3.2:

Estimated number of people (ages

) suffering from diabetes (solid line)

and its predicted growth in the near future (dashed line) [29, 30].

3.1.5 Physical Activity

Even though

physical activity

is not a typical vital sign per se, it can give useful information

when analyzed together with other vital signs. Collected data about the physical activity of a

3.2 Methods of Measurement

patient can include the

position

of his body (or at least of the body part the sensor is placed

on) relative to the floor and the positive and negative

acceleration

his body experiences (or,

again, the acceleration of the body part where the sensor is placed). More information can

be derived using these two sets of data, for example a system for

fall detection

as discussed

in Section 5.4.2 or for counting the steps taken in a day. The technical side of measuring

these parameters is described in Section 3.2.5.

3.1.6 Skin Temperature

The human body needs a relatively

constant

temperature of

36.5−37◦C

in its core, where

all the important organs are located, to function properly [

]. If it deviates to much from

this default value, the body has regulatory mechanisms it can use to bring the temperature

back to normal. One way to counteract too low core temperature is the

shivering

of muscles

that produces heat. Another is to

reduce

the blood flow to the extremities and the skin

which results in a lower

skin temperature

as the heat is kept more to the core of the body. A

way to decrease the core temperature in case it is too high is

sweating

. Sweat on the skin

evaporates which results in a

cooling effect

. The temperature of the skin is normally slightly

lower than the core temperature [

]. Since the skin temperature depends very much on

external factors and can vary substantially, it is difficult and not always possible to draw

conclusions about the core temperature from it. However, skin temperature can be a helping

indicator to prevent false diagnoses when treating patients with chronic pain [

] and in other

cases it may be used as an

additional

source of information and help to better interpret the

results. Section 3.2.6 gives information on how skin temperature can be measured.

3.2 Methods of Measurement

The previous section established a basic understanding of the different characteristic signs

used in medical diagnostic. However, these vital signs have to be measured in order to

be used as indicators and parameters of medical conditions. This section describes how

these measurements are taken and what sensors can be used to integrate them into sensor

3 Vital Signs and Sensors

devices. Having that in mind, the main focus is directed to

non-invasive

methods as these

are the important ones for long term measurements in everyday situations.

3.2.1 Blood Oxygen Saturation

The standard non-invasive method of measuring the

arterial blood oxygen saturation

pulse oximetry

. To be precise, pulse oximetry measures the amount of hemoglobin that

carries oxygen in relation to hemoglobin that has no oxygen bound at the time, as detailed in

Section 3.1.1. A pulse oximeter has two

light-emitting diodes (LEDs)

of different wavelengths.

One with

red

light at a wavelength of

660 nm

and one with

infrared (IR)

light at a wavelength

940 nm

. Placed opposite of the two LEDs is a

photodetector

, normally a photo diode, that

can sense these two wavelengths (cf. Figure 3.3). In order to measure the oxygen saturation,

the patient’s fingertip or earlobe is placed between the LEDs and the detector, usually by

clipping the pulse oximeter to either one of them. The device

alternates

between activating

the red LED, activating the infrared LED and having both LEDs off and the photodetector

takes three measurements of the light that passed through the body part. One measurement

when the red LED is on, one when the infrared LED is on, and one when both LEDs are

off in order to account for

ambient light

that may otherwise interfere with the result. Then it

starts again with the red light and goes through this loop again and again.

Red and Infrared LEDs

Ambient Light

Photodetector

Figure 3.3:

Pulse Oximeter placed on a finger with the red and infrared LEDs, interfering

ambient light and photodetector.

3.2 Methods of Measurement

Some of the red light is

absorbed

while passing through the body part and deoxygenated

blood absorbs more of the red light than oxygenated blood. In contrast, more infrared light is

being absorbed by oxygenated blood than by deoxygenated blood when passing through

the body part. The oximeter further utilizes the

changes

in absorption that happen when

the blood circulates after each heartbeat to exclude

static interference

caused by skin, fat

and other tissue, for example, so the oxygen saturation of just the arterial blood can be

determined [

]. As a byproduct, the pulse rate that is discussed in the next section can be

measured.

The general abbreviation for oxygen saturation is

SO2

. To clarify which method was used to

determine the oxygen saturation, the terms

peripheral capillary oxygen saturation (SpO2)

and

arterial oxygen saturation (SaO2)

are used. SpO

refers to the measurement by pulse

oximetry and SaO

refers to the

invasive

method of arterial blood gas, at which a blood

sample is taken from one of the arteries and the oxygen saturation is analyzed in a laboratory.

Normal values for SpO

are

96 −100%

[

]. Pulse oximetry, especially with fitness sensor

devices, is susceptible to

interference

such as nail polish on fingernails. More factors are

listed in Section 4.3.1.

3.2.2 Heart Rate and Pulse

Every time the heart beats, it pushes blood through the entire human body as described

in Section 3.1.2. The frequency of these beats is called

heart rate

and is measured in

beats per minute (bpm)

. The circulating wave of blood being pumped into the arteries and

through the body with each heartbeat results in a sudden and short increase in pressure

within the arteries and veins. This spike can be detected in the head and the extremities (i.e.

arms and legs) and the pulsating wave is called the

pulse wave

. The rate of these spikes,

the

pulse rate

, is measured in beats per minute (bpm) as well. The pulse includes further

characteristics to describe the pulsation of blood: Is the pulsation

evenly

spaced or is it

arrhythmic

? Is the pulse

weak

strong

? How

quickly

does the

increase

and

decrease

blood in the artery happen? Average pulse and heart rates for healthy humans when resting

are 60 −100 bpm [58].

The heart rate can be measured in a non-digital way by the use of a

stethoscope

. It is

placed on the upper chest of the patient in the area where the heart is located and it enables

3 Vital Signs and Sensors

the user to hear the so called

heart sounds

. Normally, two somewhat different sounds can

be heard for every beat of the heart. These sounds are caused by

muscular contractions

closing different valves in the heart. The heart rate of a healthy adult is the half the number

of sounds heard in

seconds. A digital method of measuring the heart rate and the go-to

measurement for many health issues is the

electrocardiography (EKG)

. It can give much

more detailed information about the condition of a heart than just the heart rate. The EKG

uses the fact that the beating of a heart is triggered by small electrical currents and that

there is depolarization in heart muscle cells during every heartbeat. This creates electrical

fields that can be measured outside the body by electrodes placed on the skin. These fields

are then used to derive graphs such as the one in Figure 3.4 called

electrocardiogram

which

can then be used for medical diagnoses. For example, the time lapsed from the top of one

spike (called R wave) to the next is called

RR-interval

and it can be used to calculate the

heart rate for this moment in time [17].

Figure 3.4: Electrocardiogram of a healthy young adult.

The same principles apply for fitness sensor devices that use a

chest belt

to read the user’s

heart rate. Two electrodes are placed on the inside of the belt with a little gap between

them and need direct contact with the skin, like the EKG. Then, the measured heart rate

is often displayed on a device that can be worn around the wrist, such as the device in

Section 4.3.3. Or it is sent to a smart mobile device. While the EKG and chest belt can be

used to continuously record the heart rate over a long period of time, there are also devices

like stand-alone watches that have two electrodes on their surface. The user has to touch

these two electrodes with two of his fingertips to get a reading of his heart rate. After he

removes the fingertips from the electrodes the readings stop, so this solution is not feasible

for long term recordings [50].

3.2 Methods of Measurement

Manual measurement of the pulse rate is usually done by putting two or three fingertips on

a place of the body where an artery can be felt from outside the body and applying slight

pressure. Such places are

wrists

ankles

temples

neck

and others. Then, the number of

spikes that occur within a 60 second time frame is counted and that number is the pulse

rate. Using modern technologies, the pulse rate can also be measured via pulse oximetry

as described in Section 3.2.1. Some fitness sensor devices use a slight variation to pulse

oximetry with one red and one infrared LED. They have

green

LEDs instead, as does, for

example, Apple’s

Apple Watch

, which has infrared and green LEDs. The principle stays the

same, however oxygen saturation and pulse rate are then measured with visible light.

Under normal circumstances, if the blood flow to the extremities is not obstructed by disease

or physical influences from the outside and the heart functions at its full potential, the heart

rate and the pulse rate are equal. With this in mind, the terms heart rate and pulse rate are

used interchangeably throughout this thesis if not explicitly stated otherwise.

3.2.3 Blood Pressure

The most common non-invasive method of measuring

blood pressure

in clinics and practices

is the

Riva-Rocci

method, named after the Italian physician Scipione Riva-Rocci (1863–

1937) who improved the blood pressure meter that was used at that time and whose principle

is still used today (cf. Figure 3.5(a)). The doctor places a

cuff

around the patients upper arm,

ideally at heart level to prevent hydrostatic influences, and inflates it while listening to the

arterial pulse of the arm with a stethoscope. By inflating the cuff, it applies pressure around

the arm until the pressure is high enough to stop the blood flowing through the constriction

even when the heart pumps blood into the arteries in its systole phase. No pulse can be

heard through the stethoscope at that time. The doctor then slowly releases the air from

the cuff, thereby decreasing the pressure on the upper arm, and simultaneously listens for

the blood to start flowing again. When this happens and the rushing blood can be heard

through the stethoscope, he takes note of the current pressure from a

manometer

that is

attached to the cuff. This pressure is called the systolic blood pressure as it represents the

highest blood pressure generated by the systole of the heart. The doctor proceeds to lower

the pressure until the sounds of the pulse become dull or fade rapidly. Then, the pressure

displayed on the manometer corresponds to the lowest blood pressure, called the diastolic

3 Vital Signs and Sensors

blood pressure, that happens in the diastole phase when the heart expands and refills with

blood [63, 53].

(a) Manual Riva-Rocci method. (b) Automatic blood pressure monitor.

Figure 3.5: Manual (a) and automatic (b) blood measurement methods [21, 43].

Persistently high blood pressure is called

hypertension

and the opposite, prolonged low blood

pressure is call

hypotension

. Hypertension is further categorized by its severity. Table 3.1

shows the three categories of hypertension according to the

World Health Organization

(WHO)

[

]. The unit of measurement of blood pressure is

millimeters of mercury (mmHg)

and the blood pressure of a healthy

20 −40

-year-old adult should be at around

120 mmHg

systolic and

80 mmHg

diastolic. A systolic pressure of

100 −150 mmHg

and a diastolic

pressure of 60 −90 mmHg are considered normal.

Hypertension systolic (mmHg) diastolic (mmHg)

Grade 1 140 −159 90 −99

Grade 2 160 −179 100 −109

Grade 3 ⩾180 ⩾110

Table 3.1: The three grades of severity of hypertension according to the WHO.

Digital blood pressure meters (cf. Figure 3.5(b)) that work

autonomously

without a doctor

listening with a stethoscope use the

oscillometric

method. The device typically has

piezo-

electric

sensors that convert pressure into an electrical signal. When an occluding cuff that

has these sensors integrated is placed around an arm and blood flows through the arm, the

pulse influences the pressure of the cuff and using the magnitude of the resulting oscillations

the pulse can be calculated. The actual measurement takes place in a similar way to the

3.2 Methods of Measurement

Riva-Rocci method. The cuff is filled with air until the pressure exerted on the arm completely

stops the blood flow. Then, the pressure is slowly reduced until faint oscillations are detected.

After continuing to reduce the pressure, these oscillations disappear again [61, 35].

3.2.4 Blood Glucose Level

The blood

glucose level

represents the amount of glucose that is contained in blood and

is measured in either

milligrams per deciliter (mg/dl)

millimoles per liter (mmol/l)

. Most

glucose meters available today take their measurements in a

minimally

invasive way. Never-

theless, the measurements are invasive as they need a small amount of the patient’s blood

that they then analyze. The blood sample is usually obtained from the finger by pricking

the fingertip with a small needle and is then applied on a test strip and inserted into the

glucose meter

(cf. Figure 3.6). Other places can be used to obtain the blood sample but

the one from the fingertip shows changes in the glucose level more quickly [

]. This

invasive way can cause discomfort, more so because many diabetics have to measure their

the glucose level several times a day.

Figure 3.6:

Measuring the glucose level in a small drop of blood using a glucose meter [

In consequence, non-invasive ways of measuring the glucose level are a very pursued

area of research. Some approaches try to determine the amount of glucose in other bodily

fluids such as

sweat

saliva

urine

tears

. However, the glucose concentration in these

3 Vital Signs and Sensors

fluids is very low which makes accurate measurements difficult [

]. Nonetheless, Google

announced that it is developing a

contact lens3

that will have the ability to measure the

glucose level of tears directly on the eye. They also explore the possibility of embedding

LEDs into the lens in order to display information such as warnings of low blood sugar

directly into the user’s line of sight. In spite of the ingenuity of this idea, a Google developer

stated in an interview that the glucose level measured by the contact lens will probably

also show

delayed values

compared to the invasive measurements on fingertips

. Studies

conducted in 2006 and 2012 came to the conclusion that some of the techniques explored

for noninvasive glucose monitoring look promising but their

poor accuracy

still poses a

problem [62, 55].

3.2.5 Acceleration

Measuring the

physical movement

and position of a sensor device and with it the movement

and position of the person wearing it is accomplished in the same way in which the movement

and position of smartphones is determined: by the use of

accelerometers

. An accelerometer,

as the name suggests, measures acceleration which can manifest itself in

static

forces such

as gravity or

dynamic

forces that occur during movement. Their functioning is best illustrated

on the basis of a mechanical accelerometer (cf. Figure 3.7(a)). A so called

seismic mass

is suspended by a spring which connects it to the housing of the accelerometer. If the

accelerometer is moved, the seismic mass lags behind and the direction and force of its one-

dimensional movement can be measured by how much the spring is stretched or compressed.

Thus, three-dimensional movement can be measured using three accelerometers, one for

each axis. The static position of an object such as a sensor device or smartphone can be

determined by detecting the

gravitational force

that is exerted by the earth and pulls the

seismic mass towards the ground.

Accelerometers integrated in modern devices range in the magnitude of a few hundred

micrometers

in size and they use

electrical signals

to measure their movement and position.

Figure 3.7(b) shows an accelerometer whose parts form a comb-like structure and thereby

differential capacitors, with one part that is fixed to the housing and the other part that

can slightly move. When the accelerometer is moved, the distance between those plates

3http://googleblog.blogspot.de/2014/01/introducing-our-smart-contact-lens.html

4http://www.healthline.com/diabetesmine/newsflash-google-is-developing-glucose-sensing-contact-lenses

3.2 Methods of Measurement

changes which results in a change of capacitance from which the strength of the movement

that triggered the change can be deducted. Another option is to use a

piezoelectric

crystal

such as

quartz

that creates small amounts of voltage when its crystal structures are placed

under mechanical stress during acceleration [20, 11].

Seismic Mass

Movement

(a) Mechanical Accelerometer

Fixed Plates

Movable Seismic Mass

(b) Capacitive Accelerometer

Figure 3.7:

Principles of mechanical accelerometers (a) and capacitive accelerometers (b).

