Tinnitus is a condition in which phantom sounds are perceived without a corresponding external stimulus. Those sounds usually take the form of ringing, hissing, or buzzing, but other less common types of perception have also been reported (Baguley et al., 2013; Langguth et al., 2013). The underlying causes of tinnitus are not fully clear, but auditory pathway deafferentation is commonly recognized as key factor in the etiology of tinnitus (Shore et al., 2016). Although tinnitus is generally a benign condition, its bothersome manifestation, which is estimated to affect 1% of the population (Biswas and Hall, 2020), can be both debilitating (Cima, 2018) and costly (Maes et al., 2013). Tinnitus can be subdivided into two categories: acute and chronic (de Ridder et al., 2021). The former category describes the rather common phenomenon of phantom sounds being perceived for several seconds or minutes after insult to the auditory system (e.g., listening to loud music); whereas the latter category refers to uninterrupted perception of the phantom sound for at least 3 months. In its chronic presentation, tinnitus rarely resolves entirely (Tunkel et al., 2014; Simões et al., 2021).

Available treatments are not effective at suppressing the chronic phantom perception for most sufferers. As a result, most treatment strategies seek to reduce tinnitus-related distress (Tunkel et al., 2014). There is a growing consensus that tinnitus is a heterogeneous condition and that this characteristic may significantly impact both how the condition is experienced by patients and the efficacy of treatments (Simoes et al., 2019; Kleinjung and Langguth, 2020). Factors demonstrated to affect treatment outcomes include sociodemographics, personality, and tinnitus characteristics, though many others are currently being investigated (Genitsaridi et al., 2020). For example, the heritability of tinnitus differs depending on its laterality (i.e., whether sounds are perceived in one or both ears) and the patient's gender (Maas et al., 2017). Furthermore, evidence suggests that personality traits may explain the response to online cognitive behavior therapy treatment (Kleinstäuber et al., 2018), but not acoustic stimulation (Hafner et al., 2020). It is unclear which factors are related to tinnitus-related disability, as the psychoacoustic properties of tinnitus (e.g., laterality, loudness, pitch, type of perceived sound) do not fully explain the tinnitus distress (Cederroth et al., 2019; Kleinjung and Langguth, 2020). Therefore, it is of great clinical interest to understand which factors are associated with distress, especially at the individual level.

The advent of minimally intrusive longitudinal sampling methods has allowed researchers and clinicians to develop predictive models at the individual level, while at the same time maximizing the ecological validity of assessments (Wright and Woods, 2020). Not surprisingly, there is growing interest in using ecological momentary assessment (EMA) in the fields of psychology and psychopathology in general (Wright and Woods, 2020), as it provides clinicians and researchers a tool to investigate the interplay between emotions, symptoms, and behaviors, while considering contextual associations (e.g., how mental phenomena co-occur and interact over time) in naturalistic settings (Shiffman et al., 2008). In the context of tinnitus, Probst et al. (2016), modeled patterns of daily fluctuation of tinnitus characteristics and identified a mediatorial role of tinnitus loudness on stress, while Pryss et al. used EMA to establish that patients often have recollection bias regarding tinnitus fluctuations throughout the day (Pryss et al., 2019). However, most of the studies in the tinnitus field using EMA have focused on group-level analysis. A recurring clinical topic is how uniquely tinnitus is experienced by patients, with different triggers leading to higher tinnitus distress and/or loudness in certain patients, but not others. Identification of factors that could lead to personalized interventions is of great clinical utility.

Therefore, the objective of this study was to apply EMA 206 methods to identify factors (see Table 2) that could be associated with tinnitus loudness and distress at both the individual and group levels. Furthermore, data collected used EMA sampling were used to identify whether such factors influence loudness and/or distress on a subsequent day. To achieve these objectives, we first investigated whether states such as tinnitus loudness, distress, and mood impact autocorrelates or cross-correlates throughout subsequent days at the group level. Second, we modeled data at the individual level using linear regressions with elastic net regularization for each unique time series with tinnitus distress and tinnitus loudness as dependent variables. Third, we used subgrouping within group iterative multiple model estimation (S-GIMME) to obtain unique models for each participant on contemporaneous and lagged effects between the variables collected with EMA.

