study design and data sources

This retrospective, observational, pharmacovigilance cohort study is a population-based disproportionality analysis using real-world data from the United States (US) FDA Adverse Event Reporting System (FAERS). The FAERS database contains safety reports, medication errors, and product complaints for FDA-approved drugs and biologics, spontaneously submitted by healthcare professionals, consumers, and pharmaceutical companies globally. As part of the FDA’s postmarketing safety surveillance program, its goal is to enable the timely detection of pharmacovigilance signals and provide a basis for the safe clinical use of drugs. Due to the large amount of data and diversity of information, the database is widely used for adverse event (AE) detection to identify potential drug-related AEs in real-world clinical settings. The FDA receives over 1.5 million AE reports annually, making FAERS one of the largest AE reporting databases. Robust data mining algorithms have also been developed to detect signals, where a positive signal indicates a statistical association between an AE and a drug. Approval from an institutional review board was not necessary under the Declaration of Helsinki, as the study utilized publicly available data from the US FAERS, involving anonymized and deidentified patient information.

To address potential issues with duplicate, irregular, and incomplete reports in FAERS, we used OpenVigil 2.1, a validated, web-based mining software (OpenVigil) that leverages the openFDA online interface to retrieve pharmacovigilance data. This platform retrieves only complete case reports and applies rigorous cleaning processes to remove duplicates, correct formatting errors, and consolidate drug terms. OpenVigil further refines data using demographic and drug information, following the US Adopted Name scheme to accurately map drug names. As a result, only complete, valid reports with uniquely identified case IDs and drugs that were free of overlaps in age, sex, or reported AEs were retained. As a result, only complete, valid reports with uniquely identified case IDs and drugs free of overlaps in age, sex, or reported AEs are retained. While this process may reduce the frequency of AEs compared to raw FAERS data, it significantly enhances data quality by excluding incomplete and duplicate reports. Each report is encoded using Medical Dictionary for Regulatory Activities (MedDRA) preferred terms and analyzed using structured query language to standardize AE signal detection. Additionally, clinical reviewers at the Center for Drug Evaluation and Research and the Center for Biologics Evaluation and Research ensure the accuracy and consistency of the data.

main outcome measures

We systematically searched the database and extracted data from October 2003 to March 2024 on cases with the highest incidence of IFIS associated with FDA-approved drugs, regardless of their recorded role (ie, primary suspect, secondary suspect, concomitant, or interacting). Drugs with fewer than 10 reported cases of IFIS, as well as eye drops commonly used preoperatively or intraoperatively during cataract surgery that directly affect pupil dynamics, such as size or reactivity (eg, phenylephrine, cyclopentolate, and tropicamide), were excluded from the analysis.

statistical analyses

To evaluate disproportionality between cases and noncases, we used the frequency method based on the fourfold tables alongside internationally recognized signal mining techniques (Table S2). These included the proportional reporting ratio (PRR), reporting odds ratio (ROR), and relative risk reduction, each calculated with 95% confidence intervals (CIs) (Tables S2 and S3). Together, these methods enhanced the sensitivity and specificity of AE signal detection, with higher ROR, PRR, and relative risk reduction values indicating a stronger statistical relationship between the drug of interest and AE. ^, Using the proportional imbalance method, we compared the IFIS occurrence ratio for each drug to the background reporting rates of all other drugs in the database.

Only adverse drug reaction signals that met the criteria of Evans 2001 (ie, n > 2, χ ² >4, PRR > 2) were considered true AE signals. To correct for multiple comparisons, we applied the Bonferroni correction, adjusting the significance level by dividing α (0.05) by the total number of comparisons ( n ), resulting in a corrected significance threshold of 0.05/44 = 0.0011 for 44 comparisons. In addition to using the Evans 2001 criteria for initial signal detection, we further validated the findings by calculating the IC ₀₂₅ value, which represents the lower bound of the 95% CI for the information component (IC). Statistical significance was defined as IC ₀₂₅ > 0, indicating that the AEs were reported more frequently than expected for the target drug.

Overall, we applied a multistep approach to identify meaningful drug-AE signals, starting with Evans’ 2001 criteria to filter out unlikely signals, followed by P value analysis to assess statistical significance. To enhance signal stability and reliability, IC ₀₂₅ values were calculated to reduce background noise and adjust for reporting variability, ensuring true signals are distinguished from random fluctuations, especially for rare or underreported events. Statistical analyses were performed using R version 4.3.1 (R Core Team) and Microsoft Excel 2024.