Abstract:
|
The objective was to examine the capacity of models to estimate risk of end-stage renal disease using variables from the electronic health records. We conducted a retrospective cohort study of 28,779 persons with chronic kidney disease (CKD) who received care during 1996-2009. We ascertained ESRD from the national registry thru 2011. We developed and evaluated the performance of proportional hazards models to estimate ESRD risk using criteria: 1) proportion of cases followed (PCF) and 2) proportion of the population needed to be followed (PNF). A model that included 5 common variables (age, race, sex, eGFR, dipstick proteinuria) from the EHR performed well in discriminating progressors and non-progressors, capturing 80% of cases at year 5 by following 13% of the cohort at highest ESRD risk (PNF[.80]=.13). This model performed well in subgroups with hypertension or diabetes. Performance declined substantially in subgroups with moderate (PNF[.80]=.25) and with severe CKD (PNF[.80]=.45). We concluded that a simple model using common variables from the EHR can be used to predict most cases of ESRD. This model may help to guide surveillance in cohorts with low or modest rates of ESRD.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.