3.2.6 Skin Temperature

The

Angel

sensor wristband

is one of only a few fitness sensor devices that offer the

option of

skin temperature

as one of the recorded vital signs. If they do offer it, they

use the same technology that has been used in electronic clinical thermometers for a

long time. There are materials that change their

electrical resistance

depending on their

temperature. Once they are calibrated, the skin temperature can be derived from the

resistance that the element currently has. With

metals

such as platinum, copper or nickel,

they are called

resistance temperature detectors (RTD)

and use the physical principle of

positive temperature coefficient (PTC). When the material is

ceramic

polymer

, they are

called thermistors, a combination of thermal and resistor, and the most common type

when measuring temperature is the

NTC thermistor

which stands for negative temperature

coefficient (NTC) [

]. In order for this method of measuring temperature to work, the

skin has to have direct and proper contact with the sensor.

5http://www.angelsensor.com

3 Vital Signs and Sensors

An alternative method that uses infrared does not need this close contact and allows the

temperature to be taken remotely. IR thermometers utilize the fact that all matter with

a temperature above

absolute zero

(

−273.15◦C

) emits

thermal radiation

. The amount of

emitted radiation depends on the temperature of the object. At very high temperatures, the

radiation can even be seen by the naked eye, for example the red, orange or white glow of

steel at temperatures above

600◦C

. An IR thermometer collects the radiation of the IR range

that is emitted from the circular spot it is pointed at and uses an optical lens to focus it on a

detector. The measured temperature is then shown on a display [

]. Big advantages of

this remote measuring of patient’s temperature is the eliminated risk of

spreading infections

between patients when using the same device on multiple patients and the very short time

of less than one second needed to take a reading.

4 Fitness Sensor Devices

In this chapter, the focus is on fitness sensor devices and what they are capable of. After a

brief overview and comparison of sensor devices that are currently available on the market,

there is information given about how they communicate with other devices, in particular

about wireless communication technologies such as

Bluetooth

, the newer and more energy-

saving

Bluetooth Smart

and

Near Field Communication

. In order to determine how well

fitness sensor devices can perform measurements and provide health-related information

in real-life situations, the

iHealth Pulse Oximeter PO3

was used in the execution of various

test scenarios. Furthermore, the recorded data obtained by three different sensor devices

was compared in a statistical analysis to find out if the data differs significantly or if their

means can be considered equal.

4.1 Device Overview

The range of available fitness sensor devices has widened considerably in recent years.

Once simple

single-purpose

devices that could only detect steps taken, they nowadays

usually have

multiple

sensors integrated. Table 4.1 gives an overview of currently available

consumer sensor devices and their sensory capabilities. The listing focuses on devices that

have two or more different sensors and, where applicable, represent the flagship model of

the respective manufacturer. Capabilities that depend on additional equipment such as chest

belts for heart rate measurements are excluded. Due to the vast and rapidly evolving market,

however, the list is not meant to be exhaustive. Besides

dedicated fitness sensor devices

such as the

Fitbit Surge

and

Angel’s Angel

, it also includes

smartwatches

such as

Apple’s

Apple Watch

Motorola’s Moto 360

. They may lack some of the features that dedicated

fitness sensor devices have but still can provide useful health-related information about their

owner.

4 Fitness Sensor Devices

Device Pulse Movement Skin Temp. SpO2GPS Elevation

Amiigo (Amiigo) X X X X — —

Angel (Angel) X X X X — —

Apple Watch (Apple) X X — — — —

Surge (Fitbit) X X — — X X

PO3 (iHealth) X— — X— —

Peak (Intel Basis) X X X — — —

Moto 360 (Motorola) X X — — — —

Gear S (Samsung) X X — — X X

Pulse OX(Withings) X X —X—X

Table 4.1: Overview of modern (fitness) sensor devices and their sensory capabilities.

All the devices in Table 4.1 are worn on the wrist and all but one can take their readings while

being attached to it. Only the

Pulse OX

from

Withings

has to be taken off of its wristband so

that a finger can be placed on the back of the device in order to perform

pulse rate

and

blood

oxygen saturation

measurements.

Pulse rate

and

physical movement

are characteristics

that are quite commonly available in modern fitness sensor devices. Other characteristics,

such as

skin temperature

and

blood oxygen saturation

, are not as common, most likely

because they may either be not that significant or not cost-efficient enough. Some products

such as the Angel wristband are still in an active development phase, are not yet released to

the general public and may offer certain measurements only after a firmware update from

the manufacturer. However, their features, specifications and design look very well suited to

be used in certain areas of patient care.

4.2 Wireless Communication

Early models of sensory devices depended partially on

wired connections

to the device

the data should be transferred to. Manufactures used

serial

and

Universal Serial Bus

(USB)

connections among others to transfer stored information between devices. A few

used

infrared

interfaces which made

wireless communication

somewhat possible, but the

need for an unobstructed line of sight between the devices and the very limited range and

transmission speed had a negative effect on user acceptance. Today’s fitness sensor devices,

4.2 Wireless Communication

however, are always equipped with one or more wireless communication technology which

presents the user with a much more convenient way to transfer data. The most commonly

used technologies are

Bluetooth

Bluetooth Smart

and

Near Field Communication

. This

section gives details about their specifications, their inner workings and their fields of

application.

4.2.1 Bluetooth

Bluetooth is an omnipresent wireless communication standard with its origin dating back

over 15 years. After being developed by

Ericsson

in 1994, its first official specification

was published in 1999 and the

Special Interest Group (SIG)

tasked with creating a uniform

standard for short-range wireless communication was created a year earlier [

]. Among the

founding members were leading communication technology companies such as Intel, Nokia

and the aforementioned Ericsson. The naming process of this SIG is worth mentioning

because Bluetooth was only intended to be an

interim

code name until the members could

agree on an official name. The name Bluetooth was derived from the Danish

King Harald

Gormsson

who fittingly unified tribes of Sweden, Denmark and Norway into Scandinavia and

whose nickname was Blåtand, meaning Bluetooth. The naming process, after suggestions

like “Flirt — getting close, but not touching”, resulted in the name

PAN

which stands for

Personal Area Networking

. However, the SIG ultimately was renamed to Bluetooth SIG after

concerns about trademark issues arose. The Bluetooth logo is the combination of the two

runes for H and B, Harald Blåtand’s initials [2, 31].

Bluetooth devices use a frequency spectrum of

2,400.0

MHz to

2,483.5

MHz which is further

divided into

channels with

MHz bandwidth each, a guard space of

MHz at the bottom

and one of

3.5

MHz at the top to avoid interference with other technologies. To further

minimize the impact of interference by other devices using the

2,400

MHz band, Bluetooth

performs

adaptive frequency hopping (AFH)

. It splits up the data into packets and uses one

the

channels to send one packet before it switches to another channel in a predefined

sequence at a rate of

1,600

hops per second. If AFH detects interference on one of the

channels, it skips that channel in the following transmissions. Combining AFH with the

meters transmission range for devices in Bluetooth

Class 2

, the most common among

mobile devices, makes Bluetooth very useful for wireless fitness sensor devices. It takes into

4 Fitness Sensor Devices

account that many devices may be used in a certain area such as fitness studios or sport

events and it also helps to avoid problems with Wi-Fi networks that also use

2,400

MHz

band. The average power consumption of Class 2 devices is

2.5

mW, however the following

section will present the advanced Bluetooth Smart which is even more power saving. Classic

Bluetooth offers data rates of 1−3Mb/s [8, 6].

4.2.2 Bluetooth Smart

Bluetooth Smart was officially introduced as part of the

Bluetooth 4.0

specification in the year

2010. It is also known as

Bluetooth Low Energy

and

Wibree

. Similar to the classic Bluetooth

technology, Bluetooth Smart uses the frequency spectrum

2,400.0

MHz to

2,483.5

MHz but

with only

channels of

MHz bandwidth each instead of

channels with

MHz. It offers

data rates of

Mb/s and its main distinction to classic Bluetooth is its low power consumption.

According to the Bluetooth SIG, it ranges from

50%

to as low as

when compared to

classic Bluetooth, depending on the use case. A simple beacon device that sends out static

information and that can be used as perimeter indicator in location-based services can be

powered for one or two years with a single coin cell battery

. Another advantage is the short

amount of time it takes to establish a connection. It can be as low as

ms with Bluetooth

Smart as compared to 100 ms with classic Bluetooth [7].

One additional feature that makes Bluetooth Smart a fitting communication technology for

wireless sensor devices is the principle of

profiles

which it adopted from classic Bluetooth.

Bluetooth Smart profiles are based on the generic attribute profile (GATT) which is used to

define a common data structure between senders and receivers and to discover services

that devices are offering. Using GATT as a common starting point, the derived profiles are

then fitted for their specific use case. Profiles improve

compatibility

between different device

manufactures and application developers and are an important factor as to why Bluetooth is

so widely used.

Each profile includes a mandatory

primary service

and may include a secondary

auxiliary

service

whose implementation can be optional or mandatory. For each service,

require-

ments

and

dependencies

on other services can be defined, along with one or multiple

characteristics

. Characteristics ultimately define how a data value is

represented

, its

de-

1http://www.aislelabs.com/reports/beacon-guide

4.2 Wireless Communication

scription

and

properties

[

]. Figure 4.1 shows the structure of the

heart rate profile

. The

profile consists of the primary service called

heart rate

and a secondary service called

device information

. The service heart rate is then specified with the three characteristics

heart rate measurement

body sensor location

and

heart rate control point

. And finally,

body sensor location

, for example, defines that this information is represented by an

8-bit

number

and if this number has a value of

, the sensor is located on the chest, if it has a

value of 2, it is located on the wrist, and so on [5].

Heart Rate Profile

Heart Rate Service

Body Sensor Location

Location (8bits)

Heart Rate Measurement

Flags (8bits)

Heart Rate Value (uint8or uint16)

.. .

Heart Rate Control Point

Reset Energy Expended (8bits)

Device Information Service

Manufacturer Name String

Manufacturer Name (utf8s)

Model Number String

Model Number (utf8s)

Serial Number String

Serial Number (utf8s)

.. .

Figure 4.1:

Structure of the Bluetooth Smart Heart Rate Profile (

uint8

-bit unsigned

integer; uint16 =16-bit unsigned integer; utf8s= UTF-8encoded string).

The currently available profiles already cover many fitness and health-related values and

new profiles are specified every so often. Some of the standardized profiles

relevant to

fitness and health care are:

•Blood Pressure Profile

•Cycling Power Profile

•Cycling Speed and Cadence

•Glucose Profile

•Continuous Glucose Monitoring Profile

•Health Thermometer Profile

•Heart Rate Profile

2https://www.bluetooth.org/en-us/specification/adopted-specifications

4 Fitness Sensor Devices

•Pulse Oximeter Profile

•Running Speed and Cadence Profile

This

structured

approach to data transmission helps developers when creating new applica-

tions for smart mobile devices. It ensures

interoperability

across vendors and devices. As

long as the sensor devices they want to connect with and gather data from adhere to the

respective Bluetooth profile, it does not matter from whom the device is manufactured.

4.2.3 Near Field Communication

Near Field Communication (NFC) is a wireless communication technology that was derived

from

Radio-frequency Identification (RFID)

. It uses the same physical principles as RFID but

while RFID offers different frequency ranges depending on the situation (e.g. Low Frequency

125−134

kHz, High Frequency

13.56

MHz, Ultra High Frequency

860

MHz to

960

MHz), NFC

only uses a subset of these frequencies (

13.56

MHz). It supports two-way communication

but can only operate within short distances of less than

cm and with transmission speeds

of up to 424 kbit/s [16, 45].

Despite its low transmission speed and range compared to Bluetooth and Bluetooth Smart,

NFC can be useful when an initial connection between two devices has to be established. A

user can indicate which two devices he wants to connect by holding them closely together

instead of having to chose the correct device from a listing in an application. NFC is then

used to

negotiate

parameters for another connection that uses a different technology, for

example Bluetooth. After that, this new and in most cases faster connection can be used to

transfer user data. Google uses this approach for

Android Beam

, for example. Two Android

devices have to be briefly held together back-to-back to initialize the transfer of data. Android

then establishes a Bluetooth connection for the actual transmission and the users benefit

from better speed and range from this point forward

. A well-established example that makes

use of Android Beam is Motorola Migrate4, an application marketed by Motorola that helps

users to transfer their data to a new smartphone. If both, the old and the new device, support

NFC, Motorola Migrate offers the user the options to establish the first connection between

the two devices by holding them back-to-back, after which the actual migration takes place

3https://developer.android.com/about/versions/android-4.1.html

4https://www.motorola.com/us/motorola-migrate/motorola-migrate.html

4.3 Hands-on Tests

and contacts, media files and calendars among other data are transferred via Bluetooth

while the devices can be held at a distance of more then a few centimeters.

NFC is a feature that many smartphones running Android, Windows or BlackBerry as

their operating system have been supporting for quite some time. Apple has only recently

released its first NFC enabled devices, the iPhone 6 and iPhone 6 Plus. NFC is not as

common among tablets. No model of Apple’s iPad supports it, however, there are some

Android and Windows tablets that do

. The number of NFC enabled smart mobile devices

will certainly grow as new applications are being developed and deployed. Google

, as well

as Apple

, has announced the introduction of payment systems in cooperation with large

credit card companies such as MasterCard, Visa and American Express that are based

on NFC technology with which users will be able to pay using their smartphones. This will

further add to the significance of NFC.

4.3 Hands-on Tests

This section describes the different tests that were conducted with a fitness sensor device.

In order to gain

first hand insights

on whether or not such sensor devices present a useful

source for health information that can improve medical diagnoses,

everyday situations

are

analyzed, such as working a desk job while standing or making a short trip using trains and

buses. Considering that the fitness sensor device used is not officially made for long term

measurements during physical activity, the data of the sensor device is

compared

to data of

other fitness sensor devices as a reference to gather information about the

accuracy

of the

recorded data.

4.3.1 iHealth Wireless Pulse Oximeter PO3

The fitness sensor device used to conduct the following tests was the

iHealth Wireless Pulse

Oximeter PO3

. It is manufactured by the 2009 founded California-based company

iHealth

5https://en.wikipedia.org/wiki/List_of_NFC-enabled_mobile_devices

6http://officialandroid.blogspot.de/2015/05/pay-your-way-with-android.html

7https://www.visa.co.uk/products/visa-contactless/mobile-contactless/apple-pay

4 Fitness Sensor Devices

Lab Inc.8

which has a variety of different sensor devices in its portfolio. Besides the PO3,

iHealth Lab also sells

blood pressure monitors

glucose monitors

scales

, and

activity and

sleep tracker

, all of which are capable of wireless communication and send their data to

smartphones and tables using Bluetooth or Wi-Fi connections. According to iHealth Labs,

all their products are tested and adhere to the strictest clinical validation protocols9.

Figure 4.2: The iHealth Wireless Pulse Oximeter PO3 by iHealth Lab Inc.

