Estimating baseline kidney function in hospitalized patients with impaired kidney function

Abstract

Background and objectives: Inaccurate determination of baseline kidney function can misclassify acute kidney injury (AKI) and affect the study of AKI-related outcomes. No consensus exists on how to optimally determine baseline kidney function when multiple preadmission creatinine measurements are available.

Design, setting, participants, & measurements: The accuracy of commonly used methods for estimating baseline serum creatinine was compared with that of a reference standard adjudicated by a panel of board-certified nephrologists in 379 patients with AKI or CKD admitted to a tertiary referral center.

Results: Agreement between estimating methods and the reference standard was highest when using creatinine values measured 7-365 days before admission. During this interval, the intraclass correlation coefficient (ICC) for the mean outpatient serum creatinine level (0.91 [95% confidence interval (CI), 0.88-0.92]) was higher than the most recent outpatient (ICC, 0.84 [95% CI, 0.80-0.88]; P<0.001) and the nadir outpatient (ICC, 0.83 [95% CI, 0.76-0.87; P<0.001) serum creatinine. Using the final creatinine value from a prior inpatient admission increased the ICC of the most recent outpatient creatinine method (0.88 [95% CI, 0.85-0.91]). Performance of all methods declined or was unchanged when the time interval was broadened to 2 years or included serum creatinine measured within a week of admission.

Conclusions: The mean outpatient serum creatinine measured within a year of hospitalization most closely approximates nephrologist-adjudicated serum creatinine values.

Figure 1.

Preadmission creatinine value availability. Data are shown as box plots grouped by days before admission on the x-axis and further stratified by inpatient, outpatient, or either location of measurement. Counts of measurement are indicated on the y-axis. Horizontal lines within each box denote the median number of available serum creatinine counts, with the bottom and top of each box representing the 25th and 75th percentiles, respectively. The bottom/top whisker corresponds to the lowest/highest count that falls within 1.5 times the interquartile range below/above the 25th/75th percentile of counts.

Figure 2.

Bland-Altman plots comparing selected approaches for estimating baseline kidney function with expert adjudication during the 7- to 365-day time interval. (A) Most recent outpatient. (B) Mean outpatient. (C) Nadir outpatient. (D) Most recent inpatient or outpatient. x-axis represents the average of serum creatinine (SCr) value between the estimation method and adjudicated value. y-axis indicates the difference in serum creatinine values between the adjudicated value and the estimation method. The dotted lines represent a 0.3-mg/dl deviation in either direction. The right side of each figure shows a histogram of the difference between the adjudicated value and the estimation method. The percentage values show the potential bias in AKI diagnosis indicating the estimated values at least 0.3 mg/dl deviations above or below the adjudicated value. ICC, intraclass correlation coefficient.

Figure 2.

Bland-Altman plots comparing selected approaches for estimating baseline kidney function with expert adjudication during the 7- to 365-day time interval. (A) Most recent outpatient. (B) Mean outpatient. (C) Nadir outpatient. (D) Most recent inpatient or outpatient. x-axis represents the average of serum creatinine (SCr) value between the estimation method and adjudicated value. y-axis indicates the difference in serum creatinine values between the adjudicated value and the estimation method. The dotted lines represent a 0.3-mg/dl deviation in either direction. The right side of each figure shows a histogram of the difference between the adjudicated value and the estimation method. The percentage values show the potential bias in AKI diagnosis indicating the estimated values at least 0.3 mg/dl deviations above or below the adjudicated value. ICC, intraclass correlation coefficient.