Comprehensive insights into the TrackYourTinnitus database [original]

Available online at www.sciencedirect.com

Procedia Computer Science 00 (2019) 000–000

www.elsevier.com/locate/procedia

The 17th International Conference on Mobile Systems and Pervasive Computing (MobiSPC)

August 9-12, 2020, Leuven, Belgium

Comprehensive insights into the TrackYourTinnitus database

Robin Krafta,b,∗, Michael Stacha, Manfred Reicherta, Winfried Schleec, Thomas Probstd,

Berthold Langguthc, Marc Schicklera, Harald Baumeisterb, R¨

udiger Prysse

aInstitute of Databases and Information Systems, Ulm University, Germany

bDepartment of Clinical Psychology and Psychotherapy, Ulm University, Germany

cClinic and Policlinic for Psychiatry and Psychotherapy, University of Regensburg, Germany

dDepartment for Psychotherapy and Biopsychosocial Health, Danube University Krems, Austria

eInstitute of Clinical Epidemiology and Biometry, University of W¨urzburg, Germany

Abstract

The ubiquity of smart mobile devices facilitates data collection in the healthcare domain. Two of the concepts, which can be

applied in this context, are mobile crowdsensing (MCS) and ecological momentary assessment (EMA). TrackYourTinnitus (TYT)

is an advanced mobile healthcare platform that combines both concepts enabling the monitoring and evaluation of the users’

individual variability of tinnitus symptoms. This paper describes the underlying data set and structure of the TYT mobile platform

and highlights selected issues whose investigation provides advanced insights into the users of this mobile platform as well as their

data.

2020 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Peer-review under responsibility of the Conference Program Chairs.

Keywords: Mobile crowdsensing (MCS); ecological momentary assessment (EMA); mobile healthcare application; chronic disorder; tinnitus

1. Introduction

As smart mobile devices (e.g., smartphones and smartwatches) are becoming ubiquitous, new opportunities for

collecting data emerge in the healthcare domain. Mobile crowdsensing (MCS) and ecological momentary assessment

(EMA) constitute two fundamental concepts that benefit greatly from these advancements [6,4]. Both concepts can

be used in combination in the form of mobile applications to correlate EMA data with sensor measurement data,

enabling valuable additional insights into the patients’ situation [3]. The TrackYourTinnitus (TYT) mobile platform

utilizes MCS and EMA to track the individual tinnitus of a user as well as to monitor and evaluate its variability over

∗Corresponding author.

E-mail address: [email protected]

1877-0509 c

2020 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Peer-review under responsibility of the Conference Program Chairs.

2R. Kraft et al. /Procedia Computer Science 00 (2019) 000–000

time. This work describes the data set and structure that underlies the TYT platform, highlights insights that we gained

from the gathered data, and discusses technical issues and limitations.

A recent review on mobile health crowdsensing [12] has confirmed that, although an increasing amount of research

is conducted in the field, only few experience reports exist that were gained over multiple large-scale, long-running

projects like TYT and related MCS-EMA endeavors. Although several MCS and EMA mobile applications with re-

search background (e.g., Moodpath1), open source platforms (e.g., PACO2), and commercial tools such (e.g, ilumivu3)

exist, only little information is publicly available about the data structures underlying these apps as well as the insights

that can be gained from them. This makes it quite challenging to compare existing approaches or to start new projects

and research in this area. The work at hand addresses this issue by providing the following contributions:

•A comprehensive description of the TYT data set and its structure is provided. This allows (1) better under-

standing and reproducing the results obtained from this data set, (2) comparing these results with the data from

other projects, and (3) serving as basis for new projects.

•Descriptive statistics for different aspects in the data set are provided. Particularly, insights that can be gained

from these data and that crave for additional investigations are presented.

•Future analyses, technical issues, and limitations of the data set are discussed.

The remainder of this paper is organized as follows. In Section 2, background information on the TYT platform as

well as the underlying data set is provided. Descriptive statistics and the insights gained from these data are presented

in Section 3. Section 4discusses future evaluations and analyses, technical issues and limitations of the current data

set. Finally, Section 5concludes the paper with a summary and an outlook.

