Discussion
In this pharmacovigilance analysis of disproportionality, we found that GLP-1RAs are not associated with an increased reporting rate for retinal AE, but rather with a significantly lower reporting ratio compared with other GLMs. Results are consistent across different models and subanalyses.
Together with the results of a recent meta-analysis,11 these data reassure the risk of retinopathy progression raised by the SUSTAIN-6 trial.3 In the absence of a specific drug–AE relation, we support the hypothesis that retinopathy progression in patients who received semaglutide in the SUSTAIN-6 trial may be attributed to the rapid improvement in glucose control,17 in insulin-treated patients with pre-existing retinopathy.18
Most specific retinal AEs were reported less frequently in association with GLP-1RA than other GLMs, with a few notable exceptions. Hemorrhagic retinopathy occurred more frequently in reports listing versus those not listing GLP-1RA, but this difference did not reach statistical significance and a quite similar and more common AE (retinal hemorrhage) occurred much less frequently and with high statistical significance in reports with GLP-1RA. The only AE listed significantly more often in reports with than in those without GLP-1RA was retinopexy, an intervention performed to repair retinal detachment, which can occur as a complication of advanced diabetic retinopathy. However, numbers were very small and retinal detachment occurred much less frequently and with high statistical significance in reports with GLP-1RA. Therefore, these data reasonably indicate that there was no consistent signal in the FAERS that GLP-1RA may be associated with AEs suggestive of diabetic retinopathy progression. For the evaluation of an emerging AE, drugs indicated as suspect or concomitant are considered equally because new signals may uncover for drugs not previously known to be responsible for such AE. Thus, the imbalanced distribution of suspect or concomitant drugs between GLP-1RA and other GLMs is irrelevant to the safety evaluation of retinal AE reports.
In the FAERS, up to 30% of reports listing GLM do not include a diabetes indication. This issue was noted previously,13 19 and may reflect that GLMs were being used for the treatment of pre-diabetes (eg, metformin and acarbose)20 21 or obesity (eg, liraglutide).22 However, incompleteness of reports likely contributes to the high number of reports listing GLM without a diabetes indication. Nonetheless, as retinopathy is a specific complication of diabetes, it is important to filter the diabetes indication to verify robustness of data. Vice versa, some reports (2.5%) including diabetes as a concomitant condition may not list any GLM. Our data indicate that disproportionality of retinal AEs for GLP-1RA is not affected by the diabetes indication and associated GLM. Particularly, among GLMs, insulin therapy is typically considered as a proxy of disease severity, and retinopathy occurs more frequently in patients on insulin.23 Indeed, in the FAERS retinal AEs were more than four times more frequent in reports listing (11.7/1000) than in those not listing insulin (2.9/1000). Importantly, retinal AEs were less frequently included in reports with GLP-1RAs than in those without, irrespective of concomitant insulin therapy.
It should be noted that lower rates of retinal AEs were associated with GLP-1RAs despite reports with GLP-1RAs listing more comorbid conditions and concomitant medications, including those for the treatment of hypertension, which is a risk factor for retinopathy progression. Therefore, the lower rate of retinal AEs is unlikely attributable to a lighter disease burden in GLP-1RA users. However, information on glucose and blood pressure control, which are major determinants of retinopathy, is not available in pharmacovigilance databases. As a result, differences that likely exist between patients on GLP-1RA and those on other GLMs could not be adjusted for.
The FAERS does not yet contain AE reports for semaglutide, but the analysis by individual GLP-1RAs showed consistent results. Remarkably, no difference was noted between exenatide and liraglutide, which are two GLP-1RAs with very different chemical structures. The lower PRR associated with less common GLP-1RA likely reflects the small number of AEs filed for such drugs.
Several limitations of FAERS data have to be acknowledged. Above all, in most reports there is no demonstration of a causal relationship between drug exposure and the AE, and some reports are filed by non-healthcare professionals. The inability to make causal inference is a limitation of all pharmacovigilance studies and it is shared by cohort observational studies. PRR in the FAERS does not inform on the true risk in clinical practice, because the population at risk is extensively under-represented. Furthermore, known drug AEs tend to be reported more frequently, thereby diluting new AE signals, and incompleteness of reports limits the emergence of links between concomitant conditions or drugs and new AEs.
Screening for retinopathy onset and progression should be performed routinely in patients with diabetes, but eventual drug-associated retinal AEs may not be captured in clinical practice as they are in clinical trials. In the FAERS, whether AE reports derived from clinical trials or routine pharmacovigilance cannot be reliably ascertained. Furthermore, there are no clear criteria to define whether a retinal event is to be considered part of the disease’s natural history or a suspected adverse drug reaction. These difficulties may lead to an under-reporting of drug-associated retinal AEs.
Notwithstanding these limitations, pharmacovigilance assessment is an extremely helpful tool to monitor drug safety and detect new safety signals. There are several examples of how this can be applied to diabetes pharmacotherapy. For instance, the initial warning that sodium glucose cotransporter inhibitor 2 (SGLT2) inhibitors can cause diabetic ketoacidosis was primed by a disproportionality in the FAERS.24 Furthermore, FAERS analysis has recently confirmed that the SGLT2 inhibitor canagliflozin can increase the risk of lower extremity amputations.13
In summary, our data indicate that, in the FAERS, retinal AEs are not disproportionally associated with GLP-1RA. As causal inference is impossible, we cannot conclude that GLP-1RA protects from retinopathy events, but these data are important from a safety perspective. Although we await for postmarketing information on semaglutide, our present study complements information originating from RCTs and reassures the risk of retinopathy progression associated with GLP-1RA therapy.