Chapter 1: Ecological Momentary Assessment in Mental Health Research

In this chapter, we critically examine EMA research findings from adjacent fields, such as well-being, mental health, mood and psychotic disorders, to gain a better understanding of how this methodology fills in research gaps. Further, we address both strengths and limitations of the methodology which must be considered, especially when adapting this approach to tinnitus research. The compilation is far from exhaustive, but should provide insights into the potentialities and challenges of EMA research.

Fine-Grained Evaluation of Fluctuations

The EMA methodology enables repeated sampling within-day and over longer time periods that allows to study both short-term as well as long-term patterns of self-reported symptoms. Stieger and Reips investigated daily and weekly fluctuations of well-being with three daily assessments over a two-week period in 213 participants and found that individuals felt better in the evening and on the weekends [16]. An earlier hypothesized “blue Sunday effect”, i.e. individuals feeling worse on Sundays, could be refined by a more detailed resolution of measurement times. Sunday ratings from the morning until noon followed the pattern of increasing mood of the other days. The difference started to become apparent in the afternoon: While mood increased further during afternoon on Mondays to Saturdays (Mevening = 4.53), mood level on Sundays stagnated (Mevening = 0.31; t = 4.10, p < 0.001, d = 0.23). Thus, not the whole Sunday, but only the second half might be “blue” [16].

Houben et al. conducted a meta-analysis to study the association between emotional dynamics and well-being measured with EMA across the time scales of seconds, hours, or days (max. one week) [17]. Results showed a negative correlation of well-being with emotional variability (i.e. range of emotional states across time, ρ = −0.178, p = < 0.001), emotional instability (i.e. magnitude of emotional changes from one moment to the next, ρ = −0.205, p = < 0.001), and emotional inertia (i.e. ability to predict emotional state from the previous moment, ρ = −0.151, p = < 0.001). It was concluded that people high in well-being exhibit a moderate emotional reactivity as well as an intact emotional regulation to events resulting in smaller deviations from the baseline level [17].

While these findings highlight EMA's strength in capturing subtle temporal patterns, they rely on frequent engagement from participants over days or weeks. It is suspected that participant fatigue can impact both compliance and data quality, particularly in clinical populations. Meta-analytical evidence estimates average compliance to be around 80%, with no differences between healthy and clinical populations [18,19,20,21]. Only psychotic patients were found to be less compliant than healthy participants [19]. However, the impact on data quality remains difficult to assess, and compliance rates may also be affected by self-selection bias (inclusion of only EMA-motivated participants).

High Ecological Validity of Measurement

EMA is characterized by high ecological validity catching the participant within his or her ordinary daily life. Findings known from retrospective evaluation can be replicated or refined by EMA data as well as supported with objective measures. MacKerron and Mourato performed a large study with more than 20.000 participants to investigate the relationship between self-rated momentary well-being and immediate environment obtained from GPS data of more than one million data points [22]. Controlling for potential confounders such as weather, daylight or activity as well as subject-level differences including demographic information, they found that feeling better was related to being in natural surroundings. Some confounders were identified through open-source data (e.g. weather) and were matched with the exact location and timestamp, others were measured with EMA (e.g. activity) [22].

A similar investigation that studied associations between self-rated momentary well-being and self-rated surroundings extended this finding by time-lasting effects of natural surroundings on happiness [23]. Seeing trees and the sky predicted well-being at the next assessment which appeared to be on average 2.5 h later. The feeling of being in contact with nature predicted well-being on average 5 h later in the second next measurement [23].

Myin-Germeys et al. investigated emotional reactivity to daily life stress in psychotic, depressed and bipolar patients as well as healthy controls [24]. Vulnerability to extreme life events in these psychopathological disorders is well-known from retrospective research, the authors aimed to extend this finding to daily stress reactivity using EMA methodology. Depressed and psychotic patients exhibited a higher daily stress reactivity in negative affect compared to bipolar patients and healthy controls, while psychotic and bipolar patients showed a higher daily stress reactivity in positive affect compared to depressed patients and healthy controls. Those differences were only apparent for activity-related stress but not for social stress [24]. Thus, vulnerability to stress for severe mental illnesses could be extended to more subtle daily life stressors and re-fined for different diagnoses, types of stress and affective states.