2. Background information

In this section, background information on the TYT mobile platform is provided and the underlying data set as

well as its structure is described.

The TrackYourTinnitus (TYT) platform has been in operation since 2014 and been continuously evolved. It is

composed of a registration and information website4, two native mobile applications (iOS and Android), and a central

backend that stores the collected data in a relational database. The mobile applications track the individual tinnitus

perception of users by asking them to complete ecological momentary assessment (EMA) [11] questionnaires for

tinnitus assessment at randomly selected times of the day [8]. The exact procedure was described in [3]. Tinnitus is

the perception of an internal sound when there is no corresponding external noise. The symptoms are subjective and

vary over time. Therefore, TYT was implemented to monitor and evaluate the variability of symptoms over time based

on EMA and MCS [10].

Before describing the data set itself, the data collection & processing procedure is briefly outlined. The data is

first collected by the mobile applications, stored on the mobile device and, if the device is connected to the internet,

eventually transmitted to the backend. The backend then validates and assigns the transmitted data to the model

classes, and finally stores the data in a relational database. The relational data model has been chosen to enforce data

consistency based on the well-known ACID transaction guarantees. For this work, a snapshot of the data set was

extracted on 23 March 2020 and, in order to reduce noise in the data, subsequently cleaned. This snapshot represents

the basis of the descriptive statistics of this work.

In order to illustrate the data set, an excerpt of the TYT data model representing the entered user data is depicted

as Entity-relationship model in Fig. 1. Since TYT uses an MCS approach to collect data, the user table is either in

direct or indirect relation with all other entities. To separate operational data and user data, table User Metadata (see

Fig. 1,A

) is responsible for storing demographic data (e.g., country, mobile device information, etc.); currently, it

comprises data of 4929 unique users. To conduct studies, users may be assigned to one or more user groups.

1https://mymoodpath.com/

2https://pacoapp.com/

3https://ilumivu.com/

4https://www.trackyourtinnitus.org/

R. Kraft et al. /Procedia Computer Science 00 (2019) 000–000 3

Questionnaire

Group

User Group

Data Set

Table Entries Unique Users

A 4929 4929

B 77488 2632

C 9262 3193

D 143 84

Answer Sheet

IS-A

EMA

Questionnaire

Answer Sheet B

Questionnaire

Answer Sheet

User Metadata

Therapy

User

Fig. 1: Entity-relationship model (ERM) of the TYT database. Tables A–D are considered in detail in the scope of this work. The respective number

of entries in the database and the number of unique users who submitted these data are shown in the table to the right of the figure.

As explained in Section 1, TYT utilizes MCS and EMA to collect data on the user’s individual tinnitus. To be

more precise, standardized questionnaires (see Section 3.1) are used as a data collection instrument. The answers,

in turn, are stored in Answer Sheets. The latter can be divided into an answer sheet for (1) a fixed questionnaire

related to the momentary assessment and (2) a dynamically composable questionnaire. The first type is denoted as

EMA Questionnaire Answer Sheet (see Fig. 1,B

). It consists of the user’s answers, which enable the tracking of the

individual tinnitus, an environmental sound level measurement, information about the triggered system notification

on the device, and the current user agent. Note that the EMA questionnaire is a predefined questionnaire that has not

been modified since the start of the TYT platform in order to ensure comparability of the answers. Since the EMA

questionnaire is triggered on a daily basis, it holds the largest share of the answers sheet with 77488 stored entries

of 2632 users. The second type of answers sheets, denoted as Questionnaire Answer Sheet (see Fig. 1,C

), holds the

answers of dynamically composable questionnaires. To be more precise, the latter include all other questionnaires

that, for example, gather demographic user data. Those questionnaires can be grouped into complex data collection

schemes if needed. Currently, the TYT data set consists of 9262 Questionnaire Answer Sheets of 3193 unique users.

Additionally, TYT users may store information about their on-going treatments and therapies. The Therapy table

(see Fig. 1,D

) contains information about the therapy type (e.g., dental treatment, sport, hearing aid, etc.), the current

medication of the patient, personal notes, and the start as well as end dates of the treatment. Note that this feature is

only available on the registration and information website, which is reflected in the small number of therapies in the

data set (n=143).

