An updated NCDR acute kidney injury (AKI) risk model, developed using CathPCI Registry data, refined AKI prediction following PCI and facilitated enhanced clinical care, benchmarking and quality improvement, according to a recent study, while another study that investigated the implications of race-based estimates, found that incorporating a GFR estimate without a Black race term into the NCDR AKI risk prediction model resulted in a better calibrated model to determine risk in Black patients. Both studies were published in JACC: Cardiovascular Interventions.

In the first study, excluding patients on dialysis or lacking postprocedural creatinine, Anezi Uzendu, MD, et al., identified 455,806 PCI procedures (patients’ median age 67, 68.8% men, 86.8% White) from 2020, splitting the cohort into two groups: 70% assigned to the derivation cohort and 30% to the validation cohort. The authors found that AKI occurred in 7.2% of patients following PCI and AKI requiring dialysis occurred in 0.7% of patients.

For the full logistic regression AKI model, 26 candidate variables were initially selected, with the ability “to discriminate AKI with a C-statistic of 0.801 in the deviation cohort and 0.798 in the validation cohort.” The authors then reduced the number of variables to 13. This final AKI model had a C-statistic of 0.798 and “excellent calibration (intercept = –0.03 and slope = 0.97) in the validation cohort,” retaining more than 95% of the explanatory power provided by the initial model.

The strongest predictors of AKI included the clinical severity of patients at the time of PCI such as salvage procedures (odds ratio [OR], 9.54; 95% CI, 8.41-10.82), shock (OR, 7.24; 95% CI, 6.65-7.88) and unresponsive cardiac arrest (OR, 3.85; 95% CI, 3.48-4.25) as well as the presence of preexisting severe chronic kidney disease with a glomerular filtration rate (GFR) <30 (OR, 4.57; 95% CI, 4.32-4.83).

“The current models have better discrimination than the prior NCDR risk model (C-statistic = 0.79 vs. 0.71), primarily because of the expanded data collection in version 5 of the CathPCI data collection form,” write the authors. “The new data elements for responsive and nonresponsive cardiac arrest, frailty, cardiovascular instability, and shock were independently associated with AKI risk and suggest that the increased burden of collecting these data may be offset by their value in better estimating risk, particularly if used to prospectively improve care.”

In the second study, Uzendu, et al., used the same CathPCI Registry dataset to develop four AKI models per patient for each of the following estimates of baseline renal function: the traditional GFR equation with a race term, two GFR equations without a race term, and serum creatinine alone. The authors then compared each model’s ability to predict AKI.

They noted that risk models without a race term were better at predicting AKI in Black patients than models utilizing a race term (intercept = –0.01 vs. 0.15), with models labeled as “race-agnostic” reclassifying 6% of Black patients into higher risk categories. Nevertheless, AKI occurred in Black patients 18% more often than expected, even in models without a race term. This increased risk remains unexplained by disease severity or captured processes of care.

“Although extensive assessment of additional process of care variables could further illuminate other sources of disparate AKI rates by race after PCI, that insidious nature of structural racism and the multiple ways social determinants of health disproportionately impact certain racial minorities’ outcomes after PCI required mixed-method approaches and interdisciplinary investigation to adequately quantify,” write the authors.