全球智库信息集成服务系统

Key Findings:

A successful predictive modeling strategy depends upon identifying (1) a precisely defined health care problem, (2) a promising intervention, (3) the appropriateness of a targeted approach in implementing the intervention, and (4) sufficient data and analytic capacity to develop and validate a predictive model. Importantly, only one of these four steps is statistical in nature. Once the decision has been made to build a predictive model, it is critical to remember that (1) the resulting stratification is akin to a screening—not a diagnostic—test, often requiring further risk assessment and/or monitoring; (2) there is a always a trade-off between false positives and false negatives when deciding where in the predicted risk distribution to intervene; and (3) underlying risk of elevated health care need is dynamic, as such it is important to reassess risk at clinically relevant intervals and avoid evaluating interventions based on simple pre-post comparisons of the high-risk intervention group.