Abstract P342: Assessing Interaction in GFR-risk Associations in Meta-analyses: An example of Hypertension in the CKD Prognosis Consortium
Background: Meta-analysis methods for evaluating interactions are not well described. Assessing interactions with eGFR is particularly challenging due to its non-linear association with risk. We describe methods to evaluate point-wise and overall interactions using an example of hypertension-eGFR interaction on mortality risk.
Methods: The applicability of our methods was evaluated in 28 out of 46 cohorts joining the Chronic Kidney Disease Prognosis Consortium (CKD-PC). In each study, eGFR linear splines (knots at each 15 from 15 to 105 ml/min/1.73m²) and their product terms with hypertension were fitted, providing hazard ratios (HRs) for eGFR (vs. eGFR 95) in both hypertensive and non-hypertensive groups. From this model, the interaction was evaluated as the relative HR (rHR) in hypertension vs. non-hypertension at each 1 ml/min/1.73m² of eGFR between 15 and 120 (point-wise interaction). HRs and rHRs for each eGFR value in each cohort were then pooled using random effects models. The overall interaction was assessed as the inverse-variance average of differences in all spline coefficients for the entire range of eGFR between hypertensive vs. non-hypertensive groups.
Results: There were 942,429 participants and 69,072 deaths. Low eGFR was associated with increased mortality in the hypertensive (gray line) and non-hypertensive (black line) groups (figure). Significant interactions by hypertension status were observed at eGFR range below 60 ml/min/1.73m² at each eGFR value p-interaction<0.05 is noted by “x” in the figure. In this range, the impact of low eGFR on mortality was weaker in the hypertensive group as compared with the non-hypertensive group. Overall interaction test was also significant (P<0.001).
Conclusion: We developed and implemented meta-analytic methods to assess point-wise and overall interactions in non-linear regression models. Interactions can be comprehensively evaluated using both methods and should be interpreted based on biological and clinical relevance.
- © 2012 by American Heart Association, Inc.