Introduction to Decision Panel Optimization

Introduction

The Decision Panel Optimization module in the meddecide package provides a comprehensive framework for optimizing diagnostic test combinations in medical decision-making. This vignette introduces the basic concepts and demonstrates core functionality.

Key Concepts

Testing Strategies

When multiple diagnostic tests are available, they can be combined in different ways:

Single Testing: Use individual tests independently
Parallel Testing: Perform multiple tests simultaneously
- ANY rule (OR): Positive if any test is positive
- ALL rule (AND): Positive only if all tests are positive
- MAJORITY rule: Positive if majority of tests are positive
Sequential Testing: Perform tests in sequence based on previous results
- Stop on first positive
- Confirmatory (require multiple positives)
- Exclusion (require multiple negatives)

Optimization Criteria

The module can optimize test panels based on various criteria:

Accuracy: Overall correct classification rate
Sensitivity: Ability to detect disease (minimize false negatives)
Specificity: Ability to rule out disease (minimize false positives)
Predictive Values: PPV and NPV
Cost-Effectiveness: Balance performance with resource utilization
Utility: Custom utility functions incorporating costs of errors

Installation and Loading

# Install meddecide package
install.packages("meddecide")

# Or install from GitHub
devtools::install_github("ClinicoPath/meddecide")

# Load required packages
library(meddecide)
#> Warning: replacing previous import 'jmvcore::select' by 'dplyr::select' when
#> loading 'meddecide'
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(ggplot2)
#> Warning: package 'ggplot2' was built under R version 4.3.3
library(rpart)
#> Warning: package 'rpart' was built under R version 4.3.3
library(rpart.plot)
library(knitr)
#> Warning: package 'knitr' was built under R version 4.3.3
library(forcats)

Basic Example: COVID-19 Screening

Let’s start with a simple example using COVID-19 screening data:

# Examine the data structure
str(covid_screening_data)
#> 'data.frame':    1000 obs. of  8 variables:
#>  $ patient_id   : int  1 2 3 4 5 6 7 8 9 10 ...
#>  $ rapid_antigen: Factor w/ 2 levels "Negative","Positive": 1 2 1 1 1 1 1 1 1 1 ...
#>  $ pcr          : Factor w/ 2 levels "Negative","Positive": 1 2 NA NA 1 1 NA 1 1 NA ...
#>  $ chest_ct     : Factor w/ 2 levels "Normal","Abnormal": 2 1 1 1 1 1 1 1 1 1 ...
#>  $ symptom_score: num  8 6 1 1 5 5 5 4 2 5 ...
#>  $ covid_status : Factor w/ 2 levels "Negative","Positive": 2 2 1 1 1 1 1 1 1 1 ...
#>  $ age          : num  35 33 39 28 62 32 64 23 58 36 ...
#>  $ risk_group   : Factor w/ 3 levels "High","Low","Medium": 2 2 2 2 2 3 3 3 2 2 ...

# Check disease prevalence
table(covid_screening_data$covid_status)
#> 
#> Negative Positive 
#>      851      149
prop.table(table(covid_screening_data$covid_status))
#> 
#> Negative Positive 
#>    0.851    0.149

Running Basic Analysis

# Basic decision panel analysis
covid_panel <- decisionpanel(
  data = covid_screening_data,
  tests = c("rapid_antigen", "pcr", "chest_ct"),
  testLevels = c("Positive", "Positive", "Abnormal"),
  gold = "covid_status",
  goldPositive = "Positive",
  strategies = "all",
  optimizationCriteria = "accuracy"
)

Interpreting Results

The analysis provides several key outputs:

Individual Test Performance: How each test performs alone
Optimal Panel: The best combination of tests
Strategy Comparison: Performance of different testing approaches
Decision Tree: Optimal sequence for testing

Understanding Testing Strategies

Parallel Testing Example

# Simulate parallel testing with ANY rule
# Positive if rapid_antigen OR pcr is positive
parallel_any <- with(covid_screening_data, 
  rapid_antigen == "Positive" | pcr == "Positive"
)

# Create confusion matrix
conf_matrix_any <- table(
  Predicted = parallel_any,
  Actual = covid_screening_data$covid_status == "Positive"
)

print(conf_matrix_any)
#>          Actual
#> Predicted FALSE TRUE
#>     FALSE   573    1
#>     TRUE     25  134

# Calculate metrics
sensitivity_any <- conf_matrix_any[2,2] / sum(conf_matrix_any[,2])
specificity_any <- conf_matrix_any[1,1] / sum(conf_matrix_any[,1])

cat("Parallel ANY Rule:\n")
#> Parallel ANY Rule:
cat(sprintf("Sensitivity: %.1f%%\n", sensitivity_any * 100))
#> Sensitivity: 99.3%
cat(sprintf("Specificity: %.1f%%\n", specificity_any * 100))
#> Specificity: 95.8%

