Introduction
… sometimes all that is required is a useful visualization …
Increasingly intensive longitudinal data are collected, in which people, such as patients with a mental disorder, are measured daily or multiple times a day for a period of weeks or even months [
2‐
5]. An important motivation for collecting intensive longitudinal data in patients with a mental disorder is to capture psychopathology in its natural daily environment, and by studying it at a micro level, also its dynamics [
6]. For example, depressive mood might vary from day to day depending on how much stress is experienced during the days. A further advantage is that when such intensive longitudinal data are used instead of traditional retrospective questionnaires, artefacts such as recall biases can be minimized [
3]. Gathering this kind of data are done, for example, using experience sampling methodology (ESM; [
7]) or ecological momentary assessment (EMA; [
4]). Whereas ESM assesses phenomena such as momentary mood, thoughts, symptoms and context (e.g., social company), EMA is broader and can also include, for instance, physiological assessment and sensor data. ESM and EMA are often used interchangeably; we will use the term ESM hereafter [
8].
Importantly, ESM is not just a research tool for studying psychopathology in daily life [
9], it is also finding its way to clinical practice. Studying patients intensively over time allows giving context- and individual-specific feedback based on their own ESM data [
10]. ESM-based feedback thus allows for detailed idiographic information. This can inform both therapist and patient about moods and thoughts of the patient over the measured period of time, as well as what the patient did, with whom, and where. This unique source of information is currently lacking in standard therapeutic practice, but fits well with the state-of-the-art of case conceptualization [
11]. Such ESM-based feedback can furthermore help in tracking down patterns of the individual dynamics of emotions (e.g., how much do emotions fluctuate over time) and the association between those emotions and daily life events [
10]. Therefore, ESM-based feedback can help to show possible sources that improve or worsen mood variations over time, and thus not only help to monitor if the patient is doing well or if an intervention is effective, but also give new insights on which behaviours are functional or dysfunctional in daily life. This kind of ESM-based feedback can then guide future therapy interventions. ESM-based feedback is a promising new diagnostic tool empowering both the patient as well as the therapist [
12].
The additional benefit of ESM-based feedback was shown in a pioneering randomized control trial of Kramer et al. [
13]. In this study, patients with a depressive disorder either just filled-out an ESM questionnaire or also received individual-tailored feedback based on their own ESM data. It appeared that patients with a depressive disorder who received ESM-derived feedback versus depressed patients who did not receive such feedback had a clinically relevant decrease in their depressive symptoms. This suggests that personalized ESM-derived feedback can be beneficial. More generally, it highlights that ESM-derived feedback can be a useful addition to the treatment for patients with a depressive disorder.
ESM-derived feedback comes in many forms. This can vary from real-time advice or interactive feedback [
14,
15] to a feedback report containing analyses or summary statistics summarizing a certain period in which ESM data were gathered [
10]. Regarding real-time advice, Bauer et al. [
14], for example, gave their patients feedback in real-time through person-tailored text messages promoting, for instance, social support. Similarly, Hareva et al. [
15] continuously analyzed ESM data of a patient after every four data points to check if his values were below a beforehand set threshold. In case ESM scores were below the threshold the patient would receive real-time advice by an e-mail containing the message: “Let’s take a rest”.
Other studies have given ESM-based feedback through a report. Such a report could be given face-to-face or online [
16,
17]. Several studies provided verbal feedback in a face-to-face fashion that was complemented with visual summaries of weekly affect: a pie chart showing, for instance, the amount of time spent on activities (e.g., eating and drinking or household activities) and the amount of positive affect during these different contexts or activities. After several weeks, feedback furthermore included bar graphs indicating changes in weekly affect level and depressive complaints throughout the ESM study [
13,
16]. Other popular ways of giving descriptive feedback are means and line graphs of a single or several variables showing, for instance, if there is a trend in the data [
18]. Unique is the
N = 1 study of Groot [
19], where visualizations of patterns of emotions over time were combined with information on specific physical activities (e.g., running). Van Roekel et al. [
20] gave similar face-to-face feedback as in previous studies, but additionally included specific suggestions on how participants could increase feelings of pleasure in their lives. Furthermore, individual feedback also contained comparison of one’s own scores to a norm group. In a large, still ongoing crowdsourcing study “How Nuts are the Dutch” the ESM-based feedback was similar in style as in the aforementioned studies, but was only given online, and not in face-to-face [
17].
