Prediction of mortality in the first two years of hemodialysis: results from a validation study

Chronic hemodialysis (HD) patients experience high rates of mortality. Alerting medical staff of patients at increased risk of death may support clinical decision making.

A large cohort of incident HD patients was used to develop logistic regression models to predict death in the subsequent 6 months ('derivation cohort'). Predictors were age, gender, race, ethnicity, vascular access type, diabetic status, pre-HD systolic and diastolic blood pressure, pre-HD weight, pre-HD temperature, relative interdialytic weight gain, serum albumin, hemoglobin, phosphorus, serum creatinine, serum sodium, urea reduction ratio, equilibrated normalized protein catabolic rate, and equilibrated dialytic and renal Kt/V. These logistic regression models were then applied to validation cohorts. Predictive performance of the models was described in terms of sensitivity, specificity, and area under receiver-operating characteristic curves (AUC-ROC).

A total of 6,838 incident HD patients were followed over 2 years. The derivation cohort initially comprised 4,512 patients. In the validation cohort of initially 2,326 patients, the logistic regression models were able to predict mortality in subsequent 6-month periods with a sensitivity between 58 and 69%, and a specificity of 74-77%; the respective AUC-ROC were 0.67-0.72 (all p < 0.0001). Pre-HD weight and serum albumin levels were consistently significant predictors of mortality in all models.

The results indicate that logistic regression models are useful tools in estimating incident HD patients' probability of death in 6-month intervals for at least up to 2 years after beginning dialysis. Model predictions may be used to alert medical staff to patients at increased risk of death and facilitate timely diagnostic and therapeutic interventions to improve outcomes.