Response to Comment on Flory et al. Reports of Lactic Acidosis Attributed to Metformin, 2015–2018. Diabetes Care 2020;43:244–246

We appreciate Dr. Brand’s commentary (1) on our brief report (2) and substantively agree with her points about the role of “duplication and latency” in distorting reporting rates for metformin-associated lactic acidosis (MALA). Her closing recommendation that analyses of adverse event reporting systems should be replicated across multiple databases is reasonable in many circumstances, but we argue that it would not be helpful in all cases.

With respect to duplication and latency, one of the major findings of our report was that crude analysis of MALA reporting rates in the U.S. Food and Drug Administration’s Adverse Event Reporting System (FAERS) is indeed distorted by the same major events that Dr. Brand emphasizes. We also noted the EudraVigilance Data Analysis System (EVDAS) reporting requirement change in 2017, as well as the surge of duplicate case reports associated with the publication by Angioi et al. (3,4). Dr. Brand’s additional point that exclusion of literature-derived cases may stabilize reporting rates is well taken.

Interestingly, if all reports with an associated literature citation are excluded from our analysis, the number of MALA cases in the U.S. is 36 in 2015 and rises steadily to 113 in 2018. While this sensitivity analysis reduces the overall number of reports, it shows a greater proportional increase in MALA cases (since after this restriction, annual MALA cases triple over the study period instead of doubling).

We also appreciate the point that rates of MALA reporting may vary widely across countries. At the suggestion of peer reviewers, we carried out parallel analyses using publicly reported European and Canadian safety data, although these exploratory findings were not included in our final report. Those analyses did not show any increase in MALA reporting. However, since our report was focused on assessing the consequences of a regulatory event that occurred only in the U.S., it was not clear whether this difference reflected differences in reporting between different countries or a genuine difference in MALA trends in different countries.

Because we were focused on a single country, Dr. Brand’s closing recommendation that adverse event studies be duplicated across multiple databases would have been difficult for us to follow. In other settings, her advice to analyze several pharmacovigilance data sets is very reasonable, but not a panacea. Issues of both duplication and latency affect most if not all spontaneous adverse event reporting systems, as does the more fundamental concern over lack of a denominator in these databases. A finding that is replicated across multiple such systems might replicate only because the same bias recurred in each. The best solution to such concerns is to follow up safety signals using approaches less susceptible to bias, for example, cohort studies in which denominator data are available.