< Back to previous page

Project

Signaling changes in experience sampling data in real-time: Development and evaluation of statistical process control methods for detecting early warning signals of depression

Detecting early warning signals of impending depressive episodes would pave the way for timely intervention. Since depression is characterized by altered affective experiences, recent studies have investigated whether experience sampling (ESM) data regarding affective experiences in daily life may yield such early warning signals. In ESM studies, participants are instructed to report on their momentary affect at a number of measurement occasions throughout the day, for multiple days. So far, detection of early warning signals in ESM data has mainly been addressed from a retrospective perspective, after all data have been collected. These studies point to changes in the mean and variance of the ESM data as potentially informative warning signals.

However, for effective prevention, it is crucial to detect such changes in real-time during data collection, rather than after the fact. Statistical process control (SPC) procedures, originally developed for monitoring quality characteristics of manufacturing processes, seem promising tools. SPC procedures would allow us to evaluate at each measurement occasion whether reported affective experiences fall within the person's own normal range or if a change has occurred. Such a change might signal an impendent depressive episode and thus call for intervention. However, affective ESM data also violate major assumptions of the SPC procedures: The observations are not independent across time, are often skewed distributed, are characterized by missingness, and insufficient data regarding a person's normal range is available.

In this dissertation, we therefore investigate the potential of SPC for the timely detection of imminent depressive episodes. In the first part (Chapters 2 and 3) of this dissertation, we focus on detecting mean changes in ESM data. In Chapter 2, we evaluate how well six popular univariate and multivariate SPC procedures succeed in signaling mean changes in simulated data with typical ESM characteristics (i.e., skewed distribution, autocorrelation, missing data). Results showed that the exponentially weighted moving average (EWMA) procedure performs especially well, and that it is beneficial to monitor day averages rather than the scores on the individual measurement occasions. The remaining chapters build on these results. In Chapter 3, the potential of the EWMA procedure is showcased by means of three different applications in psychology. The second part (Chapters 4 and 5) of this dissertation targets the detection of variance changes (on top of mean changes). In Chapter 4, we study whether monitoring day statistics of variability outperforms existing SPC approaches for detecting variance changes in simulated data. In Chapter 5, we investigate whether monitoring the variability of mental states, on top of the mean, can further improve the signaling of an impending depressive relapse, using the TRANS-ID data. In the third part (Chapter 6) of this dissertation, we compare person-specific and multilevel modeling approaches to deriving EWMA control limits in the case limited baseline data (i.e., data where a person's affect is within normal range) are available.

Date:1 Oct 2019 →  18 Sep 2023
Keywords:Statistical Process Control
Disciplines:Statistics and data analysis
Project type:PhD project