The same group followed schizophrenic patients and healthy controls with EMA over six consecutive days in another study [25]. They aimed to investigate momentary emotional experiences and hedonic capacities of this group of patients since they are often said to have deficits in emotion expression and hedonic experiences. Patients exhibited higher intensity in negative affect and lower intensity in positive affect compared to controls. Also, patients reported experiencing fewer positive events. However, when experiencing a positive event, hedonic experience was equally strong for patients and controls (χ2(3) = 1.69, p = 0.64). The same effect was found for social hedonic capacity: Even though patients were less often in company of others, they expressed the same level of positive affect when in company with others compared to controls (β = 0.01, 95% CI = 0.08–0.06, p = 0.76). Thus, there was no empirical evidence for specific deficits in emotional experiences nor hedonic capacities. The difference appeared at the behavioural level, i.e. patients experienced less pleasant events and social interactions resulting in more negative affect overall [25].

It is important to remember that ecological validity is not fully provided by an out-of-lab measurement. EMA is a subjective measurement method which faces similar problems as retrospective questionnaires such as individual interpretation of scales and questions or recency effects. In addition, highly intensive measurement could induce reactivity effects such as behavioural change which also threatens ecological validity [26].

Longitudinal Measurement Enables Chronological Associations

The repeated measurements of EMA contribute to a deeper understanding of temporally shifted associations even if the observational nature limits definitive conclusions regarding causal inference. Triantafillou et al. investigated the relationship between self-reported sleep quality and mood in depressed, anxious and healthy subjects [27]. The study revealed several interesting findings. First, they found similar associations in both directions, mood predicted sleep quality of the following night (β = 0.247; p < 0.001) and sleep quality predicted mood of the following day (β = 0.270; p < 0.001). Second, when taking into account individual differences answering self-report scales by applying z-score standardization, the effect of sleep on mood (β = 0.344; p < 0.001) was much larger than vice versa (β = 0.132; p < 0.001). Third, potential confounders such as self-reported stress, energy and focus levels as well as objectively measured physical activity, weather, day type (working day or day off) and weekday could not explain those effects [27]. Thus, it is very likely that sleep quality has an impact on next-day mood which has considerable clinical implications.

Thewissen et al. studied the temporal relationship of emotional experiences with paranoid episodes in a clinical sample with ten daily EMA ratings over 6 consecutive days [28]. Increased anxiety and decreased self-esteem significantly predicted the onset of a paranoid episode at the next time point. During the episode, patients exhibited higher levels of anxiety, depression, and anger as well as lower levels of self-esteem. Initial paranoia intensity, depression and anger further predicted episode length. The longitudinal design was able to sharpen earlier findings demonstrating the importance of self-esteem and negative emotions in paranoia [28].

Although EMA supports temporal modeling, it is still observational. Thus, even though potential confounders were ruled out, an effect of sleep quality on next-day mood was not tested in a controlled experimental setting. The same applies to the effect of anxiety and self-esteem on paranoid episodes.

Comparison with Retrospective Measures

An important aspect of the EMA field is validity research, which comprises different approaches such as comparison with retrospective self-reports. Ben-Zeev and Young compared momentary and retrospective ratings of depressive symptoms in hospitalized depressed patients and nonclinical controls [29]. Patients rated 7 of 13 symptoms more severe in retrospective compared to momentary assessments and showed no difference among the other 6 symptoms. Controls retrospectively rated 4 symptoms to be more severe, 4 symptoms to be less severe and showed no difference on the remaining 5 symptoms. Thus, while controls had both positive and negative biases, depressed patients tended to show a negative bias in some retrospective assessments [29].

Even if memory biases might be tackled with (close-to) momentary assessments, those assessments are not free from cognitive distortions [30]. This could be a plausible concern in tinnitus patients with comorbidities such as depression or hyperacusis where negative thought patterns or hypervigilance towards sounds might skew responses.

Treatment Effect

EMA has also been used to enhance understanding of the effect of treatment. Barge-Schaapveld and colleagues were interested in the effects of antidepressant treatment in the patients’ daily life experience [31]. Treatment responders reported spending more time on household chores and less time in passive leisure activities after treatment. They experienced more positive and less negative affect as well as more highly positive mood states during activities and an increase in sleep quality [31].