The Wireless Pulse Oximeter PO3 (cf. Figure 4.2) is a finger clip sensor that measures the

user’s pulse rate and blood oxygen saturation. It can display the two measured vital signs

on its screen, consisting of green LEDs, and send the data to a smart mobile device via

Bluetooth Smart. The PO3 has an internal memory that can store up to

100

measurements

which can then be accessed at a later point in time. Its display has further LEDs indicating

an active Bluetooth Smart connection, a low or charging battery and whether or not it can

detect a pulse rate. Figure 4.3 lists a few of the product specifications and recommendations

given in the owner’s manual [

] on how to use the device in order to get accurate readings.

The manual also contains the disclaimer that “The Pulse Oximeter PO3 is not a medical

device and should only be used by healthy individuals who are performing non-medical

sports or recreational activities. It is intended to be used for spot monitoring and not for

continuous monitoring” [

]. To what extent these two issues influenced the test runs will be

addressed in the two following sections.

8http://www.ihealthlabs.com

9http://www.ihealthlabs.eu/en/content/106-clinically-validated

4.3 Hands-on Tests

“[...]

2. Display System: LED

3. Power Source: Lithium-ion battery

4. Peak wavelength: 660nm/880nm

5. SpO2 Measuring Range: 70-99%

6. Average Root Mean Square (ARMS) of SpO2 Accuracy: 80%

∼

99%:

2%,

70% ∼79%: ±3%, <70%: no definition.

7. Pulse Rate Measuring Range: 30-250 bpm

8. Pulse Rate Accuracy: 30 ∼99 bpm: ±2 bpm, 100 ∼250 bpm: ±2%.

9. Automatic Shut-off: After 8 seconds of no indication on the sensors

10. Operation Environment: 5°C-40°C; Humidity <80%

[...]

2. Limit finger movement as much as possible when using the device.

Otherwise, the Pulse Oximeter PO3 might misinterpret excessive movement

as good pulse strength.

3. Do not use the Pulse Oximeter PO3 on the same hand/arm when using a

blood pressure cuff or monitor.

[...]

5. The Pulse Oximeter must be clean for a proper reading.

6. Your finger must be clean to ensure a proper reading.

7. Any of the following conditions may cause inaccurate measurements of the

Pulse Oximeter, including BUT NOT LIMITED TO:

•Flickering or very bright light;

•Poor blood circulation;

•Low hemoglobin;

•Hypotension, severe vasoconstriction, severe anemia or hypothermia;

•Nail polish, and/or artificial nails;

•

Any tests recently performed on you that required an injection of intravas-

cular dyes.

8. The Pulse Oximeter PO3 may not work if you have poor circulation. Rub

your finger to increase circulation, or place the device on another finger.

[...] ”

Figure 4.3: Excerpt from the iHealth Pulse Oximeter PO3 owner’s manual [27].

4 Fitness Sensor Devices

4.3.2 Measurements

The equipment used during the realization of different test scenarios was the iHealth Pulse

Oximeter PO3 and a Google Nexus 4 running Android 4.4.4. The recommended way of

collecting and accessing health data recorded by a PO3 device is via the

iHealth Cloud10

operated by iHealth Labs. The recorded pulse rate and blood oxygen saturation is uploaded

to the iHealth Cloud by the official iHealth app that has to be installed. A developer of an

application can then use a provided

application programming interface (API)

to access the

data uploaded by the user of the sensor device. In the context of this research, though, this

was not an acceptable method, especially the upload to a corporate cloud and storage of

personal data on a third party system. There are not only security issues involved but it also

poses certain risks to the users’ privacy as discussed in Section 6.1.

An alternative method is the use of an

Android software development kit (SDK)

provided

by iHealth Labs. This SDK enables the developer to

directly

receive pulse rate and blood

oxygen saturation data in his Android application without a detour via their cloud. It keeps

the compilation and storage of data locally between the pulse oximeter and the smartphone

which not only

improves

performance but also

mitigates

privacy issues. Consequently, this

access method was used in the test runs. It delivered about seven measurements for every

second but because these seven data sets normally showed the same values and to reduce

complexity, only the first measurement of every second was used in the subsequent data

analysis.

Before starting with the actual tests, an initial experiment was conducted in order to determine

possible alternative locations the iHealth Pulse Oximeter PO3 could be placed on. To this

end, the finger clip was placed on a subject’s earlobe in order to get a data sample but this

did not give any usable results. It was difficult to find a position in which the vital signs were

detected at all and the clip tended to slip off when moving even a little bit. However, even if

it had been possible to collect data, Haynes showed in [

] that data collected by a pulse

oximeter finger clip that is attached to the patient’s earlobe is not reliable enough for clinical

purposes.

10http://developer.ihealthlabs.com/index.htm

4.3 Hands-on Tests

Basic Indoor Activities

The first scenario in which vital signs were recorded was kept very simple and basic. The test

subject was sitting in a chair and remained seated while keeping movements to a minimum.

The duration of this test was

minutes and one measurement consisting of heart rate and

blood oxygen saturation was recorded for every second. Figure 4.4 visualizes the results.

The red graph represents the heart rate, the blue graph represents the oxygen saturation

in the blood. The data itself is very inconspicuous, the oxygen saturation only fluctuates

between

97%

and

98%

and has an average level of

97.52%

(

SE = 0.01

)

. The heart rate is

also fairly steady with an average of

58.32 bpm

(

SE = 0.11

), a maximum of

82 bpm

and a

minimum of

48 bpm

. In the

minutes after marker

(cf. Figure 4.4), the subject dozes off

which explains the very low heart rate up to marker

when he wakes up again. Compared

to the readings in the following scenarios, the amplitudes of the heart rate graph are very

small due to the lack of movement and relaxing characteristic of this test scenario.

120

160

200

0 10 20 30 40 50 60

100

Heart Rate [bpm]

Oxygen Saturation [%]

Time [min]

A B

Heart Rate

Oxygen Saturation

Figure 4.4:

Pulse oximetry data of

minutes sitting still. After marker

there is a period of

dozing off until marker B.

11SE ≡standard error of the mean

4 Fitness Sensor Devices

120

160

200

0 10 20 30 40 50 60

100

Heart Rate [bpm]

Oxygen Saturation [%]

Time [min]

Heart Rate

Oxygen Saturation

Figure 4.5:

Pulse oximetry data of

minutes standing at a desk and walking around a few

steps, simulating a workplace environment.

The second test scenario simulated an environment similar to an office space. Again, the

duration was

minutes in which the test subject this time had to stand at a desk and use a

computer or write things down. He also walked around several times but only a few meters.

Figure 4.5 shows the resulting data points. The oxygen saturation fluctuates a bit more

this time and is between

89%

and

98%

with an average of

96.93%

(

SE = 0.01

). The heart

rate has an average of

83.58 bpm

(

SE = 0.14

), a maximum of

146 bpm

and a minimum of

56 bpm

. However, there are only a few heart rate values in the areas around the minimum

and maximum value. The short time span in which the oxygen saturation is at its minimum

89%

is also a singular event which could indicate some

erroneous

readings because no

correlation can be made to any special physical activity.

Outdoor Cycling

Having obtained good results in a

controlled

environment in the first two scenarios, further

tests were conducted to find out about the

limitations

of the iHealth Pulse Oximeter PO3.

Therefore, the device was used during an

outdoor

bicycle ride on a clear sunny day which

4.3 Hands-on Tests

is way outside the conditions recommended by the manufacturer. Figure 4.6 shows the

recorded data of the first

minutes of this trip. As can be seen by the discontinuous graphs,

the reading was disturbed again and again and after it was obvious that no useful data could

be retrieved this way, the test run was interrupted. However, the PO3 can not be blamed for

this poor performance. It is caused by the surrounding environment that is too bright and by

the vibrations the finger and the PO3 attached to it were subject to. It is rather a quirk of the

pulse oximetry method of measuring and not device specific. As described in Section 3.2.1,

pulse oximetry utilizes red and infrared light to detect the oxygen saturation and the pulse

rate. If the ambient light is too bright and too much of it reaches the photodetector, the faint

light that the two LEDs emit is overpowered by it which makes an accurate measurement

impossible.

120

160

200

0 2 4 6 8 10

100

Heart Rate [bpm]

Oxygen Saturation [%]

Time [min]

Heart Rate

Oxygen Saturation

Figure 4.6:

Incomplete pulse oximetry data of a

minute bicycle ride in for the sensor

adverse conditions (sunny, cloudless) and with no supplemental cover used.

After some considerations, a working solution to this problem was found. The PO3 was

covered up with a

bandage

which itself was fixated with

tape

. This helped in two ways: it

prevented

the sunlight to interfere with the LEDs of the pulse oximeter and it also

secured

its position on the finger. Furthermore, the test subject tried to separate the hand to which

the pulse oximeter was attached as often as possible from the handle bar. The resulting

4 Fitness Sensor Devices

heart rate and blood oxygen data is visualized in Figure 4.7. It can be observed that there

are no data points missing which illustrates that the actions taken to modify the test setup

were sufficient. However, especially the part about avoiding vibrations turned out to be a

major

nuisance

, depending on the surface of the ground. If the road was paved and smooth,

the vibrations were weak enough so that the handle bar could be grabbed and used as usual.

If, on the contrary, the surface was rugged and bumpy like on a dirt track, the vibrations

most likely would have interfered with the measurements which effectively meant driving

one-handed

on such tracks. This is not a solution that is practicable for general use, though,

but served its purpose well during the test runs.

120

160

200

0 20 40 60

100

Heart Rate [bpm]

Oxygen Saturation [%]

Time [min]

Heart Rate

Oxygen Saturation

Figure 4.7:

Pulse oximetry data of an

minute bicycle ride in adverse conditions for the

sensor (sunny, cloudless) but with supplemental cover used. The low heart rate

between marker

and

was due to a short stop to check the equipment for

proper operation.

The resulting data in Figure 4.7 shows an average blood oxygen saturation of

97.18%

(

SE = 0.03

) with a maximum of

99%

and a minimum of

90%

. The heart rate has a maximum

184 bpm

, a minimum of

52 bpm

and averages at

136.80 bpm

(

SE = 0.32

). Naturally, it

varies a lot during the course of this test just like the level of physical strain fluctuates a lot,

too, always depending on outside influences such as the properties and condition of the

route and the direction and speed of the wind. It further is the first scenario in which the

4.3 Hands-on Tests

recorded oxygen saturation level varies considerably which is also to be expected during a

long-lasting sporting activity with varying physical strain. The low heart rate values between

marker

and marker

(cf. Figure 4.7) occurred during a short stop in which the equipment

was checked for proper operation.

Outdoor Activities of Daily Living

The first two test scenarios (sitting and standing) were kept simple to get some basic

measurements, followed by the more extreme scenario of outdoor cycling which covers

more of a fitness-related area. While it is not very hard to imagine that future sensor devices

will also function properly under adverse conditions encountered in outdoor sports, it is

even more likely that they will be suitable for

activities of everyday living

. The following two

scenarios try to represent ordinary activities that one may do outside his home during a day.

To prevent any major interference, the same

precautions

regarding ambient light were taken

as during the cycling scenario, using bandage and tape.

120

160

200

0 10 20 30 34

100

Heart Rate [bpm]

Oxygen Saturation [%]

Time [min]

A B C

Heart Rate

Oxygen Saturation

Figure 4.8:

Pulse oximetry data of a

minute visit to a friend. Point

marks the arrival

at the friend’s house, marker

the departure from it and

the arrival at home

including emptying the postbox.

4 Fitness Sensor Devices

The first of these activities of daily living (ADL) scenarios was a visit by foot to a friend.

The iHealth PO3 recordings cover

minutes of heart rate and oxygen saturation data

which represents the time span between leaving the house and returning to it again. The

corresponding visualization can be seen in Figure 4.8. The maximum heart rate of

130 bpm

was reached shortly before arriving at the friends apartment at marking

and was caused

by taking the stairs to the second floor. The stay itself, the time period between marker

and

, only consisted of some standing and walking and talking and therefore shows no

interesting spikes. The period between

and

represents the walk home.

marks the

arrival at home opening the front door and emptying the postbox, during which the minimum

heart rate of

53 bpm

occurred. The average heart rate during this test was

80.37 bpm

(

SE = 0.34

). The oxygen saturation varied between

95%

and

99%

with an average level of

97.87% (SE = 0.01).

Basically, the activities during this test scenario can be summarized as

minutes of walking,

followed by

minutes of staying in one place which is, again, followed by

minutes of

walking home. In order to introduce a wider diversity of activities into the recording, the

final test scenario was chosen to be a trip to the next major city, Ulm, in order to complete

a fictitious task and returning home. The duration of this trip is

120

minutes and includes

walking, trains and buses as means of transportation. Figure 4.9 shows the resulting blood

oxygen saturation and heart rates. The maximum saturation is

99%

, the minimum

87%

and

the average is

96.83%

(

SE = 0.03

). The saturation level, again, fluctuates considerably.

The maximum heart rate is

140 bpm

, the minimum

53 bpm

and the average heart rate is

78.19 bpm (SE = 0.20).

The recording in Figure 4.9 starts with the departure from home and the walk to the train

station. The fluctuating vital signs reflect the also varying steepness of the route and even

the short waiting period at a pedestrian light at minute

is reflected by a local minimum

for the heart rate. At marker

, the test subject boards the train and sits down. The heart

rate slows down and the oxygen saturation recovers and stays at

99%

during the first train

ride. Between

and

, the subject has to change trains which results in a rising heart rate

and falling oxygen saturation. After marker

, at which point the test subject has boarded a

connecting train and sat down, the heart rate falls and the oxygen saturation rises again.

marks the arrival at Ulm central station where the test subject gets off the train, exits the

station and makes his way to the S-Bahn. The time period between

and

represents the

stay in Ulm, including in this order: leaving the station, briefly riding on the S-Bahn while

4.3 Hands-on Tests

standing up, walking through the city to the target destination, completing a task, walking

back to the train station. Marker

is placed in the middle of the execution of the task which

included waiting in line for some minutes. This explains the low heart rate and high oxygen

saturation compared to the surrounding values.

marks the return to and arrival at Ulm

central station where the test subject sits down and waits for his train. At marker

, he

boards the train home and sits down. The spike of the heart rate at minute

occurred

during a ticket inspection. The time after

includes alighting from the train, walking to a

connecting bus, boarding it and sitting down at marker

. The bus ride ends at marker

at which point the test subject gets off the bus and walks to his home. At marker

, he

finally arrives at home. The spike of the heart rate in the end is caused by going upstairs at

home.

All the scenarios, and in particular this last one, show that very detailed and expressive

recordings of vital signs can be obtained by the use of sensor devices. Even a fitness sensor

device that is only specified for brief spot checks of heart rate and blood oxygen saturation

such as the iHealth Pulse Oximeter PO3 gives reasonable results when used for continuous

measurements.

120

160

200

0 20 40 60 80 100 120

100

Heart Rate [bpm]

Oxygen Saturation [%]

Time [min]

A BCD E F G HIKL

Heart Rate

Oxygen Saturation

Figure 4.9:

Pulse oximetry data of a

120

minute trip to the city of Ulm including the completion

of a given fictitious task. See Table 4.2 for explanations for each marker.

4 Fitness Sensor Devices

Marker Action

AHaving a seat after boarding the train.

BExiting train in order to catch a connecting train.

