Long-term outcomes with emicizumab prophylaxis in patients with hemophilia a in China

Abstract Introduction: Emicizumab, a bispecific monoclonal antibody that bridges activated factor IX and factor X, has emerged as an effective prophylactic treatment for patients with hemophilia A (PwHAs). However, real-world responses are heterogeneous, long-term data remain limited, and the determinants of long-term efficacy are unclear. Methods: Data from 132 pediatric and adult PwHAs with/without factor VIII (FVIII) inhibitors who received emicizumab prophylaxis for more than one year from 25 centers in China were pooled to establish a long-term efficacy and safety profile. Annualized bleed rate (ABR) and Annualized joint bleed rate (AJBR) were estimated using the negative binomial regression model. Machine learning models—including the Feature Tokenizer Transformer, random forest, XGBoost, LightGBM, ResNet, and CatBoost—were applied to predict post-emicizumab ABR. Results: A total of 132 patients were enrolled in our study. All patients were followed up for a median of 26.0 months (interquartile range [IQR], 16.0–40.0). Across a median efficacy period of 25.5 months (IQR, 16.0–40.0; data cutoff: May 25, 2025), the model-based ABR (for all bleeds) and treated ABR were 0.81 (95% confidence interval [CI], 0.62–1.07) and 0.30 (0.21-0.42), respectively. ABR declined compared to baseline and then stabilized at <1 in analyses based on 12-month treatment intervals. During months 1 to 12 (n=132), 62.9% of participants had 0 bleeds, 82.6% had no treated bleeds, 90.9% reported no joint bleeds, and 94.7% reported no treated joint bleeds. To determine the factors associated with long-term efficacy, we applied machine learning models. Among them, the Feature Tokenizer Transformer model yielded the best performance (AUC=0.905±0.012). Feature importance was ranked by mean absolute SHapley Additive exPlanations (SHAP) values. The top three predictors of post-emicizumab ABR > 1 were: historical inhibitor status, whether standard loading dose (3 mg/kg for 4 weeks) was administered, and baseline ABR. Injection site reactions were reported in 10 out of 132 patients (7.6%). No deaths, thromboembolic events, or thrombotic microangiopathies were reported during the study period. Conclusions: In this multicenter real-world study, we indicated that emicizumab prophylaxis maintained low bleeding rates in PwHAs of all ages and remains well tolerated. Historical inhibitor status, whether standard loading dose was administered, and baseline ABR were the most informative variables for predicting post-emicizumab ABR > 1.