Sequential Testing Example

# Simulate sequential testing
# Start with rapid test, only do PCR if rapid is positive
sequential_result <- rep("Negative", nrow(covid_screening_data))

# Those with positive rapid test
rapid_pos_idx <- which(covid_screening_data$rapid_antigen == "Positive")

# Among those, check PCR
sequential_result[rapid_pos_idx] <- 
  ifelse(covid_screening_data$pcr[rapid_pos_idx] == "Positive", 
         "Positive", "Negative")

# Create confusion matrix
conf_matrix_seq <- table(
  Predicted = sequential_result == "Positive",
  Actual = covid_screening_data$covid_status == "Positive"
)

print(conf_matrix_seq)
#>          Actual
#> Predicted FALSE TRUE
#>     FALSE   851   51
#>     TRUE      0   98

# Calculate metrics
sensitivity_seq <- conf_matrix_seq[2,2] / sum(conf_matrix_seq[,2])
specificity_seq <- conf_matrix_seq[1,1] / sum(conf_matrix_seq[,1])

cat("\nSequential Testing:\n")
#> 
#> Sequential Testing:
cat(sprintf("Sensitivity: %.1f%%\n", sensitivity_seq * 100))
#> Sensitivity: 65.8%
cat(sprintf("Specificity: %.1f%%\n", specificity_seq * 100))
#> Specificity: 100.0%

# Calculate cost savings
pcr_tests_saved <- sum(covid_screening_data$rapid_antigen == "Negative")
cat(sprintf("PCR tests saved: %d (%.1f%%)\n", 
            pcr_tests_saved, 
            pcr_tests_saved/nrow(covid_screening_data) * 100))
#> PCR tests saved: 876 (87.6%)

Cost-Effectiveness Analysis

When costs are considered, the optimal strategy may change:

# Analysis with costs
covid_panel_cost <- decisionpanel(
  data = covid_screening_data,
  tests = c("rapid_antigen", "pcr", "chest_ct"),
  testLevels = c("Positive", "Positive", "Abnormal"),
  gold = "covid_status",
  goldPositive = "Positive",
  strategies = "all",
  optimizationCriteria = "utility",
  useCosts = TRUE,
  testCosts = "5,50,200",  # Costs for each test
  fpCost = 500,            # Cost of false positive
  fnCost = 5000            # Cost of false negative
)

Visualization

Performance Comparison Plot

# Create performance comparison data
strategies <- data.frame(
  Strategy = c("Rapid Only", "PCR Only", "Parallel ANY", "Sequential"),
  Sensitivity = c(65, 95, 98, 62),
  Specificity = c(98, 99, 97, 99.9),
  Cost = c(5, 50, 55, 15)
)

# Plot sensitivity vs specificity
ggplot(strategies, aes(x = 100 - Specificity, y = Sensitivity)) +
  geom_point(aes(size = Cost), alpha = 0.6) +
  geom_text(aes(label = Strategy), vjust = -1) +
  scale_size_continuous(range = c(3, 10)) +
  xlim(0, 5) + ylim(60, 100) +
  labs(
    title = "Testing Strategy Comparison",
    x = "False Positive Rate (%)",
    y = "Sensitivity (%)",
    size = "Cost ($)"
  ) +
  theme_minimal()

Decision Trees

Decision trees provide clear algorithms for clinical use:

# Generate decision tree
covid_tree <- decisionpanel(
  data = covid_screening_data,
  tests = c("rapid_antigen", "pcr", "chest_ct", "symptom_score"),
  testLevels = c("Positive", "Positive", "Abnormal", ">5"),
  gold = "covid_status",
  goldPositive = "Positive",
  createTree = TRUE,
  treeMethod = "cart",
  maxDepth = 3
)

Interpreting the Tree

A typical decision tree output might look like:

1. Start with Rapid Antigen Test
   ├─ If Positive (2% of patients)
   │  └─ Confirm with PCR
   │     ├─ If Positive → COVID Positive (PPV: 95%)
   │     └─ If Negative → COVID Negative (NPV: 98%)
   └─ If Negative (98% of patients)
      ├─ If Symptoms > 5
      │  └─ Perform Chest CT
      │     ├─ If Abnormal → Perform PCR
      │     └─ If Normal → COVID Negative
      └─ If Symptoms ≤ 5 → COVID Negative

Advanced Features

Cross-Validation

Validate panel performance using k-fold cross-validation:

# Run with cross-validation
covid_panel_cv <- decisionpanel(
  data = covid_screening_data,
  tests = c("rapid_antigen", "pcr", "chest_ct"),
  testLevels = c("Positive", "Positive", "Abnormal"),
  gold = "covid_status",
  goldPositive = "Positive",
  crossValidate = TRUE,
  nFolds = 5,
  seed = 123
)

