Development of a time-dependent prediction model for the risk of Pneumocystis jirovecii pneumonia in ANCA-associated vasculitis

Objectives In patients with ANCA–associated vasculitis (AAV), Pneumocystis jirovecii pneumonia (PJP) is a potentially fatal complication. As treatment intensity typically changes over time, PJP risk also changes, making fixed prophylaxis strategies suboptimal. We aimed to develop a time-dependent prediction model to guide individualized decisions on PJP prophylaxis. Methods We conducted a retrospective cohort study using a nationwide claims database. Patients newly diagnosed with AAV between 2015 and 2024 were included. Candidate predictors included demographics, comorbidities, and monthly-updated medication use. A flexible parametric survival model was developed to predict the risk of PJP. Discrimination, calibration and clinical utility were evaluated by internal-external cross-validation (IECV) to investigate model’s generalizability. Results Among 15 649,036 individuals in the database, 2,751 patients with AAV were included. Of them, 170 (6.2%) developed PJP. The final model included age, monthly maximum glucocorticoid dose, use of rituximab, cyclophosphamide, and PJP prophylaxis. In the IECV, the time-dependent C-index was 0.70 (95% confidence interval [CI] 0.60–0.79) at 2 months and 0.66 (95% CI 0.61–0.71) at 12 months; calibration was generally acceptable. In the clinical utility analysis, per 1,000 patients with AAV including 61 PJP cases, 948 would be recommended prophylaxis with a monthly risk threshold of 1.0%. Of them, 643 had not received it, with 34 potentially preventable PJP cases, and 336 could have discontinued prophylaxis during follow-up. Conclusion We developed a time-dependent model to predict the risk of PJP in patients with AAV. The model may support more appropriate, individualized prophylaxis decisions.