S87 Discovery and validation of a personalised risk predictor for incident tuberculosis in settings aiming towards pre-elimination (PERISKOPE-TB)

Background The lifetime risk of tuberculosis (TB) among individuals with latent Mycobacterium tuberculosis infection (LTBI) is commonly estimated as 5–10%, but is highly variable between individuals. Validated estimates of personalised risk are needed to facilitate precise targeting of preventative treatment. We aimed to characterise population-level TB risk among people tested for LTBI, and to develop and validate a prognostic model to estimate personalised risk of disease. Methods We pooled individual-level data from 18 systematically-identified cohort studies conducted in 20 countries with low TB transmission (annual incidence ≤20/100,000 persons). We estimated population-level incident TB risk using flexible parametric survival models with random effect intercepts by source study. We then developed and validated a flexible parametric survival prediction model for incident TB using the internal-external cross-validation framework, iteratively discarding one contributing dataset from model development and using it for validation. Findings In pooled data including 80,468 individuals tested for LTBI and 803 TB cases, 5-year cumulative risk of incident TB among people with untreated LTBI was 15.6% (95% CI 8.0–29.2) for child contacts, 4.8% (3.0–7.7) for adult contacts, 5.0% (1.6–14.5) for migrants, and 4.8% (1.5–14.3) for immunocompromised groups. We found highly variable estimates within risk groups, necessitating a personalised approach to risk-stratification. We thus developed a prognostic model that combines a quantitative measure of T cell sensitisation and clinical covariates. These covariates included age, history of TB exposure (household contact of smear-positive index case, other contact, migration from high TB burden country or no exposure), HIV status and receipt of a solid organ or haematological transplant. Validation of this model achieved a random-effects meta-analysis C-statistic of 0.88 (0.82–0.93) for incident TB over 2 years. Decision curve analysis revealed that applying the model improved clinical decision-making for targeting LTBI treatment. Interpretation TB incidence rates are heterogeneous among people identified as having LTBI by current standards, even after stratification by indication for screening. We present a freely available and directly data-driven personalised risk predictor for incident TB (www.periskope.org). PERISKOPE-TB will facilitate a programmatic paradigm shift by allowing a fully evidence-based and patient-centred approach to TB risk stratification in settings aiming towards pre-elimination globally.