Bootstrap Confidence Intervals

Get uncertainty estimates for performance metrics:

# Run with bootstrap
covid_panel_boot <- decisionpanel(
  data = covid_screening_data,
  tests = c("rapid_antigen", "pcr", "chest_ct"),
  testLevels = c("Positive", "Positive", "Abnormal"),
  gold = "covid_status",
  goldPositive = "Positive",
  bootstrap = TRUE,
  bootReps = 1000,
  seed = 123
)

Best Practices

Start Simple: Begin with individual test performance before combinations
Consider Context: Screening vs. diagnosis requires different strategies
Validate Results: Use cross-validation or separate test sets
Include Costs: Real-world decisions must consider resources
Think Sequentially: Often more efficient than parallel testing
Set Constraints: Define minimum acceptable performance
Interpret Clinically: Statistical optimality isn’t everything

Conclusion

The Decision Panel Optimization module provides a systematic approach to combining diagnostic tests. By considering various strategies, costs, and constraints, it helps identify practical testing algorithms that balance performance with resource utilization.

Next Steps

See the “Clinical Applications” vignette for disease-specific examples
Review “Advanced Optimization” for complex scenarios
Check “Implementation Guide” for deploying algorithms in practice

Session Information

sessionInfo()
#> R version 4.3.2 (2023-10-31)
#> Platform: aarch64-apple-darwin20 (64-bit)
#> Running under: macOS 15.5
#> 
#> Matrix products: default
#> BLAS:   /Library/Frameworks/R.framework/Versions/4.3-arm64/Resources/lib/libRblas.0.dylib 
#> LAPACK: /Library/Frameworks/R.framework/Versions/4.3-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.11.0
#> 
#> locale:
#> [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
#> 
#> time zone: Europe/Istanbul
#> tzcode source: internal
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] forcats_1.0.0      knitr_1.50         rpart.plot_3.1.2   rpart_4.1.24      
#> [5] ggplot2_3.5.2      dplyr_1.1.4        meddecide_0.0.3.12
#> 
#> loaded via a namespace (and not attached):
#>  [1] gtable_0.3.6            xfun_0.52               bslib_0.9.0            
#>  [4] htmlwidgets_1.6.4       lattice_0.22-7          vctrs_0.6.5            
#>  [7] tools_4.3.2             generics_0.1.4          tibble_3.2.1           
#> [10] proxy_0.4-27            pkgconfig_2.0.3         Matrix_1.6-1.1         
#> [13] KernSmooth_2.23-26      checkmate_2.3.2         data.table_1.17.4      
#> [16] irr_0.84.1              cutpointr_1.2.0         RColorBrewer_1.1-3     
#> [19] desc_1.4.3              uuid_1.2-1              jmvcore_2.6.3          
#> [22] lifecycle_1.0.4         flextable_0.9.9         stringr_1.5.1          
#> [25] compiler_4.3.2          farver_2.1.2            textshaping_1.0.1      
#> [28] codetools_0.2-20        fontquiver_0.2.1        fontLiberation_0.1.0   
#> [31] htmltools_0.5.8.1       class_7.3-23            sass_0.4.10            
#> [34] yaml_2.3.10             htmlTable_2.4.3         pillar_1.10.2          
#> [37] pkgdown_2.1.3           jquerylib_0.1.4         MASS_7.3-60            
#> [40] openssl_2.3.3           classInt_0.4-11         cachem_1.1.0           
#> [43] BiasedUrn_2.0.12        iterators_1.0.14        boot_1.3-31            
#> [46] foreach_1.5.2           fontBitstreamVera_0.1.1 zip_2.3.3              
#> [49] tidyselect_1.2.1        digest_0.6.37           stringi_1.8.7          
#> [52] sf_1.0-21               pander_0.6.6            purrr_1.0.4            
#> [55] labeling_0.4.3          splines_4.3.2           fastmap_1.2.0          
#> [58] grid_4.3.2              cli_3.6.5               magrittr_2.0.3         
#> [61] survival_3.8-3          e1071_1.7-16            withr_3.0.2            
#> [64] backports_1.5.0         gdtools_0.4.2           scales_1.4.0           
#> [67] lubridate_1.9.4         timechange_0.3.0        rmarkdown_2.29         
#> [70] officer_0.6.10          askpass_1.2.1           ragg_1.4.0             
#> [73] zoo_1.8-14              lpSolve_5.6.23          evaluate_1.0.3         
#> [76] epiR_2.0.84             rlang_1.1.6             Rcpp_1.0.14            
#> [79] glue_1.8.0              DBI_1.2.3               xml2_1.3.8             
#> [82] rstudioapi_0.17.1       jsonlite_2.0.0          R6_2.6.1               
#> [85] systemfonts_1.2.3       fs_1.6.6                units_0.8-7

meddecide Development Team

2025-06-03