"Because this study is observational, it cannot be demonstrated that the observed outcomes are adverse events of medications."
More from the study:
This study also has several limitations. As for all analyses of EHR, it is subject to under-coding/coding errors (including potential misdiagnosis) and unknown completeness of data (see also Appendix, pp 3–5).
  • We could not assess patients' medication adherence nor the duration of exposure as this information is not available in their records.
  • Because most individuals receiving semaglutide were matched to individuals taking a comparator drug but not vice versa, the hazard ratios reported in this paper should be seen as estimates of the average treatment effect on the treated, rather than the average treatment effect.
  • The observed associations might be partly mediated by better diabetes control as measured by HbA1c, but this was not tested here.
  • We could not differentiate between formulations (i.e., oral vs subcutaneous)4 and doses of semaglutide, as such information is not well recorded in the data.
Many of these are problems inherent with observational studies. I think it's fine that these studies are done because they can help drive the kinds of double-blind, controlled patient research to test specific hypotheses. We need to do a better job as a society educating people to understand when a headline is from an observational study and that it shouldn't carry as much weight.
Good to highlight indeed. I didn't know this nuance. It's reassuring they emphasize it themselves in the paper too.
reply
Good catch!
reply