Prospective crowdsensing versus retrospective ratings of tinnitus variability and tinnitus–stress associations based on the TrackYourTinnitus mobile platform [original]

International Journal of Data Science and Analytics (2019) 8:327–338

https://doi.org/10.1007/s41060-018-0111-4

REGULAR PAPER

Prospective crowdsensing versus retrospective ratings of tinnitus

variability and tinnitus–stress associations based on the

TrackYourTinnitus mobile platform

Rüdiger Pryss1·Thomas Probst2·Winfried Schlee3·Johannes Schobel1·Berthold Langguth3·

Patrick Neﬀ4·Myra Spiliopoulou5·Manfred Reichert1

Received: 28 September 2017 / Accepted: 24 February 2018 / Published online: 12 March 2018

Abstract

Many symptoms of neuropsychiatric disorders, such as tinnitus, are subjective and vary over time. Usually, in interviews

or self-report questionnaires, patients are asked to retrospectively report symptoms as well as their severity, duration and

influencing factors. However, only little is known to what degree such retrospective reports reflect the actual experiences

made in daily life. Mobile technologies can remedy this deficiency. In particular, mobile self-help services allow patients

to prospectively record symptoms and their severity at the time (or shortly after) they occur in daily life. In this study, we

present results we obtained with the mobile crowdsensing platform TrackYourTinnitus. In particular, we show that there is a

discrepancy between prospective and retrospective assessments. To be more precise, we show that the prospective variation of

tinnitus loudness does not differ between the users who retrospectively rate tinnitus loudness as “varying” and the ones who

retrospectively rate it as “non-varying.” As another result, the subjectively reported stress-level was positively correlated with

tinnitus (loudness and distress) in the prospective assessments, even for users who retrospectively rated that stress reduces

their tinnitus or has no effect on it. The results indicate that mobile technologies, like the TrackYourTinnitus crowdsensing

platform, go beyond the role of an assistive service for patients by contributing to more detailed information about symptom

variability over time and, hence, to more elaborated diagnostics and treatments.

Keywords Prospective assessment of neuropsychiatric symptoms ·Retrospective assessment of neuropsychiatric symptoms ·

Self-help mobile applications ·Tinnitus ·Mobile data collection ·Mobile crowdsensing

BRüdiger Pryss

ruediger[email protected]

Thomas Probst

Winfried Schlee

Johannes Schobel

Berthold Langguth

Patrick Neff

patrick.nef[email protected]

Myra Spiliopoulou

[email protected]urg.de

Manfred Reichert

[email protected]

1 Introduction

The assessment of neuropsychiatric symptoms is essential in

psychology, medicine and neuroscience. For many neuropsy-

1Ulm University, James-Franck-Ring, 89081 Ulm, Germany

2Department for Psychotherapy and Biopsychosocial Health,

Danube University Krems, Dr.-Karl-Dorrek-Straße 30, 3500

Krems, Austria

3Clinic and Policlinic for Psychiatry and Psychotherapy,

University of Regensburg, Universitätsstraße 84, 93053

Regensburg, Germany

4Neuroplasticity and Learning in the Healthy Aging Brain,

University of Zurich, Andreasstrasse 15, 8050 Zurich,

Switzerland

5Department of Technical and Business Information Systems,

Otto-von-Guericke-University Magdeburg, Universitätsplatz

2, 39106 Magdeburg, Germany

123

328 International Journal of Data Science and Analytics (2019) 8:327–338

chiatric disorders, the severity and duration of symptoms

constitute essential criteria for diagnosis. For example, a

major depressive episode can only be diagnosed if the

patients have suffered from depressive symptoms for at least

2 weeks. Moreover, factors making the symptoms more or

less strong (i.e., correlates of the symptoms) need to be

identified for case conceptualization and treatment planning.

Hence, psychologists, physicians and researchers need to

reliably assess not only symptoms and their severity, but also

their fluctuations over time. However, most indicators of the

symptoms as well as their severity, duration and correlates

are subjective. In current practice, usually, their assessment

is based on retrospective reports of the patients. In turn, this

raises the question to what degree patients are able to remem-

ber the severity, duration and correlates of the symptoms they

have actually experienced.

