Background: Sentinel clinics often rely on rapid antigen testing and sparse laboratory confirmation, yet policy decisions still require credible vaccine effectiveness (VE) estimates and interpretable why it works signals. We aimed to quantify VE using resource-light data streams while anchoring measurement error and transportability with targeted serology and severity-aware phenotyping.
Methods: We specify a sentinel-compatible test-negative design using linked routinely collected surveillance and vaccination registries as demonstrated in Brazil, with SI-PNI vaccination registry linked to e-SUS Notifica and SIVEP-Gripe respiratory surveillance, and symptomatic testing during Omicron-era circulation. Cases were symptomatic test-positive and controls symptomatic test-negative (RT-PCR or lateral flow), with specimen collection constrained to ≤10 days from symptom onset to limit outcome misclassification. VE was estimated with doubly robust nuisance modeling include propensity and outcome, and transportability reweighting to new clinic catchments; targeted dried blood spot serology was used as a calibration subset for prior infection and exposure history, leveraging established DBS feasibility for vaccine-preventable serology.
Results: In Brazil’s symptomatic-testing stream,276,385 individuals were analyzed and the symptomatic-test positivity attack proportion was 57.0% (59,052/76,385), higher in unvaccinated (61.7%) than vaccinated (56.7%). From these reported strata, the log-odds contrast for test positivity (vaccinated vs unvaccinated) was β = −0.207 with SE = 0.00369, corresponding to OR 0.81. The independence test on the 2×2 table yielded LLR = 3171.3, with Monte Carlo p < 5×10⁻⁵ under a hypergeometric null. Using vaccination status alone as a risk score produced AUC = 0.706 for identifying positives (minimal discrimination, consistent with immune-escape-era symptomatic infections). VE showed clear peak attenuation and duration shortening, in a nationally anchored test-negative analysis, booster VE against symptomatic infection peaked at 51.6% (14–29 days) but waned to 4.2% by ≥4 months, while protection against severe outcomes remained substantially higher over the same horizon, an actionable policy contrast for booster timing. For the laboratory-light biomarker layer, antigen test signal can be quantified by reader-based intensity with strong dose–response to antigen input (R²≈0.97), supporting a defensible Ct-proxy calibration in the serology subset rather than assuming PCR availability everywhere.