Re-evaluation of the comparative effectiveness of bootstrap-based optimism correction methods in the development of multivariable clinical prediction models

Multivariable predictive models are important statistical tools for providing synthetic diagnosis and prognostic algorithms based on multiple patients' characteristics. Their apparent discriminant and calibration measures usually have overestimation biases (known as 'optimism') relative to the actual performances for external populations. Existing statistical evidence and guidelines suggest that three bootstrap-based bias correction methods are preferable in practice, namely Harrell's bias correction and the .632 and .632+ estimators. Although Harrell's method has been widely adopted in clinical studies, simulation-based evidence indicates that the .632+ estimator may perform better than the other two methods. However, there is limited evidence and these methods' actual comparative effectiveness is still unclear. In this article, we conducted extensive simulations to compare the effectiveness of these methods, particularly using the following modern regression models: conventional logistic regression, stepwise variable selections, Firth's penalized likelihood method, ridge, lasso, and elastic-net. Under relatively large sample settings, the three bootstrap-based methods were comparable and performed well. However, all three methods had biases under small sample settings, and the directions and sizes of the biases were inconsistent. In general, the .632+ estimator is recommended, but we provide several notes concerning the operating characteristics of each method.