State-Level Validation of Syphilis and Chlamydia Transmission Models in the United States: An Open Surveillance Panel (2000–2024) and a Held-Out-Year Benchmark

Transmission models of sexually transmitted infections (STIs) are usually calibrated to national surveillance and are rarely tested against the case history of the individual states that run prevention programs. We assembled an open, documented panel of reported chlamydia and syphilis (primary and secondary, early non-primary non-secondary, unknown duration or late, and congenital) for the 50 U.S. states and the District of Columbia, 2000–2023, and for the nation, 2000–2024, by sex and age group, joined to a single consistent series of Census Bureau population estimates (262,958 rows). The panel reproduces the national totals exactly, and 21 of 22 named values match CDC’s published tables. We then calibrated two published compartmental models, a syphilis model with reinfection, two-stage treatment and treatment failure, and a chlamydia model with behavioral heterogeneity, to each jurisdiction’s 2011–2020 cases. We scored every fit on 2021–2023, years it never saw. With its published fixed settings, and the behavior-response strength scaled per capita to each state, the syphilis model had a median held-out mean absolute percentage error (MAPE) of 41.3%. It under-predicted in 44 of 51 jurisdictions (median bias-40.6%), because its prevalence-dependent behavior term capped incidence growth. Estimating the behavior-response strength reduced the median error to 17.6% (bias-1.2%; Wilcoxon p = 7.7 × 10−6). The fitted state values clustered around the published absolute value, not the per-capita one. The chlamydia model had a median held-out MAPE of 20.6% with a one-year infectious period, and 10.1% with the one-week period implied by treatment alone. In every fit, several parameters were not identifiable from annual counts. The panel, parameter sets with uncertainty ranges, validation scorecard, code, and an assumption register are openly released (DOIs 10.5281/zenodo.23742262 and 10.5281/zenodo.23742264).