Abstract
Background/Aims Single time-point GFR estimates are used for risk prognostication among those with CKD. Various applications of longitudinal eGFR using linear models, including time-independent slope and variability, have been shown to independently predict cardiovascular events. We wished to extend these applications by modeling longitudinal eGFR–forming estimates of current eGFR value and time-dependent eGFR slope at any point in time–and then test the independent association with cardiovascular event risk in an adjusted, joint longitudinal-survival model.
Methods We retrospectively assembled a cohort of Geisinger primary care patients with stage G3–G4 CKD(1/1/2001 to 6/30/2012) who had a minimum of 4 outpatient eGFR results and 6-months follow-up. Exclusion criteria included a history of dialysis or renal transplantation. Patients were followed through 12/31/2012 for the first occurrence of MI, HF, or stroke (CVD event), or were censored at ESRD, death or end of study. EGFR was modeled using a random intercept and slope model with a flexible spline transformation of time to capture non-linear trends. Time to first CVD event post-CKD diagnosis was modeled using a Cox Proportional Hazard regression model. The longitudinal and survival models were then jointly estimated.
Results A total of 24,777 patients met cohort entry criteria. A random sample of 90% (22,300) was used for model development (mean age 69.3 y, 42% (9409) male, 28% (6277) with diabetes, mean baseline eGFR 51 mL/min) and contributed 119,165 person-years of follow up. 6% (1339) had a prior CVD event. During follow-up 4216 (19%) patients had a post-CKD CVD event. In fully adjusted models, each 5 mL/min/1.73m2 decrease in current eGFR was associated with a hazard ratio (HR) of 1.05 (95% CI 1.04, 1.06). For the time-dependent slope of eGFR trend, each 2 mL/min/1.73m2/year decrement was associated with a HR of 1.06 (95% CI 1.03, 1.10). Both diabetes and gender modified the eGFR slope associations, such that the associations were significantly greater among non-diabetics (HR = 1.17) and females (HR = 1.11).
Conclusions Longitudinally modeled eGFR and time-dependent eGFR slope each independently associates with CVD risk among patients with moderate and advanced CKD. The relative prognostic value of static, single time point eGFR vs. joint modeling of eGFR warrants further investigation.
