Dynamic time warp analysis of individual symptom trajectories in patients with bipolar disorder

Abstract

Introduction Manic and depressive mood states in bipolar disorder (BD) may emerge from the non-linear relations between constantly changing mood symptoms exhibited as a complex dynamic system. Dynamic Time Warp (DTW) is an algorithm that may capture symptom interactions from panel data with sparse observations over time. Objectives The current study is the first to analyze a time series of depression and manic symptoms using DTW analyses in patients with BD. We studied interactions and relative changes in symptom severity within and between participants. Methods The Young Mania Rating Scale and Quick Inventory of Depressive Symptomatology were repeatedly assessed in 141 patients with BD, with on average 5.5 assessments per patient every 3 to 6 months. DTW calculated the distance between each of the 27*27 pairs of standardized symptom scores. The changing profile of standardized symptom scores of BD patients was analyzed in individual patients, yielding symptom dimensions in aggregated group-level analyses. Using an asymmetric time-window, symptom changes that preceded other symptom changes (i.e., Granger causality) yielded a directed network. Results The mean age of the patients was 40.1 (SD 13.5) years old, and 60% were female. Idiographic symptom networks were highly variable between patients. Yet, nomothetic analyses showed five symptom dimensions: core (hypo)mania (6 items), dysphoric mania (5 items), lethargy (7 items), somatic/suicidality (6 items), and sleep (3 items). Symptoms of the ‘Lethargy’ dimension showed the highest out-strength, and its changes preceded those of ‘somatic/suicidality’, while changes in ‘core (hypo)mania’ preceded those of ‘dysphoric mania’. Image: Image 2: Image 3: Conclusions DTW may help to capture meaningful BD symptom interactions from panel data with sparse observations. It may increase insight into the temporal dynamics of symptoms, as those with high out-strength (rather than high in-strength) could be promising targets for intervention. Disclosure of Interest None Declared