Mobile technologies can effectively contribute to shed

light on this question. In particular, they allow complement-

ing the retrospective reports of the patients with prospective

assessments of symptom variation over time. In this arti-

cle, we describe how the mobile crowdsensing platform

TrackYourTinnitus [31,32,36–38,42], which we developed

during the last years, contributes to prospectively monitor

symptom variability over time for individuals with tinni-

tus. Tinnitus can be described as the phantom perception

of sound. Depending on its definition and duration as well

as on the patient age and birth cohort, between 5.1 and

42.7% of the population worldwide experience tinnitus at

least once during their lifetime [26]. On the one hand, tinni-

tus varies among patients (i.e., inter-individual variability);

on the other hand, it may vary for a particular patient (i.e.,

intra-individual variability) as well. Moreover, the diagnosis

and treatment of tinnitus and potential comorbidities require

assessments of several symptoms like loudness and varia-

tion of the perceived sound(s), stress-level, depressive and

anxiety symptoms as well as concentration. In this work,

we aim to compare findings from prospective and retrospec-

tive assessments of tinnitus symptoms (loudness and distress)

as well as the potential influencing factor/correlate “stress-

level,” which is often reported in the context of tinnitus

[24].

The limited (ecological) validity of retrospective self-

reports has been shown in several studies on other neu-

ropsychiatric disorders. For example, [1] assessed physical

activities of patients with eating disorders by retrospec-

tive self-reports as well as prospective assessments with an

accelerometer. Patients reported significantly less physical

activity retrospectively compared to the prospective mea-

surements with the accelerometer. In turn, [23] investigated

retrospectively as well as prospectively assessed anxiety and

related cognition in patients with agoraphobia. While anx-

iety did not differ between retrospective and prospective

assessments, cognition did.

Suchresultshighlighttheimportanceofecological momen-

tary assessment (EMA; also known as ambulatory assess-

ment & experience sampling) to support clinicians in assess-

ing neuropsychiatric symptoms as well as their correlates

accurately in real time. In EMA, the variable in question

(e.g., symptoms) is assessed repeatedly in daily life [46].

Instead of retrospectively asking the individuals, through an

interview or questionnaire, how strongly they experienced a

symptom in a given past time interval, the individuals are

asked how they currently experience it as well as its sever-

ity and potential correlates. In turn, this is accomplished at

several time points within the given time interval.

In the aforementioned studies, prospective and retrospec-

tive assessments were juxtaposed manually. To effectively

exploit the prospective assessments in a clinical setting, how-

ever, an integrated solution is needed, i.e., the EMA of a

patient should be transferred automatically to a database and

be made available to the responsible clinician(s) [34,43],

given the consent of the patient. Note that it has been already

reported for a long time that electronic systems are appreci-

ated by study participants [17], increase data accuracy [29],

lead to more complete datasets [22] and reduce costs [30]

compared to traditional paper-based methods. However, the

exploitation of the prospective assessments next to the ret-

rospective reports has not been investigated in the area of

tinnitus yet.

This paper presents the TrackYourTinnitus (TYT) mobile

crowdsensing platform [37,38] for the juxtaposition of retro-

spective and prospective assessments, with a focus on tinnitus

loudness, tinnitus distress and psychological stress. We elab-

orate how the prospective data are collected and how they

should be maintained for further usage. The paper provides

a significant extension of the work we presented in [35]. In

particular, [35] did not include the analysis of tinnitus dis-

tress and psychological stress. As the latter is associated with

several disorders (in general [3]; for tinnitus [24]), this anal-

ysis constitutes another fundamental comparison between

real-time assessments and retrospective reports. We provide

detailed backgrounds on the gathered and evaluated data as

well as the results for psychological stress. This additional

analysis reconfirms that mobile crowdsensing services will

become increasingly important for collecting large and eco-

logically valid longitudinal datasets in the context of clinical

research.

The remainder of the paper is organized as follows.

