Output details

Automatic covariate selection in logistic models for chest pain diagnosis: A new approach

Improving decision making in management of chest pain is a strongly active area. This international collaboration provides world-leading results in data-driven techniques in terms of diagnostic performance, probabilistic calibration and portability from hospital-to-hospital. This paper overcomes the known sub-optimality of conventional model building so that covariate selection becomes part of the optimization and avoids the arbitrariness of forward/backward selection. Resulting models are compact, requiring less user interaction (crucial to professional acceptance), with only marginal compromise in performance across four hospitals.