MedDecide 01: Introduction to Medical Decision Analysis for Pathologists

Introduction

The meddecide module provides comprehensive tools for medical decision analysis, enabling healthcare professionals to make evidence-based clinical decisions through systematic evaluation of diagnostic tests, treatment options, and health economic outcomes. This module combines traditional decision tree analysis with advanced Markov modeling for complex medical scenarios.

Learning Objectives:

Understand the fundamentals of medical decision analysis
Learn diagnostic test evaluation and ROC analysis
Master decision tree construction and interpretation
Apply Markov modeling for chronic disease scenarios
Conduct cost-effectiveness analysis for healthcare interventions
Implement precision medicine decision frameworks

Module Overview

meddecide encompasses four main areas of medical decision analysis:

1. Diagnostic Test Evaluation

ROC Analysis: Receiver operating characteristic curves and AUC calculation
Diagnostic Accuracy: Sensitivity, specificity, predictive values
Test Performance: Likelihood ratios and clinical utility measures
Biomarker Validation: Companion diagnostic development

2. Decision Tree Analysis

Simple Decision Trees: Binary choice scenarios with immediate outcomes
Complex Decision Trees: Multi-branch decisions with sequential outcomes
Cost-Effectiveness Trees: Economic evaluation integrated with clinical outcomes
Sensitivity Analysis: Parameter uncertainty assessment

3. Markov Modeling

State Transition Models: Chronic disease progression modeling
Markov Chains: Time-dependent health state transitions
Cohort Simulation: Population-level outcome prediction
Economic Evaluation: Long-term cost-effectiveness analysis

4. Precision Medicine Decisions

Biomarker-Guided Therapy: Treatment selection algorithms
Companion Diagnostics: Test-treatment combinations
Pharmacoeconomics: Personalized medicine cost-effectiveness
Real-World Evidence: Population-based decision modeling

library(ClinicoPath)
library(dplyr)
library(pROC)

Getting Started

Installation and Setup

meddecide is part of the comprehensive ClinicoPath suite and provides specialized tools for medical decision analysis.

In jamovi:

Install the ClinicoPath module from the jamovi library
Navigate to meddecide in the analysis menu
Choose from: Agreement, Decision, ROC, or Power Analysis

In R:

# Install from GitHub
if (!requireNamespace("devtools", quietly = TRUE)) {
  install.packages("devtools")
}
devtools::install_github("sbalci/meddecide")

# Load the package
library(ClinicoPath)

Sample Datasets

meddecide includes comprehensive datasets for decision analysis training:

# Load decision analysis datasets
data(basic_decision_data)
data(markov_decision_data) 
data(precision_oncology_data)

# Decision tree data overview
cat("Basic decision data dimensions:", nrow(basic_decision_data), "×", ncol(basic_decision_data), "\n")
cat("Key variables:", paste(names(basic_decision_data)[1:8], collapse = ", "), "...\n")

# Markov model data overview
cat("Markov decision data dimensions:", nrow(markov_decision_data), "×", ncol(markov_decision_data), "\n")
cat("Key variables:", paste(names(markov_decision_data)[1:8], collapse = ", "), "...\n")

# Precision oncology data overview
cat("Precision oncology data dimensions:", nrow(precision_oncology_data), "×", ncol(precision_oncology_data), "\n")
molecular_vars <- names(precision_oncology_data)[grepl("Mutation|Status|TPS", names(precision_oncology_data))]
cat("Biomarker variables:", paste(molecular_vars[1:5], collapse = ", "), "...\n")

Core Analysis Workflows

1. ROC Analysis and Diagnostic Test Evaluation

Foundation for biomarker validation and diagnostic test assessment.

