Nonlinear time-series approaches in characterizing mood stability and mood instability in bipolar disorder

Abstract

Bipolar disorder is a psychiatric condition characterized by episodes of elevated mood interspersed with episodes of depression. While treatment developments and understanding the disruptive nature of this illness have focused on these episodes, it is also evident that some patients may have chronic week-to-week mood instability. This is also a major morbidity. The longitudinal pattern of this mood instability is poorly understood as it has, until recently, been difficult to quantify. We propose that understanding this mood variability is critical for the development of cognitive neuroscience-based treatments. In this study, we develop a time-series approach to capture mood variability in two groups of patients with bipolar disorder who appear on the basis of clinical judgement to show relatively stable or unstable illness courses. Using weekly mood scores based on a self-rated scale (quick inventory of depressive symptomatology-self-rated; QIDS-SR) from 23 patients over a 220-week period, we show that the observed mood variability is nonlinear and that the stable and unstable patient groups are described by different nonlinear time-series processes. We emphasize the necessity in combining both appropriate measures of the underlying deterministic processes (the QIDS-SR score) and noise (uncharacterized temporal variation) in understanding dynamical patterns of mood variability associated with bipolar disorder.

Figures

Figure 1.

A schematic of mood patterns in bipolar disorder: the disorder does not simply feature full-blown episodes of mania and depression with periods of normality. Rather, ongoing inter-episodic mood instability is also a clinically common, yet a poorly understood feature of the disorder.

Figure 2.

Illustration of the type of mood score charts used to rate patients in clinic as either stable or unstable. Mood stability in bipolar disorder was assessed over the course of six months using depressed mood score data (QIDS-SR) for individuals and used to define as either (a) an ‘unstable’ mood profile or (b) a ‘stable’ mood profile.

Figure 3.

Patient attrition rates from the study over 220 weeks. Plot showing (a) survival curves for the stable (solid lines) and unstable (dashed lines) group and (b) the overall survival group (with confidence intervals as dotted lines).

Figure 4.

Mood score time series for individual patients from the stable group (black lines) with fitted threshold autoregressive model (equations (3.1) and (3.2); red and blue points). RMSE values (see the electronic supplementary material) are used to describe the overall correspondence between the points predicted by the equation (circles) and the actual values reported by patients (black lines). Note that the scale on the y-axes vary between plots.

Figure 5.

Mood score time series for individual patients from the unstable group (black lines) with fitted threshold autoregressive model (equations (3.3) and (3.4); red and blue points). RMSE values (see the electronic supplementary material) are used to describe the overall correspondence between the points predicted by the equation (circles) and the actual values reported by patients (black lines). Note that the scale on the y-axes vary between plots.