Objective To construct a prediction model of interstitial pneumonia(IP) in patients with diffuse large B-cell lymphoma(DLBCL)based on gradient boosting machine (GBM) and to verify its efficacy.
Methods The clinical data of 220 patients with DLBCL were retrospectively analyzed, including 51 cases(23.18%) with IP and 169 cases without IP. The patients were divided into training set (154 cases) and test set(66 cases) according to a 7 to 3 ratio. The prediction model was constructed based on GBM algorithm. The receiver operating characteristic (ROC) curve was used to evaluate model differentiation, and model fitting was represented by a calibration curve.
Results Five optimal features including age, disease stage, international prognostic index (IPI) score, smoking history, and lactate dehydrogenase (LDH) were involved in. The descending order of their relative importance was as follows: age, staging of disease, LDH, IPI score and smoking history. The ROC curve showed that the area under the curve (AUC) of the GBM model was 0.872(95%CI, 0.800 to 0.945) in the training set and 0.891(95%CI, 0.755 to 1.000) in the test set, respectively. The calibration curve showed that the GBM predicted probabilities in the test set and training set were in agreement with the observed outcomes.
Conclusion The incidence of IP in DLBCL patients after treatment is 23.18%, which is mainly related to age, disease stage, IPI score, smoking history and LDH level. The GBM model built based on these factors has high accuracy and differentiation, and could provide a reference for decision-making of clinical treatment in DLBCL patients.
DownLoad: