Format

Send to

Choose Destination
Clin Nephrol. 2019 Feb;91(2):65-71. doi: 10.5414/CN109310.

External validation of prediction models for mortality in an incident dialysis population
.

Abstract

AIMS:

Different prediction models have been established to estimate mortality in the dialysis population. This study aims to externally validate the different available mortality prediction models in an incident dialysis population.

MATERIALS:

This was a retrospective cohort study of incident hemodialysis and peritoneal dialysis patients at two academic tertiary care centers.

METHODS:

Three previously published prediction models were used: the Liu index, the Urea5 score, and a predictive model estimating the survival probability by Hemke et al. [6]. Models were compared using the C-statistic, net reclassification index, and integrated discrimination improvement. Only the subgroup of 193 patients with enough data to be included in all models was used.

RESULTS:

377 patients were started on dialysis in both institutions between 2006 and 2011. Median follow-up was 787 days. 104 patients (27.6%) died during follow-up and 181 were admitted to the hospital (48.0%). All three models were predictive of mortality and hospital admissions. The survival probability model by Hemke et al. [6] performed better than the other two models for mortality (C-statistic 0.72). The Liu index had the highest performance for hospital admissions (C-statistic 0.65). Using reclassification statistics (reference = Urea5), the only model to improve discriminatory ability was the Liu index for the outcome of hospital admission.

CONCLUSION:

The survival probability model by Hemke et al. [6] may be preferred for mortality prediction in incident dialysis patients. The Liu index could be used to predict hospital admissions in the same population. Available models demonstrated only modest performance in predicting either outcome. Therefore, alternative models need to be developed.
.

PMID:
30526813
DOI:
10.5414/CN109310
[Indexed for MEDLINE]

Supplemental Content

Full text links

Icon for Dustri-Verlag Dr. Karl Feistle GmbH & Co. KG
Loading ...
Support Center