library(pROC)
# label: 金标准，0 1 变量
# pred: 模型预测值，连续变量

# 方法一
cal_metrics <- function(label, pred){
  roc.p=pROC::roc(label, pred)
  if (roc.p$auc>0.5){
    cutoff=roc.p$thresholds[which.max(roc.p$sensitivities+roc.p$specificities)]
    sensitivity=roc.p$sensitivities[which.max(roc.p$sensitivities+roc.p$specificities)]
    specificity=roc.p$specificities[which.max(roc.p$sensitivities+roc.p$specificities)]
    df=data.frame(type='positive classification',
                  auc=round(roc.p$auc,3),cutoff=cutoff,
                  sensitivity=sensitivity,specificity=specificity)
    return(df)
  }
  else{
    cutoff=roc.p$thresholds[which.min(roc.p$sensitivities+roc.p$specificities)]
    sensitivity=roc.p$sensitivities[which.min(roc.p$sensitivities+roc.p$specificities)]
    specificity=roc.p$specificities[which.min(roc.p$sensitivities+roc.p$specificities)]
    df=data.frame(type='negative classification',
                  auc=1-round(roc.p$auc,3),cutoff=cutoff,
                  sensitivity=1-sensitivity,specificity=1-specificity)
    return(df)
  }
}

# 方法二：
roc.p=pROC::roc(label, pred)
cutoff=roc.p$thresholds[which.max(roc.p$sensitivities+roc.p$specificities)]

pred_b = as.integer(pred > cutoff)
cft = table(pred_b, label)

cm = confusionMatrix(cft, positive = "1", mode = "everything")
paste0('cutoff: ', cutoff)
print(roc.p$auc)
print(cm)

参考文献

1. https://github.com/cran/cutoff/blob/master/R/roc.R
2. https://www.jianshu.com/p/c61feeb8b36f