CHaving a seat after boarding the connecting train.

DExiting train and leaving Ulm central station towards the S-Bahn.

EIn the middle of completing the given task.

FArriving at the train station and waiting for the train home.

GHaving a seat after boarding the train.

HExiting train in order to catch a bus.

IHaving a seat on the bus.

KGetting off the bus and walking home.

LArrival at home, getting out the key and opening front door.

Table 4.2:

Detailed explanation of the actions taken during the 120 minute recording of a trip

to a major city (cf. Figure 4.9).

4.3.3 Comparing Sensor Devices

In three additional test runs, the heart rate displayed by the PO3 finger clip was compared

to those displayed by other devices from different manufacturers using different methods of

measurement. The second device was a

Sigma PC912

heart rate monitor which is comprised

of a

chest belt

that gathers the heart rate data and a

watch

that displays it (cf. Figure 4.10(a)).

The third device was a

Motorola Moto 36013

smartwatch (cf. Figure 4.10(b)) which measures

the heart rate using pulse oximetry with two green LEDs as described in Section 3.2.2.

All three tests were conducted over a time period of

minutes and the heart rate values

displayed on each of the devices were recorded manually with a time interval of

seconds.

The heart rate app

Moto Body Heart Rate

that is pre-installed on the Moto 360 does not

support continuous measurements but only spot checks. Nevertheless, this app and its

spot checks are used due to the lack of better alternatives. The measuring process of this

app normally took

7−8

seconds so it was possible to time the readings to fit into the

second intervals. In some cases, the Moto 360 app was not able to successfully complete

a measurement and it displayed the error message “Couldn’t get it — Check your watch

12http://www.sigmasport.de/en/produkte/pulscomputer/topline/pc_9

13http://moto360.motorola.com

4.3 Hands-on Tests

placement and keep your arm still”. In these cases, the readings of all three devices were

disregarded and the measurement was repeated adhering to the given

second interval.

(a) Sigma PC9. (b) Motorola Moto 360.

Figure 4.10:

The watch component of the Sigma PC9 heart rate monitor showing a heart

rate of

98 bpm

(a) and the Motorola Moto 360 smartwatch showing its Moto

Body Heart Rate application currently measuring (b).

Sitting

The first comparison scenario consisted of the test subject calmly sitting in a chair. He did

not stand up and kept activities such as arm and head movement to a minimum during the

minutes test period. The blue graph in Figure 4.11 visualizes the

heart rate values taken

with the iHealth PO3 finger clip attached to the left index finger, the red graph represents

the data taken with the Sigma PC9 heart rate monitor and the green graph represents the

heart rate values displayed by the Motorola Moto 360. The PO3 showed a minimum heart

rate of

59 bpm

, a maximum of

85 bpm

and the calculated average heart rate is

73.20 bpm

(

SE = 0.76

). The Sigma PC9 showed a minimum heart rate of

62 bpm

, a maximum of

86 bpm

and the calculated average is

73.28 bpm

(

SE = 0.81

). The Motorola Moto 360 showed a

minimum heart rate of

54 bpm

, a maximum of

93 bpm

and the calculated average is also

73.28 bpm

(

SE = 1.03

). A total of

readings had to be repeated after the Moto 360 failed

to get a heart rate measurement. One possible reason for this poor performance can be a

weak pulse when sitting down and resting which makes measurements using pulse oximetry

more prone to errors.

4 Fitness Sensor Devices

The interesting characteristics in these comparison tests are the deviations between the

measured values of each of the devices. Since they all measure the

same

vital sign, they

should display the

same

value. This is clearly not the case and Figure 4.12 shows the

calculated deviation of all three devices to each of the other two devices. The blue graph

represents the number of beats per minute the iHealth PO3 reading deviates from the Sigma

PC9 reading. The number is positive if the PO3 displayed a higher value than the PC9,

negative if the value was lower than the one on the PC9 and the deviation is

if they both

showed the identical value. The red graph visualizes the deviation of values displayed by

the Motorola Moto 360 to the ones displayed by the Sigma PC9 and the green graph the

deviation of the Moto 360 to the PO3. The blue graph and with it the deviation between

PO3 and PC9 seem to describe the smallest values compared to the other two graphs.

Calculating the average of the deviations is not very meaningful since the values vary

between being positive and negative. For example, the mean deviation between PO3 and

PC9 is

−0.08 bpm

and the one between Moto 360 and PC9 is

0bpm

. However, it is obvious

that the latter varies more than the former. A more significant characteristic in this situation

is the standard deviation (SD), for whose calculation each single deviation is squared first so

values with opposite signs will not cancel each other out when summing them up. The SD of

the difference between PO3 and PC9 is

2.42 bpm

and the SD of the difference between Moto

360 and PC9 is

4.56 bpm

, nearly two times as high. This corresponds more closely with the

graphs in Figure 4.12 and their represented heart rate values. The average deviation of the

third pair of devices, the Moto 360 and the PO3, is

0.08 bpm

(

SD = 4.93

). The maximum

absolute deviation of a single pair of values was

16 bpm

between Moto 360 and PO3,

13 bpm

between Moto 360 and PC9 and only

8bpm

between PO3 and PC9. The heart rate values

displayed by the iHealth PO3 and the Sigma PC9 matched in

(

24.6%

) of all

cases.

There are only

(

6.6%

) matches between the Moto 360 and the PC9 and with

(

4.9%

)

even one less match between the Moto 360 and the PO3. In only

(

3.3%

) cases was the

displayed value identical on all three devices.

Furthermore, statistical analyses were conducted to answer the question if the mean

differences between devices vary significantly (

) or if they can be considered statistically

zero (

). The statistical test used was either the parametric

paired-samples t

-test if the

differences between two devices were normally distributed or the non-parametric

Wilcoxon

signed-rank

test if the differences were not normally distributed. The tests used to check for

normal distribution were the

Kolmogorov-Smirnov

and the

Shapiro-Wilk

test. All statistical

4.3 Hands-on Tests

tests were conducted using

IBM SPSS Statistics 21

[

]. If their results were not conclusive

or contradictory, additional data such as the corresponding

frequency distribution

and

Q–Q

plot (cf. Appendix A.2) were consulted [14].

In this scenario, the deviation values between Moto 360 and PC9 and between Moto 360

and PO3 are normally distributed, however, the deviation values between PO3 and PC9

are not (cf. Appendix A.1). The reason for this are probably the many times the deviation

between the PO3 and PC9 is

0bpm

±1bpm

which results in a high kurtosis. A Wilcoxon

signed-rank test used due to the absence of normal distribution showed that, on average, the

displayed values of the PO3 (

Mdn. = 71

) and the PC9 (

Mdn. = 72

) do not differ significantly

(

z=−.593, p > .05

). Paired-samples

-tests showed that neither is the average difference

between PO3 (

M= 73.20, SE = 0.76

) and Moto 360 (

M= 73.28, SE = 1.03

) significant

(

t(60) = −.130, p > .05

), nor is, on average, the difference between Moto 360 and PC9

(M= 73.28, SE = 0.80) significant (t(60) = .000, p > .05).

120

160

200

00 05 10 15 20 25 30

Heart Rate [bpm]

Time [min]

iHealth PO3

Sigma PC9

Moto 360

Figure 4.11:

Comparison of heart rates gathered from iHealth PO3, Sigma PC9 and Motorola

Moto 360 while sitting calmly in a chair.

4 Fitness Sensor Devices

-20

-15

-10

-5

00 05 10 15 20 25 30

Deviation [bpm]

Time [min]

PO3 to PC9

Moto 360 to PC9

Moto 360 to PO3

Figure 4.12: The deviations of the displayed values for each pair of devices while sitting.

Standing

The second comparison scenario is similar to the previously used second measurement

scenario but it only included standing at a desk, working with a computer and pen and paper.

The test subject did not move around other than normal foot movement while standing.

Again, the three devices PO3, PC9 and Moto 360 were used to display the heart rate every

30 seconds over a period of 30 minutes. The PO3 showed a minimum heart rate of

68 bpm

a maximum of

86 bpm

and the calculated average heart rate is

77.28 bpm

(

SE = 0.47

). The

PC9 showed a minimum heart rate of

70 bpm

, a maximum of

87 bpm

and the calculated

average is

76.87 bpm

(

SE = 0.51

). The Moto 360 showed a minimum heart rate of

68 bpm

, a

maximum heart rate of

93 bpm

and an average of

77.85 bpm

(

SE = 0.68

). The corresponding

graphs can be seen in Figure 4.13. This time, the Moto 360 failed to conduct a successful

reading only in

cases, compared to the

times in the previous scenario. Figure 4.14

shows the deviations between each pair of devices. The average deviation from iHealth

PO3 to Sigma PC9 is

0.41 bpm

(

SD = 2.67, Max. = 6

), from PC9 to Moto 360 is

0.98 bpm

(SD = 4.47, Max. = 15) and from PO3 to Moto 360 0.57 bpm (SD = 4.44, Max. = 14).

4.3 Hands-on Tests

120

160

200

00 05 10 15 20 25 30

Heart Rate [bpm]

Time [min]

iHealth PO3

Sigma PC9

Moto 360

Figure 4.13:

Comparison of heart rates gathered from iHealth PO3, Sigma PC9 and Motorola

Moto 360 while standing and working at a desk.

-20

-15

-10

-5

00 05 10 15 20 25 30

Deviation [bpm]

Time [min]

PO3 to PC9

Moto 360 to PC9

Moto 360 to PO3

Figure 4.14:

The deviations of the displayed values for each pair of devices while standing.

4 Fitness Sensor Devices

The values displayed by the PO3 and the PC9 matched in

(

16.4%

) of the total of

cases. Only

(

9.8%

) matches occurred between the Moto 360 and the PC9 and with

(

6.6%

) matches between the Moto 360 and the PO3. In

readings the three devices showed

matching values.

In this scenario, the tests of normality for all three device pairs were inconclusive (cf.

Appendix A.1) however a visual inspection of the corresponding frequency distributions and

Q–Q plots (cf. Appendix A.2) warrant the assumption of normally distributed deviations.

Consequently, a paired-samples

-test was done for each pair of devices, however, the

mean difference between the Sigma PC9 and the iHealth PO3 is not significant (

t(60) =

−1.197, p > .05

). Nor is the mean difference between PC9 and Motorola Moto 360 significant

(

t(60) = −1.719, p > .05

) and also the mean difference between the PO3 and the Moto 360

is not significantly different from zero (t(60) = −1.010, p > .05).

Cycling

The third test was conducted while cycling on a stationary indoor bicycle with increasing and

decreasing intensity. This test run was only conducted with two of the three devices, the

iHealth PO3 and the Sigma PC9. The Motorola Moto 360 was not suited for such an intense

sporting activity due to its leather wristband. The high intensity during this test, compared to

the first two scenarios, is also the reason for the high minimum and maximum values. In

light of that, the minimum heart rate according to the iHealth PO3 data was

135 bpm

, the

maximum

184 bpm

and the calculated average

154.48 bpm

(

SE = 1.74

), which is represented

by the blue graph in Figure 4.15. The Sigma PC9, represented by the red graph, showed

a minimum heart rate of

131 bpm

, a maximum of

180 bpm

and a calculated average heart

rate is

153.38 bpm

(

SE = 1.65

). In

(

16.4%

) out off the

cases both devices displayed

identical heart rate values and the maximum deviation between them was

5bpm

. All the

deviation values between the two devices can be seen in Figure 4.16. They average at

1.10 bpm

(

SE = 0.23, SD = 1.79

) and are not normally distributed. The thereby required

Wilcoxon signed-rank test shows that the values displayed on the iHealth PO3 (

Mdn. = 156

compared to the values displayed by the Sigma PC9 (

Mdn. = 153

), are significantly higher

(z=−4.195, p < .001).

4.3 Hands-on Tests

120

160

200

00 05 10 15 20 25 30

Heart Rate [bpm]

Time [min]

iHealth PO3

Sigma PC9

Figure 4.15:

Comparison of heart rate values gathered from iHealth PO3 and Sigma PC9

while stationary cycling indoors.

-20

-15

-10

-5

00 05 10 15 20 25 30

Deviation [bpm]

Time [min]

PO3 to PC9

Figure 4.16:

The deviations of the values displayed on the iHealth PO3 and the Sigma PC9

while cycling indoors.

4 Fitness Sensor Devices

-20

-15

-10

-5

00 05 10 15 20 25 30

Deviation PO3 to PC9 [%]

Time [min]

sitting

standing

cycling

Figure 4.17:

Comparison of the percentage deviations between the values displayed by the

iHealth PO3 and the Sigma PC9 during the three scenarios.

In conclusion, all three devices report heart rate values that, on average, do not differ

significantly from each other in activities of daily living with a relatively low heart rate. At the

same time, singular outliers with high deviation can be observed, such as the

15 bpm

(

20.5%

)

deviation from the Motorola Moto 360 to the Sigma PC9 in the second scenario. Depending

on the use case in which sensor devices are adopted, the occurrence of outliers has to be

factored in to prevent erroneous conclusions that can result in false positives. One possible

solution is to always look at series of values and compare each value to the preceding and

succeeding mean level.

The cycling scenario showed that there can be a significant mean difference in measured

values between different sensor devices. In

(

60.7%

) readings the iHealth PO3 displayed a

higher value than the Sigma PC9, compared to

(

23.0%

) cases in which it showed a slightly

lower value (cf. Figure 4.16). However, Figure 4.17 shows that even though the deviation is

significant in the cycling scenario, the mean percentage deviation (using absolute values)

while cycling (

M= 1.08, SE = 0.10

) is less than while sitting (

M= 2.60, SE = 0.27

) or

standing (

M= 2.84, SE = 0.26

). That means that in these scenarios and using the iHealth

PO3 and the Sigma PC9, the deviation between the devices did not increase proportionally

to the increasing recorded heart rates.

5 Medical Applications

Having discussed the basic vital signs and their measurement techniques in Chapter 3 as

well as the capabilities and limitations of fitness sensor devices in Chapter 4, this chapter

builds on this foundation and explores possible

real-life

medical use cases in which the

deployment of sensor devices may improve patient care. This can be beneficial not only for

the treated patients but also for the doctors involved in the treatment process. Well-known

corporations such as Google are pursuing this area of research with great effort. Google not

only plans to integrate a glucose meter in contact lenses as mentioned before but is also

currently developing their own sensor wristband for the use in a medical environment1.

The three scenarios examined here are the improvement of existing

fall detection systems

an easier

verification

that the prescribed

dosage

of certain

medications

is adequate and

an adaptive system for

dynamic oxygen regulation

. Important factors that are crucial for a

successful adoption of sensor devices in medical applications are

patient compliance

and

the

willingness of doctors

to utilize this technology. Furthermore, the concepts of

sensitivity

and

specificity

of medical testing are briefly introduced as they are relevant when discussing

possible problems and challenges of the proposed solutions.