Related work is discussed in Sect. 2. In Sect. 3, we describe

the TYT mobile crowdsensing platform and explain the

workflow we implemented for collecting and maintaining

prospective assessments. Section 4presents the data as well

as the statistics used for juxtaposing the prospective with

the retrospective assessments. In Sect. 5, we present the

results of the statistical analyses, which suggest that integrat-

ing prospective assessments into the diagnostic–therapeutic

123

International Journal of Data Science and Analytics (2019) 8:327–338 329

processis crucial for optimizing diagnostics and, hence, treat-

ments. The paper concludes with a summary and outlook in

Sect. 6.

2 Related work

Mobile crowdsensing is an emerging research topic in various

application domains [19,20,39,45]. In the medical domain,

however, this research direction has been neglected so far.

The fact that the medical domain is less considered by con-

temporary crowdsensing approaches might be explainable

by legal and data privacy issues [2]. Nevertheless, mobile

crowdsensing offers promising perspectives for the medical

domain [8], as it exhibits unique features for gathering valu-

able patient data in the large scale [5]. In particular, mobile

crowdsensing allows for the effective, context-aware gather-

ing [21] of daily life patient data [33], which, in turn, will

shift clinical research to a new level.

Besides TrackYourTinnitus (TYT), other studies have

applied EMA approaches to track tinnitus in daily life

[9,11,27,48]. Yet, their focus was not to compare retrospec-

tive ratings and prospectively crowdsensed data. [48] and

[11] were pilot studies that showed for example that tinni-

tus tracking is feasible without negative consequences for

the participants and that tinnitus varies within and between

participants. [9] investigated fluctuations of tinnitus as well

as associations between tinnitus and stress. [27] tracked tin-

nitus and other symptoms (e.g., dizziness) in patients with

Meniere’s syndrome. Contrary to these approaches, TYT is

an open-source application, available for download in the

iOS app store or Google play store, so that there is a larger

and more representative sample compared to the participants

of the cited studies.

Beyond tinnitus, EMA approaches capturing many other

aspects such as pain [14] or feelings [15] in daily life were

scientifically evaluated. In addition, EMA approaches were

studied in the areas of mood disorders and mood dysregula-

tion [40,47] as well as in the context of substance use [44]

and eating disorders [7]. In psychotherapy research, EMA

has been used to predict patient progress [13]. Although most

neuropsychiatric symptoms are subjective experiences and,

thus, most EMA approaches use self-reports to capture these

symptoms, some neuropsychiatric symptoms are behavioral

(e.g., avoidance in anxiety disorders) or physiological (e.g.,

increase of heart rate in anxiety disorders). Note that mobile

systems offer opportunities to measure behavioral or physi-

ological data in daily life [6].

Altogether, EMA approaches provide unprecedented

opportunities to study neuropsychiatric symptoms under eco-

logically valid conditions [28], even though the utilization of

its possibilities is still in its infancy, especially in the medical

domain.

3 The TrackYourTinnitus platform

TrackYourTinnitus (TYT) is a mobile crowdsensing platform

that comprises a Web site for user registration, two mobile

applications (for iOS and Android) and a relational database

(MySQL) as the central repository storing the collected data

[37]. In particular, the anonymized or pseudonymized TYT

data from the repository are made available to clinicians as

well as researchers. The Web site further provides two funda-

mental features: First, users can visualize recorded tinnitus

data; second, they can report on their current tinnitus treat-

ment. In general, TYT was developed to track the individual

tinnitus perception of users. In this context, the procedure

depicted in Fig. 1is applied to the TYT users.1

Following this procedure, TYT pursues three goals. First,

data shall be collected on a daily basis (cf. Fig. 1,3

). How-

ever, a crowd user shall not foresee the times he or she is

asked to sense data (cf. Fig. 1,2

). This is ensured by asking

the crowd users in various daily life situations. Second, the

collected data shall enable new kinds of data analytics like

juxtaposing real-time assessments and retrospective reports

(cf. Fig. 1,1

). Third, gathered data shall be used to provide

feedback to the mobile crowd users.

To enable the use of TYT as well as to provide data being

appropriate for the data analysis applied in the context of this

paper, the following procedure has to be accomplished by the

users (cf. Fig. 1).

First, users have to create a TYT account, by using either the

TYT Web site or the TYT mobile applications.

