6E-06 - Heterogeneity of treatment effects for GLP1-RA on kidney outcomes in patients with type 2 diabetes mellitus

Background: Clinical trials suggest that, on average, glucagon-like peptide-1 receptor agonists (GLP1-RA) lower the risk of kidney outcomes in T2DM patients. However, doctors intuitively know that the average treatment effect doesn’t apply to all patients. Heterogeneity of treatment effects (HTE) recognizes that there are individual variations in treatment response due to differences in baseline health status and various other factors. Identifying and understanding HTE is a crucial step to personalized medicine, where treatments are tailored to the individual patient.

Objectives: To investigate HTE for GLP1-RAs on kidney outcomes in patients with type 2 diabetes by using novel risk-based and machine learning approaches.

Methods: We included adults who newly initiated GLP-1RA or dipeptidyl peptidase-4 inhibitors (DPP-4i) between 2008 and 2021, in Stockholm, Sweden. We quantified the effect of starting one or another treatment on a composite kidney outcome and addressed confounding by PS matching. Three models at time of treatment initiation were used to predict the outcome risk or treatment effect: the established kidney failure risk equation (KFRE), an internal risk-based model using Random Survival Forest (RSF), and an effect-based model using Causal Survival Forest (CSF). 5-year risk differences (RDs) and hazard ratios (HRs) were estimated in the overall cohort and across quartiles of outcome risk or treatment effect estimated by three models.

Results: We identified a total of 16,775 GLP-1RA initiators and 11,772 DPP-4i initiators. After PS matching, 8,043 patient pairs remained. Starting GLP-1RA (vs. DPP-4i) was associated with a lower 5-year risk of kidney outcomes (RD -0.51%; 95% CI -1.08%, 0.05%), corresponding to a HR of 0.75 (95% CI 0.57, 0.99). Patients in the highest quartiles of distribution for the KFRE (RD -2.83% [-5.29%, -0.51%], HR 0.61 [0.40, 0.91]), the RSF risk-based model (RD -3.64% [-6.41%, -1.18%], HR 0.62 [0.43, 0.86]), and the CSF effect-based model (RD [95% CI] -1.90% [-3.85%, -0.27%], HR [95% CI] 0.67 [0.44, 0.93]) demonstrated the largest absolute and relative risk reductions for GLP-1RA vs. DPP-4i. Conversely, there was no meaningful benefit of GLP1-RA for patients within the lower quartiles of outcome risks or treatment effects. Model discrimination of the CSF effect-based model was higher than KFRE and the RSF risk-based model, with c-for-benefit values of 0.683, 0.547, and 0.588 respectively.

Conclusions: In this large cohort of patients starting GLP-1RA vs. DPP-4i, treatment benefits for a composite kidney outcome were largely explained by baseline characteristics. Drug prescription based on expected risks or treatment effect may maximize medication benefits, avoid unnecessary side effects or drug-drug interaction and reduce costs.