Re: “Association Between Thionamides and Acute Pancreatitis: A Case–Control Study” by Guo et al.

Dear Editor:

We read with interest the article by Guo and colleagues entitled “Association between thionamides and acute pancreatitis: A case–control study” (1). For >70 years the thionamide antithyroid drugs methimazole, its prodrug carbimazole (hereafter MMI for both drugs), and propylthiouracil have been the cornerstone in the treatment of hyperthyroidism (2). Recently, acute pancreatitis (AP) was added to the list of potential adverse drug reactions to MMI (3,4). In contrast to our recent nationwide study (4), which found ongoing MMI use to be associated with a 56% increased risk of AP, Guo et al. found no association between ever usage of thionamides and AP (1). Besides reasons of power (our study comprises three times as many cases; 459 versus 149), the differences in risk estimates between the two studies could also be explained by limitations in the study design used by Guo et al.

The question their study (1) seeks to answer can be summarized as “Do individuals who at any point in time have been exposed to thionamide compounds have an increased risk of AP compared to individuals who have never used thionamide compounds?” This disregards the duration of the exposure, the time since last exposure and whether exposure is ongoing. Thus, this exposure definition assumes that the increased risk of AP after exposure to thionamides is permanent, which is a very unlikely scenario. All reports of MMI-related AP concern ongoing MMI users, usually with onset of AP within 4 weeks of initiation of therapy. Our recent study supported these observations by showing that ongoing MMI exposure is associated with an increased risk of AP, while no evidence of a cumulative dose–response effect was found (4). Substituting the relevant ongoing MMI exposure with the less relevant ever-exposure, as in the study by Guo et al., will introduce a bias toward the null, which may explain why they could not demonstrate an association.

Moreover, cases and controls were matched based on the propensity score (PS) (1). Guo et al. do not report how PSs were estimated, rendering it impossible to assess the validity of their approach. A PS is an estimated probability of being a drug user, given the subject's characteristics. Thus, matching cases and controls on PSs results in similar frequencies of exposure between both groups, and odds ratios obtained using unconditional logistic regression will, therefore, be strongly biased toward the null, which may contribute to the null-association found in their study. Given the low number of covariates included in the PS model, a much less problematic approach would have been risk-set sampling followed by multivariate regression.

Overall, due to a number of methodological shortcomings, the study by Guo et al. (1) neither contradicts our recent findings nor the European Medicines Agency's warning of an association between ongoing MMI use and AP as a serious adverse effect.