Dataset and Repository for Selection on Testing Engagement: A Structural Bias in Cross-Sectional Incidence Estimation Using Recency Assays

Code, data, and compiled manuscript materials for a stand-alone methodological letter correcting the Kassanjee/Gao cross-sectional HIV incidence estimator for competing-risk structural censoring, applied to 34 high-burden US metropolitan areas using AIDSVu 2023 surveillance data. The Kassanjee cross-sectional incidence estimator, used in counterfactual-controlled pre-exposure prophylaxis (PrEP) efficacy trials including PURPOSE 1 (NCT04994509) and PURPOSE 2 (NCT04925752), assumes closed-system observability of the at-risk population: every recently-infected individual within the recency window is equally likely to be present at screening. Populations experiencing structural censoring — competing-risk hazards from overdose mortality, incarceration, displacement, carceral disruption of healthcare, and related mechanisms disproportionately affecting marginalized groups — violate this assumption in a systematic and directional way. This archive contains the Python implementations, AIDSVu-derived 34-MSA surveillance datasets, and compiled manuscript materials (letter + supplement, both LaTeX sources and PDF) for a correction framework deriving: (i) the effective mean duration of recent infection under structural censoring, Omega*(gamma) = integral of P_R(t) * S_c(t) dt; (ii) the joint incidence rate ratio bias factor B_IRR incorporating both screening-cohort and intervention-arm observation probabilities; (iii) the structural amplification produced by the 90-day no-prior-testing eligibility criterion common to Phase 3 PrEP trials; and (iv) application to 34 high-burden US metropolitan areas using AIDSVu 2023 late-diagnosis percentage as the empirical proxy for structural hazard. Key empirical result. Across 34 MSAs, the Kassanjee denominator is inflated by 8.7% to 27.3% with monotone scaling in late-diagnosis percentage (Pearson r = 0.9995, p < 10^-48). The joint IRR bias factor ranges 0.969 to 0.999 under severity-coupled retention assumptions, corresponding to reported IRR attenuations of up to 3.1% in the highest-structural-severity cohort. The bias is structurally guaranteed rather than merely possible when trial-design eligibility criteria select the Incidence Phase cohort on the testing-engagement axis. Reproducibility. All results are fully reproducible from publicly available AIDSVu 2023 data (no registration required, no individual-level information). The complete reproduction pipeline executes in under 90 seconds. See README.md for instructions. Companion work. This methodological letter is part of a broader research program on HIV prevention trial methodology. The biological ceiling framework (Finite Prevention Windows for HIV PEP, under review at Science Advances) and the city-level structural barrier framework (Meyer/Kamitani collaboration, under review at BMC Public Health) provide the biology and epidemiology that sit alongside this measurement-bias contribution. No external funding. Work conducted independently under Nyx Dynamics LLC. The author reports prior employment with Gilead Sciences, Inc. (January 2020 – November 2024); all Gilead stock was fully divested December 2024 prior to initiation of this research. Gilead Sciences had no role in conception, analysis, interpretation, or decision to archive.