# ROC Analysis Example
# In jamovi: meddecide > ROC > ROC Analysis

# Using precision oncology data for biomarker evaluation
data(precision_oncology_data)

# Create binary outcome for treatment response
response_binary <- as.numeric(precision_oncology_data$Treatment_Response %in% 
                             c("Complete_Response", "Partial_Response"))

cat("ROC Analysis Example - PD-L1 TPS for Treatment Response\n")
cat("=====================================================\n\n")

# Basic ROC statistics
if(requireNamespace("pROC", quietly = TRUE)) {
  library(pROC)
  
  # ROC analysis for PD-L1 TPS
  roc_pdl1 <- roc(response_binary, precision_oncology_data$PD_L1_TPS, quiet = TRUE)
  
  cat("PD-L1 TPS ROC Results:\n")
  cat("AUC:", round(auc(roc_pdl1), 3), "\n")
  cat("95% CI:", paste(round(ci.auc(roc_pdl1), 3), collapse = " - "), "\n")
  
  # Optimal cutpoint
  optimal_cutpoint <- coords(roc_pdl1, "best", ret = "threshold")
  cat("Optimal cutpoint:", round(optimal_cutpoint, 1), "%\n")
  
  # Clinical cutpoints evaluation
  cutpoints <- c(1, 10, 20, 50)
  cat("\nPerformance at Clinical Cutpoints:\n")
  
  for(cutpoint in cutpoints) {
    sens <- coords(roc_pdl1, cutpoint, ret = "sensitivity")
    spec <- coords(roc_pdl1, cutpoint, ret = "specificity")
    cat(paste0("PD-L1 ≥", cutpoint, "%: Sensitivity = ", round(sens, 3), 
              ", Specificity = ", round(spec, 3), "\n"))
  }
}

# Diagnostic test characteristics
n_positive <- sum(precision_oncology_data$PD_L1_TPS >= 50)
n_total <- nrow(precision_oncology_data)
prevalence <- round(n_positive / n_total * 100, 1)

cat("\nDiagnostic Test Characteristics:\n")
cat("PD-L1 ≥50% prevalence:", prevalence, "%\n")
cat("Sample size:", n_total, "patients\n")

2. Simple Decision Tree Analysis

Evaluate treatment alternatives with immediate outcomes.

# Decision Tree Analysis Example
# In jamovi: meddecide > Decision > Decision Tree Graph

# Using basic decision data
data(basic_decision_data)

cat("Decision Tree Analysis Example - Surgery vs Medical Treatment\n")
cat("===========================================================\n\n")

# Decision analysis summary
decision_summary <- basic_decision_data %>%
  group_by(treatment) %>%
  summarise(
    n = n(),
    mean_prob_success = round(mean(prob_success_surgery + prob_success_medical)/2, 3),
    mean_cost = round(mean(cost_surgery + cost_medical)/2, 0),
    mean_utility_success = round(mean(utility_success), 3),
    clinical_response_rate = round(mean(clinical_outcome == "Complete_Response") * 100, 1),
    .groups = 'drop'
  )

cat("Decision Alternatives:\n")
print(decision_summary)

# Expected value calculation example
surgery_cases <- basic_decision_data[basic_decision_data$treatment == "Surgery", ]
medical_cases <- basic_decision_data[basic_decision_data$treatment == "Medical Treatment", ]

if(nrow(surgery_cases) > 0 && nrow(medical_cases) > 0) {
  # Surgery expected values
  surgery_ev_cost <- mean(surgery_cases$cost_surgery * surgery_cases$prob_success_surgery + 
                         surgery_cases$cost_complications * surgery_cases$prob_complications)
  
  surgery_ev_utility <- mean(surgery_cases$utility_success * surgery_cases$prob_success_surgery + 
                            surgery_cases$utility_failure * (1 - surgery_cases$prob_success_surgery))
  
  # Medical treatment expected values  
  medical_ev_cost <- mean(medical_cases$cost_medical * medical_cases$prob_success_medical)
  
  medical_ev_utility <- mean(medical_cases$utility_success * medical_cases$prob_success_medical + 
                            medical_cases$utility_failure * (1 - medical_cases$prob_success_medical))
  
  cat("\nExpected Value Analysis:\n")
  cat("Surgery - Expected Cost: $", round(surgery_ev_cost, 0), 
      ", Expected Utility: ", round(surgery_ev_utility, 3), "\n")
  cat("Medical - Expected Cost: $", round(medical_ev_cost, 0), 
      ", Expected Utility: ", round(medical_ev_utility, 3), "\n")
  
