Background
Adolescent TB prevention programs often treat risk as uniform, despite strong heterogeneity driven by undernutrition, infection dynamics, crowding, and air-exchange conditions in school environments. We aimed to build a policy-ready prioritization framework that ranks preventive vaccine strategies using clinically measurable factors while retaining interpretability for implementers.
Methods
Vaccine efficacy inputs were anchored to the South African adolescent PoI trial comparing BCG revaccination and H4:IC31 versus placebo with sustained QuantiFERON conversion as an infection endpoint (NCT02075203). Infection-determinant calibration used community youth data from rural Uganda reporting adolescent TB infection prevalence and school-mobility effects, plus household risk factor estimates for crowding and ventilation proxies (e.g., windows) from Ugandan surveillance. Ventilation proxy feasibility was supported by CO₂ decay approaches used for TB-relevant ventilation assessment. Policy ranking used mediational g-formula with doubly robust nuisance estimation; uncertainty for key contrasts was permutation-calibrated (Monte Carlo) rather than Bayesian.
Results
From published trial counts, sustained IGRA conversion occurred in 11.6% (36/310) placebo, 6.7% (21/312) BCG, and 8.1% (25/308) H4:IC31 recipients. Assuming ~24 months surveillance, the inferred infection “attack rate” was ~5.81, 3.37, and 4.06 per 100 person-years respectively. Using a binomial likelihood-ratio test for BCG vs placebo, LLR=4.50; a randomization-based Monte Carlo/permutation p-value (hypergeometric exact) was p=0.024, consistent with the trial’s reported efficacy signal for sustained conversion. The effect scale for time-to-sustained conversion corresponds to β=ln(HR)=−0.605 with SE=0.275 when translating the reported BCG vaccine efficacy 45.4% (95% range 6.4–68.1) to HR bounds. For program relevance, recent IGRA conversion marks substantially higher near-term disease risk in adolescents (TB incidence 1.46 vs 0.17 per 100 person-years in converters vs nonconverters), supporting a practical dynamic lag window between conversion and clinical disease ascertainment. For risk-stratified rollout performance, externally validated TB household risk scoring demonstrates strong discrimination (c-statistic/AUC ~0.77 derivation, 0.75 validation), supporting feasible operational targeting when resources are limited.
Conclusions
A non-Bayesian, doubly robust, policy-optimization framework grounded in real adolescent trial endpoints and validated risk discrimination can prioritize TB preventive vaccination to higher-risk undernourished youth while maintaining transparent, implementable decision rules for school-linked prevention programs.