【摘要】 目的 探讨不同算法筛选的糖尿病足溃疡（DFU）截肢预测模型。方法 收集2015年1月至2020年12月新疆医科大学第一附属医院收治的364例DFU患者的临床资料，按照截肢情况将其分为截肢组（n=213）和非截肢组（n=151），分别通过单因素分析、Boruta算法和随机森林-递归特征消除（RF-RFE）算法进行截肢危险因素分析，并构建临床预测模型，比较模型的c指数、F1分数和Brier分数，评估模型的预测效能和临床意义。结果 两组患者年龄、高血压病程、冠心病病程、Wagner评分、部位-缺血-神经病变-细菌感染-面积-深度（SINBAD）评分、国际糖尿病足工作组（IWGDF）分级比较，差异均有统计学意义（P﹤0.05）。实验室指标中截肢组患者低密度脂蛋白（LDL）、高密度脂蛋白（HDL）、甘油三酯（TG）、血钙、血磷、白蛋白与球蛋白比值（A/G）、平均血小板分布宽度（PDW）、血红蛋白（Hb）均低于非截肢组患者，截肢组患者球蛋白（GB）、中性粒细胞比例（N）、纤维蛋白原（FIB）、国际标准化比值（INR）、平均红细胞分布宽度（RDW）/白蛋白比率、中性粒细胞/淋巴细胞比值（NLR）、血小板与淋巴细胞比值（PLR）均高于非截肢组患者，差异均有统计学意义（P﹤0.05）。多因素分析结果显示，Wagner分级﹥2级、SINBAD评分﹥3分、FIB、Hb、PDW、INR、年龄均是DFU患者截肢的独立危险因素（P﹤0.05）。传统Logistic回归模型c指数、F1分数和Brier分数分别为0.771、0.809、0.163。采用Boruta算法得出对截肢影响最大的影响因素为年龄、Wagner分级﹥2级、SINBAD评分﹥3分、IWGDF分级﹥3级、A/G、INR、FIB、N、Hb、RDW比白蛋白比率、NLR和PLR，模型c指数、F1分数、Brier分数分别为0.686、0.744、0.163.RF-RFE算法得出DFU截肢危险因素为NLR、PLR、N、肌酐和PDW，模型c指数、F1分数和Brier分数分别为0.748、0.769、0.220。结论 不同算法从不同逻辑对DFU患者截肢的危险因素进行评估，可与传统统计学方法结合，为DFU的治疗决策提供依据互补。更多还原

【Abstract】 Objective To explore the prediction model of amputation of diabetic foot ulcers by different algorithms.Method Clinical data of 364 DFU patients admitted to the First Affiliated Hospital of Xinjiang Medical University from January 2015 to December 2020 were collected and divided into amputation group(n=213) and non-amputation group(n=151) according to the conditions of amputation. The risk factors of amputation were analyzed by univariate analysis, Boruta algorithm and random forest-recursive feature elimination(RF-RFE) algorithm, respectively. The clinical prediction model was constructed, and the c index, F1 score and Brier score of the model were compared to evaluate the prediction efficiency and clinical significance of the model. Result Compared the age, hypertension course,coronary heart disease course, Wagner score, site-ishamia-neuropathy-bacterial infection-area-depth(SINBAD) score and International Working Group on the Diabetic Foot(IWGDF) grade of two groups, the differences were statistically significant(P<0.05). Laboratory indicators included low density lipoprotein(LDL), high density lipoprotein(HDL),triglyceride(triglyceride) in the amputee group. The levels of TG, blood calcium, blood phosphorus, albumin/globulin rate(A/G), average platelet distribution width(PDW), and hemoglobin(Hb) were lower than those in the nonamputation group.The levels of globulin(GB), proportion of neutrophils(N), fibrinogen(FIB), international normalized ratio(INR), average red blood cell distribution width(RDW)/albumin ratio, neutrophils to lymphocytes ratio, NLR and platelet to lymphocyte ratio(PLR) were higher than those in the non-amputation group, with statistical significance(P<0.05). Multivariate analysis showed that Wagner grade>2, SINBAD score>3, FIB, Hb, PDW, INR and age were all independent risk factors for amputation in DFU patients(P<0.05). The c index, F1 score and Brier score of traditional Logistic regression model were 0.771, 0.809 and 0.163, respectively. According to Boruta algorithm, the most influential factors on amputation were age, Wagner grade>2, SINBAD score>3, IWGDF grade>3, A/G, INR, FIB, N, Hb, RDW ratio of albumin, NLR and PLR. Model c index, F1 score and Brier score were 0.686, 0.744 and 0.163, respectively.According to RF-RFE algorithm, the risk factors for DFU amputation were NLR, PLR, N, crinine and PDW, and model c index, F1 score and Brier score were 0.748, 0.769 and 0.220, respectively. Conclusion Different algorithms evaluate the risk factors of amputation in DFU patients from different logic, which can be combined with traditional statistical methods to provide complementary basis for DFU treatment decisions.更多还原