Prediction of coagulation factor inhibitor levels using Random Forest model based on the activated partial thromboplastin time (aPTT) mixing study: model development and prototype application

Abstract Objectives This study aimed to develop a machine learning approach to predict factor VIII inhibitor titers from activated partial thromboplastin time (aPTT) mixing study data. Methods Records of 173 patients who underwent factor VIII inhibitor testing at Songklanagarind Hospital from January 2023 to December 2025 were retrospectively analyzed. Predictor variables included mixing aPTT, age, and sex, with Bethesda assay inhibitor titer as the outcome. A two-stage hurdle Random Forest model was developed, consisting of a classifier for inhibitor positivity (>0.6 BU) followed by a regressor for quantitative titer prediction. The final model was implemented in a prototype application. Results Using out-of-fold predictions, the model achieved an AUC of 0.972 (95 % CI, 0.944–0.994) for inhibitor classification. Sensitivity, specificity, positive predictive value, and negative predictive value were 85.7 % (95 % CI, 74.3–92.6 %), 99.1 % (95 % CI, 95.3–99.8 %), 98.0 % (95 % CI, 89.3–99.6 %), and 93.5 % (95 % CI, 87.8–96.7 %), For titer prediction, the model showed a mean absolute error of 5.189 BU, R 2 of 0.629, and Pearson correlation coefficient of 0.794. The model showed a higher AUC than the Rosner Index and percent correction. Conclusions A Random Forest model using routine aPTT mixing study data may support factor VIII inhibitor identification and titer estimation. The prototype application highlights its potential as a practical laboratory decision-support tool, although external validation is required before clinical implementation.