5.1 Sensitivity vs. Specificity

An important aspect of medical diagnostics is the

validity

of tests. A test and its results

have to be accurate in order to be useful and valid. Diagnostic tests are called

binary

tests

because their results can lead to one of two conclusion: the disease or illness is

either

present

absent

. The objects of these tests are symptoms of diseases, which can,

http://www.bloomberg.com/news/articles/2015-06-23/google-developing-health-tracking-wristband-for-health-

research

5 Medical Applications

however, vary from patient to patient to a certain degree. Their accuracy can be described

by the concepts of sensitivity and specificity.

Sensitivity describes the probability that a test can detect a disease, given that the disease

is in fact present in the test subject. Table 5.1 shows the four distinct possible outcomes

of a test. On the one hand, there are those test subjects that are in fact diseased and the

test will detect it, called

true positives

. In contrast to that, there are

false negatives

. Cases

in which the test will not detect the disease, even though the test subject is diseased. The

sensitivity of a test is calculated with Equation 5.1.

Sensitivity =True Positives

True Positives +False Negatives (5.1)

If a test detects all

diseased as positive and misses none, its sensitivity is

X+0 = 100%

which is the best possible sensitivity. If, however, the test detects

diseased but misses

the sensitivity X

X+Ywill decrease to a value less than 100%.

Disease Present Disease Absent

Test Positive True Positive False Positive

Test Negative False Negative True Negative

Table 5.1: Possible outcomes of a medical test.

Just like sensitivity characterizes how accurately a test can detect the truly diseased,

specificity is used to describe how accurately it can identify the test subjects that are not

diseased. The two remaining categories in Table 5.1 of test subjects that are not diseased

and are either tested negative (i.e.,

true negatives

) or wrongly tested positive (i.e.,

false

positives) are now used. Equation 5.2 is used to calculate specificity of a test.

Specificity =True Negatives

True Negatives +False Positives (5.2)

Similar to sensitivity, the best specificity a test can have is

X+0 = 100%

, in which no test

subject is wrongly detected positive. If the number of false positives gets bigger than

, the

level of specificity decreases below 100%.

5.1 Sensitivity vs. Specificity

For the sake of completeness, it should be mentioned that there are two more additional

values that can be derived from Table 5.1: the

positive predictive value

and the

negative

predictive value. The positive predictive value is calculated with Equation 5.3.

Positive Predictive Value =True Positives

True Positives +False Positives (5.3)

It represents the likelihood that a test subject really is diseased, given that the test results

came back positive. The negative predictive value, in return, is calculated using Equation 5.4.

Negative Predictive Value =True Negatives

True Negatives +False Negatives (5.4)

The negative predictive value and stands for the likelihood that a test subject really is non-

diseased, given that the test results were negative. In other words, they stand for how likely

it is that a given positive or negative test result is correct [34, 13, 48].

The

ideal

diagnostic test has a sensitivity of

100%

and a specificity of

100%

, meaning that

every single diseased test subject is tested positive and every single non-diseased subject

is dismissed. No false positives or false negatives occur. However, this is normally not

the case and a

reasonable balance

between sensitivity and specificity values has to be

reached. These principles and the need to find the right balance between them also have to

be factored in when trying to adopt fitness sensor devices in a medical environment and to

create monitoring applications and notification systems around them.

In some of the medical use cases in which sensor devices could be deployed, the initial

decision whether the patient is in an urgent emergency situation that needs immediate

attention or not is made

programmatically

by an application. Firstly, in order to achieve

the best possible test sensitivity and depending on the use case, the recorded data has to

be as

accurate

as possible and of high temporal resolution. This reduces the number of

false negatives as the necessary information is available to correctly detect disease-relevant

characteristics in the data. Secondly, the specificity level has to be increased as high as

possible by

safeguarding

the application and its notification system against false positives.

This can be accomplished by choosing sensible

thresholds

at which an alarm is triggered, by

implementing

grace periods

to allow for temporary interruptions and by ignoring

statistical

outliers

in the data. These specific values depend heavily on the intended use and in some

cases may even have to be adjusted to the individual patient.

5 Medical Applications

5.2 Patient Compliance

The compliance of patients is a medical term that is used to describe to what extent patients

adhere to directions given to them by their doctors. This includes a variety of different

aspects of patient care: Does a patient take his medication? Does he take it regularly and

at the right time? Is the patient following a recommended diet? Is he doing his physical

exercises? A key element of a successful treatment is good compliance. Only if medication

is taken as prescribed and exercise is done correctly, they can have their intended positive

effects on patients’ diseases.

In spite of its importance, good compliance is all but a given. A report published by the

WHO states that only about

50%

of patients in developed countries that suffer from chronic

diseases comply with the instructions for their treatment. Numbers of developing countries

are suspected to be even worse. Poor compliance not only has a negative effect on the

chances for

success

of treatments but it can also prolong their

duration

and reduce their

effectiveness. It can result in increased costs for health insurance companies and patients,

not to mention the emotional implications. The reasons for poor compliance are diverse.

According to the WHO, there are five interacting areas that affect patient compliance.

There are

social and economic factors

such as gender, age and race and especially in

developing countries poverty, lack of social support networks and the level of education.

Health care team and system-related factors such as poor medication distribution systems

and inadequately educated or overworked health care professionals. The third area of

factors is referring to the

condition

such as severity of the disease, its symptoms, rate of

progression and whether or not there are treatment options available. In contrast, there are

therapy-related factors

such as the duration of the treatment and previous failures. The lack

of immediate benefits and the appearance of adverse or side effects [

] and also a too

complex medication regimen can hinder compliance. The more pills and the more frequently

a patient has to take them, the more likely he is to take only a reduced dosage, take is less

frequently or stop taking them at all [

]. The fifth and last type of factors are

patient-related

factors

such as the patient’s knowledge about the disease, his motivation and confidence to

cure or manage it and whether he believes in the diagnosis or not.

Forgetfulness

is another

patient-related factor of great importance in regards to the elderly. They often have multiple

chronic diseases and have to take a variety of medication throughout the day, however, their

mental ability decreases with age and they have to be reminded to take them.

5.3 Doctor’s Acceptance

The WHO further suggests to differentiate between the terms

compliance

and

adherence

with adherence meaning that patients follow the given instructions, agree with them and are

integrated

in the development and planning of a personalized treatment plan. In contrast,

compliance should be used when patients do obey the instructions given to them but don’t

necessarily understand, approve or support them. This distinction may be helpful when it

comes to finding methods for improvement because they each represent a different set of

problems to solve [66].

Of course, the important role of patient compliance is also applicable to the use of sensor

devices in a medical context. Such devices can be looked at as just additional tools for

analysis and treatment that have to be used exactly in the way the doctor has suggested.

Only then are the results reliable and can have the best possible influence on the patient’s

recovery. Research shows that in America, over one third of people that own a fitness sensor

device

stopped

using it six months after receiving it [

]. There are measures that can be

taken to improve compliance in general such as

including

the patient in decisions regarding

the treatment and keeping him

informed

about all important aspects of the disease. Besides

that, there are also measures that specifically target compliance when using sensor devices.

If the patient has to worry about keeping the device

charged

, feels that it is

too heavy

, has

his movements

hindered

or is

bothered

by it in any way, he is more likely to dismiss it. These

are all points that have to be considered. Furthermore, the patient should be made aware of

in what concrete way the sensor device is expediting his recovery.

5.3 Doctor’s Acceptance

While the patient compliance as discussed in Section 5.2 will be an important aspect, it

is only part of the issues. Another big part will be played by the treating physicians and

whether or not they are

willing to participate

in this new development. The patients may be

bothered by having to wear the sensor device all the time but the physician has to handle all

the accumulated data he receives from it. The amount of data that will be available, even if

only a fraction of patients share their personal health information, poses a

serious challenge

in terms of storage and analysis. Furthermore, some of the data may even be wrong due

to inaccurate sensors or plain misuse of the sensor device by patients. Will the physician

liable

[

] if the data for a certain diagnosis was available, but he did not act on it as he

5 Medical Applications

may not have seen it yet? He might delegate some of the analyses to an outside firm or

laboratory, however that just intensifies the liability issues. A last and maybe one of the most

important issues are the costs. The

initial investment

for the necessary infrastructure and

possibly additional employees are going to

discourage

physicians to adopt the technology in

their practice. Especially so if they can not be sure whether their patients will embrace the

new technology or not. However, the patients can not try it if their physician does not make

the upgrade which leads to a typical chicken-and-egg problem.

5.4 Use Cases

The remainder of this chapter discusses three scenarios in which the introduction of fitness

sensor devices could benefit patients as well as the medical professionals treating them

and contribute to an improved patient care as compared to present conditions. These

improvements include

saved time

increased flexibility

and

faster response times

in emer-

gency situations. The scenarios were worked out during a course of multiple interviews

with doctors and medical students. First, they were presented with the types of information

modern fitness sensor devices can provide. Then, the situations in which this additional and

more detailed information would be most beneficial were discussed further. The scenarios

chosen are the

regulation of supplemental oxygen

fall detection

for the elderly and

dosage

adjustment of medication.

5.4.1 Automatic Oxygen Regulation

There are certain medical conditions that result in lungs which are left with only a reduced

capability of processing oxygen. Two examples of such conditions are

cardiac insufficiency

and

chronic obstructive pulmonary disease (COPD)

, which is often caused by smoking

tobacco or adverse environmental factors. As of today, COPD can not be cured but there

are treatments available to alleviate the patient’s symptoms such as shortness of breath

and chronic cough. Treatment options include

oxygen therapy

in which the patient is given

pure oxygen as a supplement to the existing breathable air around him. And even though

the human body depends on oxygen to survive, too much of it can also be harmful. The

5.4 Use Cases

exact amount of supplemental oxygen needed, however, is different with every patient. The

current approach is to give the patient a certain amount of oxygen using an oxygen cylinder

that is connected via a regulator and a thin tube to a nasal cannula or face mask. After a

short period of time, his blood oxygen saturation is checked and the flow of oxygen adjusted

if necessary. These steps are repeated until the desired saturation level is reached.

Sensor Device

Patient

Doctor

Regulator

ears

oxyge

vides

cur

lue

s t

satu

Figure 5.1:

Overview of persons and devices taking part in the oxygen regulation scenario

and their interactions (Icons [46]).

With the introduction of a sensor device to this scenario, this repetitive task can be automated.

The required devices are a

sensor device

capable of measuring blood oxygen saturation

and a

regulator device

that can communicate wirelessly with the sensor and regulate the

flow of oxygen. Figure 5.1 shows all the main parties and devices involved in this scenario

and the flowchart in Figure 5.4 visualizes the sequence of events that a program controlling

the regulator device would have to go through. After the device is switched on by medical

personnel, the

target level

of blood oxygen saturation that the patient should have is entered

and the device tries to establish a connection with the sensor device that is attached to the

patient. If it fails, it tries again while showing an error message. If it is successful, it reads

the

current saturation level

and compares it to the previously entered target value. If the

values are equal, the device changes nothing and waits for a certain amount of time after

which it re-scans the patient’s saturation level and compares it again. If the target and the

current oxygen saturation are not equal, there are two options. In case the current value is

lower than the target value, the patient needs a higher amount of supplemental oxygen and

the regulating device

increases

the flow of oxygen. In the opposite case, if the current value

is higher than the target value, the regulator device can

decrease

the flow of oxygen to the

5 Medical Applications

patient. Regardless of which case occurs, the next step is to wait a certain amount of time

until the changes can take effect and re-scan the current level.

Optionally, a smart mobile device such as a tablet computer could be used as

intermediary

It can display information about all the patients that are using this kind of equipment including

their current oxygen saturation and what their target level is set to. It can also be used as a

gateway

in cases where the regulator device and sensor device do not share a common

wireless communication technology.

5.4.2 Improved Fall Detection

An important issue of elderly people living alone in their homes are falls. Not only can the fall

itself cause injuries, if the fallen person is no longer capable of standing up by his own and

call for assistance, it also has the potential to result in

psychological trauma

and additional

physiological damage

. A matter of most importance after a fall is to get help as quickly as

possible. A system already available in some German cities is called, among other names,

“

Heimnotruf

”

which corresponds to

medical alert

personal emergency response system

(PERS)

in English speaking countries. The elderly person carries a device with a single

button on it and if that button is pressed, a telephone communication is established between

emergency services and a base station that is in the home of the user. If the elderly person

is responsive and can communicate, further actions that need to be taken are discussed. If

he does not respond, emergency personnel is dispatched immediately to his location. A big

problem with this system is the situation in which the elderly person falls and is not able to

press the emergency button because he is unconscious or can not reach the button on the

device.

As mentioned in Chapter 2, research in this area tries to automatically detect a fall event so

that further steps can be initiated without the need of user initiative. This is done either using

dedicated sensor devices such as accelerometers or using smartphones. The approach

proposed in this thesis uses both types of devices in order to increase sensitivity as well

as specificity. The required hardware is a

smartphone

and a

fitness sensor device

that

measures acceleration, pulse rate, blood oxygen saturation and blood pressure and can be

worn

on the wrist

. Figure 5.2 shows devices and people involved in this scenario. The initial

2http://www.samariterbund.net/pflege-betreuung/notrufsysteme

5.4 Use Cases

detection is done via the accelerometer data from the fitness sensor device in which possible

falls are identified. If such an identification occurs, it is tested whether or not the smartphone

is being worn on the body or not, using the sensor for ambient light in the smartphone. If

it is not worn at the time of the fall as detected by the fitness sensor device, the data from

the smartphone is dismissed and a decision whether or not to alert emergency services

is done solely on the basis of all the different bits of information gathered by the fitness

sensor device. If, however, the smartphone was worn on the body, its accelerometer data

is also consulted and on the basis of the

combined data

, a

more accurate

decision can be

made. The flowchart in Figure 5.5 visualizes the steps involved in this improved fall detection

process. A great advantage of this approach is that many people already have a smartphone

with them for most of the time in their daily lives so they do not have to remember to take it

with them, in contrast to the fitness sensor device which they may have remember to put on

every morning.

Sensor Device

Elderly Person

Family

Emergency Service

ears

Smartphone

ito

ifi

orts fa

Figure 5.2:

Overview of devices interacting with the elderly person and other participants in

context of the fall detection scenario.

5.4.3 Dosage of Medication

Overcrowded waiting rooms in hospitals and private practices are an inconvenience for both

patients and doctors. The patients are more likely to get irritated due to the time wasted

while waiting. Because of that, the doctors then have to work with discontent patients and

may feel pressured into completing examinations as fast as possible. A solution that targets

the problem of long distances that patients sometimes have to travel to see a necessary

5 Medical Applications

specialist is

telemedicine3

. The term describes a form of examination and treatment in

which patient and doctor are not in the same room, sometimes not even in the same country.

The patient still has to be in a medical facility that is equipped for this kind of procedure

but the doctor can remain in his own office. Their communication and necessary visual

examinations are conducted via video conference using cameras and monitors.

