Developing a Risk Prediction Model for the Occurrence of Heart Failure in Patients on Maintenance Hemodialysis
Objective: To construct a risk prediction model for heart failure (HF) in patients on maintenance hemodialysis (MHD) and to explore the risk factors for heart failure (HF) in MHD patients so as to identify and reduce adverse prognosis at an early stage. Methods: Data of MHD patients hospitalized in the Affiliated Hospital of Qinghai University from 2021 to 2023 were collected. These patients were divided into the HF group (n = 59) and the non-HF group (n = 297) based on whether they developed HF as the outcome event. The baseline characteristics of the two groups were compared, and the patients were randomly divided into a modeling set (n = 250) and a validation set (n = 106) in a ratio of 7:3. The predictor variables were determined by LASSO regression, and a prediction model for HF in MHD patients was constructed by binary logistic regression and a nomogram was drawn. The area under the receiver operating characteristic (ROC) curve was used to evaluate the discrimination of the model. Results: HF occurred in 42 (16.8%) MHD patients in the modeling set and 17 (16.04%) patients in the validation set. LASSO regression combined with Logistic regression analysis showed that hypertension (OR = 4.05, 95% CI = 1.86~9.30), abnormal C-reactive protein (OR = 3.04, 95% CI = 1.45~6.71), dialysis frequency 3 times/week (OR = 4.08, 95% CI = 1.80~9.97), and dialysis age (OR = 1.18, 95% CI = 1.02~1.36) were independent influencing factors for SF in MHD patients (P < 0.05). A risk prediction model including the above four influencing factors was constructed and a nomogram was drawn. The areas under the ROC curve of the prediction model in the modeling set and validation set were 0.785 (95% CI = 0.7124~0.8575) and 0.746 (95% CI = 0.6103~0.8821), respectively, with good discrimination. Conclusion: Hypertension, abnormal C-reactive protein, dialysis frequency 3 times/week, and dialysis age are independent risk factors for HF in MHD patients. The prediction model has good discrimination.
Patients on Maintenance Hemodialysis
维持性血液透析(maintenance hemodialysis, MHD)是终末期肾病患者的主要肾脏替代治疗手段
纳入2021年至2023年青海大学附属医院的356名MHD患者。纳入标准:(1) 符合美国肾病协会关于改善全球肾脏病预后标准的CKD5期的透析病人;(2) 年龄 ≥ 18岁;(3) 规律透析 ≥ 3月,≥2次/周。排除标准:(1) 资料不完整;(2) 合并自身免疫疾病;本研究已获得青海大学医学院伦理委员会审核批准(伦理批件编号:PJ202402-10,受理编号:SL202402-10)。
采用Rstudio 4.3.1进行统计分析。血生化指标按照是否在医学参考值范围内转为二分类资料,符合正态的定量资料以( )表示,使用t检验做组间比较;不符合正态的定量资料以(P50 + IQR)表示,组间采用Wilcoxon检验比较。计数资料以相对数表示,组间比较采用卡方检验。对缺失值小于20%的变量进行多重插补,数据预处理后按7:3划分为建模集(n = 250)和验证集(n = 106),LASSO回归筛选变量,二分类logistic回归构建模型并绘制列线图,采用受试者工作特征(ROC)曲线下面积评估模型的区分度。
根据纳入排除标准,356例MHD患者被纳入到研究中,其中建模集250例,验证集106例,发生HF的59例,未发生HF的297例,筛选流程见
变量 |
总体数据(n = 356) |
训练集(n = 250) |
验证集(n = 106) |
统计量 |
P |
|
干体重,mean ± SD |
61.89 ± 11.17 |
62.03 ± 11.27 |
61.54 ± 10.97 |
0.382 |
0.703 |
|
透析龄,median [IQR] |
1.00 [0.50, 3.00] |
1.00 [0.58, 3.00] |
1.00 [0.44, 3.00] |
14060 |
0.358 |
|
年龄,median [IQR] |
56.00 [47.00, 66.00] |
56.00 [46.00, 66.00] |
56.00 [50.25, 67.00] |
12797.5 |
0.610 |
|
性别,n (%) |
