Bayesian Surveillance Methods for Healthcare Performance Monitoring

Provides Bayesian surveillance methods for prospective monitoring of healthcare performance, patient safety, and clinical quality indicators. The package implements beta-binomial monitoring for binary outcomes, gamma-Poisson monitoring for count outcomes, posterior predictive alert probabilities, Bayesian early-warning signal detection, risk-adjusted surveillance, simulation tools, decision-support methods, and graphical summaries. These methods support continuous performance monitoring and timely detection of adverse trends in healthcare systems. The methodology is motivated by established risk-adjusted monitoring, sequential surveillance, and healthcare quality-improvement frameworks , , , and .

BayesSurveillance

BayesSurveillance is an R package for Bayesian adaptive surveillance and intervention learning.

Core workflow

library(BayesSurveillance)

dat <- simulate_surveillance_data(seed = 1)
fit <- fit_bewrs(dat)
risk <- compute_dynamic_bewrs(fit)
peib <- estimate_peib(risk)
policy <- recommend_action(peib)
evaluate_policy(policy)

Adaptive update workflow

old_dat <- simulate_surveillance_data(seed = 1)
new_dat <- simulate_surveillance_data(seed = 2)

updated_policy <- update_policy(old_dat, new_dat)
evaluate_policy(updated_policy)

# Or run the full pipeline directly
policy2 <- adaptive_update(new_dat)

Core methodological idea

The package extends BEWRS from early-warning risk prediction to adaptive intervention learning:

1. Estimate posterior underperformance risk.
2. Convert risk into Dynamic BEWRS using persistence and deterioration.
3. Estimate Provider-specific Expected Intervention Benefit (PEIB).
4. Recommend Bayes-optimal action by expected-loss minimisation.
5. Update the policy as new surveillance data arrive.