Abstract Rationale In progressive pulmonary fibrosis (PPF), clinicians consider treatment intensification. However, whether all patients with PPF should receive antifibrotic therapy is unclear. We previously suggested antifibrotic therapy may confer survival advantage among patients who meet PPF criteria under standard care. Accordingly, we aimed to develop and externally validate a prediction model, restricted to routinely available variables, to estimate the 24-month risk of further progression under standard therapy after a PPF diagnosis and to inform antifibrotic treatment decisions. Methods We analyzed a retrospective, multicenter cohort from Osaka University Hospital and Osaka Prefectural Toneyama Medical Center. PPF followed the 2022 proposed definition. PPF under standard therapy was defined as disease-specific care without antifibrotics, with treatment adequacy retrospectively confirmed by multidisciplinary discussion. We modeled 24-month absolute risk using a Fine-Gray subdistribution hazards model treating death as a competing event. The primary development used complete-case (CC) data; robustness was examined through multiple imputation by chained equations in sensitivity analyses. External validation used the Toneyama CC set. Primary metrics were time-dependent AUC and Brier score at 2 years. Calibration was evaluated and locally aligned using pseudo-observation-based linear recalibration. Clinical utility was assessed with an IPCW-based time-to-event Decision Curve Analysis (DCA), comparing the model to a “treat-all” antifibrotic strategy; report net benefit (NB) and ΔNB (NBₘodel − NBₜreat-all) across decision thresholds. Results The development cohort comprised 223 CC patients (Osaka University), and the external validation cohort comprised 236 CC patients (Toneyama). In the validation set, 24-month events occurred in 105/236 (44. 5%). In external validation, the 2-year discrimination was AUC 0. 72 (bootstrap 95% CI 0. 55-0. 90). After linear recalibration, the 2-year Brier score improved from 0. 098 to 0. 080, indicating better absolute-risk calibration. DCA showed ΔNB became clearly positive at thresholds around 0. 55; at 0. 60, ΔNB ≈ +0. 438 (≈ +43. 8 per 100 patients) compared with treat-all, with an estimated intervention reduction of ∼29 per 100 while maintaining higher net benefit (Figure 1). Time-varying effects were suggested for FVC and KL-6, indicating that the proportional subdistribution hazards assumption may not be fully satisfied. Conclusion This model provides reasonable 24-month risk stratification in PPF and achieves improved calibration after local linear recalibration. DCA supports a clinically coherent decision-making strategy: lower-risk patients may be prioritized for further assessment, whereas intermediate-to-higher-risk patients may benefit from prompt antifibrotic initiation, outperforming a treat-all approach in net benefit while reducing unnecessary interventions. Prospective application and model updating to address time-dependent effects are warranted. This abstract is funded by: No
Niitsu et al. (Fri,) studied this question.