女 |
126 (35.39) |
89 (35.60) |
37 (34.91) |
0.016 |
0.900 |
男 |
230 (64.61) |
161 (64.40) |
69 (65.09) |
|||
高血压,n (%) |
无 |
172 (48.31) |
119 (47.60) |
53 (50.00) |
0.172 |
0.679 |
续表
有 |
184 (51.69) |
131 (52.40) |
53 (50.00) |
|||
糖尿病,n (%) |
无 |
210 (58.99) |
151 (60.40) |
59 (55.66) |
0.691 |
0.406 |
有 |
146 (41.01) |
99 (39.60) |
47 (44.34) |
|||
血红蛋白,n (%) |
正常 |
103 (28.93) |
72 (28.80) |
31 (29.25) |
0.007 |
0.932 |
异常 |
253 (71.07) |
178 (71.20) |
75 (70.75) |
|||
血钙,n (%) |
正常 |
213 (59.83) |
159 (63.60) |
54 (50.94) |
4.962 |
0.026 |
异常 |
143 (40.17) |
91 (36.40) |
52 (49.06) |
|||
血磷,n (%) |
正常 |
113 (31.74) |
73 (29.20) |
40 (37.74) |
2.503 |
0.114 |
异常 |
243 (68.26) |
177 (70.80) |
66 (62.26) |
|||
SF, n (%) |
正常 |
207 (58.15) |
149 (59.60) |
58 (54.72) |
0.729 |
0.393 |
异常 |
149 (41.85) |
101 (40.40) |
48 (45.28) |
|||
HDL, n (%) |
正常 |
61 (17.13) |
45 (18.00) |
16 (15.09) |
0.443 |
0.506 |
异常 |
295 (82.87) |
205 (82.00) |
90 (84.91) |
|||
TG, n (%) |
正常 |
252 (70.79) |
172 (68.80) |
80 (75.47) |
1.602 |
0.206 |
异常 |
104 (29.21) |
78 (31.20) |
26 (24.53) |
|||
CHDL, n (%) |
正常 |
327 (91.85) |
227 (90.80) |
100 (94.34) |
1.246 |
0.264 |
异常 |
29 (8.15) |
23 (9.20) |
6 (5.66) |
|||
CRP, n (%) |
正常 |
178 (50.00) |
131 (52.40) |
47 (44.34) |
1.934 |
0.164 |
异常 |
178 (50.00) |
119 (47.60) |
59 (55.66) |
|||
透析频率,n (%) |
2次/周 |
96 (26.97) |
68 (27.20) |
28 (26.42) |
0.023 |
0.879 |
3次/周 |
260 (73.03) |
182 (72.80) |
78 (73.58) |
|||
透析通路,n (%) |
通路 |
116 (32.58) |
80 (32.00) |
36 (33.96) |
0.130 |
0.718 |
AVF |
240 (67.42) |
170 (68.00) |
70 (66.04) |
|||
iPTH,n (%) |
正常 |
27 (7.58) |
19 (7.60) |
8 (7.55) |
0.000 |
0.986 |
异常 |
329 (92.42) |
231 (92.40) |
98 (92.45) |
|||
心衰,n (%) |
无 |
297 (83.43) |
212 (84.80) |
85 (80.19) |
1.145 |
0.285 |
有 |
59 (16.57) |
38 (15.20) |
21 (19.81) |
|||
LDL, n (%) |
正常 |
344 (96.63) |
240 (96.00) |
104 (98.11) |
0.475 |
0.491 |
异常 |
12 (3.37) |
10 (4.00) |
2 (1.89) |
对所有变量(18个变量)进行共线性诊断,VIF值均 < 5,不存在共线性,LASSO回归分析将变量压缩为4个(透析龄、高血压、C反应蛋白、透析频次),见
以HF (赋值:心衰组 = 1,非心衰组 = 0)为因变量,LASSO回归分析筛选的4个变量作为自变量纳入二分类logistic回归分析,结果显示高血压、透析龄、C反应蛋白、透析频率是MDH患者发生HF的独立危险因素,多因素结果见
变量 |
β |
SE |
z value |
OR (95% CI) |
P |
高血压(有) |
1.398 |
0.407 |
3.43 |
4.05(1.86, 9.30) |
<0.001 |
C反应蛋白(异常) |
1.110 |
0.388 |
2.86 |
3.04(1.45, 6.71) |
0.004 |
透析频率(3次/周) |
1.405 |
0.433 |
3.24 |
4.08(1.80, 9.97) |
<0.001 |
透析龄 |
0.163 |
0.726 |
2.25 |
1.18(1.02, 1.36) |
0.025 |
绘制建模集和验证集的ROC曲线并计算曲线下面积AUC值,结果显示,建模集和验证集中得到的AUC值分别为0.785和0.746。验证集的AUC仅比建模人群下降0.039,表明该预测模型在建模集和验证集中的区分度均较好,见
在长期MHD治疗过程中,会对机体细胞免疫和体液免疫造成损伤,肾脏疾病增加了循环血量,加剧了心力衰竭(HF)的症状和疾病快速进展
本研究通过回顾性分析,建立了一个适用于MHD患者心力衰竭(HF)风险预测的列线图模型。结果表明,高血压病史、透析龄、每周透析频次达到3次以及C反应蛋白水平是MHD患者发生心力衰竭的独立危险因素。这些结果表明,MHD患者并发心力衰竭是一个多因素共同作用的结果。一项针对透析患者的观察性研究结果表明,动脉血压是心力衰竭的独立危险因素
在MHD患者中,C反应蛋白(CRP)水平的升高可能反映了慢性炎症状态,这种状态与心力衰竭的发生和进展有关。炎症不仅是心衰的结果,也是心衰的原因,并可能贯穿于心衰的发生和进展过程
本研究建立了适用于MHD患者心力衰竭风险预测的列线图模型,确定高血压病史、透析龄、每周透析频次 ≥ 3次及C反应蛋白水平为独立危险因素。MHD患者并发心力衰竭是多因素作用结果,连续测量血压、个体化管理血压及监测CRP水平可降低心衰风险。但研究存在局限性,基于小样本回顾性研究可能有选择偏倚,且纳入指标有限。未来应纳入更多MHD风险指标以改善模型预测效果,从而识别高风险群体,采取有效预防和治疗措施。
基于贝叶斯时空模型的青海省健康差异及影响机制研究项目(项目编号:2021-sk-1)。
*通讯作者。