This systematic review of observational studies aimed to calculate the absolute risk reduction (ARR) of heart failure events in people treated with sodiumglucose co-transporter-2 inhibitors (SGLT2-i).

The authors meta-analyzed 14 studies and found greater ARR in subjects with baseline cardiovascular disease (CVD), with a number needed to treat ranging from 70 (CVD group) to 580 (without CVD), in line with size effect from RCTs and supporting current guideline recommendations.

The systematic review is well executed (protocol registered in advance in PROSPERO), clear elegibility criteria, search strategy, and various pre-planned analyses according to CVD at baseline, large cohorts, new-user design, geographical origin, comparators.

The topic is of interest, although the following major issues deserve attention by the authors:

• The authors mentioned in the discussion only one recent meta-analysis on real world data (failing to address the issue of CVD at baseline) [PLoS One. 2021 Feb 19;16(2):e0244689].

However, by simply performing a general search in Medline, I found at least 3 additional systematic reviews accounting for observational research potentially addressing heart failure with SGLT2-i [Expert Rev Clin Pharmacol. 2019 Apr;12(4):299-308. Pharmacol Res. 2021 Oct;172:105836. doi: 10.1016/j.phrs.2021.105836. Diabet Med. 2021 Sep;38(9):e14600].

The authors are invited to carefully verify the existence of previous similar systematic reviews, and anticipate them in the introduction to better convey the added value of their study.

• The authors used an adapted NOS scale to assess the quality of studies and claimed it is endorsed by the Cochrane collaboration (albeit the reference is dated 2011). Which changes have been performed? Moreover, and most importantly, Cochrane recommends the ROBINS-I tool for observational studies. Please comment on this issue.

• Observational studies are subjects to confounders (even unmeasurable) and additional bias such as channeling and immortal-time biases [Diabetes Obes Metab. 2018 Dec;20(12):2711-2723].

Although the overall quality rated by the authors was high, the studies differ in terms of follow-up, comparators and covariates used in the propensity score matching.

Therefore, I recommend a table to be added in the text (not only in the supplementary mmaterial) to provide an overview on key features. In fact, although (statistical) heterogenity was moderate, it is important to verify whether the studies are sufficiently comparable to allow meta-analytical approach.

• Sensitivity analyses depending on the various comparators are not shown, in particular results on DPP-4 inhibitors and GLP1-RAs, stratified by CVD. These data would be very useful and deserve discussion.

The authors have successfully addressed comments, including updated search and subgroup/secondary analyses (e.g., accounting for immortal time bias), with consistent results.

The large heterogeneity was expected and discussed as a limitation, in line with a recent review published in DOM (Diabetes Obes Metab. 2018 Dec;20(12):2711-2723)