3. Potential insights into the TrackYourTinnitus database

Based on the extracted data set described in Section 2, different aspects of the TYT platform are investigated and

explored for potential insights in the following.

3.1. Socio-demographics and sub-populations

In order to assess the socio-demographic data of the user population of TYT, the data collected with the three

initial questionnaires when users start the mobile application for the first time was evaluated. The Mini Tinnitus Ques-

tionnaire (n=3179) assesses the initial distress of users, whereas the Tinnitus Sample Case History Questionnaire

(TSCHQ) (n=3009) corresponds to a standardized questionnaire for tinnitus case history [5], including several ques-

tions on the socio-demographic background of patients as well as their tinnitus history. Finally, the Worst Symptom

questionnaire (n=3072) asks the users to select their subjectively worst symptom from a list of symptoms (or to

indicate that none of the symptoms apply to them).

The answers for the Mini Tinnitus Questionnaire have a mean sum score of 13.64 (s=6.11). The correlation

between the individual sum scores and the answers of the EMA questionnaires can then be investigated (e.g., the

perceived tinnitus loudness or distress). Furthermore, socio-demopraphic data assessed with the TSCHQ are shown in

Table 1. For different questions, it can be examined if there are differences between the sub-populations. For instance,

it can be investigated whether there are differences in the perception of tinnitus between female and male users. Finally,

the distribution of the subjectively worst symptom of each user is shown in Fig. 2a. It can be investigated whether

4R. Kraft et al. /Procedia Computer Science 00 (2019) 000–000

there is a correlation between this worst symptom and the course of the tinnitus assessed with the EMA questionnaire.

Additionally, as the user is asked in the EMA questionnaire whether he perceives this worst symptom at that very

moment, it can be investigated how the various symptoms evolve over time.

Characteristic Value Frequency (n=3009) %

Gender Male 2032 67.96

Female 958 32.04

Handedness Right 2176 72.7

Left 364 12.16

Both Sides 453 15.14

Familiy history of tinnitus complaints Yes 720 24.11

No 2266 75.89

Table 1: Socio-demographic data assessed with the Tinnitus Sample Case History Questionnaire (TSCHQ)

Another aspect worth considering is the number of submissions for the EMA questionnaire. As can be seen in

Fig. 2b, 2632 of the 4929 users (53.4%) submitted at least one answer sheet, but only 30.85% of these users have

10 or more submissions (812), with only a few users having 100 or more (163), 500 or more (26), or even 1000 or

more (8) submissions. Note that the latter group of 8 users submitted 14.37% of the total EMA answer sheets in the

database, i.e., a few users make up a large part of the total data set. For these users, so-called n-of-1 studies [2] could

be conducted to gain in-depth insights on the course of their tinnitus. Furthermore, it can be investigated whether

there are differences in the course of the tinnitus between these sub-populations of users. Finally, as adherence and

incentives constitute critical topics in EMA and MCS [13,1,3], it could be investigated what the difference between

sub-populations is, also with regard to other aspects discussed in this paper, and ultimately, what might motivate users

to submit more data over a longer period of time.

0100 200 300 400 500 600

Harder to relax.

Difficult to follow conversation/music/film.

Hard to sleep.

Difficult to concentrate.

Strong worries.

Depression.

None.

More sensitive to environmental noise.

More irritable with family/friends/colleagues.

Number of users

(a) Number of users by their reported worst symptom (n=3072).

500

1000

1500

2000

2500

3000

At le ast one ≥ 10 ≥ 1 00 ≥ 50 0 ≥ 10 0 0

Number of users

Number of submissions

(b) Number of users with different numbers of EMA submissions.

Users with no submissions are omitted in the figure.

Fig. 2: Number of users by (a) their worst symptom and (b) their number of EMA submissions.

