Detecting mean changes in experience sampling data in real time: A comparison of univariate and multivariate...

Area of Science:

• Psychiatry and Mental Health
• Statistical Monitoring
• Data Science

Background:

• Early detection of mood disorders is crucial for timely intervention.
• Experience Sampling Method (ESM) data offers continuous affective insights but poses statistical challenges.
• Traditional Statistical Process Control (SPC) methods require adaptation for ESM data's unique characteristics (autocorrelation, skewness, missingness).

Purpose of the Study:

• To evaluate the performance of various SPC procedures for detecting mood changes in affective ESM data.
• To adapt and assess SPC methods considering the specific challenges of ESM data, such as missingness and autocorrelation.
• To identify the most effective SPC procedures for early warning signal detection in mood disorder monitoring.

Main Methods:

• Introduced six univariate and multivariate SPC procedures: Shewhart, Hotelling's T², EWMA, MEWMA, CUSUM, and MCUSUM.
• Utilized day averages of affective ESM data to address missingness, autocorrelation, and skewness.
• Evaluated SPC performance on simulated data mimicking ESM features and on real patient data.

Main Results:

• EWMA and CUSUM procedures demonstrated superior performance in detecting small to moderate mean changes compared to Shewhart and Hotelling's T².
• Monitoring day averages proved more effective than using individual measurement occasions for SPC in ESM data.
• (M)EWMA and (M)CUSUM methods are recommended for their robustness with ESM data.

Conclusions:

• EWMA and CUSUM-based SPC methods are effective for early detection of mood disorder warning signals in ESM data.
• Using daily averages is a viable strategy to handle common issues in ESM data for SPC.
• Further research is recommended to optimize SPC for continuous affective monitoring and mood disorder prevention.