PredictivePerformance: Calculates various predictive performance measures.

In pjnewcombe/Pmisc: A collection of miscellaneous functions.

Description

Given a list of predictive scores, and a list of corresponding "true" outcomes, this function calculates various predictive performance measures. Note that this function can be used in the context of multiple imputation (if the argument n.mi.chains is used), in which case the performance measures are calculated within each chain, then averaged.

Usage

PredictivePerformance(predictions, outcomes)

Arguments

predictions	A list of predictions for each cross validation fold. If multiple imputation was used this should be a list of lists, the first level corresponding to multiple imputation chains, and the second level to cross validation folds. I.e. predictions[[1]][[2]] would contain predictions from the 1st multiple imputation chain, for patients in the 2nd cross-validation fold.
outcomes	A list of "true" binary outcomes for each cross validation fold. Note the patients ordering must be identical to that in predictions above.

Value

A list containing various measures of predictive performance: rocr - the ROCR base object roc - ROCR object containing ROC curve information ppvTpr - ROCR object containing positive predictive value vs Trupe positive rate information auc - The average ROC auc over all folds (and, if relevant, MI chains) obs.risk.qs.e - Average observed risk in the predicted risk quantiles

Author(s)

Paul Newcombe

pjnewcombe/Pmisc documentation built on March 26, 2020, 2:09 p.m.