3.2. Therapies

Another secondary data set that has yet to be evaluated are Therapies. A feature on the website allows users to

document treatments that might influence their tinnitus in any way. For each treatment, a therapy type (e.g., medical

treatment, physical therapy, auditory stimulation), name, time period (i.e., start and end date (optional)) and a personal

note can be entered. As this feature was primarily implemented for the user himself, this data is currently not inter-

preted in any way. The data set contains 143 therapy entries from 84 different users. The ten most frequent therapy

R. Kraft et al. /Procedia Computer Science 00 (2019) 000–000 5

types are shown in Fig. 3. The mean therapy duration is 139.17 (s=335.3). It can be investigated whether there is a

correlation between the course of the tinnitus and a specific accompanying therapy or type of therapies.

0 5 10 15 20 25 30 35 40

Medical treatment

Tinnitus ma sker

Hearing aid

Auditory stimulation

Psychotherapy

Biofeedback

Tinnitus Retraining Therapy (TRT)

Sports

Dental treatment

Physical therapy

other

Number of therapies

Fig. 3: The ten most frequent therapy types. The remaining therapy types are summarized with ’other’ (n=143).

3.3. Countries and regions

For each user, the mobile application stores the country extracted from the device locale on registration. Users that

different countries. The distribution of the ten countries with the most users are shown in Fig. 4. It can be seen that the

largest number of users is from Germany, followed by the United States, the Netherlands and Great Britain. This can

be explained by the fact that TYT was initially launched in Germany and has been available in German and English.

Since 2017, the mobile application has been available in Dutch as well. It can now be investigated whether there are

differences in the perception of tinnitus between countries and regions.

200

400

600

800

1000

1200

1400

1600

1800

DE US NL GB CH CA ES IT FR AU other

Number of users

Fig. 4: Number of users by country code. The ten countries with the most users are displayed, users from other countries are grouped under ’other’

(n=4929).

To this end, for each EMA answer sheet, the country of the respective users who submitted the sheet was recorded.

As only users who submitted at least one answer sheet are considered, this results in 101 different countries (of the

initial 193). The distribution of countries for answer sheets is shown in Fig. 5a. The countries were then grouped

by their continent, as shown in Fig. 5b. For the resulting sub-populations, it can be investigated whether there are

6R. Kraft et al. /Procedia Computer Science 00 (2019) 000–000

differences in the course of the tinnitus between countries and continents respectively. Similarly, other implications,

e.g., for the hemispheres (northern vs. southern, western vs. eastern), could be investigated.

5000

10000

15000

20000

25000

30000

35000

40000

other

Number of answer sheets

(a) Countries

10000

20000

30000

40000

50000

60000

70000

Europe

North America

Asia

South America

Oceania

Africa

Antarctica

Number of answer sheets

(b) Continents

Fig. 5: Number of answer sheets by (a) country and (b) continents (n=77488). For (a), the ten countries with the most answer sheets are displayed,

answer sheets from other countries are grouped under ’other’.

3.4. Smart mobile devices

Another meta data that is automatically extracted and stored for each user is the mobile user agent, which is used

for (a) the submission of the initial socio-demographic questionnaires (cf. Section 3.1) and (b) each time an EMA

questionnaire is submitted. For the initial questionnaires, 27.67% (1364) of the users use an iOS device, whereas

21.57% (1063) use an Android device. For the remaining 50.76% (2502) of users, there is no information on the

mobile operating system, indicating that they did not complete the initial questionnaires (1736) or submitted the

questionnaires before introducing this feature (766). As shown in Fig. 6a, 40.78% of EMA answer sheets are submitted

by an iOS device, whereas 59.22% are submitted by an Android device. It can be investigated whether there is a

difference in the perception and course of tinnitus between users of these operation systems. In addition, for Android,

the exact device model is stored. The ten most used Android device models are shown in Fig. 6b. Note that 32 users

have changed their device one or more times during the use of the TYT mobile application and are therefore included

multiple times in this consideration. For the latter users, it could be further investigated whether the device model

directly influences their EMA data.

