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

Flory et al. (1) analyzed cases of lactic acidosis (LA) reported to the U.S. Food and Drug Administration’s Adverse Event Reporting System (FAERS) for 2015–2018 to evaluate whether changes to the U.S. metformin label in 2016 altered the reporting rate of LA.

LA is rare, and large electronic databases are useful in studying its incidence. The Originator Pharmacovigilance (PV) Database (Merck KGaA) includes all individual case study reports on all metformin tablets manufactured by Merck, from patients, health care practitioners, health authorities, and the literature. The EudraVigilance Data Analysis System (EVDAS) and FAERS collect cases from marketing authorization holders, health authorities, and the literature in the European Union and U.S., respectively.

A new EVDAS system requirement (November 2017) requires manufacturers to pull all LA cases in their countries of marketing authorization from the system into their respective company PV database. This increased the number of reports from, for example, France in the Merck PV database, as few identify the manufacturer of the product in EVDAS. The new requirement subsequently created a spillover to FAERS for all companies with marketing authorizations and respective case reporting obligation in the European Union and U.S.

The reporting rates for metformin-associated LA (MALA) have fluctuated quite significantly between 2015 and 2019 (Table 1). Cases from the literature are prone to duplication and latency; e.g., 28 cases in a publication in 2018 (2) resulted in 54 cases in EVDAS in 2018, causing a spike in reporting (Table 1B). Reporting rates are more stable if literature-derived cases are excluded (Table 1D).

In the U.S., the percentage of all reported cases is low, but it is much higher in Italy and, especially, France, where MALA is a topic of health authority interest and where physicians are encouraged to report cases. Other biases are also at play. For example, an ongoing PhD thesis in northern France increased reporting of MALA in 2018–2019, again causing a reporting spike for that period (Table 1B–D).

Social media is used increasingly by health care practitioners and patients to identify adverse events (AE) (3), with further potential for bias and nocebo effects. For example, a high frequency of reports of muscle AE with statins in observational data, news media, and social media has not been substantiated in randomized controlled trials (4,5).

Thus, reporting of AE to PV databases is subject to multiple sources of distortion and bias, and caution is needed when using these reports to interpret changes in AE incidence. We recommend that two or more PV databases should be used in such studies.