Recent studies often go beyond descriptive statistics and try to additionally offer feedback based on inferential statistics [
16,
20‐
23]. Such inferential statistics mostly takes the form of Vector AutoRegressive (VAR) modelling, with which one can detect, for example, whether certain activities at a given time point lead to more or less positive affect at a later time point [
21,
24,
25]. Inspired by the network approach, such models are presented and visualized as personalized networks [
26,
27]. According to the network approach, causal interactions between variables such as symptoms should be the focus of study in understanding what mental disorders are and how these disorders develop [
28‐
30]. The idea behind providing personalized networks then is that the potentially causal relationships between momentary self-reported affect states, context and behaviours could provide possible intervention targets for therapy [
18].
Although increasing the complexity of ESM-based feedback may hold great promise for getting the most out of ESM data, the usefulness of this approach for patients is still unclear. The interpretation of VAR-based network models is highly depended on the exact network model used and different choices regarding a VAR based network model can lead to widely different recommendations for the patient [
31,
32]. This not only results in conflicting advice on where and how to perform further interventions, but also in the additional complication that at the moment we do not have well developed tools to evaluate these different outcomes. Furthermore, the complexity of what is exactly modelled makes it more difficult for patients and therapists to fully understand the ESM-derived feedback.
As such complex statistical methods still face many issues that need to be overcome [
33], in this paper we turn our attention to further developing descriptive measures for personalized feedback. We introduce a new framework for giving descriptive feedback. Currently used descriptive statistical measures, such as histograms, only present a summary of the data and do not reflect the dynamic fluctuations in affect and factors that influence it. To address this shortcoming, we give back to the patient and therapist the data as it was filled-out by the patient. In this new ESM visualization approach, or ESMvis, the raw ESM data can be visualized in a dynamic way using the freely available software
R [
34]
. Thus, instead of using summary descriptives such as averages, or only showing selected variables that were measured, this visualization technique presents both the overall trajectory and the specific time moments in a movie format. This includes not only quantitative measures, such as how much a patient scored on a certain variable, but also qualitative information, such as written information about an experienced (un)pleasant event.
In order to showcase this new visualization approach, we illustrate all aspects of the visualization with ESM data of a clinical patient, who monitored herself over 1 year as part of a relapse prevention plan. We show that the method was experienced as insightful both by the therapist and the patient, and can potentially function as an add-on tool to care-as-usual. Both the data and the R-code is freely accessible and links will be provided in the “
Methods” section.
Discussion
We have provided a first demonstration of a new tool for the dynamic visualization of raw ESM data. The feedback was experienced as insightful by both the therapist and the patient, indicating patterns that could potentially help in future treatment of the patient. Our tool is freely available, adjustable, and easy to use, making it widely applicable to different kinds of ESM data. In addition to potential applications in clinical practice, gaining insights into the data is also a crucial first step before running more complex analyses. As such, ESMvis can work as an exploratory tool that can lead to new hypotheses, and in the end, inform more complex techniques [
37,
38].
One of the main advantages of ESMvis is that, in contrast to most statistical methods such as VAR, there are no restrictions regarding the number of time points that are needed. Any ESM dataset, short or long, can be visualized and reported back to the patient and therapist. Furthermore, not only the raw data but also the missingness is explicitly represented, giving an immediate grasp on how much missingness there is and when. In addition, there is a complete representation of all the data, including the context in which the emotions fluctuate, and the qualitative information and commentaries that the patient fills out. ESMvis is a comprehensive tool for visualization of complex time-series data collected for a long period (e.g., weeks, months) during daily life of an individual patient.
The wealth of information in ESM data, however, can also lead to problems for ESMvis. For example, when many (e.g., 10) variables are represented in the line plots, it will be difficult to discern them from each other. Fortunately, this issue is largely solved using the Shiny app, in which you can click some information on and off (such as variables in the line plots, or the commentaries in the circle figure), or zoom in to a certain time period if wanted. Additionally, adjusting the ESMvis for other studies with different setups might require extensive recoding. First, for new ESM questionnaires, one has to always decide upfront what will be classified as positive, negative and event variables. Second, when the data form is not exactly the same as in the current study (e.g., large number of variables with different scales, multiple events), the current code needs to be adjusted requiring some further programming. Thus, at the moment, ESMvis is not a standalone tool for clinical practice.
Another limitation is that even though ESMvis was experienced as insightful by the therapist and patient, a more systematic way of studying its usefulness is needed to confirm the clinical relevance of the tool. This could take the form of a randomized controlled trial or a qualitative study consisting of structured interviews with therapists using the tool [
39]. Relatedly, more research is needed that explicitly focuses on the clinical benefit of ESM feedback for the patient over and above just filling in the ESM questionnaires (along the lines of [
13,
16,
40]).
To conclude, we hope to have shown that ESMvis can be used in addition to current personalized feedback methods, and can be informative for researchers, therapists and patients to get a full grasp of the complexity of ESM data and daily life in general.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.