3.5. Notifications

Notifications are a crucial part of the EMA data collection methodology, since they remind users to fill out the

questionnaires and, therefore, help to establish a continuous sampling rate. In order to correlate the triggered system

notifications with the questionnaire filled out, the TYT mobile application stores meta data about the notification (i.e.,

timestamp of the trigger event), meta data about the answer sheet (i.e., timestamp of the moment of storage), and the

user’s current notification scheme. In TYT, the latter is either a schedule of fixed points in time (i.e., fixed) or the

result of an algorithm that produces random points in time (i.e., random). The notification scheme can be changed

by the user at any time [9]. Currently, the TYT data sets consists of 77488 EMA Questionnaire Answer Sheets. With

respect to the notification settings, 60100 answer sheets (77.5%) are stored with a random notification setup and

17388 answer sheets (22.5%) with a fixed scheme of points in time. In addition to the number of answer sheets, Fig. 7

illustrates the distribution of answer sheets that can be attributed to triggered notifications and the non-attributable

answer sheets. With less than 25%, both of the notification schemes show a rather low amount of attributable answer

R. Kraft et al. /Procedia Computer Science 00 (2019) 000–000 7

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

iOS Android

Number of answer sheets

(a) Operating systems

Samsung Galaxy S7

Samsung Galaxy A5 (2017)

Samsung Galaxy S5

Google Nexus 5

Samsung Galaxy S7 Edge

Samsung Galaxy S8

Samsung Galaxy S9

Samsung Galaxy S5 mini

Google Pixel

Samsung Galaxy A5

Number of users

(b) Android device models

Fig. 6: (a) Number of answer sheets by operating systems and (b) number of users by the ten most used Android device models (n=77488).

sheets. Interestingly, the amount of attributable answer sheets is 10% higher for a fixed notification scheme compared

to a random notification scheme. The effect of the notification scheme for the notification adherence could be subject

of further investigations.

Fig. 7: Numbers of answer sheets that are associated with a fixed or random notification

4. Discussion

Based on the aspects and sub-populations considered in this work, more advanced analyses and evaluations could

be conducted in order to gain in-depth insights on TYT, its users and, ultimately, the tinnitus disorder. For instance,

machine learning (ML) can be applied to the large data set of different features for both supervised and unsupervised

learning (e.g., [7]). Furthermore, advanced statistical analyses, e.g., multilevel analyses on the hierarchical EMA

data or n-of-1 studies on the large data sets for some of the users (cf. Section 3.1), can be applied to the data set. In

addition, as the same (or similar) structure of data can be found in other related projects like TrackYourHearing (TYH),

TrackYourDiabetes (TYD),TrackYourStress (TYS) or TinnitusTipps [3], the aforementioned analyses can be applied to

these data sets equivalently.

There are several peculiarities and technical considerations about what data is stored and when it is collected. For

instance, when recording the user agent (i.e., the operating system and smart mobile device models) of users, one

should take into account that users might change their device in the course of using the application. For this purpose,

8R. Kraft et al. /Procedia Computer Science 00 (2019) 000–000

it is good practice to store the user agent during registration and each time data is collected thereafter. Furthermore,

the TYT platform and its data structure has several limitations. For example, in case of notifications (cf. Section 3.5),

the assessment of the notification adherence (i.e., how often users answer the EMA questionnaire after receiving a

notification) is aggravated by the fact that notification schedules are only stored locally on the devices and, therefore,

no information on how many notifications the user has received in total is available during the analysis. Furthermore,

it is noteworthy that notifications are only optional reminders and users are able to switch or close the TYT application

at any point in time. In both cases, the answer sheet can no longer be attributed to a notification with the consequence

that the data set does not contain any notification meta data.

5. Summary & outlook

In this work, we presented the underlying data set and its structure of the TrackYourTinnitus (TYT) platform. We

then provided descriptive statistics on various aspects of the data set and highlighted potential insights that could be

gained from this data. Finally, we discussed in-depth analyses and evaluations that could be conducted on the data set

based on the highlighted aspects as well as technical issues and limitations of this and other similar data sets.

