Clear evidence indicated that the protective effects of ARBs against progression to diabetes were superior to those of placebo or other active anti-hypertensive treatment [1–16]. Meta-analyses showed that, overall, ARBs were associated with a 15 ~ 25% reduction in the risk of diabetes [3, 28]. However, no solid evidence shows that all ARBs pose the same protective effect against diabetes. Instead, our findings indicate differential associations of ARBs with diabetes risk. In a large nationwide cohort, we found that olmesartan posed a slightly higher diabetes risk while candesartan seem to be associated with reduced diabetes risk.
Our data was in part in line with previous studies showing consistent protective effects of candesartan on diabetes risk. In comparison with placebo, candesartan therapy was associated with fewer newly diagnosed diabetes in patients with heart failure  and possibly also in the elderly patients . When compared with active treatment (i.e., amlodipine and non-ARB antihypertensive agents), candesartan was also associated with a lower risk of diabetes in hypertensive patients with or without coronary artery disease [10, 11].
Previous studies demonstrated that ARBs possess various PPARγ activation activities. Telmisartan seems to have strongest activity, followed by candesartan and irbesartan; while other ARBs have little activities [20–23]. In our study, however, telmisartan was not associated with lower diabetes incidence as expected. This is consistent with a large randomized clinical trial showing no effect of telmisartan on diabetes incidence as compared to placebo . The explanation for this discrepancy is not clear but may be due to pleiotropic effects of ARBs on glucose metabolism beyond PPARγ activation. Indeed, candesartan was shown to improve insulin sensitivity through PPARγ-independent mechanism and to increase insulin content in pancreatic beta-cells by attenuating oxidative stress [27, 30]. Losartan, irbesartan and valsartan, and telmisartan have been shown to exert various anti-diabetic effects other than PPARγ activation, including augmentation of blood flow to muscle, direct modulation of insulin signaling and up-regulation of glucose transporter expression in muscle, reduction of islet fibrosis through inhibition of TGF-β, and activation of AMPK/SIRT1 pathway [31–35]. These data, together with our observation, support heterogeneous anti-diabetic effects of ARBs.
Our study has some unique strength. First, this is currently the only study comparing diabetes risk for individual ARB. Second, this study is a nationwide population-based cohort study. The huge sample size and long duration of follow-up enable this study adequately powered to detect difference among ARBs. Third, the intention-to-treat analysis preserves the baseline comparability of the treatment groups, reduces the potential bias due to drug switching or discontinuation, and provides a conservative estimate. Therefore, the results provided real-life estimations of diabetes risk associated with individual ARB, which have important clinical implications. Diabetes is a strong risk factor of cardiovascular disease, renal failure, and retinopathy which imposes enormous economic burden to the health care system [36–38]. The changed diabetes incidence with different ARB therapy would translate directly to mortality and medical costs. Fourth, different analytic strategies, including the ITT and exclusive user analyses, showed similar results and indicated an internal consistency among these results.
Our study has some limitations. First, this is an observational study but not a randomized clinical trial. Therefore, the association between individual ARBs and new-onset diabetes may be affected by unknown or unmeasured confounders. Laboratory and anthropometric data such as fasting glucose or body mass index were not available in the NHI database. However, these factors are unlikely to confound the results since physicians did not prescribe specific ARB according to these factors. Furthermore, our analyses have been fully adjusted for these baseline co-morbidities. Second, imperfect adherence to therapy or discontinuation of therapy could have attenuated our intention-to-treat effect estimate. Third, the definition of new-onset diabetes was made according to the ICD-9-CM code but not by screening tests. Although the definition algorithm using ICD-9-CM diagnosis code has a high positive predictive value, some diagnoses of diabetes might be missed. Nevertheless, the magnitude of increased risk associated with olmesartan changed little (from 7% to 5%) when more strict definition (diagnosis code plus anti-diabetic therapy) was applied in the sensitivity analysis. The under-estimation due to un-diagnosed diabetes or latent diabetes should be minimal.
In conclusion, we found a small but significantly increased diabetes incidence in olmesartan initiators as compared to losartan initiators in a large nationwide population-based cohort. Telmisartan was not associated with reduced diabetes incidence. These findings suggest heterogeneous diabetes risks associated with different ARBs beyond a class effect, but the difference in diabetes risk does not seem to correlate with PPARγ activation activities. Further study is needed affirm this observation.