Second, users have to fill in three registration questionnaires

(cf. Fig. 5). First of all, they have to fill in the “Mini-TQ-12”

questionnaire (cf. Fig. 5, Mini-TQ-12 [12]), which measures

tinnitus-related psychological problems. Second, they have

to fill in the “Tinnitus Sample Case History Questionnaire”

(TSCHQ) (cf. Fig. 5,TSCHQ[18]), in which details about

the current tinnitus status, relevant co-morbidities and the

tinnitus history are assessed. Note that TSCHQ comprises

two questions being crucial for the results of this paper. To

be more precise, the 11th item of TSCHQ asks the user

retrospectively whether or not the tinnitus loudness varies

from day to day. The 26th item of TSCHQ, in turn, requests

from users to rate retrospectively whether stress is associ-

ated with their tinnitus. Third, users have to fill in the “Worst

Symptom” questionnaire (cf. Fig. 5, Worst Symptom ques-

tionnaire), which asks the users about the worst symptom

currently caused by their tinnitus. While the first two ques-

tionnaires constitute already used instruments, the third one

was newly developed in the context of the presented research.

Altogether, the completion of the three questionnaires with

1More detailed information about the procedure can be found at https://

www.trackyourtinnitus.org/process.pdf

123

330 International Journal of Data Science and Analytics (2019) 8:327–338

Fig. 1 Mobile crowdsensing collection procedure

Fig. 2 Impression of assessment questionnaire in iOS

their 58 questions in total is a fundamental prerequisite for

users who want to access the TYT Web site features as well

as the TYT mobile applications (Fig. 2).

Third, after registering and completing the required ques-

tionnaires, users may exploit the mobile applications to

track their tinnitus and potential correlates during daily life.

For this purpose, a user needs to log in to the Android

or iOS mobile application. Then, he/she is asked to fill

in the assessment questionnaire developed for TYT (cf.

Fig. 5, assessment questionnaire). This repeatedly admin-

istered assessment questionnaire comprises 8 items (cf.

Table 1) including questions on the current tinnitus loudness,

tinnitus distress and subjective stress-level. Figure 3gives an

impression of how the questionnaire looks like in iOS.

Fourth, the assessment questionnaire is provided in two

ways. Either the mobile application automatically displays

the questionnaire to the user or the user himself makes the

conscious decision to fill out the questionnaire (cf. Fig. 5,

conscious decision). The first procedure is the preferred one

and is realized as follows: The assessment questionnaire is

randomly presented to the user up to 12 times per day. Fig-

ure 3gives an impression of the notification settings in iOS.

Table 1 TrackYourTinnitus

assessment questions Question Scale M

Did you perceive the tinnitus right now? BS Perception

How loud is the tinnitus right now? VAS Loudness

How stressful is the tinnitus right now? VAS Strain

How is your mood right now? VAS Mood

How is your arousal right now? VAS Arousal

Do you feel stressed right now? VAS Stress

How much did you concentrate on the

things you are doing right now?

VAS Con.

Do you feel irritable right now? BS Irritability

BS binary scale, VAS visual analogue scale, Mmeasurement of, Con. concentration

123

International Journal of Data Science and Analytics (2019) 8:327–338 331

Fig. 3 Impression of notification settings in iOS

For the application of the assessment questionnaire, noti-

fication features for both Android and iOS as well as a

notification algorithm were realized [37]. We only present

the algorithm running on iOS (cf. Algorithm 1) and the cal-

culated notifications for a single day. In practice, notifications

are calculated in advance. The algorithm, in turn, works as

follows:

1. It partitions the time window a user has specified with

respect to a particular day into ntime intervals of equal

length. ncorresponds to the number of notifications the

user has chosen.

2. The algorithm then calculates exactly one notification for

each interval. Thereby, it ensures that for each notification

the points in time are randomly calculated.

3. Finally, it is ensured that there is an interval of at least

15 min between two notifications.

On the one hand, the procedure ensures that users cannot

foresee the time when being asked; on the other hand, it

ensures that they are sensed in various daily situations. Note

that this randomized approach was realized to improve the

ecological validity of the applied method. The approach to

randomly apply the assessment questionnaire is illustrated in

2arc4random_uniform(upper_bound): iOS internal function to return

a uniformly distributed random number less than upper_bound.