In a follow-up investigation, the authors studied the feasibility to predict treatment effects by early changes in momentary positive and negative affect [32]. A change in positive affect during the first week of treatment predicted HDRS (Hamilton Depression Rating Scale) score change, response and remission rates at the end of treatment (OR = −0.6/4.3/9.3, p < 0.001). Changes in negative affect predicted HDRS score change and remission with smaller effects (OR = −0.3/3.6, p < 0.01). Adding early change in positive affect to early change in HDRS score improved prediction accuracy of all treatment outcomes at the end of treatment which holds considerable implications for clinical decision making [32].

Another investigation focused on the ability to measure antidepressant treatment-related early side effects with EMA [33]. It was found that patients reported more side effects using EMA than to the General Practitioner (GP) within the first week, e.g. dizziness was reported by 5 times as many patients according to EMA. Side effects reported by EMA were associated with lower momentary quality of life and higher dropout risk. Another EMA finding of this study was that intraindividual fluctuations of momentary quality of life decreased greater for patients than for healthy controls during the treatment course [33].

Measuring treatment effects based on EMA is very promising, but there are some things to consider: Missing data could be NMAR (not missing at random) if patients drop out of treatment due to absent or negative treatment response [30]. Repeated assessments could lead to increased self-monitoring or self-reflection which may reduce the symptoms on its own [26]. Both effects, if valid, conflate treatment effects.

Phenotyping

Longitudinal self-rated data is further suitable to identify subgroups of patients based on their within- and between-day fluctuations. van Genugten et al. analysed EMA mood data from a period of seven days to identify latent groups of depressed patients based on their average mood, mood variability and emotional inertia [34]. Results revealed a four-profile model that differed in terms of average mood and variability, but not inertia. The first group (5% of patients) was characterized by the most negative and least variable mood, the second group (71%) by a moderately positive and variable mood, the third group (14%) by the most positive and moderate variable mood, and the fourth group (10%) by a moderately positive and high variability in mood. At baseline, the third group was less depressed than the other groups. The authors speculate the first group to represent the melancholic depression type with persistent negative mood and limited mood reactivity, while the second group could represent atypical depression which is a more common form with some degree of mood reactivity [34].

Conclusion: EMA in Mental Health Research

In summary, findings from mental health research fields were able to underline the power of EMA to enhance our understanding of mood, well-being, emotional experiences, and psychopathological symptoms by extending or refining results from traditional retrospective methods. As demonstrated, this methodology enabled the precise detection of short- and long-term patterns, captured differences within- and between subjects, could be combined with objective measures via passive sensing, was scalable to large sample sizes, revealed delayed or prolonged associations, captured the nuances of daily life, enhanced the understanding and early detection of treatment response, and distinguished unique symptom profiles among subgroups. It can be a valuable addition to the toolkit of researchers and clinicians as it reveals insights with significant implications for personalised interventions and more accurate diagnosis. Yet, this chapter also highlights important limitations such as participant burden influencing data quality and protocol compliance, selection bias, ecological validity, reactivity and cognitive distortions which could influence validity and reliability of EMA data. More EMA research findings can be obtained from reviews on suicidal thoughts [35], psychopathology [36], mood disorders [37], depression [38], well-being [39], substance use [40] and mobile crowdsensing [41].

Chapter 2: Ecological Momentary Assessment in Tinnitus Research

This chapter focuses on the application of EMA within tinnitus research. It begins with an exploration of the methodological evolution of EMA and is followed by a comprehensive overview of existing studies and study designs.

Chronological Development of the Literature

In 1985, several years before the concept of EMA was introduced, Scott and colleagues pioneered by using daily self-recordings of tinnitus loudness, discomfort, depression and irritation to evaluate the effects of a psychological intervention [42]. Symptom severity was rated at pre-specified times on paper–pencil forms. The wider dissemination of EMA in tinnitus research began in 2012 with the publication of Henry’s pilot study investigating the feasibility of measuring within- and between-day variability of tinnitus symptoms [43]. In this work, EMA questions were thoroughly selected in a two-stage process involving focus-groups and prompts were sent out by a personal digital assistant (PDA; which can be considered the predecessor of the smartphone). Three years later, the spread of smartphones into our daily lives had enabled the authors of the second study on EMA feasibility in tinnitus to send out alerts via text messages which contained a link to the EMA survey [44]. Since then, at least one paper on EMA in the tinnitus field has been published per year (see Fig. 1). The first smartphone app that fully integrated EMA was “TrackYourTinnitus” (TYT), which collected longitudinal data on tinnitus symptoms from over 4000 patients since 2013 using a crowdsensing framework. Subsets of this data were used not only for feasibility evaluation [45], but also to enhance understanding of the longitudinal relationship and daily fluctuation of tinnitus symptoms [46, 47]. The feasibility of EMA in tinnitus has consistently been subject to research by looking at construct validity [48], compliance with dense EMA schedule [49], differences to end-of-day data (EDD) [50], and potential influence on tinnitus symptoms [51]. Further, two publications compared momentary tinnitus symptoms with measures from retrospective questionnaires [52, 53]. In 2018, the first EMA study using machine learning (ML) got published that investigated methods to capture similarities between tinnitus patients on the basis of one-time questionnaire and longitudinal EMA data [54]. Subsequently, many other ML-based analyses followed [55,56,57,58,59,60,61,62]. In recent years, the feasibility to use EMA as an instrument for collecting outcome measures in clinical studies [63,64,65] or to combine EMA with process mining tools to analyse temporal process patterns and further improve the understanding of tinnitus symptom variability has been explored [66]. An overview of EMA studies published over the years is depicted in Fig. 1.

