目的  探讨基于Logistic回归分析和列线图构建维持性血液透析(maintenance hemodialysis，MHD)患者死亡的风险预测模型。 方法  回顾性收集贺州市人民医院2019年1月—2021年1月收治的319例MHD患者临床资料(作为建模组)，另外收集2021年2月—2022年3月收治的80例MHD患者临床资料(作为验证组)。回顾2年结果，根据死亡情况将患者分为死亡组和存活组；采用Lasso回归分析筛选协变量，Logistic分析患者死亡的影响因素；R软件构建MHD患者死亡列线图预测模型；采用Bootstrap进行内部验证；受试者工作特征(receiver operating characteristic，ROC)和校准曲线评估模型的区分度和一致性；决策曲线分析(decision curve analysis，DCA)评估列线图模型的临床应用价值。 结果  本研究建模组死亡81例，存活238例；验证组死亡16例，存活64例。建模组中死亡组和存活组年龄(t=5.497，P＜0.001)、糖尿病肾病(χ2=28.016，P＜0.001)、血管通路类型(χ2=34.491，P＜0.001)、超敏C反应蛋白(high-sensitivity c-reactive protein，hs-CRP)(t=42.746，P＜0.001)、血红蛋白(hemoglobin，Hb)(t=15.038，P＜0.001)、白蛋白(albumin，ALB)(t=11.973，P＜0.001)、全段甲状旁腺激素(intact parathyroid hormone，iPTH)(t=2.937，P=0.004)、B型利尿钠肽(B-type natriuretic peptide，BNP)(t=2.937，P=0.011)比较差异有统计学意义。Lasso分析筛选出6个预测因素，将Lasso回归筛选出的自变量进行Logistic分析，结果显示年龄(OR=1.269，95% CI：1.146～1.405，P＜0.001)、糖尿病肾病(OR=3.347，95% CI：1.149～9.749，P=0.001)、血管通路类型(OR=3.205，95% CI：1.212～8.476，P=0.016)、hs-CRP(OR=2.842，95% CI：1.922～4.204，P＜0.001)、Hb(OR=0.940，95%CI：0.910～0.971，P＜0.001)、ALB(OR=0.814，95%CI：0.742～0.894，P＜0.001)是MHD患者死亡的影响因素。ROC曲线显示建模组模型预测MHD患者死亡的AUC为0.830，采用Bootstrap法显示C-index为0.830，H-L检验χ2=7.621，P =0.704。DCA曲线显示模型在0.13～0.92概率范围内有较好的正向净获益。ROC曲线内部交叉验证显示验证组模型预测MHD患者死亡的AUC为0.844，采用Bootstrap法，显示C-index为0.844，H-L检验χ2=7.412，P=0.681。DCA曲线显示内部交叉验证模型在0.12～0.89概率范围内有较好的正向净获益。 结论  年龄、糖尿病肾病、血管通路类型、hs-CRP、Hb、ALB是MHD患者死亡的影响因素，基于Logistic回归分析构建MHD患者死亡的列线图预测模型区分度较好，净收益值较高，可帮助临床医师优化治疗策略。

Objective  To develop a risk prediction model for mortality in maintenance hemodialysis (MHD) patients based on Logistic regression analysis and a nomogram. Methods Clinical data of 319 MHD patients admitted to Hezhou People's Hospital from January 2019 to January 2021 were retrospectively collected as the training cohort, and data of 80 MHD patients admitted from February 2021 to March 2022 were collected as the validation cohort. Two‑year follow‑up outcomes were reviewed, and patients were divided into a death group and a survival group based on mortality status. Lasso regression was used to screen covariates, and Logistic analysis was performed to identify factors influencing mortality. A nomogram prediction model for mortality in MHD patients was constructed using R software. Internal validation was conducted using the Bootstrap method. Receiver operating characteristic (ROC) curves and calibration curves were used to assess the model's discrimination and calibration, and decision curve analysis (DCA) was applied to evaluate the clinical utility of the nomogram. Results In the training cohort, there were 81 deaths and 238 survivors; in the validation cohort, 16 deaths and 64 survivors. In the training cohort, significant differences were observed between the death group and the survival group in age (t=5.497, P<0.001), diabetic kidney disease (χ²=28.016, P<0.001), vascular access type (χ²=34.491, P<0.001), high‑sensitivity C‑reactive protein (hs‑CRP) (t=42.746, P<0.001), hemoglobin (Hb) (t=15.038, P<0.001), albumin (ALB) (t=11.973, P<0.001), intact parathyroid hormone (iPTH) (t=2.937, P=0.004), and B‑type natriuretic peptide (BNP) (t=2.937, P=0.011). Six predictors were screened by Lasso regression and subsequently included in Logistic analysis, which identified age (OR=1.269, 95% CI: 1.146～1.405, P<0.001), diabetic kidney disease (OR=3.347, 95% CI: 1.149～9.749, P=0.001), vascular access type (OR=3.205, 95% CI: 1.212～8.476, P=0.016), hs‑CRP (OR=2.842, 95% CI: 1.922～4.204, P<0.001), Hb (OR=0.940, 95% CI: 0.910～0.971, P<0.001), and ALB (OR=0.814, 95% CI: 0.742～0.894, P<0.001) as independent factors influencing mortality in MHD patients. The ROC curve showed an AUC of 0.830 for the training model, with a C‑index of 0.830 by Bootstrap and a Hosmer–Lemeshow χ² of 7.621 (P=0.704). The DCA curve indicated a favorable net benefit within the probability range of 0.13–0.92. Internal cross‑validation in the validation cohort yielded an AUC of 0.844, a C‑index of 0.844, and a Hosmer–Lemeshow χ² of 7.412 (P=0.681), with the DCA curve showing a positive net benefit within the probability range of 0.12–0.89.  Conclusion Age, diabetic kidney disease, vascular access type, hs‑CRP, Hb, and ALB are factors influencing mortality in MHD patients. The nomogram prediction model based on Logistic regression demonstrates good discrimination and high net benefit, which may assist clinicians in optimizing treatment strategies.