Clinical Prediction Models & ML Interpretability — clinicalprediction • ClinicoPath

Develop and validate clinical prediction models using machine learning algorithms with interpretability analysis. Includes comprehensive model comparison, feature selection, cross-validation, and explainable AI through SHAP and LIME methods. Designed for clinical research applications with regulatory compliance features for model validation and documentation.

Usage

clinicalprediction(
  data,
  outcome_var,
  predictor_vars,
  patient_id,
  time_var,
  event_var,
  stratify_var,
  model_type = "random_forest",
  problem_type = "classification",
  outcome_type = "binary",
  feature_selection = TRUE,
  selection_method = "recursive_fe",
  feature_engineering = TRUE,
  handle_missing = "mice_imputation",
  train_proportion = 0.7,
  validation_method = "cv_10fold",
  hyperparameter_tuning = TRUE,
  tuning_method = "random_search",
  random_seed = 42,
  interpretability = TRUE,
  shap_analysis = TRUE,
  lime_analysis = FALSE,
  permutation_importance = TRUE,
  partial_dependence = TRUE,
  individual_explanations = FALSE,
  n_explanations = 10,
  performance_metrics = TRUE,
  calibration_analysis = TRUE,
  clinical_metrics = TRUE,
  roc_analysis = TRUE,
  threshold_optimization = TRUE,
  compare_models = FALSE,
  baseline_models = TRUE,
  ensemble_models = FALSE,
  risk_stratification = TRUE,
  n_risk_groups = 3,
  nomogram = TRUE,
  decision_curve = TRUE,
  external_validation = TRUE,
  bootstrap_ci = 1000,
  stability_analysis = TRUE,
  bias_analysis = TRUE,
  detailed_output = TRUE,
  clinical_report = TRUE,
  save_model = FALSE,
  export_predictions = FALSE,
  regulatory_documentation = TRUE
)

Arguments

data: the data as a data frame
outcome_var: Target variable for prediction (binary, continuous, or survival)
predictor_vars: Variables to use as predictors in the model
patient_id: Patient identifier for tracking predictions
time_var: Time to event variable for survival prediction models
event_var: Event indicator for survival prediction models
stratify_var: Variable for stratified sampling and validation
model_type: Type of machine learning model to fit
problem_type: Type of prediction problem
outcome_type: Specific type of outcome variable
feature_selection: Perform automated feature selection
selection_method: Method for automatic feature selection
feature_engineering: Perform automated feature engineering
handle_missing: Method for handling missing data
train_proportion: Proportion of data for training (70 percent = 0.7)
validation_method: Method for model validation
hyperparameter_tuning: Perform automated hyperparameter optimization
tuning_method: Method for hyperparameter tuning
random_seed: Random seed for reproducibility
interpretability: Generate model interpretability analysis
shap_analysis: Generate SHAP (SHapley Additive exPlanations) values
lime_analysis: Generate LIME (Local Interpretable Model-agnostic Explanations)
permutation_importance: Calculate permutation feature importance
partial_dependence: Generate partial dependence plots for key features
individual_explanations: Explain individual predictions using SHAP/LIME
n_explanations: Number of individual predictions to explain in detail
performance_metrics: Calculate comprehensive performance metrics
calibration_analysis: Assess model calibration
clinical_metrics: Calculate clinical decision analysis metrics
roc_analysis: Perform ROC curve analysis with confidence intervals
threshold_optimization: Optimize prediction threshold for clinical use
compare_models: Compare multiple model types
baseline_models: Include simple baseline models for comparison
ensemble_models: Create ensemble of best performing models
risk_stratification: Create risk stratification groups
n_risk_groups: Number of risk stratification groups (e.g., low/medium/high)
nomogram: Create clinical nomogram for risk calculation
decision_curve: Perform decision curve analysis for clinical utility
external_validation: Prepare model for external validation
bootstrap_ci: Number of bootstrap samples for confidence intervals
stability_analysis: Assess model stability across different samples
bias_analysis: Analyze model bias across demographic groups
detailed_output: Include detailed model diagnostics and explanations
clinical_report: Generate clinical interpretation report
save_model: Save trained model for future predictions
export_predictions: Export individual predictions with probabilities
regulatory_documentation: Include documentation for regulatory submission

Value

A results object containing:

`results$overview`					a table
`results$dataset_info`					a table
`results$feature_selection_results`					a table
`results$feature_engineering_summary`					a table
`results$performance_summary`					a table
`results$classification_metrics`					a table
`results$survival_metrics`					a table
`results$feature_importance`					a table
`results$shap_summary`					a table
`results$individual_explanations`					a table
`results$model_comparison`					a table
`results$risk_stratification`					a table
`results$decision_curve_analysis`					a table
`results$cross_validation_results`					a table
`results$stability_analysis`					a table
`results$bias_fairness_analysis`					a table
`results$hyperparameter_results`					a table
`results$clinical_interpretation`					a table
`results$regulatory_summary`					a table
`results$roc_plot`					an image
`results$calibration_plot`					an image
`results$feature_importance_plot`					an image
`results$shap_summary_plot`					an image
`results$shap_dependence_plot`					an image
`results$decision_curve_plot`					an image
`results$risk_distribution_plot`					an image
`results$model_comparison_plot`					an image
`results$stability_plot`					an image
`results$nomogram_plot`					an image

Tables can be converted to data frames with asDF or as.data.frame. For example:

results$overview$asDF

as.data.frame(results$overview)

Examples

# \donttest{
data('clinical_data')
#> Warning: data set ‘clinical_data’ not found

clinicalprediction(
    data = clinical_data,
    outcome_var = "mortality",
    predictor_vars = c("age", "biomarker", "stage"),
    model_type = "random_forest",
    interpretability = TRUE,
    validation_method = "cv_10fold"
)
#> Error in clinicalprediction(data = clinical_data, outcome_var = "mortality",     predictor_vars = c("age", "biomarker", "stage"), model_type = "random_forest",     interpretability = TRUE, validation_method = "cv_10fold"): argument "event_var" is missing, with no default
# }