Predicting incident heart failure in CKD using eGFR slopes: head-to-head comparison of four estimating equations
C Kaufman, S E Claudel, A VermaAbstract
Background
Longitudinal eGFR decline predicts incident heart failure (HF) in chronic kidney disease (CKD), but prognostic utility may vary by filtration marker. Creatinine-based eGFR can be biased by non-GFR determinants (e.g., low muscle mass), whereas cystatin C–based eGFR may reflect systemic inflammation. Evidence comparing which equation yields the most informative eGFR slope for HF risk stratification is limited.
Purpose
The primary aim was to quantify and compare the hazard ratio (HR) per 1–standard deviation (SD) faster eGFR decline across all four equations, followed by head-to-head tests of prognostic superiority of the 2012 eGFR_Cys and 2021 eGFR_Cr-Cys slopes versus the 2021 eGFR_Cr slope.
Methods
We studied 4,236 adults with CKD in CRIC; the outcome was centrally adjudicated incident HF after baseline. Annualized eGFR slopes were estimated with mixed-effects models for four CKD-EPI equations (2009 eGFR_Cr, 2021 eGFR_Cr, 2021 eGFR_Cr-Cys, 2012 eGFR_Cys) and modeled per 1-SD faster decline in multivariable Cox models adjusted for demographics, cardiovascular risk factors, medications, hemoglobin, proteinuria, and baseline eGFR. Prognostic performance was compared using prespecified metrics for fit (LR χ²/AIC), discrimination (C-statistic), calibration stability (bootstrap calibration slope/linearity; R²), and reclassification (net reclassification improvement [NRI] and integrated discrimination improvement [IDI]).
Results
Participants had a mean age of 59.5 years (SD 10.6); 55.6% were men and 42.6% were Black. Over a median follow-up of 5.8 years, 507 incident HF events occurred. Participants who developed HF were older and had higher body mass index and systolic blood pressure, with a greater prevalence of diabetes (66.3% vs 46.0%) and prior cardiovascular disease (50.1% vs 27.0%) (all P<0.001). Baseline eGFR was lower across all equations among those who developed HF, and absolute annual eGFR decline was steeper. In adjusted models, all standardized slopes were independently associated with incident HF (Table 1): 2021 eGFR_Cr HR 1.35 (95% CI 1.21–1.51), 2009 eGFR_Cr HR 1.36 (95% CI 1.22–1.52), 2021 eGFR_Cr-Cys HR 1.43 (95% CI 1.27–1.60), and 2012 eGFR_Cys HR 1.47 (95% CI 1.30–1.66). In head-to-head comparisons (Table 2), the 2012 eGFR_Cys slope demonstrated the strongest model fit (likelihood ratio χ² 475.7), highest discrimination (C-statistic 0.772), and greatest calibration stability (bootstrap R² 0.93). The 2021 eGFR_Cr-Cys slope showed improved performance relative to 2021 eGFR_Cr but with greater calibration instability (R² 0.86).
Conclusion
In CKD, the eGFR slope derived from the 2012 cystatin C–based equation was most strongly associated with incident HF and demonstrated superior overall prognostic performance compared with creatinine-based and combined equations. These findings support cystatin C–based eGFR trajectories as a preferred kidney biomarker for longitudinal HF risk stratification in CKD.For image description, please refer to the figure legend and surrounding text.For image description, please refer to the figure legend and surrounding text.