Adequate sample size for developing prediction models is not simply related to events per variable

Higher EPV is needed when low-prevalence predictors are present in a model to eliminate bias in regression coefficients and improve predictive accuracy.