Estimating Kidney Failure Risk Using Electronic Medical Records.

Area of Science:

• Nephrology
• Biostatistics
• Health Informatics

Background:

• The four-variable kidney failure risk equation (KFRE) predicts kidney failure risk in patients with reduced GFR.
• Electronic medical record implementation of KFRE is hindered by low urine albumin-to-creatinine ratio (ACR) testing rates.
• Alternative methods for estimating ACR are needed to enhance KFRE utility.

Purpose of the Study:

• To evaluate the utility of imputed ACR from protein-to-creatinine ratio (PCR) or dipstick protein in the four-variable KFRE.
• To compare the performance of the three-variable KFRE versus the four-variable KFRE with imputed ACR.
• To assess the feasibility of integrating imputed ACR into KFRE for clinical practice.

Main Methods:

• Utilized electronic health records from OptumLabs Data Warehouse for patients with eGFR <60 ml/min/1.73 m².
• Categorized patients based on the availability of ACR, PCR, or dipstick protein testing within three years.
• Compared the predictive discrimination and calibration of the three-variable KFRE against the four-variable KFRE using imputed ACR values.

Main Results:

• Out of 976,299 patients, only 19% had ACR testing; PCR and dipstick protein were available for 2% and 36%, respectively.
• The four-variable KFRE demonstrated significantly better discrimination than the three-variable KFRE across all tested groups, including those with imputed ACR.
• While calibration was acceptable for the four-variable KFRE, the three-variable KFRE showed systematic bias in groups lacking ACR or PCR data.

Conclusions:

• The four-variable KFRE performs better than the three-variable KFRE, even when ACR is imputed.
• Imputing ACR from PCR or dipstick protein is a viable strategy for enhancing KFRE implementation in electronic medical records.
• Integrating ACR, even imputed, into the KFRE is recommended for improved kidney failure risk prediction.