Validating Biobehavioral Technologies for Use in Clinical Psychiatry

Area of Science:

• Psychiatry
• Biomedical Engineering
• Data Science

Background:

• Advanced biobehavioral and genetic measures offer potential for psychiatric disorder insight.
• Clinical sciences face challenges implementing objective measures, often due to flawed psychometric evaluation.
• Traditional reliability and validity metrics are insufficient for novel biobehavioral technologies.

Purpose of the Study:

• To define and evaluate "resolution" as a critical metric for psychiatric measurement.
• To demonstrate the impact of resolution on the reliability and validity of acoustic features for predicting self-injurious thoughts/behaviors (SITB).

Main Methods:

• Acoustic features and self-reported symptoms were collected from 124 psychiatric patients via smartphone and clinical interviews.
• Analysis focused on the influence of temporal (minutes, weeks) and spatial (smartphone vs. interview) resolution on data.
• Machine learning models were used to predict SITB from acoustic features at varying resolutions.

Main Results:

• Acoustic feature reliability was unstable until temporal/spatial resolution was specified.
• Prediction accuracy for SITB reached ~87% but was dependent on resolution.
• Model generalizability required "temporally-matched" resolution between acoustic and SITB measures.

Conclusions:

• Resolution is a crucial, yet often overlooked, factor in evaluating psychiatric measures.
• Optimizing resolution is essential for reliable and valid biobehavioral data in clinical psychiatry.
• Careful consideration of resolution is necessary to unlock the potential of biobehavioral technologies.