calibrate3: Competing Risk Calibration

In mattwarkentin/precogs: Risk Prediction Model Assessment

Description

This function will allow you to compute calibration metrics: - predicted and observed probabilities - predicted and observed event counts

Usage

calibrate3(data, horizon, ng, predvar, time, cens, cause)

Arguments

data	Dataframe to use for absolute-risk predictions
horizon	Time horizon for predictions
ng	Number of deciles based on model-predicted risks (default is 10)
predvar	Unquoted name of predicted risks in data frame
time	Time
cens	Censor status
cause	Integer value for the event-of-interest to calibrate prediction

Details

These functions use non-standard evaluation (tidy_eval). Thus, you can pass predvar, time, and cens as unquoted.

calibrate uses ntile to place observations into groups based on predicted risks. calibrate2 uses quantile and cut, instead. This produces calibration metrics equivalent to validstats::cical. Lastly, calibrate3 uses the same discretization as calibrate but uses cuminc instead of validstats::cif to compute the cumulative incidence function.

Value

A tibble containining the decile index along with the predicted and observed cumulative incidences, by decile, and the predicted and observed events across the n risk quantiles. Risk and event count differences by decile are also included.

Author(s)

Matthew T. Warkentin

Examples

## Not run: 
library(riskRegression)
library(prodlim)

# Simulate Data
sim.data <- sampleData(100, outcome = 'competing.risks')

# Fit Fine-Gray Model
fgr.mod <- FGR(formula = Hist(time, event) ~ X6 + X7 + X8 + X9,
               data = sim.data, cause = 1)

# Make predictions
horizon <- median(sim.data$time)
sim.data$preds <- predictRisk(fgr.mod, sim.data, cause = 1, time = horizon)

calibrate(sim.data, horizon, 10, preds, time, event, 1)
calibrate2(sim.data, horizon, 10, preds, time, event, 1)
calibrate3(sim.data, horizon, 10, preds, time, event, 1)

## End(Not run)

mattwarkentin/precogs documentation built on Jan. 12, 2020, 6:24 p.m.