Algorithm 1: iOS algorithm for daily notifications of a user

Data:

timeInterval: time interval a user has specified for a day

numberOf NotificationsPerDay: notifications specified for a day

Result:

scheduleLocalNotification: calculated random notifications for a day

1begin

2lengthOfIntervall = timeInterval/numberOfNotificationsPerDay;

3lastNotification = 900; /* the 15 minutes */

4foreach n∈numberOf Notif icationsPerDay do

5secondsSinceStartOfInterval =

arc4random_uniform2(lengthOfIntervall);

6absoluteInterval=

7secondsSinceStartOfInterval+(n*lengthOfIntervall);

/* check the 15 minutes */

8if absoluteInterval −lastNoti f ication <900 then

9absoluteInterval = 2*absoluteInterval - lastNotification;

10 end

11 lastNotification = absoluteInterval;

/* check if notification is in

absoluteInterval */

12 if absoluteInterval <timeInterval then

/* notification found */

13 scheduleLocalNotification =

scheduleLocalNotification ∪absoluteInterval;

14 end

15 end

16 end

Fig. 4 Possible user action after a notification

Fig. 4. It works on both mobile operating systems in exactly

the same ways.

After a notification appears, the user may click on it. In the

latter case, the TYT mobile app is started (if not already run-

ning) and the assessment questionnaire is directly displayed

to the user. Then, he or she can fill out the questionnaire and

finally save the entered data. After saving the questionnaire

data, the mobile app is terminated 3 s later. Within these 3 s

the result is transferred to the TYT backend (if the mobile

app is online; otherwise, the result is locally stored until the

device gets an online connection).

The procedure to automatically terminate the app shall

speed up the process to fill out questionnaires after being

notified. Note that the user feedback we have received so far

supports this technical procedure. If the user does not save

123

332 International Journal of Data Science and Analytics (2019) 8:327–338

Fig. 5 TrackYourTinnitus crowdsensing platform

the questionnaire data or switches to the main menu before

saving it, the questionnaire will be not saved and transferred

to the TYT backend.

Finally, anotheraspectrelatedtonotifications is important.

As can be seen from Fig. 4, the user might click on a current

notification or previous ones. The latter situation might occur

if the user did not click on a previous notification. Each time

a notification is not clicked, the TYT mobile app memorizes

and stores it in the local message store. If a user clicks on

the local message store, he or she may see these previous

notifications. Note that the allowed amount of previous noti-

fications messages differs between the two mobile operating

systems. As we store separate timestamps for the notifica-

tion as well as the saving time of each processed assessment

questionnaire, users can also fill out questionnaires related

to previous notifications.

Fifth, while filling in the assessment questionnaire, the smart

mobile device of a user records the environmental sound level

if this option has been activated.

Sixth, results gathered with the assessment questionnaire and

the sound-recording component are transferred to the TYT

database. The latter, in turn, provides comprehensive features

that enable researchers to evaluate the gathered user data.

Note that we exactly used these features to obtain the results

presented in Sect. 5.

The TYT crowdsensing platform provides multilingual

support. Currently, English, German and Dutch are sup-

ported. The current number (September 2017) of processed

questionnaires and registered users is shown in Fig. 5.As

can be seen, the TYT mobile crowdsensing platform is fre-

quently used. So far, users from 80 countries have entered

data to the platform. It is noteworthy that users from seven

countries provide the major part of all data gathered. To be

more precise, most of the collected data is provided by users

from Germany, followed by the USA, Netherlands, the UK,

Switzerland, Canada and France.

In the following, technical insights into the TYT plat-

form, which are relevant in the context of this work, will be

sketched. Concerning domain-specific requirements (from

the medical and psychological domains), the TYT platform

stores user data in an anonymized or pseudonymized man-

ner. Only if users want to use the password reset function,

they have to provide personal data (i.e., the e-mail address),

which are saved separately from the app data. However, users

are informed about the fact that the e-mail address must be

stored for resetting a password.

To ensure privacy, TYT presents a consent form to users.