When running an MCS and EMA project like TYT, increased insights can be gained if additionally, socio-

demographic and meta data about the users of the platform is stored. This allows identifying various sub-populations

and investigate correlations between characteristics of users and their EMA data in order to gain a better understand-

ing of the assessed phenomenon (e.g., a disorder like tinnitus). As an outlook, we intend to conduct further analyses

as discussed in this work. For instance, we are currently investigating the environmental sound level measured with

the mobile application and how it interacts with other characteristics described in this paper. We hope that providing

a comprehensive description of the data set to the community allows to better understand and reproduce the results

from our research that are based on the TYT data set or similar projects. Finally, we believe that the insights gained in

the scope of this work can serve as a basis for other and future projects and facilitate the comparability between their

results.

References

[1] Agrawal, K., Mehdi, M., Reichert, M., Hauck, F., Schlee, W., Probst, T., Pryss, R., 2018. Towards incentive management mechanisms in the

context of crowdsensing technologies based on trackyourtinnitus insights. Procedia computer science 134, 145–152.

[2] Duan, N., Kravitz, R.L., Schmid, C.H., 2013. Single-patient (n-of-1) trials: a pragmatic clinical decision methodology for patient-centered

comparative effectiveness research. Journal of clinical epidemiology 66, S21–S28.

[3] Kraft, R., Schlee, W., Stach, M., Reichert, M., Langguth, B., Baumeister, H., Probst, T., Hannemann, R., Pryss, R., 2020. Combining mobile

crowdsensing and ecological momentary assessments in the healthcare domain. Frontiers in Neuroscience 14, 164.

[4] Kubiak, T., Smyth, J.M., 2019. Connecting domains—ecological momentary assessment in a mobile sensing framework, in: Digital Pheno-

typing and Mobile Sensing. Springer, pp. 201–207.

[5] Langguth, B., Goodey, R., Azevedo, A., Bjorne, A., Cacace, A., Crocetti, A., Del Bo, L., De Ridder, D., Diges, I., Elbert, T., et al., 2007.

Consensus for tinnitus patient assessment and treatment outcome measurement: Tinnitus research initiative meeting, regensburg, july 2006.

Progress in brain research 166, 525–536.

[6] Pryss, R., 2019. Mobile crowdsensing in healthcare scenarios: Taxonomy, conceptual pillars, smart mobile crowdsensing services, in: Digital

Phenotyping and Mobile Sensing. Springer, pp. 221–234.

[7] Pryss, R., John, D., Reichert, M., Hoppenstedt, B., Schmid, L., Schlee, W., Spiliopoulou, M., Schobel, J., Kraft, R., Schickler, M., et al.,

2019. Machine learning findings on geospatial data of users from the trackyourstress mhealth crowdsensing platform, in: 2019 IEEE 20th

International Conference on Information Reuse and Integration for Data Science (IRI), IEEE. pp. 350–355.

[8] Pryss, R., Reichert, M., Herrmann, J., Langguth, B., Schlee, W., 2015a. Mobile crowd sensing in clinical and psychological trials–a case study,

in: 2015 IEEE 28th International Symposium on Computer-Based Medical Systems, IEEE. pp. 23–24.

[9] Pryss, R., Reichert, M., Langguth, B., Schlee, W., 2015b. Mobile Crowd Sensing Services for Tinnitus Assessment, Therapy and Research, in:

IEEE 4th International Conference on Mobile Services (MS 2015), IEEE Computer Society Press. pp. 352–359.

[10] Schlee, W., Pryss, R.C., Probst, T., Schobel, J., Bachmeier, A., Reichert, M., Langguth, B., 2016. Measuring the moment-to-moment variability

of tinnitus: the trackyourtinnitus smart phone app. Frontiers in aging neuroscience 8, 294.

[11] Stone, A.A., Shiffman, S., 1994. Ecological momentary assessment (ema) in behavorial medicine. Annals of Behavioral Medicine .

[12] Tokosi, T.O., Scholtz, B.M., 2019. A classification framework of mobile health crowdsensing research: A scoping review, in: Proceedings of

the South African Institute of Computer Scientists and Information Technologists 2019, ACM. p. 4.

[13] Zhang, X., Yang, Z., Sun, W., Liu, Y., Tang, S., Xing, K., Mao, X., 2015. Incentives for mobile crowd sensing: A survey. IEEE Communications

Surveys & Tutorials 18, 54–67.