The approach proposed in this thesis tries to tackle the problem of overcrowded waiting

rooms from a different angle. If the number of times a patient has to visit the doctor’s office

could be

reduced

, less people would be waiting and the queue for those waiting would be

shorter. One process in which the patient could save himself the trip to the doctor is the

adjustment of the

dosage

newly prescribed medication

. Since finding the right dosage

depends on a multitude of factors and it differs with every patient, it is, to a certain extent, a

trial and error

process. The current approach is to estimate the initial dosage, taking into

account the doctor’s personal experience as well as common reference values. The patient

takes the medicine in this initial dose for a certain amount of time, after which he has to

return to his doctor, get examined and possibly have the dosage adjusted. This is repeated

until an adequate dose is determined.

Sensor Device

Patient

ears

asur

Smartphone

pro

Doctor`s Office

Doctor

data

accum

ate

Figure 5.3:

Overview of interactions between participants in the medication dosage scenario.

The scenario that was chosen to illustrate the proposed solution is the administration of

blood pressure medication. Even though current multi-sensor fitness devices do not support

the measurement of blood pressure, there are already stand-alone devices that not only

can communicate with smart mobile devices via Bluetooth but also are designed to take the

3http://www.nytimes.com/2012/10/09/health/nantucket-hospital-uses-telemedicine-as-bridge-to-mainland.html

5.5 Treat the Patient, Not the Disease

measurement on the wrist

. Therefore, it is very likely that future multi-sensor fitness devices

will be capable of also measuring blood pressure. The requirements for this scenario are a

sensor device that measures blood pressure and a smartphone or tablet. Furthermore, the

doctor has to provide the infrastructure necessary to transfer the blood pressure data, store

and analyze it. Figure 5.3 gives an overview of the interacting parts in this scenario.

The initial steps stay the same as in the current procedure, up to the point when the patient

returns back home for the first time and starts taking the medicine in the initially prescribed

dosage. From this point forward, the smart mobile device periodically gathers the patient’s

blood pressure value from the sensor device and stores it. After a certain amount of time, the

patient sends the accumulated data to his doctor who can then analyze the data and decide

whether to keep the current medication dosage, increase or decrease it. The flowchart in

Figure 5.6 gives a visual summary of all steps involved. This way, the patient saves himself

a trip to the doctor’s office and the doctor can organize his time in a more efficient way.

5.5 Treat the Patient, Not the Disease

The examples presented above, especially the automatic oxygen regulation and the ad-

justment of medication doses, describe situations in which it is technologically possible to

transfer tasks that are currently performed by doctors to computer systems that depend on

information from sensor devices. One aspect that came up on multiple occasions during

the conducted interviews of doctors and medical students can be summarized with the

saying “

treat the patient, not the disease

”. The idea behind this saying is that, when treating

a patient diagnosed with a disease, the doctor should not only focus on the symptoms and

causes of the disease but also look at the patient as a whole. This includes information such

as his

psychological state

, his

social environment

and possible

side-effects

of or

interac-

tions

between medication. This medical premise conflicts with the absence of interpersonal

communication and direct contact between doctor and patient in the suggested scenarios.

When fitness sensor devices are adopted into medical treatment processes and automate

some of the tasks, this limiting factor has to be taken into consideration. If necessary,

measures have to be devised to make up for the lack of direct interaction.

4http://www.ihealthlabs.com/blood-pressure-monitors/wireless-blood-pressure-wrist-monitor

5 Medical Applications

Activate Machine

Input Target SpO

Connect to Sensor

Successful?

Get Current SpO

Show Error Message

Target Value

Current Value

Wait Briefly

Target Value

Current Value

Increase O

Flow

Decrease O

Flow

yes

Figure 5.4:

Flowchart of the automated oxygen regulation process from the point of view of

the regulator device that controls the flow of oxygen.

5.5 Treat the Patient, Not the Disease

Activate Mobile App

Connect to FSD

Successful?

Get FSD Data

∗

Show Error Message

Fall possible?

Get SP Light Data

Smartphone

on Body?

Assess

with

Data

from Smartphone

Assess

without

Data

from Smartphone

Get SP Data

∗

Fall Detected?

Wait Briefly

Call

Emergency Services

yes

Figure 5.5:

Flowchart of the improved fall detection system using fitness sensor devices and

smartphone in combination (FSD = Fitness Sensor Device; SP = Smartphone;

Data∗= Data of multiple sensors).

5 Medical Applications

Activate Mobile App

Connect to Sensor

Successful?

Get Current BP

Show Error Message

Store BP Value

Doctor

Requested

Data

Wait Briefly

Transfer Data to Doc.

Analyze Data

good Keep Dosage

is too

low

Lower Dosage Raise Dosage

yes

Patient Domain

Doctor Domain

Figure 5.6:

Flowchart of the remote examination and adjustment of blood pressure medica-

tion (BP = Blood Pressure).

6 Concerns

The growing number of users of fitness sensor devices and their applicability for the medical

field and health care brings a multitude of challenges along with it. Besides the inherent

technical issues when developing the devices such as battery lifetime and durability, there are

social aspects that warrant closer consideration. As a prerequisite for accurate conclusions,

the sensor devices have to be worn over a prolonged period of time. Given the context of

fitness and health, the data poses risks to people’s privacy and everyday living. This chapter

discusses security and privacy issues as well as social challenges that may occur.

6.1 Privacy and Security Issues

The best and easiest way to preserve one’s privacy is not to collect any sensitive data at

all. When no private information is divulged, there is no risk of it ending up in the wrong

hands. Alan Westin, a former professor at Columbia University and well-known researcher

in this field, defined privacy as “the claim of individuals, groups, or institutions to determine

for themselves when, how, and to what extent information about them is communicated to

others” [65].

Despite the constantly growing need to protect one’s privacy against private corporations and

federal governments, there are situations where the possible gains outweigh the risks. The

medical use cases outlined in Chapter 5 can be considered as such exceptions, provided

that there are precautions put in place to minimize the risk of information leaks and to help

users to stay in control of their private information. The three main attack vectors that have

to be addressed are

acquisition

transfer

and

storage

of patient’s health data. This section

gives a short analysis of these

attack vectors

, following possible countermeasures that can

be taken against them.

6 Concerns

6.1.1 Data Gathering

The first possible leak of patient data (i.e. vital signs) occurs at the fitness sensor device

itself

or rather in the immediate

vicinity

around it. Nowadays, devices typically use Bluetooth

or Bluetooth Low Energy connections to transmit their data to the user’s smart mobile device

to be processed and visualized. These transmissions may be

intercepted

by a third party,

either by actively

hijacking

the connecting to the fitness device and receiving all data or

by passively

listening

to the communication between fitness device and its owner’s smart

device.

Some sensor devices store measured vital signs on an internal memory and keep it available

for a certain amount of time. The next time a smart mobile device is connected, all the

collected data can be transferred. In this way,

continuous

measurements are possible even

if there is

no connection

established to a smart mobile device at the time of recording. This

can pose a privacy problem, however, if the sensor device is lost or gets stolen because the

finder or thief has access to previously recorded data.

6.1.2 Data Transfer

The same risks hold true for the loss and theft of smart mobile devices that are used in

conjunction with sensor devices as even more accumulated measurements and evaluations

are stored on it. Considering what else can be found on today’s smartphones such as

messages

photos

and other

private

information, the disclosure of health data might be of

less importance for the owner in comparison.

The next weak link in the processing of health data appears when patient data is transferred

away from the user’s smart device. The destination of this transfer can be a dedicated

server or cloud storage owned by

private corporations

or a

doctor’s office

, or a central data

storage operated by the

Federal Department of Health

. If the data is handled carelessly, it

will most likely be sent via some kind of

HTTP communication

. In other words, it will be sent

unencrypted

as plain text. There are a number of scenarios in which this method poses

vulnerability

such as public Wi-Fi access points where unencrypted data traffic can be

easily intercepted by anyone inside the area of radio reception.

6.1 Privacy and Security Issues

6.1.3 Data Storage

Significant diagnoses and conclusions about health issues may be based, among other

things, on data that has been recorded over a considerable long period of time and stored

persistently

. The comparison of past data with current values allows medically trained

professionals to draw conclusions about the

progress of treatments

and about the

course

of diseases

. Because it obviously can not be avoided to store recorded patient health

information somewhere to be consulted at a later point in time, the huge amount of data

that accumulates will be a promising target for attackers. The two basic and most common

scenarios to be considered for data management are

centralized

and

decentralized

storage

of patient health data.

Centralized data storage could be provided and owned by the Federal Department of Health

or by large corporations such as health insurance companies. The number of individuals

whose personal health information is stored would be in the

millions

or even

tens of millions

The risk-reward ratio an attacker is faced with is low because if he can circumvent the

security measures put in place to protect the system, the amount of information he gains

access to is

enormous

. Even the threat of punishment may not be enough to discourage

attempts. Despite that, one big advantage of central data management is that specialized

companies can be tasked with the administration of the storage system and that imposed

security policies can be monitored and enforced more easily and efficiently.

In contrast, implementation and enforcement of such strict policies would not be feasible

with any decentralized solution. For example, if every doctor’s office had to store the health

data of their patients on their own, they would also be the ones responsible for installation

and maintenance of the storage infrastructure, as it is currently the case with existing patient

information. They would either handle it themselves or, more likely, delegate it to a local

IT company. With

thousands

of those small stand-alone solutions, it would be difficult to

enforce

consistent

policies and to ensure an overall high

level of security

. Even though it

would result in a higher likelihood for some of those isolated storage infrastructures to be

compromised, the impact on the general public would be minimal because each intrusion

only affects hundreds or thousands of individual patients. It would still be a bad situation for

those affected but less bad compared to the centralized scenario where a system break-in

would affect millions.

6 Concerns

6.1.4 Attackers and Countermeasures

Besides thieves that steal sensor and smart mobile devices and thereby gain access to

health data as

coincidental

by-product, there are two different kinds of main adversaries:

government

sanctioned organizations and

private

groups or

foreign

organizations. Whether

or not there are ways to protect patient data from local government institutions depends

mostly on the country and its respective legislation. In the United States of America, for

example, the FBI can subpoena records from companies using so called

National Security

Letters

. The companies have to comply and there is little to no leeway for them to oppose

it, especially as they are not

allowed

to publicly address the incident. However, National

Security Letters only cover

metadata

, not the information itself, so they, in their current form,

would not apply to actual health data.

In order to defend against foreign government institutions such as intelligence agencies

and private hacker groups that can not use the legal system to access the data, there

are technologies and approaches available that can help reduce the previously mentioned

vulnerabilities.

The initial bonding between sensor device and smart mobile device could be made more

secure. For example, by

supplementing

the bonding process with NFC. The close proximity

in which both devices have to be held next to each other in order for NFC to work, makes it

very difficult for an attacker to bond with a sensor device without the owner noticing it.

Probably the most important security measure in today’s world is encryption, meaning that

on the one hand,

every single time

any kind of patient health data is being transferred, the

communication channel has to be

encrypted

. For HTTP communication, there is

HTTPS

use instead. The Bluetooth and Bluetooth Smart transmissions between sensor device and

smart mobile device also have to happen encrypted. Furthermore, end-to-end encryption

such as

Off-the-Record1

should be used in order to prevent

information leaks

in cases

where vulnerabilities of HTTPS are exploited and the encryption of the communication

channel is breached. On the other hand, data also has to always be encrypted

before

it is

stored persistently. The key or passphrase used to encrypt and decrypt the data has to be

kept secret or in a separate and safe place and only authorized people are allowed to know

it or have access to it, otherwise the encryption would be futile.

1https://otr.cypherpunks.ca

6.2 Ethical Issues

The following part covers ethically very delicate topics. Especially Section 6.2.1 about the

adoption of sensory devices for the care of elderly people is difficult to approach. Decisions

pertaining to medical and social problems often not only concern the elderly person himself

but also

family members

close relatives

and

friends

. Notwithstanding the extreme personal

nature of such questions and the fact that there is no universally valid solution, their existence

is important enough to warrant mentioning them.

6.2.1 Geriatric Care

As presented in Chapter 5, sensor devices can be deployed in geriatric care. In particular,

when elderly people, widowed or single, are

living alone

in their own homes. Considering the

increased likelihood of

household accidents

and the scenario of fall detection and assuming

that the person fell because of a minor mishap such as stumbling or feeling dizzy. If

the person needs assistance but can not call for it himself because of general weakness

or broken bones, a sensor device that alarms medical personnel can

minimize

the time

the fallen person lies on the ground before help arrives and therefore minimizes possible

physiological

and

psychological

damage. In the depicted scenario, the use of a sensory

device is definitely helpful and poses no ethical problems.

Taking the increasing average life expectancy

into account, the term

quality of life

gains

more and more importance for the elderly population. For some people, getting as old

as possible is not everything. Other aspects of life such as, for example, their level of

independence

from assistance of other people or medical equipment or their own

mobility

have to be taken into account as well. The ethically challenging question is to find the right

balance between

living

and being

kept alive

and everyone has to find the best possible

answer for themselves.

With previous considerations in mind, there might be scenarios of emergencies in which a

person’s life is saved by the presence of a sensor device that would have died without it.

However, the person never fully recovers and experiences a lower quality of life from this

2http://data.worldbank.org/indicator/SP.DYN.LE00.IN

6 Concerns

point forward or in extreme cases is possibly

dependent

on life support system. The resulting

ethical considerations are similar to those of the do-not-resuscitate order in which a patient

can explicitly state his wish to

refuse

possibly life-saving measures such as cardiopulmonary

resuscitation (mouth-to-mouth and nose ventilation or chest compressions).

The two important item to take away from this discussion are: There are ethical ramifications

to be considered in that specific use of sensor devices and the decision whether or not to

use them should be left to everybody’s own discretion.

6.2.2 Quantified Self and Loss of Liberties

Another ethical problem could arise in a business environment. Obviously, a big objective

for many corporations is the need to make profit to return to their shareholders. One way

to achieve that objective is to gain the best performance out of their employees. With an

increasing amount of specific data about a person’s health and current physical condition

available, an attempt could be made to

optimize

the relation between time working and

time off in favor of work. This may be instigated either by individuals

themselves

who

intentionally make that choice or by

corporations

that urge their employees to do so. Fitbit,

for example, offers companies a special

Corporate Wellness

program that they advertise

for with slogans such as “create culture of well-being”, “increase employee productivity”,

“improve employee health status” and “boot acquisition & retention” [

]. The program also

includes statistical analysis and visualization of the gathered data on a

personal

group-

wide

and even

company-wide

scale. Well-known companies that take part in this program

are IBM

, BP and Adobe. In general, the phenomenon of trying to gather numerical and

categorical information about one’s own body is termed quantified self.

Furthermore, a development that can be currently observed in the car insurance business

has the potential to also be applied in the field of health insurance. There are car insurance

companies that offer incentives if the insured installs a device in his car that monitors certain

aspects of his driving. This data is

sent back

to the insurance company and analyzed. If

the manner of driving satisfies a given set of requirements, the insured person then pays a

reduced monthly fee.