Only if users confirm the consent form, the respective func-

tion is used by TYT. For example, if access to mobile device

features is required (e.g., to record the sound level by the

mobile device microphone), users must explicitly confirm

123

International Journal of Data Science and Analytics (2019) 8:327–338 333

this through a consent form. Another feature provided by

TYT is the participation in studies. Users may enter or leave

studies. Each study, in turn, may rely on questionnaires, con-

sent forms and notifications settings. Results of processed

questionnaires are always locally stored as JSON-file if there

is no Internet connection. If the mobile device has an Inter-

net connection, results are transferred as JSON-files through

a REST-API to the TYT backend. The latter stores them in a

relational database.

For researchers, a backend function is provided that

enables them to export the data from the relational database

to CSV files. The export feature, in turn, particularly consid-

ers that exported data are processed by different statistical

software systems. For example, all the data of a user can

be stored in one row with multiple columns or data can be

stored in a way so that one assessment point represents one

row. Recently, TYT was enhanced with data from wearables

(e.g., measuring physiological data like the heart rate). We

integrated the iOS smart watch and fitness trackers offered by

GARMIN and MIO to the TYT platform. Respective track-

ers can be used to track the heart rate while filling out the

assessment questionnaire. First results related to these heart

rate measurements are promising. In addition, we provide a

feature to determine the GPS position while filling out the

assessment questionnaire.

In the meantime, the basic design principles, techniques

and components of TYT were adopted in other contexts as

well. For example, in the myKind project [39], risk factors

during pregnancy are tracked and evaluated based on the TYT

technology. Another project using the TYT technology is

TrackYourHearing2, which aims to gain new insights into the

moment-to-moment variability of the hearing loss of users.

Altogether, TYT constitutes a powerful, generic frame-

work for measuring the moment-to-moment variability of

user aspects in everyday life.

4 Data and statistics

The analysis presented in this section is based on an export

of the TYT database made in August 2017.3In the current

study, we apply different statistical approaches to investigate

the following two research questions:

1. Does the variation of prospectively assessed tinnitus

loudness differ between TYT users who rated their tin-

nitus loudness retrospectively as varying and TYT users

who rated their tinnitus loudness retrospectively as non-

varying?

2https://www.trackyourhearing.org

3N=1491 users completed at least one assessment questionnaire and

provided meaningful datasets for the analysis.

2. How is the prospectively assessed subjective stress-level

associated with (a) prospectively assessed tinnitus loud-

ness and (b) prospectively assessed tinnitus distress in the

following three user groups: users retrospectively rating

that stress worsens their tinnitus, users retrospectively

rating that stress reduces their tinnitus and users retro-

spectively rating that stress has no effect on their tinnitus?

Regarding research question 1, two variables were inves-

tigated. The first one assessed the variability of tinnitus

loudness retrospectively. The following TSCHQ item, which

was asked once during the TYT registration process, was

used to measure retrospectively whether or not the tinnitus

loudness varies: “Does the LOUDNESS of the tinnitus vary

from day to day?” with the options to respond with either

“yes” or “no.” The second variable measured tinnitus loud-

ness prospectively; at several points in time during their daily

life, the users rated the following question of the assessment

questionnaire: “How LOUD is the tinnitus right now?,” with

a slider ranging from 0 (i.e., not audible) to 1 (i.e., maxi-

mal loudness). To enable a comparison of the retrospective

assessments on day-to-day variation with the variation of

the prospective assessments, we had to focus our analysis on

the day-to-day variation (i.e., excluding within-day variation)

of the prospective assessments. Therefore, for each patient

with more than one prospective tinnitus loudness assessment

per day, we calculated the mean of the within-day prospec-

tive assessments in order to have one prospective loudness

assessment per day. Then, the variability of these prospective

loudness ratings was calculated for each user as the standard

deviation (SD) of all their prospective loudness ratings (on

a day-to-day level). To obtain a meaningful day-to-day vari-

ability of the prospective tinnitus loudness ratings, only those

users who provided a prospective tinnitus loudness assess-

ment at least at 10 days were investigated. We used a t test

for independent samples (two-tailed, with p<0.05 indicat-

ing a statistically significant result) to evaluate whether the

SD of the prospective tinnitus loudness assessments differs

between users retrospectively describing their tinnitus loud-

ness as non-varying and users retrospectively describing their

tinnitus loudness as varying.

