A Data-Driven Clustering Method for Discovering Profiles in the Dynamics of Major Depressive Disorder Using a Smartphone-Based Ecological Momentary Assessment of Mood

Abstract

Background: Although major depressive disorder (MDD) is characterized by a pervasive negative mood, research indicates that the mood of depressed patients is rarely entirely stagnant. It is often dynamic, distinguished by highs and lows, and it is highly responsive to external and internal regulatory processes. Mood dynamics can be defined as a combination of mood variability (the magnitude of the mood changes) and emotional inertia (the speed of mood shifts). The purpose of this study is to explore various distinctive profiles in real-time monitored mood dynamics among MDD patients in routine mental healthcare.

Methods: Ecological momentary assessment (EMA) data were collected as part of the cross-European E-COMPARED trial, in which approximately half of the patients were randomly assigned to receive the blended Cognitive Behavioral Therapy (bCBT). In this study a subsample of the bCBT group was included (n = 287). As part of bCBT, patients were prompted to rate their current mood (on a 1-10 scale) using a smartphone-based EMA application. During the first week of treatment, the patients were prompted to rate their mood on three separate occasions during the day. Latent profile analyses were subsequently applied to identify distinct profiles based on average mood, mood variability, and emotional inertia across the monitoring period.

Results: Overall, four profiles were identified, which we labeled as: (1) "very negative and least variable mood" (n = 14) (2) "negative and moderate variable mood" (n = 204), (3) "positive and moderate variable mood" (n = 41), and (4) "negative and highest variable mood" (n = 28). The degree of emotional inertia was virtually identical across the profiles.

Conclusions: The real-time monitoring conducted in the present study provides some preliminary indications of different patterns of both average mood and mood variability among MDD patients in treatment in mental health settings. Such varying patterns were not found for emotional inertia.

Keywords: cluster analysis; depression; ecological momentary assessment; heterogeneity; mood dynamics; mood instability.

Figure 1

Generated mood dynamic patterns that show different combinations of mood variability and emotional inertia. There were three measurement points over the course of a day. The gray background represents a negative mood (score < 6), while the white area represents a positive mood (≥6). The horizontal black line represents the Average Mood across the 7-day monitoring period. For each panel, the average mood was ~6. (A) High variability, high emotional inertia. (B) High variability, low emotional inertia. (C) Low variability, high emotional inertia. (D) Low variability, low emotional inertia.

Figure 2

Generated EMA-response patterns of the profiles of the two- and four-profile models. The data for these graphs were generated using the standardized mean scores for the indicators of the profiles using the estimated parameters shown in Table 2. Three EMA assessments were completed each day, at 10 a.m., 8 p.m., and at a random time between 10 a.m. and 10 p.m. The horizontal black line represents the AM across the 7-day monitoring period, while the gray background represents negative mood (score < 6) and the white area represents positive mood (≥6). (a) Model 2. (b) Model 4.