Development of a prediction model to estimate the 5-year risk of cardiovascular events and all-cause mortality in haemodialysis patients: a retrospective study

Abstract

Background Cardiovascular disease (CVD) is a major cause of mortality in patients on haemodialysis. The development of a prediction model for CVD risk is necessary to help make clinical decisions for haemodialysis patients. This retrospective study aimed to develop a prediction model for the 5-year risk of CV events and all-cause mortality in haemodialysis patients in China. Methods We retrospectively enrolled 398 haemodialysis patients who underwent dialysis at the dialysis facility of the General Hospital of Northern Theater Command in June 2016 and were followed up for 5 years. The composite outcome was defined as CV events and/or all-cause death. Multivariable logistic regression with backwards stepwise selection was used to develop our new prediction model. Results Seven predictors were included in the final model: age, male sex, diabetes, history of CV events, no arteriovenous fistula at dialysis initiation, a monocyte/lymphocyte ratio greater than 0.43 and a serum uric acid level less than 436 mmol/L. Discrimination and calibration were satisfactory, with a C-statistic above 0.80. The predictors lay nearly on the 45-degree line for agreement with the outcome in the calibration plot. A simple clinical score was constructed to provide the probability of 5-year CV events or all-cause mortality. Bootstrapping validation showed that the new model also has similar discrimination and calibration. Compared with the Framingham risk score (FRS) and a similar model, our model showed better performance. Conclusion This prognostic model can be used to predict the long-term risk of CV events and all-cause mortality in haemodialysis patients. An MLR greater than 0.43 is an important prognostic factor.

Risk prediction models for cardiovascular events in hemodialysis patients: A systematic review

“…After reviewing the titles and abstracts, 3224 articles were excluded, leaving 86 articles to be read in full text. Ultimately, nine articles [14][15][16][17][18][19][20][21][22] containing 12 models were included (Figure 1). indicates not mentioned.…”

“…Three prospective cohort studies 14,17,19 and six retrospective cohort studies 15,16,18,[20][21][22] were included in the nine articles. The total sample size comprised 16,529 patients, with 3651 patients experiencing an outcome event.…”

“…The nine included papers reported 12 prediction models, Cox regression was used in six papers 14,[16][17][18][19]22 to develop the models, while logistic regression was employed in three papers. 15,20,21 For model validation, four papers 15,18-20 conducted internal validation exclusively, one paper 16 focused solely on external validation, one paper 14 employed a combination of internal and external validation, and three papers 17,21,22 did not include any validation. The AUC of the included models ranged from 0.70 to 0.88.…”

“…The AUC of the included models ranged from 0.70 to 0.88. Four papers 14,15,19,20 considered the overfitting and performed model calibration (Table 2).…”

“…The PROBAST inventory evaluation revealed that all but one 14 of the nine included studies had a high risk of overall bias. The high risk of bias was mainly concentrated in the area of data analysis, particularly in the following five areas: (1) The presence of nonlinear relationships between continuous variables during the transformation process was not tested in five studies 15-17,19,20 ; (2) All participants were not included in the analysis in six studies 15,16,18,[20][21][22] ;…”