While I am admittedly fatigued with seeing these types of analyses for every available drug & COVID, the authors do a nice job framing the study in the Introduction.

Not being familiar with the underlying data, I will focus my comments primarily on the methods.

1. While it's mentioned in the discussion, I'd recommend adding how DPP-4 inhibitors were handled during the hospitalization and after discharge. Also clarify that care & treatment during the hospitalization was not recorded and could not be accounted for in your analyses.

2) Please provide more clarity on how covariates were identified for creating the propensity score. The methods say "... selected based on their relevance in clinical practice and on the literature." but no references from the literature were provided. Are these factors known to be related to DPP4 use or the related outcomes in COVID? Why was HbA1c not included in the main model?

1. Please give more details about the IPTW. Were any type of trimming methods were used for handling individuals with particularly higher PS and weights? Were any type of robust sandwich estimation used to adjust standard errors when using weighting?

2. Why were time to event analyses not used for outcomes such as mortality (vs. traditional
   logistic regression)?

3. Please included weighted values to Table 1 and 2 as well as absolute standardized differences. I know they are graphically shown in Figure 2, but the scale makes them uninterpretable.

4. In the results, please avoid using the term "trend" which has neither a statistical or clinically-relevant meaning and can be misleading.

5. Discussion: I worry about so much emphasis and conclusions being made on the 28-day mortality finding, as it was secondary and only in a sub-group of the population who remained hospitalized at 7 days.

Thanks to the reviewers for adequately addressing all of my comments.

I remain confused if the authors truly used IPTW or the more recently-introduced overlap propensity weighting (the latter truly resulting in a zeroing of absolute standardized differences).

I also disagree with their assertion that robust SE estimation is not required, however don't believe it would materially alter the studies findings and/or conclusions; therefore it isn't a big issue.