3http://www-01.ibm.com/services/socomm/shared/pdf/2015beg.pdf

6.2 Ethical Issues

Applying this to health insurance, companies could offer special plans for members that wear

sensor devices and share their health data with them. Then again, if the recorded values

satisfy certain requirements, the monthly fee would be reduced. The person wearing sensor

devices might alter his behavior in some way, which on the one hand can have a positive

effect on the individual’s health. On the other hand, he might also adopt some changes

to his lifestyle only because he feels, and in fact is, constantly

monitored

and

supervised

by his insurance company. Depending on the set of constraints given by the company, the

insured might be limited in the liberties he previously enjoyed. This trend is already taking

place. The German insurance company Generali, for example, announced the introduction

of such an insurance plan. It offers reduced fees if the insured records certain health, fitness

and food related information and reports them back to Generali [18].

7 Conclusion and Future Work

Nowadays, technically oriented customers are presented with a wide variety of fitness sensor

devices. Most of them offer at least tracking of pulse rate and physical activities, some of

them have more uncommon sensors built in, such as for measuring skin temperature or

blood oxygen saturation. They can further chose the type and design best suited for their

individual needs, such as wristbands, chest belts and clips. Even though blood pressure

monitors and glucose meters are only available as stand-alone versions and are not yet

integrated in multi-sensor devices, it is only a matter of time until this changes. Assuming

that the current technological trend observed in recent years in this area continues, fitness

sensor devices are not only going to become smaller, more robust and more precise, they

are also going to combine even more different sensors.

The chances of this happening are promising since large technology corporations have

recognized the potential that lays in area of research. For example, Google has announced

that they are working on their own version of a sensor wristband that is not meant to be used

for fitness purposes but rather for tracking the users’ health and providing doctors with more

information about their patients [

]. Taking the minimization of sensor devices even one

step further, first sensors have already been integrated and sewn into clothing to be worn by

the user like every other piece of clothing. The trend is further aided by advancements in

wireless communication technologies and their more and more widespread use in everyday

activities such as in payment methods.

In addition, the number of manufactures that facilitate access to the sensor data recorded

by their sensor devices, in contrast to using proprietary protocols, is increasing. This is

a prerequisite so that developers and researchers can freely integrate these devices into

their own smart mobile applications. This will open up new opportunities to evaluate their

applicability in other areas. One of which may be sports medicine, particularly sports in

which athletes pilot fast and powerful machinery, such as race cars, airplanes and boats.

7 Conclusion and Future Work

Blackouts or even short dizzy spells can have disastrous consequences and have to be

prevented. Sensor devices may help by providing crucial information about the athlete’s

state of health.

However, the three medical use cases presented in this thesis show that the capabilities

of fitness sensor devices are also quite suited for classic medical applications. They can

improve patient care not only for the patients undergoing examinations or treatments but

also for the doctors and medical personnel treating them. The devices can help save time

and increase flexibility by reducing the number of times patients have to go see their doctors

which is also an advantage for the doctors. They have more time available for the patients

that still have to come in. The devices can help make better decisions regarding treatment

options and reduce costs by providing doctors with additional and more complete health

data. Further, they can improve quality of life, especially for the elderly living home alone, by

providing them with rapid assistance in cases of falls.

There are, however, some limitations. From a medical point of view, automation of patient

interaction may not always be the best option when treating patients. There is information

the doctor depends on for his decisions that sensor devices can not provide. From a

technical point of view, fitness sensor devices still have to be made more resilient against

outside interference such as ambient light, movements or vibrations. Only then are they

able to collect reliable health data in everyday situations. Nevertheless, as shown during the

hands-on tests, current sensor devices such as the iHealth Pulse Oximeter PO3 perform

reasonable well, even in situations that go beyond their specifications. But it was also shown

that, depending on the situation these devices are used in, their measurements can deviate

significantly. Future research of adoption of sensor devices in health and patient care would

benefit from close interdisciplinary collaboration between computer scientists, hardware

manufactures and medical personnel. Useful solutions can only be found if technical as well

as medical aspects are taken into account.

A further technical aspect that needs to be considered are the risks to users’ privacy. New

sensor devices such as the Angel sensor wristband should be tested in regard to accuracy

of the recorded data, and in regard to the privacy implications that may occur by recording

health data non-stop. In order to address some of the privacy concerns, further research

may be able to derive specific characteristics from the recorded data that is sufficiently

anonymized but still allows doctors to draw conclusions about the patient’s health. Thereby,

it could be prevented that raw health data has to be divulged. Also,

Bluetooth 4.2

and its

feature

LE Privacy 1.2

[

] warrant further investigation whether it can be used to improve

the privacy of users of sensor devices.

Finally, an important medical factor that is imperative to improve is patient compliance. Under

which conditions are patients willing to wear a sensor device over a prolonged period of

time? Are there situations in which the device becomes cumbersome? Can patients be

bothered with daily recharging of their device? Do they realize what benefits these devices

have for them? A survey among potential users could bring valuable insights about relevant

features that are needed for patients’ acceptance. Another survey that may give interesting

results could be conducted among doctors, asking about how they think sensor devices

might help the most when treating their patients, whether they would use them in their

patient care and if not, why and how the devices have to be improved to convince them

otherwise.

A SPSS Output

A.1 Tests of Normality

Kolmogorov-SmirnovaShapiro-Wilk

Statistic df Sig. Statistic df Sig.

Difference PC9 to PO3

Difference PC9 to Moto 360

Difference PO3 to Moto 360

,159 61 ,001 ,939 61 ,005

,091 61 ,200*,978 61 ,341

,084 61 ,200*,973 61 ,197

This is a lower bound of the true significance.*.

Lilliefors Significance Correctiona.

Figure A.1: Test of normality for the differences in the sitting scenario.

Kolmogorov-SmirnovaShapiro-Wilk

Statistic df Sig. Statistic df Sig.

Difference PC9 to PO3

Difference PC9 to Moto 360

Difference PO3 to Moto 360

,117 61 ,036 ,970 61 ,141

,121 61 ,026 ,973 61 ,196

,128 61 ,014 ,963 61 ,062

Lilliefors Significance Correctiona.

Figure A.2: Test of normality for the differences in the standing scenario.

Kolmogorov-SmirnovaShapiro-Wilk

Statistic df Sig. Statistic df Sig.

Difference PC9 to PO3 ,136 61 ,007 ,953 61 ,021

Lilliefors Significance Correctiona.

Figure A.3: Test of normality for the differences in the cycling scenario.

A SPSS Output

A.2 Q–Q Plots and Frequency Histograms

Observed Value

420-2-4-6

Expected Normal

-1

-2

-3

(a) Q–Q Plot PC9 to PO3

Difference PC9 to PO3

420-2-4-6

Frequency

12,5

10,0

7,5

5,0

2,5

0,0

(b) Frequency Histogram PC9 to PO3

Figure A.4: Q–Q Plots and Frequency Histograms for the cycling comparison scenario.

A.2 Q–Q Plots and Frequency Histograms

Observed Value

1050-5-10

Expected Normal

-2

-4

(a) Q–Q Plot PC9 to PO3

Difference PC9 to PO3

1050-5-10

Frequency

(b) Frequency Histogram PC9 to PO3

Observed Value

151050-5-10-15

Expected Normal

-2

-4

Difference PC9 to Moto 360

1050-5-10-15

Frequency

12,5

10,0

7,5

5,0

2,5

0,0

(d) Frequency Histogram PC9 to Moto 360

Observed Value

20100-10-20

Expected Normal

-2

(e) Q–Q Plot PO3 to Moto 360

Difference PO3 to Moto 360

20100-10-20

Frequency

(f) Frequency Histogram PO3 to Moto 360

Figure A.5: Q–Q Plots and Frequency Histograms for the sitting comparison scenario.

A SPSS Output

Observed Value

7,55,02,50,0-2,5-5,0

Expected Normal

-1

-2

(a) Q–Q Plot PC9 to PO3

Difference PC9 to PO3

7,55,02,5,0-2,5-5,0

Frequency

(b) Frequency Histogram PC9 to PO3

Observed Value

100-10-20

Expected Normal

-2

-4

Difference PC9 to Moto 360

1050-5-10-15-20

Frequency

12,5

10,0

7,5

5,0

2,5

0,0

(d) Frequency Histogram PO3 to Moto 360

Observed Value

100-10-20

Expected Normal

-2

-4

(e) Q–Q Plot PO3 to Moto 360

Difference PO3 to Moto 360

151050-5-10-15

Frequency

12,5

10,0

7,5

5,0

2,5

0,0

(f) Frequency Histogram PO3 to Moto 360

Figure A.6: Q–Q Plots and Frequency Histograms for the standing comparison scenario.

List of Figures

3.1

Simplified blood flow through the human heart. Red arrows stand for oxy-

genated blood, blue arrows represent deoxygenated blood [47]. . . . . . . . . 9

3.2

Estimated number of people (ages

) suffering from diabetes (solid

line) and its predicted growth in the near future (dashed line) [29, 30]. . . . . 10

3.3

Pulse Oximeter placed on a finger with the red and infrared LEDs, interfering

ambient light and photodetector. . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.4 Electrocardiogram of a healthy young adult. . . . . . . . . . . . . . . . . . . . 14

3.5 Manual (a) and automatic (b) blood measurement methods [21, 43]. . . . . . 16

3.6

Measuring the glucose level in a small drop of blood using a glucose meter [

3.7

Principles of mechanical accelerometers (a) and capacitive accelerometers (b).

4.1 Structure of the Bluetooth Smart Heart Rate Profile. . . . . . . . . . . . . . . 25

4.2 The iHealth Wireless Pulse Oximeter PO3 by iHealth Lab Inc. . . . . . . . . . 28

4.3 Excerpt from the iHealth Pulse Oximeter PO3 owner’s manual. . . . . . . . . 29

4.4 Pulse oximetry data of 60 minutes sitting. . . . . . . . . . . . . . . . . . . . . 31

4.5 Pulse oximetry data of 60 minutes standing and walking. . . . . . . . . . . . . 32

4.6 Incomplete pulse oximetry data of a 10 minute bicycle ride. . . . . . . . . . . 33

LIST OF FIGURES

4.7 Pulse oximetry data of an 60 minute bicycle ride. . . . . . . . . . . . . . . . . 34

4.8 Pulse oximetry data of a 34 minute visit to a friend. . . . . . . . . . . . . . . . 35

4.9 Pulse oximetry data of an 120 minute trip to the city of Ulm. . . . . . . . . . . 37

4.10

The watch component of the Sigma PC9 heart rate monitor showing a heart

rate of

98 bpm

(a) and the Motorola Moto 360 smartwatch showing its Moto

Body Heart Rate application currently measuring (b). . . . . . . . . . . . . . . 39

4.11

Comparison of heart rates gathered from iHealth PO3, Sigma PC9 and

Motorola Moto 360 while sitting calmly in a chair. . . . . . . . . . . . . . . . . 41

4.12 The deviations of the displayed values for each pair of devices while sitting. . 42

4.13

Comparison of heart rates gathered from iHealth PO3, Sigma PC9 and

Motorola Moto 360 while standing and working at a desk. . . . . . . . . . . . 43

4.14 The deviations of the displayed values for each pair of devices while standing. 43

4.15

Comparison of heart rate values gathered from iHealth PO3 and Sigma PC9

while stationary cycling indoors. . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.16

The deviations of the values displayed on the iHealth PO3 and the Sigma

PC9 while cycling indoors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.17

Comparison of the percentage deviations between the values displayed by

the iHealth PO3 and the Sigma PC9 during the three scenarios. . . . . . . . 46

5.1

Overview of persons and devices taking part in the oxygen regulation scenario

and their interactions (Icons [46]). . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.2

Overview of devices interacting with the elderly person and other participants

in context of the fall detection scenario. . . . . . . . . . . . . . . . . . . . . . . 55

5.3

Overview of interactions between participants in the medication dosage sce-

nario. ........................................ 56

LIST OF FIGURES

5.4

Flowchart of the automated oxygen regulation process from the point of view

of the regulator device that controls the flow of oxygen. . . . . . . . . . . . . . 58

5.5

Flowchart of the improved fall detection system using fitness sensor devices

and smartphone in combination. . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.6

Flowchart of the remote examination and adjustment of blood pressure medi-

cation......................................... 60

A.1 Test of normality for the differences in the sitting scenario. . . . . . . . . . . . 73

A.2 Test of normality for the differences in the standing scenario. . . . . . . . . . 73

A.3 Test of normality for the differences in the cycling scenario. . . . . . . . . . . 73

A.4 Q–Q Plots and Frequency Histograms for the cycling comparison scenario. . 74

A.5 Q–Q Plots and Frequency Histograms for the sitting comparison scenario. . . 75

A.6 Q–Q Plots and Frequency Histograms for the standing comparison scenario. 76

List of Tables

3.1 The three grades of severity of hypertension according to the WHO. . . . . . 16

4.1 Overview of modern (fitness) sensor devices and their sensory capabilities. . 22

4.2 Detailed explanation of the actions taken during the 120 minute recording of

atriptoamajorcity. ................................ 38

5.1 Possible outcomes of a medical test. . . . . . . . . . . . . . . . . . . . . . . . 48

List of Abbreviations

ADL . . . . . . . . . . . . . . . . activities of daily living

AFH . . . . . . . . . . . . . . . . adaptive frequency hopping

API . . . . . . . . . . . . . . . . . application programming interface

bpm . . . . . . . . . . . . . . . . beats per minute

CO2. . . . . . . . . . . . . . . . carbon dioxide

COPD . . . . . . . . . . . . . . . chronic obstructive pulmonary disease

DoS . . . . . . . . . . . . . . . . denial-of-service

EKG . . . . . . . . . . . . . . . . electrocardiography

GATT . . . . . . . . . . . . . . . generic attribute profile

Hb . . . . . . . . . . . . . . . . . hemoglobin

IR . . . . . . . . . . . . . . . . . infrared

LED . . . . . . . . . . . . . . . . light-emitting diode

mg/dl . . . . . . . . . . . . . . . . milligrams per deciliter

mmHg . . . . . . . . . . . . . . . millimeters of mercury

mmol/l . . . . . . . . . . . . . . . millimoles per liter

NFC . . . . . . . . . . . . . . . . Near Field Communication

List of Abbreviations

NTC . . . . . . . . . . . . . . . . negative temperature coefficient

O2.................oxygen

PERS . . . . . . . . . . . . . . . personal emergency response system

PTC . . . . . . . . . . . . . . . . positive temperature coefficient

RFID . . . . . . . . . . . . . . . . Radio-frequency Identification

RTD . . . . . . . . . . . . . . . . resistance temperature detectors

SaO2. . . . . . . . . . . . . . . . arterial oxygen saturation

SD . . . . . . . . . . . . . . . . . standard deviation

SDK . . . . . . . . . . . . . . . . software development kit

SE . . . . . . . . . . . . . . . . . standard error of the mean

SIG . . . . . . . . . . . . . . . . Special Interest Group

SO2. . . . . . . . . . . . . . . . oxygen saturation

SpO2. . . . . . . . . . . . . . . . peripheral capillary oxygen saturation

TUG . . . . . . . . . . . . . . . . Timed Up and Go test

USB . . . . . . . . . . . . . . . . Universal Serial Bus

WHO . . . . . . . . . . . . . . . . World Health Organization

WSN . . . . . . . . . . . . . . . . wireless sensor network

References

[1] BERLIT, P.: Klinische Neurologie. Springer Berlin Heidelberg, 2011, p. 483

[2]

