Comment on: Drugs That Interact With Colchicine via Inhibition of Cytochrome P450 3A4 and P-Glycoprotein: A Signal Detection Analysis Using a Database of Spontaneously Reported Adverse Events (FAERS)

We read with great interest the research article by Gómez-Lumbreras et al. ¹ As the authors state, their article may help properly use colchicine since colchicine has a narrow therapeutic range and its toxicity may be increased by concomitant use with drugs that inhibit its metabolic pathway.

However, the signal detection algorithm (SDA) used in this study is questionable. The authors created a subset of colchicine use groups. They used the disproportionality between concomitant and non-concomitant drug users in that subset as a drug-drug interaction (DDI) signal (Table 1b in the original article). ¹

In this study, the subset analysis used detects signals from the interest adverse event (iAE) when the patient group using drug D₁ (Colchicine) takes drug D₂ (cytochrome P450 3A4 or P-glycoprotein-inhibiting drugs). However, suppose the iAE signal score for drug D₂ is large in all patient groups. In that case, the signal is detected whether or not drug D₁ is used, so it may be impossible to determine whether the signal is due to concomitant use.^2,3

Furthermore, in general, the number of reports of concomitant use tends to be small. (In fact, the number of reports of concomitant use shown in Table 3 in the original article ¹ is minimal compared with the number of reports of single drug use.) If the number of non-iAEs reports among them is small (the number of reports of single drug use tends to include many non-iAEs in which no signal is detected), the proportion of iAEs reported will naturally increase and thus the disproportionality ratio (= signal score) to single drug is likely to rise. ⁴ Thus, the reporting odds ratio (ROR) based on frequency-based statistical methods is prone to signal score inflation when the number of reports is small, ⁵ leading to unstable detection results. ⁶

The authors attempted to solve this problem by signal scores using the O/E shrinkage. The authors applied the general O/E ratio calculation using the obsexp_shrink_signal function provided by the R pharmsignal package (https://github.com/tystan/pharmsignal). ¹ When we checked these R codes, they contained only codes for detecting single-drug signals (accessed 9 March 2023). Since Norén et al, ⁷ who proposed the O/E shrinkage, also described the SDA for DDIs in the same paper, we wonder why the authors did not use that SDA (the Ω shrinkage measure, interaction signal score). ⁸ Of course, there is no gold standard for signal detection, including DDI signals, and each SDA has its advantages, disadvantages, and limitations. ⁹ Therefore, the choice of the SDA to be used is left to the researcher’s discretion. Therefore, this letter does not deny the signals detected in the original article. However, when researchers use the SDAs likely to detect false positives, it might be difficult for many readers to accept the detected signal willingly.