After applying these criteria, plus the exclusion of assess-

ments with an inter-assessment interval of 15 mins or less

[37], N=305 users remained for the statistical analysis.

Of these N=305 users, n=54 (18%) reported retrospec-

tively that the tinnitus loudness does not vary from day to

day and n=251 (82%) reported retrospectively that the

tinnitus loudness varies from day to day. While the users ret-

rospectively rating the tinnitus loudness as stable provided

prospective assessments of tinnitus loudness for M=40.48

(SE =5.79) days (in an average interval of M=4.25

months), the users retrospectively rating the tinnitus as vary-

123

334 International Journal of Data Science and Analytics (2019) 8:327–338

ing provided data for M=44.26 (SE =3.62) days (in an

average interval of M=4.34 months).

We also ran the statistics for users with prospective loud-

ness assessments at least at 25 days in order to investigate

whether the results can be replicated. After applying these

stricter inclusion criteria, N=158 users were remaining

for the statistical analysis. Of these users, n=29 (18%)

reported retrospectively that the tinnitus loudness does not

vary from day to day, whereas n=129 (82%) reported ret-

rospectively that the tinnitus loudness varies from day to day.

Here, the users retrospectively rating the tinnitus loudness as

stable provided prospective assessments of tinnitus loudness

for M=61.38 (SE =9.17) days (average time interval the

assessments were provided in: M=4.33 months), whereas

the users retrospectively rating the tinnitus as varying pro-

vided data for M=71.80 (SE =6.13) days (average

time interval the assessments were provided in: M=4.98

months). In order to investigate whether the TYT users retro-

spectively rating the tinnitus loudness as non-varying differ in

baseline variables from users retrospectively rating the tinni-

tus loudness as varying, independent t tests and chi-squared

tests were performed (again, two-tailed with p<0.05 as

statistical significance level).

Regarding research question 2, four variables were eval-

uated. The first one assessed the impact of stress on tinnitus

retrospectively. The following TSCHQ item, again asked

once during the TYT registration procedure, was used to

measure retrospectively how the stress-level affects tinnitus:

“Does stress influence your tinnitus?” with the options to

respond “has no effect,” “reduces my tinnitus” or “worsens

my tinnitus.” This first variable was available only for a sub-

group of the TYT users (N=675). Of these N=675 users,

n=419 (62%) rated “worsens my tinnitus,” n=20 (3%)

“reduces my tinnitus” and n=236 (35%) “has no effect.”

The second variable is prospectively measured tinnitus loud-

ness. As described above, at several time points during their

daily life, TYT users rated the following question: “How

LOUD is the tinnitus right now?” with a slider ranging from

0 (i.e., not audible) to 1 (i.e., maximal loudness). The third

variable is prospectively measured tinnitus distress. Hereby,

the following question was rated at several time points by

TYT users during their daily life “How STRESSFUL is the

tinnitus right now?” with a slider ranging from 0 (i.e., not

distressed) to 1 (i.e., maximal distress). The fourth variable

is the prospectively assessed stress-level. The question “Do

you feel stressed right now?” was rated with a slider ranging

from 0 (i.e., no stress-level) to 1 (i.e., maximal stress-level)

at several time points by TYT users during their daily life.

Thereby, the 2nd, 3rd and 6th variable was contained in the

assessment questionnaire.