BLUETOOTH SIG:

15 Things You Didn’t Know About Bluetooth Technol-

ogy

http://blog.bluetooth.com/15-things-you-didnt-know-about-

bluetooth-technology. – Accessed July 15, 2015

[3]

BLUETOOTH SIG:

15 Years of Bluetooth Technology

http://www.bluetooth.

com/Pages/15-Years-of-Bluetooth-Technology.aspx

. – Accessed July 15,

2015

[4]

BLUETOOTH SIG:

Bluetooth Core Specification 4.2: Frequently Asked Questions

https://www.bluetooth.org/en-us/Documents/Bluetooth4-2FAQ.pdf

. –

Accessed August 28, 2015

[5]

BLUETOOTH SIG:

Heart Rate Profile

https://developer.bluetooth.org/

gatt/profiles/Pages/ProfileViewer.aspx?u=org.bluetooth.profile.

heart_rate.xml. – Accessed July 20, 2015

[6]

BLUETOOTH SIG:

A Look at the Basics of Bluetooth Technology

http://www.

bluetooth.com/Pages/Basics.aspx. – Accessed July 15, 2015

[7]

BLUETOOTH SIG:

The Low Energy Technology Behind Bluetooth Smart

http:

//www.bluetooth.com/Pages/low-energy-tech-info.aspx

. – Accessed

July 16, 2015

[8] BLUETOOTH SIG: Bluetooth Core Specification V4.0. 2010

REFERENCES

[9]

CHEN, Caroline ; WOMACK, Brian:

Google Reveals Health-Tracking Wristband

http://www.bloomberg.com/news/articles/2015-06-23/google-

developing-health-tracking-wristband-for-health-research

. –

Accessed August 25, 2015

[10]

CLAXTON, Ami J. ; CRAMER, Joyce ; PIERCE, Courtney: A systematic review of

the associations between dose regimens and medication compliance. In:

Clinical

Therapeutics 23 (2001), Nr. 8, pp. 1296–1310

[11]

DIMENSION ENGINEERING:

A beginner’s guide to accelerometers

http://www.

dimensionengineering.com/info/accelerometers

. – Accessed August 15,

2015

[12]

ENDEAVOUR PARTNERS:

Inside Wearables – How the Science of Behavior Change

Offers the Secret to Long-Term Engagement

http://endeavourpartners.net/

white-papers. – Accessed July 9, 2015

[13]

FALLER, H. ; LANG, H.:

Medizinische Psychologie und Soziologie

. 3. Springer Berlin

Heidelberg, 2011, pp. 324–328 (Springer-Lehrbuch)

[14]

FIELD, A.:

Discovering Statistics using IBM SPSS Statistics

. 3. SAGE Publications,

2009

[15]

FITBIT:

Fitbit Corporate Wellness

http://www.fitbit.com/fitbit-wellness

– Accessed August 13, 2015

[16]

FRENCH NATIONAL RFID CENTER:

RFID frequency ranges

http:

//www.centrenational-rfid.com/rfid-frequency-ranges-article-

16-gb-ruid-202.html. – Accessed July 12, 2015

[17]

FÜESSL, H. ; MIDDEKE, M.:

Anamnese und klinische Untersuchung

. Thieme, 2010,

p. 169 (Duale Reihe)

[18]

GENERALI DEUTSCHLAND:

Vitality

http://www.wolfgang.schreiber.service.

generali.de/online/portal/gdinternet/de/content/311198/1150478

– Accessed August 25, 2015

REFERENCES

[19]

GOLD, Deborah T. ; MCCLUNG, Betsy: Approaches to Patient Education: Emphasizing

the Long-Term Value of Compliance and Persistence. In:

The American Journal of

Medicine 119, Nr. 4, pp. 32–37

[20]

HAMMACK, W.S. ; HAMMACK, B. ; RYAN, P. ; ZIECH, N.:

Eight Amazing Engineering

Stories: Using the Elements to Create Extraordinary Technologies

. Articulate Noise

Books, 2012

[21]

HARRISON, Eddie:

Lt. Cmdr. Hung Tran checks the blood pressure of a Viet-

namese patient during a Pacific Partnership 2010 medical civic action pro-

gram

https://upload.wikimedia.org/wikipedia/commons/9/9d/US_

Navy_100606-N-4044H-159_Lt._Cmdr._Hung_Tran_checks_the_blood_

pressure_of_a_Vietnamese_patient_during_a_Pacific_Partnership_

2010_medical_civic_action_program.jpg. – Accessed August 25, 2015

[22]

HAYNES, Jeffrey M.: The Ear as an Alternative Site for a Pulse Oximeter Finger Clip

Sensor. In: Respiratory Care 52 (2007), Nr. 6, pp. 727–729

[23]

HIXON, Todd:

Are Health And Fitness Wearables Running Out Of Gas?

http://www.forbes.com/sites/toddhixon/2014/04/24/are-health-

and-fitness-wearables-running-out-of-gas. – Accessed July 26, 2015

[24]

HUPPELSBERG, J. ; WALTER, K.:

Kurzlehrbuch Physiologie

. 4. Thieme, 2013

(Kurzlehrbuch)

[25]

IBM:

SPSS Statistics 21, 2012

http://www-01.ibm.com/software/

analytics/spss. – Accessed August 2, 2015

[26]

IGUAL, R. ; MEDRANO, C. ; PLAZA, I.: Challenges, issues and trends in fall detection

systems. In: BioMedical Engineering OnLine 12 (2013)

[27]

IHEALTH LAB INC.:

iHealth Wireless Pulse Oximeter Owner’s Manual

http://www.ihealthlabs.com/files/9414/0192/1752/wireless_

pulse_oximeter_User_Manual.pdf. – Accessed July 24, 2015

REFERENCES

[28]

INSTRUMENTATIONTODAY:

Radiation Pyrometer

http://www.

instrumentationtoday.com/radiation-pyrometer/2011/08

. – Accessed

July 22, 2015

[29] INTERNATIONAL DIABETES FEDERATION:IDF Diabetes Atlas. 1st – 6th Edition

[30]

INTERNATIONAL DIABETES FEDERATION:

IDF Diabetes Atlas Key Findings 2014

http:

//www.idf.org/diabetesatlas/update-2014. – Accessed July 20, 2015

[31]

KARDACH, Jim:

Tech History: How Bluetooth Got Its Name

http://www.eetimes.

com/document.asp?doc_id=1269737. – Accessed July 24, 2015

[32]

KARGL, F. ; LAWRENCE, E. ; FISCHER, M. ; LIM, Y.: Security, Privacy and Legal Issues

in Pervasive eHealth Monitoring Systems. In:

7th International Conference on Mobile

Business ICMB, pp. 296–304, 2008

[33]

KELLY, Greg: Body temperature variability (Part 1): a review of the history of body

temperature and its variability due to site selection, biological rhythms, fitness, and

aging. In: Alternative Medicine Review 11 (2006), December, Nr. 4, pp. 278–293

[34]

KESSLER, H.:

Kurzlehrbuch Medizinische Psychologie und Soziologie

. 3. Thieme,

2015, p. 67 (Kurzlehrbuch)

[35]

LARKIN, K. T. ; ZAYFERT, C.:

Oscillometric Method - Methods of Blood

Pressure Measurement

http://www.severehypertension.net/hbp/more/

oscillometric-method. – Accessed July 19, 2015

[36]

LONGO, D. ; FAUCI, A. ; KASPER, D. ; HAUSER, S. ; JAMESON, J. ; LOSCALZO, J.:

Harrison’s Principles of Internal Medicine, 18th Edition. Mcgraw-hill, 2011

[37]

MAIER, Christoph:

Langzeitmessung der Hauttemperatur hilft Schmerzsyndrom diag-

nostizieren

http://www.dgss.org/news-detail/?tx_ttnews[tt_news]=

259&cHash=c53cf99fa1304f48561f502cee987bde. – Accessed July 26, 2015

[38]

MAJUMDER, A. J. A. ; RAHMAN, F. ; ZERIN, I. ; EBEL, W. Jr. ; AHAMED, S. I.: iPrevention:

Towards a Novel Real-time Smartphone-based Fall Prevention System. In:

Proceedings

of the 28th Annual ACM Symposium on Applied Computing, pp. 513–518, 2013

REFERENCES

[39]

MALAN, D. ; FULFORD-JONES, T. ; WELSH, M. ; MOULTON, S.: Codeblue: An ad hoc

sensor network infrastructure for emergency medical care. In:

International workshop

on wearable and implantable body sensor networks 5 (2004), pp. 12–14

[40]

MARSCHOLLEK, M. ; REHWALD, A. ; WOLF, K.-H. ; GIETZELT, M. ; NEMITZ, G. ;

SCHWABEDISSEN, H. Meyer z. ; SCHULZE, M.: Sensors vs. experts - A performance

comparison of sensor-based fall risk assessment vs. conventional assessment in a

sample of geriatric patients. In:

BMC Medical Informatics and Decision Making

(2011)

[41]

MEINGAST, M. ; ROOSTA, T. ; SASTRY, S.: Security and Privacy Issues with Health

Care Information Technology. In:

Engineering in Medicine and Biology Society. EMBS

’06. 28th Annual International Conference of the IEEE, pp. 5453–5458, 2006

[42]

MICRO-CHIP TECHNOLOGIES:

NTC Thermistors

http://www.microchiptechno.

com/ntc_thermistors.php. – Accessed July 24, 2015

[43]

NAGY, Csaba:

Blood Pressure Monitor

https://pixabay.com/en/blood-

pressure-monitor-bless-you-838109. – Accessed August 25, 2015

[44]

NATIONAL HEART, LUNG AND BLOOD INSTITUE:

What Are the Signs and

Symptoms of Hypotension? https://www.nhlbi.nih.gov/health/health-

topics/topics/hyp/signs. – Accessed July 20, 2015

[45]

NFC FORUM:

NFC and Contactless Technologies

http://nfc-forum.org/what-

is-nfc/about-the-technology. – Accessed July 12, 2015

[46]

OPEN SECURITY ARCHITECTURE:

OSA Icon Library 13.05

http://www.

opensecurityarchitecture.org/cms/library/icon-library

. – Accessed

August 29, 2015

[47]

OPENCLIPARTVECTORS:

Heart and Blood Flow

https://pixabay.com/en/

heart-blood-flow-arteries-veins-148157. – Accessed August 25, 2015

[48]

PARIKH, R. ; MATHAI, A. ; PARIKH, S. ; CHANDRA SEKHAR, G. ; THOMAS, R.: Under-

standing and using sensitivity, specificity and predictive values. In:

Indian J Ophthalmol

56 (2008), Nr. 1, pp. 45–50

REFERENCES

[49]

RAYTEK CORP.:

The Principles of Noncontact Temperature Measurement

http://support.fluke.com/raytek-sales/Download/Asset/9250315_

ENG_D_W.PDF. – Accessed July 22, 2015

[50]

RUNTASTIC TEAM:

Herzfrequenzmesser: Wie sie funktionieren & dein Train-

ing unterstützen

http://blog.runtastic.com/de/technologie/

herzfrequenzmesser-wie-sie-funktionieren. – Accessed July 20, 2015

[51]

SALOMON, R. ; LUDER, M. ; BIEBER, G.: iFall - a new embedded system for the detection

of unexpected falls. In:

Pervasive Computing and Communications Workshops (PER-

COM Workshops), 2010 8th IEEE International Conference on, pp. 286–291, 2010

[52]

SAVICKYTE, Terese:

Blood Glucose Meter

https://pixabay.com/en/diabetes-

blood-finger-glucose-777002. – Accessed August 25, 2015

[53]

SCHMIDT, Robert F. ; LANG, Florian:

Physiologie des Menschen mit Pathophysiologie

30. Springer Berlin Heidelberg, 2011 (Springer-Lehrbuch)

[54]

SCHOBEL, Johannes ; SCHICKLER, Marc ; PRYSS, Rüdiger ; NIENHAUS, Hans ;

REICHERT

, Manfred: Using Vital Sensors in Mobile Healthcare Business Applications:

Challenges, Examples, Lessons Learned. In:

9th Int’l Conference on Web Informa-

tion Systems and Technologies (WEBIST 2013), Special Session on Business Apps,

pp. 509–518, 2013

[55]

SO, Chi-Fuk ; CHOI, Kup-Sze ; WONG, Thomas K. ; CHUNG, Joanne W.: Recent

advances in noninvasive glucose monitoring. In:

Medical Devices (Auckland, N.Z.)

pp. 45–52 (2012)

[56]

SPOSARO, F. ; TYSON, G.: iFall: An android application for fall monitoring and response.

In:

Engineering in Medicine and Biology Society. Annual International Conference of

the IEEE, pp. 6119–6122, 2009

[57]

TEMPERATURES.COM INC.:

Contact Sensors

http://www.temperatures.com/

sensors/csensors/. – Accessed July 24, 2015

REFERENCES

[58]

THE AMERICAN HEART ASSOCIATION:

All About Heart Rate (Pulse)

http:

//www.heart.org/HEARTORG/Conditions/More/MyHeartandStrokeNews/

All-About-Heart-Rate-Pulse_UCM_438850_Article.jsp

. – Accessed

July 24, 2015

[59]

TILAKARATNA, Prasanna:

How pulse oximeters work explained simply

http://www.

howequipmentworks.com/pulse_oximeter. – Accessed July 26, 2015

[60]

TIWARI, A. ; PATRA, H.K. ; CHOI, J.W.:

Advanced Theranostic Materials

. Wiley, 2015

(Advanced Material Series)

[61] TOWNSEND, Neil: Oscillometry. In: Digital Medicine, pp. 48–54 (2010)

[62]

TURA, Andrea ; MARAN, Alberto ; PACINI, Giovanni: Non-invasive glucose monitoring:

Assessment of technologies and devices according to quantitative criteria. In:

Diabetes

Research and Clinical Practice 77 (2007), Nr. 1, pp. 16–40

[63]

WARD, Matthew ; LANGTON, Jeremy A.: Blood pressure measurement. In:

Continuing

Education in Anaesthesia, Critical Care & Pain 7 (2007), Nr. 4, pp. 122–126

[64]

WEBMD:

How and When to Test Your Blood Sugar With Diabetes

http://www.

webmd.com/diabetes/how-test-blood-glucose. – Accessed July 20, 2015

[65] WESTIN, Alan: Privacy and Freedom. Atheneum, 1967

[66]

WHO:

Adherence to Long-Term Therapies: Evidence for Action

http://www.

who.int/chp/knowledge/publications/adherence_full_report.pdf

. –

Accessed July 20, 2015

[67]

WHO:

Guidelines for the Management of Hypertension

http://www.who.int/

cardiovascular_diseases/guidelines/hypertension_guidelines.pdf

– Accessed July 19, 2015

Name: Christoph Bachmaier Matrikelnummer: 677055

Erklärung

Ich erkläre, dass ich die Arbeit selbständig verfasst und keine anderen als die angegebenen

Quellen und Hilfsmittel verwendet habe.

Ulm,den ..............................................................................

Christoph Bachmaier