Detecting warning signs for psychopathology in real time while accounting for context: Two novel statistical process control applications

Statistical process control (SPC) may detect whether and when repeatedly assessed emotions reach unusual levels, which holds promise for the real-time detection of imminent depression. However, SPC does not account for contextual effects on emotions, such as people feeling systematically worse during stressful events and better during weekends. This may cause false alarms (e.g., presence of warning signs during stressful events) as well as false negatives (e.g., absence of warning signs during weekends). We therefore present two novel context-sensitive SPC methods, which adjust the monitored score according to contextual factors. A simulation study showed that these context-sensitive methods outperform the standard SPC method when contextual effects are large while the effect of depression on emotions is relatively small, but lose their advantage when contextual factors are biased. An empirical illustration confirmed these findings. Context-sensitive SPC methods are thus recommended when contextual factors can be accurately pinpointed, which may hold, for instance, for location and temporal cycles (seasons, menstrual cycles, week/weekend days).