  # Cost-effectiveness ratio
  if(medical_ev_utility != surgery_ev_utility) {
    icer <- (surgery_ev_cost - medical_ev_cost) / (surgery_ev_utility - medical_ev_utility)
    cat("Incremental Cost-Effectiveness Ratio: $", round(abs(icer), 0), " per utility unit\n")
  }
}

3. Markov Model Analysis

Long-term disease progression modeling for chronic conditions.

# Markov Model Analysis Example
# In jamovi: meddecide > Decision > Decision Tree Graph (Markov Model Tree)

# Using markov decision data
data(markov_decision_data)

cat("Markov Model Analysis Example - Chronic Disease Management\n")
cat("=========================================================\n\n")

# Treatment strategy comparison
strategy_summary <- markov_decision_data %>%
  group_by(treatment_strategy) %>%
  summarise(
    n_states = n(),
    mean_healthy_to_sick = round(mean(prob_healthy_to_sick), 4),
    mean_sick_to_recovered = round(mean(prob_sick_to_recovered), 4),
    annual_cost_healthy = round(mean(cost_healthy_state), 0),
    annual_cost_sick = round(mean(cost_sick_state), 0),
    annual_utility_healthy = round(mean(utility_healthy), 3),
    annual_utility_sick = round(mean(utility_sick), 3),
    .groups = 'drop'
  )

cat("Treatment Strategy Comparison:\n")
print(strategy_summary)

# Transition probability validation
cat("\nTransition Probability Validation:\n")
sample_state <- markov_decision_data[1, ]
prob_sum <- sample_state$prob_healthy_to_sick + (1 - sample_state$prob_healthy_to_sick)
cat("Probability sum check (should = 1.0):", prob_sum, "\n")

# Long-term outcome simulation (simplified)
time_horizon <- 10  # years
initial_cohort <- 1000

cat("\nSimulated 10-Year Outcomes (per 1000 patients):\n")
for(strategy in unique(markov_decision_data$treatment_strategy)) {
  strategy_data <- markov_decision_data[markov_decision_data$treatment_strategy == strategy, ][1, ]
  
  # Simplified simulation
  annual_transition_rate <- strategy_data$prob_healthy_to_sick
  annual_recovery_rate <- strategy_data$prob_sick_to_recovered
  
  # Estimate patients progressing over time horizon
  patients_progressing <- round(initial_cohort * annual_transition_rate * time_horizon)
  
  # Estimate total costs
  total_cost <- strategy_data$cost_healthy_state * time_horizon + 
                patients_progressing * strategy_data$cost_sick_state
  
  cat(paste0(strategy, ": ", patients_progressing, " progressions, $", 
            round(total_cost, 0), " total cost\n"))
}

4. Precision Medicine Decision Analysis

Biomarker-guided treatment selection with cost-effectiveness evaluation.

# Precision Medicine Decision Analysis
# Combining diagnostic testing with treatment selection

data(precision_oncology_data)

cat("Precision Medicine Decision Analysis\n")
cat("===================================\n\n")

# Biomarker testing strategy evaluation
biomarker_strategy <- precision_oncology_data %>%
  mutate(
    # Define testing strategy
    test_egfr = TRUE,  # Universal EGFR testing
    test_pdl1 = TRUE,  # Universal PD-L1 testing
    
    # Treatment assignment based on biomarkers
    recommended_treatment = case_when(
      EGFR_Mutation == "Positive" ~ "EGFR_TKI",
      PD_L1_TPS >= 50 ~ "Anti_PD1_Mono",
      PD_L1_TPS >= 1 ~ "Combo_Immuno",
      TRUE ~ "Chemotherapy"
    ),
    
