Objectives: Clinical prediction rules and models are developed by applying statistical techniques to find combinations of predictors that categorize a heterogeneous group of patients into subgroups of risk. Our goal is to teach clinicians how to evaluate the validity, results, and applicability of articles describing clinical prediction tools. CLINICAL EXAMPLE: An article describing a rule to predict the need for intensive care unit care admission in patients presenting to the emergency room with chest pain.
Recommendations: Valid clinical prediction tools are developed by completely following up a representative group of patients, by evaluating all potential predictors and testing the independent contribution of each predictor variable, and by ensuring that the outcomes were independent of the predictors. To evaluate the results of an article describing a clinical prediction tool, clinicians need to know what the prediction tool is, how well it categorizes patients into different levels of risk, and what the confidence intervals are around the risk estimates. Valid prediction tools are not applicable in every patient population. Before patient care application, the clinician should ensure that the tool maintains its prediction power in a new sample of patients, that the patients are similar to patients used to test the tool, and that the tool has been shown to improve clinical decision-making.
Conclusions: There has been an increase in the development and validation of clinical prediction rules and models. It is important to evaluate the validity and reliability of these prediction tools before application.