Using predictive analytics to improve pragmatic trial design.

Area of Science:

• Health Services Research
• Clinical Trial Design
• Mental Health Analytics

Background:

• Embedded clinical trials leverage real-world data and automated systems for large-scale randomization and outcome assessment.
• A previous suicide prevention trial randomized 18,868 individuals using patient-reported data on thoughts of death or self-harm (Patient Health Questionnaire item 9).

Purpose of the Study:

• To evaluate the efficiency of using predictive analytics for suicide risk estimation in improving clinical trial design.
• To compare trial designs based on different eligibility criteria: Patient Health Questionnaire item 9 responses versus suicide risk prediction scores.

Main Methods:

• Utilized outpatient mental health visit data from seven health systems between January 1, 2009, and September 30, 2017.
• Compared eligibility criteria including Patient Health Questionnaire item 9 scores (2-3 or 3) and suicide risk score percentiles (90th, 95th, 99th).
• Assessed sample size requirements for detecting relative reductions in 90-day suicide attempt rates across criteria.

Main Results:

• Out of 24,355,599 outpatient visits, 86.3% lacked Patient Health Questionnaire data.
• Suicide risk score criteria (90th-99th percentile) identified higher proportions of the eligible population compared to Patient Health Questionnaire item 9 criteria.
• The 99th percentile risk score criterion required randomizing only 19.7% of the eligible population, versus 40.2% for a Patient Health Questionnaire item 9 score of 2-3.

Conclusions:

• Predictive analytics for suicide risk can enhance clinical trial efficiency by identifying more individuals with elevated risk than traditional patient-reported measures.
• Further research is needed to determine if individuals identified via predictive analytics respond differently to interventions compared to those identified through standard methods.