摘要
目的 探讨机器学习和解释模型在类风湿性关节炎合并骨质疏松症患者预后预测中的应用价值。方法 选取2021年6月至2023年7月新疆维吾尔自治区维吾尔医医院收治的类风湿性关节炎合并骨质疏松症患者194例,按预后情况将其分为预后不良组(46例)和预后良好组(148例),通过两组临床资料的差异构建随机森林、支持向量机、朴素贝叶斯、BP神经网络、XGBoost预测模型,同时采用多因素logistic回归模型分析患者预后的影响因素。通过受试者操作特征(ROC)曲线及PR曲线筛选出最佳预测模型后,采用SHAP解释模型对其进行特征解释,并随机抽取1例患者进行模型评估。结果 两组年龄、吸烟史、职业、类风湿因子、抗链球菌溶血素、Ig M、红细胞沉降率、谷草转氨酶、热盐包治疗、针灸治疗、推拿治疗、骨质疏松仪治疗、关节功能状态分期、患者健康评定量表评分、视觉模拟评分法评分比较,差异有统计学意义(P<0.05)。多因素分析结果显示,年龄(OR=1.066,95%CI:1.021~1.113)、职业(OR=16.711,95%CI:5.499~50.787)、骨质疏松仪使用情况(OR=6.836,95%CI:2.362~19.782)、关节功能状态分期(OR=2.756,95%CI:1.388~5.474)、患者健康评定量表评分(OR=6.287,95%CI:2.514~15.718)是类风湿性关节炎合并骨质疏松症患者预后不良的独立影响因素(P<0.05)。ROC及PR曲线结果显示,随机森林预测模型性能最好,可信性最高。SHAP解释模型显示,类风湿因子水平、患者健康评定量表评分、职业等均为类风湿性关节炎合并骨质疏松症患者预后不良的影响因素。患者模型评估结果显示,类风湿因子水平、职业、患者健康评定量表评分、年龄、是否推拿治疗为该例患者预后的主要影响因素。结论 基于机器学习的预后预测模型可预测类风湿性关节炎合并骨质疏松症患者的预后情况,可针对相关因素进行预防和规范性治疗,减少不良预后发生。
Objective To explore the application value of machine learning and interpretive models in prognostic prediction of patients with rheumatoid arthritis combined with osteoporosis.Methods A total of 194 patients with rheumatoid arthritis combined with osteoporosis admitted to Xinjiang Uyghur Autonomous Region Uyghur Hospital from June 2021 to July 2023were selected and divided into poor prognosis group(46 cases) and good prognosis group(148 cases) according to prognosis.The prediction models of random forest,support vector machine,naive Bayes,BP neural network,and XGBoost were constructed by the differences of two groups of clinical data,and the influencing factors of patient prognosis were analyzed by multi-factor logistic regression model.After the optimal prediction model was selected by receiver operating characteristic(ROC) curve and PR curve,SHAP interpretation model was used to interpret its characteristics,and one case of patients was randomly selected for model evaluation.Results There were statistically significant differences in age,smoking history,occupation,rheumatoid factor,antstreptolysin,Ig M,erythrocyte settlement rate,glutamic oxalacetic transaminase,thermal salt packet treatment,acupuncture treatment,massage treatment,osteoporosis instrument treatment,joint function stage,patient health rating scale score,and visual simulation score between the two groups(P<0.05).The results of multivariate analysis showed that age(OR=1.066,95%CI:1.021-1.113),occupation(OR=16.711,95%CI:5.499-50.787),the use of osteoporosis instrument(OR=6.836,95%CI:2.362-19.782),joint function stage(OR=2.756,95%CI:1.388-5.474),health assessment questionnaire score(OR=6.287,95%CI:2.514-15.718) were the independent influencing factors of poor prognosis in patients with rheumatoid arthritis combined with osteoporosis(P<0.05).ROC and PR curve results show that the random forest prediction model had the best performance and the highest credibility.SHAP interpretation model showed that the level of rheumatoid factor,patient health rating scale score,occupation,etc.,were all factors affecting the poor prognosis of patients with rheumatoid arthritis combined with osteoporosis.The results of patient model evaluation showed that rheumatoid factor level,occupation,health assessment questionnaire score,age,and whether massage treatment were the main factors affecting the prognosis of this patient.Conclusion The prognosis prediction model based on machine learning can predict the prognosis of patients with rheumatoid arthritis combined with osteoporosis,and can provide targeted prevention and normative treatment for related factors to reduce the occurrence of adverse prognosis.
作者
阿不都许克尔·阿不都卡地尔
玉苏甫·买提努尔
尔西丁·买买提
Abuduxukeer Abudukadier;Yusufu Maitinuer;Erxiding Maimaiti(School of Public Health,Xinjiang Medical University,Xinjiang Uygur Autonomous Region,Urumqi830011,China;Xinjiang Uyghur Autonomous Region Uyghur Hospital,Xinjiang Uygur Autonomous Region,Urumqi830049,China)
出处
《中国医药导报》
CAS
2024年第4期1-5,20,共6页
China Medical Herald
基金
国家重点研发计划基金资助项目(2017YFC 1704003)。
