Tufts PACE Clinical Predictive Model Registry: update 1990 through 2015.

Area of Science:

• Cardiovascular disease research
• Biostatistics
• Clinical informatics

Background:

• Clinical predictive models (CPMs) aid decision-making by estimating clinical outcome probabilities.
• The Tufts Predictive Analytics and Comparative Effectiveness (PACE) CPM Registry tracks cardiovascular disease (CVD) CPMs.
• The field has seen substantial growth in CPMs since the registry's last update.

Purpose of the Study:

• To systematically review and update the registry of CPMs for CVD.
• To characterize newly available CPMs based on population, performance, covariates, and outcomes.
• To assess trends in CPM development and reporting.

Main Methods:

• Systematic review of CPMs for CVD published from January 1990 to March 2015.
• CVD scope includes coronary artery disease, congestive heart failure, arrhythmias, stroke, venous thromboembolism, and peripheral vascular disease.
• Registry updated to characterize CPMs by population, performance metrics, covariates, and predicted outcomes.

Main Results:

• The updated registry includes 1083 models from 747 articles, a threefold increase from 2005-2014 compared to 1995-2004.
• Most models originate from North American (42%) or European (32%) populations, with significant numbers for coronary artery disease (265) and stroke (158).
• Median reported c-statistic is 0.77, but only 57% of validated models report calibration; discrimination reporting is improving, but calibration reporting is not.

Conclusions:

• A significant redundancy exists among CPMs for various CVD conditions.
• Despite increasing numbers, CPMs often suffer from inadequate reporting of performance and infrequent calibration assessment.
• Further research is necessary to determine the real-world impact of the existing CPM literature.