Missing Data Analysis and Multiple Imputation

Comprehensive missing data analysis and multiple imputation using mice and ggmice packages. This function provides a complete workflow for analyzing missing data patterns, performing multiple imputation by chained equations (MICE), and evaluating imputation quality. Designed specifically for clinical research applications where missing data is common and proper handling is critical for valid statistical inference.

Comprehensive clinical prediction model builder with advanced validation and performance assessment. Creates multiple logistic regression models optimized for integration with Decision Curve Analysis. Provides robust error handling, comprehensive validation, and clinical interpretation guidance.

Usage

x %||% y

x %||% y

Details

The missing data analysis function provides three main analysis types:

1. Pattern Analysis: Explores missing data structure and patterns

2. Multiple Imputation: Performs MICE imputation with convergence diagnostics

3. Complete Analysis: Combines pattern analysis and imputation

Key features include:

• Visual and tabular missing data pattern analysis

• Multiple imputation methods (PMM, Bayesian regression, logistic regression)

• Convergence diagnostics with trace plots

• Quality evaluation comparing observed vs imputed data

• Flexible parameter customization

• Clinical research focused interpretations

Common clinical applications:

• Data quality assessment for clinical trials

• Missing data handling in observational studies

• Regulatory compliance for pharmaceutical research

• Sensitivity analysis for missing data assumptions

The Prediction Model Builder supports multiple modeling approaches:

• Basic Clinical Models: Core demographic and primary risk factors

• Enhanced Clinical Models: Extended clinical variables and interactions

• Biomarker Models: Integration of laboratory values and advanced diagnostics

• Custom Models: User-defined variable combinations

Key features include:

• Automatic data splitting for unbiased validation

• Advanced missing data handling with multiple imputation

• Comprehensive performance metrics (AUC, calibration, NRI, IDI)

• Cross-validation and bootstrap validation

• Stepwise selection and penalized regression

• Seamless integration with Decision Curve Analysis

• Clinical risk score generation

• Robust error handling and validation

Examples

if (FALSE) { # \dontrun{
# Basic pattern analysis
result <- missingdata(
  data = clinical_data,
  analysis_vars = c("age", "bmi", "biomarker"),
  analysis_type = "pattern"
)

# Multiple imputation
result <- missingdata(
  data = clinical_data,
  analysis_vars = c("age", "bmi", "biomarker"),
  analysis_type = "imputation",
  n_imputations = 10,
  imputation_method = "pmm"
)

# Complete analysis
result <- missingdata(
  data = clinical_data,
  analysis_vars = c("age", "bmi", "biomarker"),
  analysis_type = "complete",
  n_imputations = 5,
  max_iterations = 10
)
} # }

if (FALSE) { # \dontrun{
# Basic clinical model
result <- modelbuilder(
  data = clinical_data,
  outcome = "cardiovascular_event",
  outcomePositive = "Yes",
  basicPredictors = c("age", "sex", "diabetes"),
  buildBasicModel = TRUE
)

# Enhanced model with biomarkers
result <- modelbuilder(
  data = clinical_data,
  outcome = "cardiovascular_event",
  outcomePositive = "Yes",
  biomarkerPredictors = c("age", "sex", "diabetes", "troponin"),
  buildBiomarkerModel = TRUE,
  crossValidation = TRUE,
  splitData = TRUE
)
} # }