A prediction model for severe hematological toxicity of BTK inhibitors

Abstract

Bruton's tyrosine kinase inhibitor (BTKi) has revolutionized the treatment of B-cell lymphomas. However, BTKi-related hematological toxicity hinders treatment continuity and may further affect clinical efficacy. To identify risk factors and predict the likelihood of BTKi-related hematological toxicities, we constructed and validated a prediction model for severe hematological toxicity of BTKi. Approved by the hospital medical science research ethics committee (No. M2022427), we collected real-world data in patients treated with BTKi from a Lymphoma Research Center in China. The outcome of interest was severe hematological toxicity caused by BTKi. 36 candidate variables were categorized into demographics, diagnostic and treatment information, laboratory data, and medical history. The study sample was randomly divided into training (70%) and validation (30%) sets. We developed and compared the performance of various modelling methods, including decision tree (DT), random forest (RF), gradient boosting decision tree (GBDT), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and logistic regression (LR). Finally, we constructed a Web-calculator of the optimal model to estimate the risk of hematological toxicity. This study was designed, conducted and reported strictly in compliance with the TRIPOD checklist. Data from a total 121 patients were included [median age, 65 years (range, 56-73 years); 74 (61.15%) men; 47 (38.84%) severe hematological toxicity]. The XGBoost model demonstrated better overall properties than other models, achieving high discrimination (AUC: 0.671; accuracy: 0.730; specificity: 0.913) and clinical benefit. The following 10 variables were used to develop the XGBoost model: white blood cell count (WBC), neutrophil count (Neut), red blood cell count (RBC), platelet count (PLT), fibrinogen (Fib), total protein (TP), aspartate aminotransferase (AST), lactate dehydrogenase (LDH), gender and type of BTKi. SHAP values demonstrated insightful associations between these variables and hematological toxicity. Finally, to facilitate clinical and research use, we also deploy the XGBoost model on a web-calculator for free access. The XGBoost model with promising accuracy was developed to predict the severe hematological toxicity of BTKi. It helps to strengthen the proactive monitoring and management of patients with hematological toxicity, and thus achieve long-term continuous BTKi treatment.

Keywords: BTK inhibitor; Hematological toxicity; Logistic regression; Machine learning; Prediction model; XGBoost.