Usage
idi(
data,
outcome,
baseline_risk,
new_risk,
time_var,
followup_time = 5,
idi_type = "standard",
discrimination_measure = "mean_diff",
trim_proportion = 0.1,
confidence_level = 0.95,
bootstrap_samples = 1000,
bootstrap_method = "bca",
hypothesis_test = TRUE,
test_method = "bootstrap",
decompose_idi = TRUE,
risk_distribution = TRUE,
discrimination_slope = TRUE,
relative_improvement = TRUE,
cross_validation = FALSE,
cv_folds = 5,
sensitivity_analysis = FALSE,
outlier_detection = FALSE,
outlier_method = "iqr",
show_summary = TRUE,
show_distributions = TRUE,
show_discrimination = TRUE,
show_validation = FALSE,
plot_risk_distributions = TRUE,
plot_discrimination = TRUE,
plot_scatter = TRUE,
plot_bootstrap = FALSE,
plot_sensitivity = FALSE,
plot_outliers = FALSE,
competing_risks = FALSE,
stratified_analysis = FALSE,
subgroup_var,
missing_handling = "complete",
calibration_aware = FALSE,
time_dependent = FALSE,
alpha_level = 0.05,
random_seed = 123
)Arguments
- data
The data as a data frame.
- outcome
Binary outcome variable (0/1) indicating event occurrence. For time-to-event data, this should be event status at the specified time point.
- baseline_risk
Predicted risks or probabilities from the baseline (reference) model. Should be on probability scale (0-1) for optimal IDI interpretation.
- new_risk
Predicted risks or probabilities from the new (enhanced) model that includes additional predictors or biomarkers.
- time_var
Time to event variable for survival data analysis. Required when using time-to-event outcomes with IDI.
- followup_time
Time point for survival analysis (years). Events occurring after this time are censored for IDI calculation.
- idi_type
Type of IDI calculation. Standard uses absolute differences, relative scales by baseline discrimination, scaled normalizes by outcome prevalence, all provides comprehensive assessment.
- discrimination_measure
Method for calculating discrimination differences. Mean difference is standard, median is robust to outliers, trimmed mean balances both approaches, robust uses M-estimators.
- trim_proportion
Proportion of extreme values to trim when using trimmed mean. 0.1 trims 10\
confidence_levelConfidence level for IDI confidence intervals and hypothesis testing. 0.95 provides 95\bootstrap_samplesNumber of bootstrap samples for confidence interval estimation. More samples provide more stable estimates but increase computation time.bootstrap_methodBootstrap confidence interval method. BCa provides bias-corrected intervals with better coverage, studentized accounts for variance changes, percentile is simple and robust.hypothesis_testPerform hypothesis test for IDI = 0. Tests whether the new model provides significant discrimination improvement.test_methodMethod for hypothesis testing. Bootstrap is recommended for most situations, asymptotic assumes normality, permutation is distribution-free, robust handles outliers.decompose_idiDecompose IDI into contributions from events and non-events. Provides insight into which population drives the improvement.risk_distributionAnalyze and compare risk score distributions between baseline and new models for events and non-events separately.discrimination_slopeCalculate discrimination slopes (difference in means) for baseline and new models with confidence intervals.relative_improvementCalculate relative improvement as percentage increase in discrimination over baseline model performance.cross_validationPerform k-fold cross-validation to assess stability of IDI estimates and reduce optimism bias in model comparisons.cv_foldsNumber of folds for cross-validation when enabled. 5-fold CV provides good balance of bias and variance.sensitivity_analysisPerform sensitivity analysis by varying follow-up times and examining IDI stability across different time horizons.outlier_detectionDetect and analyze outliers in risk predictions that may disproportionately influence IDI calculations.outlier_methodMethod for outlier detection. IQR is simple and robust, Z-score assumes normality, modified Z-score is more robust, isolation forest handles complex patterns.show_summaryDisplay comprehensive IDI summary including overall IDI, decomposed components, and statistical significance.show_distributionsDisplay risk distribution summaries for baseline and new models separately for events and non-events.show_discriminationDisplay discrimination slopes and related measures for both baseline and new models with improvements.show_validationDisplay cross-validation and sensitivity analysis results when these options are enabled.plot_risk_distributionsCreate plots showing risk score distributions for baseline and new models separated by outcome status.plot_discriminationVisualize discrimination improvement showing mean differences and IDI components with confidence intervals.plot_scatterCreate scatter plot of baseline vs new model risk scores colored by outcome status to visualize improvement patterns.plot_bootstrapPlot bootstrap distribution of IDI estimates showing sampling variability and confidence interval construction.plot_sensitivityCreate plots showing IDI variation across different analysis parameters and time points.plot_outliersVisualize detected outliers and their influence on IDI calculations with and without outliers.competing_risksAccount for competing risks in survival analysis. Modifies IDI calculation for settings with multiple event types.stratified_analysisPerform stratified IDI analysis by important subgroups to assess consistency of improvement across populations.subgroup_varVariable defining subgroups for stratified analysis. Each level will receive separate IDI calculation.missing_handlingMethod for handling missing data in risk predictions. Multiple imputation provides most robust results.calibration_awareAdjust IDI calculation to account for model calibration. Uses calibrated probabilities for more accurate assessment.time_dependentCalculate time-dependent IDI for survival models showing how discrimination improvement varies over time.alpha_levelType I error rate for hypothesis testing and confidence intervals. Standard value is 0.05 for 95\random_seedRandom seed for bootstrap sampling and cross-validation. Ensures reproducible results across analyses.A results object containing:
Tables can be converted to data frames withresults$instructionsa html results$todoa html results$summarya table results$distributionSummarya table results$discriminationAnalysisa table results$decompositiona table results$validationResultsa table results$subgroupAnalysisa table results$outlierAnalysisa table results$sensitivityTablea table results$riskDistributionsPlotan image results$discriminationPlotan image results$scatterPlotan image results$bootstrapPlotan image results$sensitivityPlotan image results$outliersPlotan image asDForas.data.frame. For example:results$summary$asDFas.data.frame(results$summary)Integrated Discrimination Improvement (IDI) analysis for assessing the incremental value of new biomarkers or risk factors in prediction models. IDI quantifies the improvement in model discrimination by measuring the difference in predicted probabilities between events and non-events for the new model compared to the baseline model. Unlike NRI which focuses on reclassification, IDI provides a numeric measure of discrimination improvement without requiring risk categories. The analysis includes confidence intervals through bootstrap methods, statistical significance testing, and decomposition into event and non-event components. Essential for biomarker validation, model enhancement assessment, and demonstrating clinical utility of new predictors in medical research. Particularly valuable when numeric risk improvement is more relevant than categorical reclassification, such as in precision medicine applications.result <- idi( data = mydata, outcome = "event_indicator", baseline_risk = "baseline_predictions", new_risk = "new_model_predictions", time_point = 5, bootstrap_samples = 1000 )