    # Estimate treatment costs
    treatment_cost = case_when(
      recommended_treatment == "EGFR_TKI" ~ 120000,
      recommended_treatment == "Anti_PD1_Mono" ~ 150000,
      recommended_treatment == "Combo_Immuno" ~ 200000,
      TRUE ~ 80000
    ),
    
    # Testing costs
    testing_cost = 3000,  # Comprehensive biomarker panel
    
    # Total cost
    total_cost = treatment_cost + testing_cost
  )

# Strategy outcomes
strategy_outcomes <- biomarker_strategy %>%
  group_by(recommended_treatment) %>%
  summarise(
    n_patients = n(),
    percentage = round(n() / nrow(biomarker_strategy) * 100, 1),
    response_rate = round(mean(Treatment_Response %in% c("Complete_Response", "Partial_Response")) * 100, 1),
    median_pfs = median(PFS_Months),
    mean_cost = round(mean(total_cost), 0),
    .groups = 'drop'
  )

cat("Biomarker-Guided Treatment Strategy Outcomes:\n")
print(strategy_outcomes)

# Cost-effectiveness summary
total_responses <- sum(biomarker_strategy$Treatment_Response %in% c("Complete_Response", "Partial_Response"))
total_cost <- sum(biomarker_strategy$total_cost)
cost_per_response <- round(total_cost / total_responses, 0)

cat("\nOverall Strategy Performance:\n")
cat("Total patients:", nrow(biomarker_strategy), "\n")
cat("Total responses:", total_responses, "\n")
cat("Response rate:", round(total_responses / nrow(biomarker_strategy) * 100, 1), "%\n")
cat("Cost per response: $", cost_per_response, "\n")

# Compare with standard care (no biomarker testing)
standard_care_response_rate <- 25  # Assumed standard chemotherapy response rate
standard_care_cost <- 80000  # Standard chemotherapy cost
cost_per_response_standard <- round(standard_care_cost / (standard_care_response_rate/100), 0)

cat("\nComparison with Standard Care:\n")
cat("Standard care cost per response: $", cost_per_response_standard, "\n")
cat("Biomarker strategy cost per response: $", cost_per_response, "\n")

if(cost_per_response < cost_per_response_standard) {
  cat("Biomarker strategy is more cost-effective\n")
} else {
  cat("Standard care is more cost-effective\n")
}

Advanced Applications

Agreement Analysis

Evaluate diagnostic test reliability and inter-observer agreement.

cat("Agreement Analysis Applications\n")
cat("==============================\n\n")

# Load biomarker validation data for agreement analysis
data(biomarker_validation_study)

# Multi-platform agreement
platform_agreement <- biomarker_validation_study %>%
  summarise(
    n_cases = n(),
    correlation = round(cor(Platform_A_PD_L1, Platform_B_PD_L1), 3),
    mean_difference = round(mean(Platform_B_PD_L1 - Platform_A_PD_L1), 2),
    agreement_within_10pct = round(mean(abs(Platform_B_PD_L1 - Platform_A_PD_L1) <= 10) * 100, 1),
    overall_agreement = round(mean(Platform_Agreement) * 100, 1)
  )

cat("Multi-Platform Biomarker Agreement:\n")
cat("Cases analyzed:", platform_agreement$n_cases, "\n")
cat("Correlation:", platform_agreement$correlation, "\n")
cat("Mean difference:", platform_agreement$mean_difference, "%\n")
cat("Agreement within ±10%:", platform_agreement$agreement_within_10pct, "%\n")
cat("Overall agreement:", platform_agreement$overall_agreement, "%\n")

# Clinical significance
if(platform_agreement$correlation >= 0.90 && platform_agreement$overall_agreement >= 85) {
  cat("Conclusion: Platforms show excellent agreement for clinical use\n")
} else {
  cat("Conclusion: Platform harmonization may be needed\n")
}

