Background: This observational study uses rigorous causal inference methods of comparative effectiveness research to emulate a trial evaluating the cardiovascular safety of analog (AI) vs. human insulins (HI).

Methods:On date of first insulin use, 127,600 adults with type 2 diabetes were classified as using: AI with or without HI, or HI only. Separate analyses assessed occurrence of: myocardial infarction (MI), stroke (CVA), congestive heart failure hospitalization (CHF), cardiovascular (CV) death, or death, censored at disenrollment, start of pregnancy, or 12/31/2013. Marginal structural modeling (MSM) with Inverse Probability Weighting (IPW) was used to compare event free survival in per-protocol analyses in which follow-up data was right-censored for cross-over or interrupted insulin therapy. Adjusted and unadjusted effect estimates were based on two logistic MSM parameterizations for counterfactual hazards. To address potential residual confounding, we conducted four sensitivity analyses with three nested covariate adjustment sets and machine learning for data-adaptive selection of covariates.

Results:Mean follow-up time was 21 months. The Table shows hazard ratios (95% CI) of AI vs. HI users for events of interest.

Conclusions: We observed no statistically significant differences in 5 major clinical outcomes between users of AI and HI.


E.B. Schroeder: None. P.J. O'Connor: None. J. Schmittdiel: None. K. Reynolds: Research Support; Self; Amarin Corporation, Novartis Pharmaceuticals Corporation, Amgen Inc., CSL Behring. J.R. Desai: None. M. Ho: None. G. Vazquez Benitez: None. J.P. Anderson: None. N. Pimentel: None. L.M. Loes: None. R. Neugebauer: None.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at