To evaluate how the prospectively assessed stress-level

is associated with prospectively assessed tinnitus (loudness

and distress) for users who retrospectively rate that stress

worsens their tinnitus, for users who retrospectively rate that

stress reduces their tinnitus and for users who retrospectively

rate that stress has no effect on their tinnitus, linear multi-

level models were performed due to the nested structure of

the data. The multilevel models had 2 levels: assessments

as level-1 and users as level-2. The multilevel models were

performed with the full maximum likelihood estimation to

handle missing data in the dependent variables (tinnitus loud-

ness and tinnitus distress), and a random intercept term was

included. The prospectively measured stress-level functioned

as time-varying covariate in these multilevel models. Due

to missing data in the covariate (no prospective stress-level

assessment), n=406 users retrospectively reporting “stress

worsens my tinnitus,” n=19 users retrospectively reporting

“stress reduces my tinnitus” and n=230 users retrospec-

tively reporting “stress has no effect” could be analyzed with

the multilevel models. Moreover, a chi-squared tests and

ANOVAs were performed to compare the three groups (TYT

users retrospectively rating that stress worsens their tinnitus,

TYT users retrospectively rating that stress reduces their tin-

nitus and TYT users retrospectively rating that stress has no

effect on their tinnitus) in baseline variables. Again, the sta-

tistical tests were performed two-tailed with p<0.05 as

statistical significance level.

5 Results

This section presents the results we obtained with respect to

the aforementioned two research questions.

Concerning research question 1, the comparisons between

the users retrospectively rating the tinnitus as not varying

and the users retrospectively rating the tinnitus as varying

are given in Table 2. As can be seen, TYT users who retro-

spectively rated their tinnitus loudness as non-varying from

day to day did not significantly differ in the average variation

(SD) of the prospective tinnitus loudness assessments from

the TYT users who retrospectively rated their tinnitus loud-

ness as varying from day to day. This result emerged when

TYT users providing prospective tinnitus loudness assess-

ments at least at 10 days were analyzed as well as when TYT

users providing prospective tinnitus loudness assessments at

least at 25 days were analyzed.

Moreover, no significant differences in baseline variables

(i.e., gender, age and tinnitus duration) were found between

users with retrospectively varying and users with retrospec-

tively non-varying tinnitus loudness. However, male users

tended to be more often in the group retrospectively rating

tinnitus loudness as non-varying than in the group retrospec-

tively rating tinnitus loudness as varying (p=0.053) in the

sample consisting of TYT users with at least 10 prospec-

tive tinnitus loudness assessments; in the sample of TYT

users with at least 25 prospective tinnitus loudness assess-

123

International Journal of Data Science and Analytics (2019) 8:327–338 335

Table 2 Results of research question 1

Retrospective rating

of tinnitus loudness

variation =no

Retrospective rating

of tinnitus loudness

variation =yes

Statistics

TYT users with prospective tinnitus loudness ratings at least at 10 days

Male gender n(%) 44 (83.0) 174 (69.9) χ2(1)=3.757; p=0.053

Age M(SE) 38.262 (3.337) 44.426 (1.306) t(292)=−1.597; p=0.111

Tinnitus duration (subjective report in

years) M (SE)

13.093 (2.032) 11.142 (0.752) t(65.666)=0.901; p=0.371

Mini-TQ-12 [12] score M (SE) 12.462 (0.749) 13.544 (0.356) t(298)=−1.275; p=0.203

Variation (SD) of the prospective

tinnitus loudness assessments M (SE)

0.151 (0.009) 0.150 (0.004) t(303)=0.106; p=0.916

TYT users with prospective tinnitus loudness ratings at least at 25 days

Male gender n(%) 24 (82.8) 93 (72.7) χ2(1)=1.271; p=0.260

Age M(SE) 34.385 (5.603) 45.657 (1.794) t(31.534)=−1.916; p=0.064

Tinnitus duration (subjective report in

years) M (SE)

16.364 (2.974) 11.353 (1.047) t(33.991)=1.589; p=0.121

Mini-TQ-12 [12] score M (SE) 12.793 (0.986) 13.024 (0.451) t(154)=−0.219; p=0.827

Variation (SD) of the prospective

tinnitus loudness assessments M (SE)

0.150 (0.013) 0.136 (0.006) t(156)=1.053; p=0.294

Results of the comparisons between TYT users retrospectively rating the tinnitus loudness as non-varying and TYT users retrospectively rating the

tinnitus loudness as varying

TYT TrackYourTinnitus, Mmean, SD, standard deviation, Mini-TQ tinnitus questionnaire short form

Table 3 Results of research question 2

Variable Parameter Estimate SE df T statistic pvalue

Users retrospectively rating that stress worsens their tinnitus

Tinnitus loudness Intercept (tinnitus loudness when