Fig. 1

Scientific articles on EMA in tinnitus

Overview of Study Designs

The design of EMA studies in tinnitus are displayed in Table 1 (N = 28). 15 studies (54%) retrieved data from the TYT app, 6 studies (21%) sampled unique data sets, 3 studies (11%) used data from the TinnitusTipps app, 3 studies (11%) used data from the UNITI app and 2 studies (7%) used data from the TinNots app. Note that using the same app may or may not involve the use of the same data set. The studies’ sample size ranges from 3 to 3691 (Mean = 460.3, Median = 203.5). The big data sets (N > 100) used in 14 studies all came from the TYT app, with two exceptions being sampled from the UNITI app. 10 publications (36%) reported the average degree of tinnitus bother of their samples. Those who did, studied on average mild, moderate and severely bothered samples. EMAs were scheduled with a range of one to seven times daily over periods of 2 weeks to 4 months. The TYT app had no pre-defined schedule which allowed the users to fill in the assessment according to their own preferences. The repeated assessments contained between 6 and 19 questions, which can be broadly grouped into tinnitus-related questions such as distress and loudness, mental health-related questions such as mood and anxiety, somatic health-related questions such as neck tension and hearing ability, and context-related questions such as activity and location. The majority of those questions focused on the current condition, while some considered the condition of the whole day (EDD questions). Compliance to the EMA protocol was reported in 8 publications (29%) and ranged between 78.3%—91.7%, however, it was not always specified how compliance was defined or how percentage was calculated.

Table 1 Overview and design of EMA tinnitus studiesChapter 3: Insights from EMA Research in Tinnitus

In the following, we summarize the insights gained from EMA in tinnitus research, following the topic structure presented in Fig. 1. At the end of each section the main findings are summarized.

Feasibility

The perception of tinnitus is at least partly modulated by psychological processes, with attentional mechanisms playing a central role. Psychological interventions aim to facilitate habituation, enabling patients to integrate the perception of tinnitus into their sensory background [68]. A main concern about repeated questioning of tinnitus symptoms during EMA is the possibility that increased symptom awareness may exacerbate the severity of the tinnitus [43]. Yet, empirical findings could not confirm this effect. Henry et al. reported no significant differences in mean Tinnitus Handicap Inventory (THI) scores after a two-week EMA period involving four daily prompts in a clinical sample (N = 24, mean difference = 1.71, p = 0.50) [43]. 90% of the participants indicated increased tinnitus awareness, with most viewing this as a positive change (e.g., gaining better self-understanding). Additionally, the majority experienced no disruption to their daily routine during the study. In a cohort of regular TYT app users with at least one month of usage (N = 66), no significant differences were found between the first and last five EMA responses regarding tinnitus loudness (p > 0.2) and tinnitus distress (p > 0.7) [45]. Similarly, no differences were observed for participants using the app for less than one month (tinnitus loudness: p > 0.5, tinnitus distress: p > 0.4), thereby ruling out the possibility of a self-selection bias, i.e. patients experiencing EMA reactivity might cease app usage earlier than those who are not affected [45]. Further, Goldberg et al. demonstrated that two consecutive EMA periods had no significant effect on tinnitus distress measured by the Tinnitus Functional Index (TFI) and the Overall Global Rating of Bother Scale (GBS; N = 40) [48]. When queried, 85% of patients reported no change in their tinnitus perception