多发性骨髓瘤患者肾损伤预测模型的构建

目的 探讨多发性骨髓瘤（MM）患者发生肾损伤的危险因素，构建预测模型。方法回顾性分析血液科收治的初诊MM患者临床资料，以血肌酐结果为标准把患者分为肾损伤组和非肾损伤组，对两组患者临床资料进行统计学分析，采用logistic回归法分析影响MM患者发生肾损伤的独立危险因素，建立预测MM发生肾损伤风险的logistic回归模型，绘制受试者工作特征曲线（ROC）、曲线下面积（AUC）评估预测效能。结果共纳入206例初诊MM患者，肾损伤组（n=75）患者的血红蛋白明显低于非肾损伤组（n=131），而白细胞、中性粒细胞/淋巴细胞比值（NLR）、血钙、血尿酸、血尿素氮和尿蛋白阳性比例均高于非肾损伤组，差异有统计学意义（P＜0.05）；logistic回归显示NLR、血钙、血尿酸、血尿素氮是影响MM发生肾损伤的独立危险因素，得到回归方程：Y=-22.770+1.765×NLR+2.642×血钙+0.023×血尿酸+0.542×血尿素氮，Hosmer-Lemeshow检验提示模型具有良好的校准度（P=0.230），AUC=0.978，当最佳截断点为4.542 1时，敏感度为85.33%，特异度为96.95%。结论本研究建立的logistic回归模型有较好的预测价值，有助于血液科医生对MM患者发生肾损伤进行预测。

Construction of predictive model of renal injury in patients with multiple myeloma

Objective To investigate the risk factors of renal injury in patients with multiple myeloma (MM) and to construct a predictive model. Methods The clinical data of newly diagnosed MM patients admitted to the Department of Hematology were retrospectively analyzed. The patients were divided into renal injury group and non-renal injury group based on blood creatinine results. The data was statistically analyzed, the logistic regression method was used to analyze the independent risk factors that affect the renal injury of MM patients. The logistic regression model to predict the risk of MM renal injury was established, the receiver operating characteristic curve (ROC) was drawn and area under the curve (AUC) was also calculated. Results A total of 206 newly diagnosed MM patients were enrolled. The hemoglobin of the renal injury group (n=75) was significantly lower than that of the non-renal injury group (n=131), while the white blood cells, neutrophil-to-lymphocyte ratio (NLR), blood calcium, the positive ratios of blood uric acid, blood urea nitrogen and urine protein were significantly higher than those of the non-renal injury grou (P<0.05). Logistic regression showed that NLR, blood calcium, blood uric acid, and blood urea nitrogen affected the occurrence of MM. The regression equation is obtained: Y=-22.770+1.765×NLR+2.642×blood calcium+0.023×blood uric acid+0.542×blood urea nitrogen. The model also showed good calibration in Hosmer-Lemeshow test (P=0.230), with AUC of 0.978; when the best cut-off point was 4.542 1, the sensitivity was 85.33% and the specificity was 96.95%. Conclusion The logistic regression model established in this study has good predictive value, which is helpful for hematologists to predict the occurrence of renal injury in MM patients.