Background: We explored the extent to which neutrophil gelatinase-associated lipocalin (NGAL) cutoff value selection and the acute kidney injury (AKI) classification system determine clinical AKI-phenotype allocation and associated outcomes.
Methods: Cutoff values from ROC curves of data from two independent prospective cardiac surgery study cohorts (Magdeburg and Berlin, Germany) were used to predict Kidney Disease: Improving Global Outcome (KDIGO)- or Risk, Injury, Failure, Loss of kidney function, End-stage (RIFLE)-defined AKI. Statistical methodologies (maximum Youden index, lowest distance to [0, 1] in ROC space, sensitivity≍specificity) and cutoff values from two NGAL meta-analyses were evaluated. Associated risks of adverse outcomes (acute dialysis initiation and in-hospital mortality) were compared.
Results: NGAL cutoff concentrations calculated from ROC curves to predict AKI varied according to the statistical methodology and AKI classification system (10.6-159.1 and 16.85-149.3 ng/mL in the Magdeburg and Berlin cohorts, respectively). Proportions of attributed subclinical AKI ranged 2%-33.0% and 10.1%-33.1% in the Magdeburg and Berlin cohorts, respectively. The difference in calculated risk for adverse outcomes (fraction of odds ratios for AKI-phenotype group differences) varied considerably when changing the cutoff concentration within the RIFLE or KDIGO classification (up to 18.33- and 16.11-times risk difference, respectively) and was even greater when comparing cutoff methodologies between RIFLE and KDIGO classifications (up to 25.7-times risk difference).
Conclusions: NGAL positivity adds prognostic information regardless of RIFLE or KDIGO classification or cutoff selection methodology. The risk of adverse events depends on the methodology of cutoff selection and AKI classification system.
Keywords: AKI phenotypes; Acute kidney injury; Cardiac surgery; Cutoff; Dichotomization; Neutrophil gelatinase-associated lipocalin; ROC; Risk assessment; Risk prediction; Subclinical AKI.