Power Analysis

Sample size and statistical power calculations for diagnostic studies.

cat("\nPower Analysis for Diagnostic Studies\n")
cat("====================================\n\n")

# Sample size calculation for ROC analysis
# Based on AUC difference detection

auc_null <- 0.5      # Null hypothesis (no discrimination)
auc_alternative <- 0.75  # Alternative hypothesis (good discrimination)
alpha <- 0.05        # Type I error rate
power <- 0.80        # Desired power

# Simplified sample size estimation
# Using normal approximation for AUC
z_alpha <- qnorm(1 - alpha/2)  # 1.96 for 95% confidence
z_beta <- qnorm(power)         # 0.84 for 80% power

# Approximate sample size (simplified formula)
estimated_n <- ((z_alpha + z_beta)^2 * 2) / ((auc_alternative - auc_null)^2 * 4)

cat("Sample Size Estimation for ROC Study:\n")
cat("Null AUC:", auc_null, "\n")
cat("Alternative AUC:", auc_alternative, "\n") 
cat("Alpha:", alpha, "\n")
cat("Power:", power, "\n")
cat("Estimated sample size:", round(estimated_n), "per group\n")
cat("Total sample size:", round(estimated_n * 2), "\n")

# Recommendations
if(estimated_n < 50) {
  cat("Recommendation: Small study feasible\n")
} else if(estimated_n < 200) {
  cat("Recommendation: Moderate-sized study required\n")
} else {
  cat("Recommendation: Large study required, consider multi-center collaboration\n")
}

Integration with Other Modules

meddecide works seamlessly with other ClinicoPath modules:

Connection to ClinicoPathDescriptives

Descriptive statistics inform decision model parameters
Quality metrics validate input data reliability
Baseline characteristics define patient populations

Connection to jsurvival

Survival analysis provides time-to-event data for Markov models
Hazard ratios inform transition probabilities
Survival curves validate decision model outcomes

Connection to jjstatsplot

Statistical plots support decision analysis communication
ROC curves and forest plots enhance interpretation
Sensitivity analysis visualization

Best Practices

Decision Model Development

Problem Definition: Clearly define the decision context and alternatives
Model Structure: Choose appropriate model type (tree vs. Markov)
Parameter Estimation: Use high-quality data sources
Validation: Test model predictions against real-world outcomes
Sensitivity Analysis: Assess parameter uncertainty impact

ROC Analysis Guidelines

Study Design: Prospective cohort with appropriate controls
Sample Size: Adequate power for clinically meaningful differences
Cutpoint Selection: Balance sensitivity and specificity based on clinical needs
Validation: Independent validation cohort required
Clinical Utility: Demonstrate impact on patient outcomes

Economic Evaluation

Perspective: Clearly state economic perspective (payer, societal)
Time Horizon: Appropriate to capture all relevant costs and benefits
Discounting: Apply standard discount rates for future costs/benefits
Uncertainty: Comprehensive sensitivity and scenario analysis
Transparency: Report all assumptions and data sources

Conclusion

The meddecide module provides comprehensive tools for evidence-based medical decision making, from simple diagnostic test evaluation to complex health economic modeling. The integration of traditional decision analysis with modern precision medicine applications makes it an essential tool for contemporary healthcare research and practice.

Next Steps

To explore specific applications in detail:

Precision Oncology: See meddecide-precision-oncology-biomarkers.Rmd
ROC Analysis: See meddecide-roc-analysis.Rmd
Decision Trees: See meddecide-decision-tree-examples.Rmd
Markov Modeling: See meddecide-decision-tree-vs-markov-analysis.Rmd

Support and Resources

Comprehensive Documentation: Detailed function references
Real-World Examples: Clinical case studies and applications
Best Practice Guidelines: Evidence-based recommendations
Community Support: Active user forums and expert guidance

This introduction provides the foundation for using meddecide in medical decision analysis. Explore the detailed vignettes for specific techniques and advanced applications.

ClinicoPath Development Team

2025-06-30