基于自动机器学习的针灸治疗功能性消化不良疗效预测模型与决策支持研究OA

Objective To develop machine learning-based models for predicting the efficacy of acupuncture in patients with functional dyspepsia(FD)and to explore their utility in clinical decision support,thereby enhancing the personalization and effectiveness of acupuncture therapy.Methods This retrospective study collected clinical data from 2 584 patients with FD treated at the 72nd Group Army Hospital of the Chinese People's Liberation Army between April 2023 and October 2024,which served as the internal dataset[internal training set(n=2 067);internal testing set(n=517)].Concurrently,data from 276 patients with FD at the First People's Hospital of Huzhou were gathered as the external dataset(external testing set).All patients received 20 acupuncture sessions over a 4-week period and completed the Nepean Dyspepsia Symptom Index(NDSI)and the Nepean Dyspepsia Life Quality Index(NDLQI).Automated machine learning(AutoML)was used to construct predictive classification and regression models based on 28 distinct features.Comparative analyses were performed against conventional machine learning algorithms,including logistic regression,support vector machine,and extreme gradient boosting(XGBoost).Results ①In the predictive classification model,the critical features influencing response to acupuncture were ranked as follows:education level,sex,disease type,NDSI upper abdominal bloating score,NDSI upper abdominal pressure score,and NDSI irregular diet score.In the predictive regression model,the primary features impacting quality of life were education level,sex,disease type,NDLQI eating dimension score,and NDLQI knowledge/control dimension score.②AutoML achieved the superior predictive performance across both internal and external testing sets for both classification and regression tasks.③In the external testing set,the AutoML classification model yielded an area under the receiver operating characteristic curve(ROC-AUC)of 0.914 7 and a precision-recall AUC(PR-AUC)of 0.915 7,while the AutoML regression model achieved a coefficient of determination(R²)of 0.863 3.④For classification prediction on the internal testing set,AutoML retained a high sensitivity of 98.15%while improving specificity to 77.72%compared with conventional models,significantly mitigating false-positive risks.⑤ For regression prediction on the external testing set,AutoML demonstrated a mean absolute error(MAE)of only 2.651 4,representing a 23.11%reduction compared with XGBoost,and a lower root mean square error-to-MAE ratio(RMSE/MAE)of 1.488.Conclusions This study established AutoML models to predict the clinical efficacy of acupuncture in patients with functional dyspepsia and developed a corresponding clinical decision support system.These predictive models hold promise for assisting clinicians in predicting patient responsiveness to acupuncture,thereby enabling the tailoring and adjustment of personalized acupuncture regimens,which may improve clinical outcomes in functional dyspepsia and optimize healthcare resource utilization.