Table 1

Analyses of association between use of GLP-1 receptor agonists versus DPP-4 inhibitors and risk of diabetic retinopathy complications

GLP-1 receptor agonistsDPP-4 inhibitorsUnadjusted HR (95% CI)Adjusted HRb (95% CI)
NEventsEvents per 1,000 patient-yearsNEventsEvents per 1,000 patient-years
Primary analysis
 Total cohorta6,65090969.411,6301,16259.31.25 (1.15–1.36)1.07 (0.95–1.20)
Subgroup analyses
 Country
  Denmark2,82137053.73,49732247.91.25 (1.08–1.45)1.02 (0.82–1.26)
  Sweden3,82953986.68,13384065.21.34 (1.20–1.49)1.12 (0.98–1.28)
 Sex
  Women2,76235464.54,73944858.71.20 (1.05–1.38)1.04 (0.87–1.25)
  Men3,88855572.86,89171459.71.29 (1.15–1.44)1.08 (0.93–1.25)
 Age (years)
  <653,94258971.74,26652168.41.12 (0.99–1.26)1.00 (0.87–1.15)
  ≥652,70832065.57,36464153.51.28 (1.12–1.47)1.13 (0.94–1.35)
 Insulin use
  No1,55415848.37,19060646.41.12 (0.94–1.33)1.18 (0.95–1.45)
  Yes5,09675176.44,44055684.91.00 (0.89–1.11)0.98 (0.86–1.11)
 Glycated hemoglobin (Sweden)c
 <8.7% (72 mmol/mol)1.17 (0.94–1.46)
 ≥8.7% (72 mmol/mol)1.04 (0.87–1.26)
Additional analysis
 Truncated follow-up time at 6 months after cohort entry6,650410135.911,630602116.71.17 (1.03–1.33)1.04 (0.88–1.24)
Sensitivity analyses
 Intention-to-treat exposure definitiond6,6501,15962.411,6301,51553.31.23 (1.14–1.33)1.03 (0.93–1.14)
 Truncated weightse6,65090969.411,6301,16259.31.25 (1.15–1.36)1.08 (0.97–1.20)
 Additionally adjusted model (Sweden)f3,82953986.68,13384065.21.34 (1.20–1.49)1.09 (0.95–1.25)
  • a Mean follow-up time (SD): 1.8 (1.5) years [2.0 (1.6) years for GLP-1 receptor agonists and 1.7 (1.4) years for DPP-4 inhibitors]. Total follow-up time: 13,105 patient-years for GLP-1 receptor agonists and 19,602 patient-years for DPP-4 inhibitors.

  • b Inverse probability of treatment weighting based on a propensity score that included 60 covariates.

  • c Analysis of patients in Sweden. There were missing data on glycated hemoglobin, and multiple imputation (Markov chain Monte Carlo method) was used to create 10 imputed data sets. Because each imputation yielded different subgroups of the total population, the number of patients, number of events, incidence rates, and unadjusted HRs are not presented.

  • d Patients were considered exposed to the study drug throughout follow-up.

  • e Inverse probability of treatment weighting can generate very large weights for patients with low probability of treatment. In this analysis, weights >5 were set to 5.

  • f Analysis of patients in Sweden additionally adjusted for glycated hemoglobin, blood pressure, albuminuria, estimated glomerular filtration rate, BMI, and smoking. Because there were missing data for all these variables, multiple imputation (Markov chain Monte Carlo method) was used to create 10 imputed data sets.