Construction of a clinical prediction model for non-adherence to biologic therapy in ankylosing spondylitis pa-tients based on machine learning
CAI Xu
XIAO Jian-wei
GUO Fen-lian
CHENG Fan-fan
HU Xin-yu
XU Chu-hua
CHEN Ze-jian
Abstract：Objective To develop a clinical prediction model for non-adherence to biologic therapy in ankylo-sing spondylitis(AS)patients.Methods A total of 201 confirmed AS patients treated in outpatient and inpatient de-partments from January 2020 to October 2022 were included in this study(220 cases were collected,and 19 were exclu-ded).Non-adherence was determined based on the proportion of treatment-covered days six months later.Feature vari-ables were selected using LASSO regression and support vector machines,and the intersection was taken.A multivariate logistic regression analysis was conducted to construct a clinical prediction model for non-adherence.The predictive abil-ity and clinical utility of the model were evaluated using the C-index,receiver operating characteristic(ROC)curve,calibration plot,and clinical decision curve.Adaboost and Lightgbm algorithms were used to validate the constructed bina-ry classification model,and ROC and PR curves were plotted to assess the model's prediction ability.An internal valida-tion set was constructed through internal sampling,and validation was performed using the C-index,calibration curve,and ROC curve.Results The study showed that the non-adherence rate to biologic therapy in AS was 46.8％.Machine learning results yielded six feature variables,including education level,monthly income,anxiety level,drug use frequen-cy,disease activity,and age,as factors for constructing the prediction model.The model had a C-index of 0.739,and the area under the ROC curve was 0.715.Decision curve analysis showed that the model could benefit approximately 90％of patients.Adaboost algorithm showed an area under the ROC curve of 0.643 and an area under the PR curve of 0.634,while the Lightgbm algorithm showed an area under the ROC curve of 0.633 and an area under the PR curve of 0.676.In-ternal validation results showed a C-index of 0.755 and an area under the ROC curve of 0.733.Conclusion The clini-cal prediction model for non-adherence to biologic therapy,based on six feature variables,demonstrates high predictive ability and practicality.It helps identify AS patients with poor adherence early on.
Keywords：ankylosing spondylitisclinical prediction modelmachine learningnonadherencebiological agents
Publication Date：2023-11-28
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 1321-1327 )