Table 1 summarizes the demographics of the samples used for both analyses. Of the original dataset, 57% of the data were excluded from the analysis as the data were not obtained from that sequence for at least 10 uninterrupted days (see Supplementary Figure 1). Thus, the sample of study 1 consisted of 488 unique sequences from 278 users. Following the guidelines of the authors of the S-GIMME package, only sequences with at least 60 days of uninterrupted usage were included in the second analysis (Lane et al., 2019). Thus, the sample consisted of 32 sequences from 32 unique users. Table 2 shows how the EMA questions were formulated (translated into English from German), with their abbreviations. Two questions, namely questions 1 and 8, were excluded from the analysis as they were dichotomous and continuous variables were necessary for computing auto- and cross-correlations.

Figure 1 illustrates how uniquely tinnitus is uniquely experienced by four arbitrarily selected TYT users. The left column of the figure shows the mean values and their dispersion with density plots; the middle column shows the fluctuation of those variables through time with time series plots; the right column depicts the contemporaneous relation between variables with correlational heat maps. Overall, these four examples highlight how symptoms may be burdensome, how they fluctuate over time, and how they interact with each other.

In this study, we identified individuals whose tinnitus loudness and distress are affected by behavioral and emotional processes of the same and of the previous day. We started by showing that at the group level, no evidence of autocorrelation or cross-correlation could be observed between the six EMA variables. However, modeling data at the individual level revealed that tinnitus loudness and distress were auto- and cross-lagged from one day to another for several individuals (see Figures 3–5). Lastly, we used S-GIMME to explore the unique interplay of these variables, effectively modeling how heterogeneously tinnitus manifests itself over time.

Our results provide a template for how to model tinnitus loudness and tinnitus distress at the individual level (see Figures 3–5). The interaction between these variables constitutes a complex mosaic of how tinnitus is experienced; Although the uniqueness of experiencing tinnitus has been widely acknowledged (Tunkel et al., 2014; Elgoyhen et al., 2015; Cederroth et al., 2019), empirical studies that demonstrate this complex relationship were lacking. Furthermore, using S-GIMME, we were able to incorporate the effect of time into the models, a critical component of experiencing tinnitus (Probst et al., 2017) often neglected in empirical studies due to the challenges of conducting multiple samplings in the traditional ambulatory setting.

Corroborating previous findings, we showed a positive association between stress, tinnitus loudness, and tinnitus distress (Scott et al., 2009; Probst et al., 2016). Interestingly, we also observed that emotional arousal, mood and concentration had ambivalent associations with both loudness and distress (see Figures 4C,F). Such associations could explain why tinnitus is uniquely experienced, and future research could further investigate factors associated with tinnitus distress and loudness, such as behavior, emotional and cognitive dynamics samples from EMA. Tinnitus is known to have potential negative consequences on cognition (Andersson and McKenna, 2009; Mohamad et al., 2016; Neff et al., 2021), but this is the first time a positive association between concentration and tinnitus distress/loudness has been shown. Future studies should further investigate this seemingly paradoxical relationship, especially considering that concentration problems are one of the core domains that tinnitus patients would like to have as outcome measures from clinical interventions (Hall et al., 2018, 2019). For example, this positive association could be related to compensatory mechanisms.

Other core domains include the ability to ignore the phantom perception, its intrusiveness, the sense of control over one's body, quality of sleep, and negative thoughts and beliefs. The first three domains could be investigated by adapting questions from the Tinnitus Functional Index, a widely used, validated questionnaire that captures different dimensions of tinnitus distress (Meikle et al., 2012). Negative thoughts could be investigated by adapting the positive and negative affect schedule (Watson et al., 1988), similar to other studies in the field of psychopathology using EMA (Wright and Woods, 2020; Heller et al., 2021). Sleep problems remain one of the main complaints of clinical tinnitus patients (Tunkel et al., 2014; Asnis et al., 2018; Lu et al., 2020; Inagaki et al., 2021; Richter et al., 2021), but no study to date investigated its effect using an EMA design, despite the evidence of its relevance on other chronic, disabling conditions (Short et al., 2017).

Our work suggests that only a subset of patients experience lagged effects between the variables sampled through EMA (see Figures 4–6). This finding could be leveraged to deliver personalized interventions through mobile apps, especially among users where tinnitus distress and loudness and auto or cross-correlated. For example, the results suggest that patients with tinnitus may benefit from just-in-time adaptive intervention (JITAI) (Nahum-Shani et al., 2016; Wang and Miller, 2019). JITAI uses mobile sensing to deliver customized interventions based on unique fluctuations recorded by the EMA. Such a system has been used in several fields, including physical health, addiction, and mental care research, but not in tinnitus (Wang and Miller, 2019). For example, customized interventions could be delivered to patients whose tinnitus loudness and/or distress are auto- or cross-correlated over days: Once the algorithm anticipates a potential spike in tinnitus loudness or stress on the next day, an intervention such as psychoeducation tips for coping with tinnitus through push notifications (Unnikrishnan et al., 2020), sound therapy (Tyler et al., 2018; Kutyba et al., 2019, 2021), online delivered cognitive behavior techniques (Weise et al., 2016; Kleinstäuber et al., 2018; Andersson et al., 2019), or meditation techniques (Fitzgerald et al., 2021) could be activated. Future studies could evaluate whether these results are replicated with different sampling frequencies. For example, an EMA every 8 h has been shown to be well tolerated in clinical settings (Shiffman et al., 2008) and would considerably reduce the duration of the study, and potentially increase the adherence to using the APP (Schleicher et al., 2020).

Methodological limitations should be considered when interpreting these results. For example, missing values continue to constitute a significant challenge for researchers, including those using EMA. We implemented a popular and robust method for data imputation; however, empirical evidence that this is the optimal method for imputing data from EMA is not available. A recent study investigated the causes of discontinuation of TYT use (Schleicher et al., 2020), but no clear predictors of adherence to app use were found. Furthermore, results suspected of bias cannot be easily discarded, as only a fraction of users used the app for more than 10 consecutive days (Supplemental Material). Therefore, it is unlikely that our sample is representative of the whole population, and caution should be exercised when interpreting the results and its generalizability. The possibility of low stability from the coefficients obtained from elastic net should also be considered, especially when covariates are highly correlated (Shen et al., 2016). Although providing daily information about tinnitus could increase user distress by repeatedly directing their attention to their tinnitus, previous work has suggested that using the TYT app does not harm users (Schlee et al., 2016). However, randomized control trials investigating the potential side-effects of recording intensive longitudinal data are still missing.

In summary, we show that tinnitus loudness and tinnitus distress are affected by both contemporaneous and lagged behavioral and emotional processes measured through EMA. Additionally, we showed that S-GIMME can distinguish users whose tinnitus distress and loudness are modulated by contemporaneous effects from those where those dynamics are modulated by both contemporaneous and lagged effects. Distinguishing these two subgroups could be therapeutical value, especially when aligned with just-in-time adaptive interventions to mitigate or prevent future peaks of tinnitus distress or increased loudness.

The raw data supporting the conclusions of this article will be made available upon request.

The studies involving human participants were reviewed and approved by the Ethics Committee of the Faculty of Medicine of the University of Regensburg (Study approval number 15-101-0204). The patients/participants provided their written informed consent to participate in this study.

JS, WS, JB, and RP designed the study. JS, WS, and JB analyzed the data. JS, WS, PN, BL, and SM interpreted the results and conclusions. JS wrote the manuscript. All authors contributed to the article and approved the submitted version.

This project has received funding from the European Union's Horizon 2020 Research and Innovation Programme, Grant Agreement Number 848261.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.