Definition of Adverse Events
The Food and Drug Administration (FDA) defines the term ‘adverse event’ as: “any untoward medical occurrence associated with the use of a drug in humans, whether or not considered drug related, including the following: an adverse event occurring in the course of the use of a drug product in professional practice; an adverse event occurring from drug overdose whether accidental or intentional; an adverse event occurring from drug abuse; an adverse event occurring from drug withdrawal; and any failure of expected pharmacological action”19,20.
Multidimensional Database Sources
The data used in this study have been curated from multiple publicly available data sources for hypertensive patients, including the FDA’s Adverse Event Reporting System (FAERS), which houses all ADEs reported to the FDA by pharmaceutical companies, healthcare providers, andconsumers. The data, including the hypertension dataset, is updated quarterly by the FDA and currently includes reports submitted from the first quarter of 2004 to the last quarter of 2019. This dataset focuses on drugs and their ADEs but includes additional data such as disease, drug, and demographic information as well as information related to patient outcome.
The data structure of these ADEs is organized in accordance with the Medical Dictionary for Regulatory Activities (MedDRA) terminology, along with the International Safety Reporting Guidance Database. We utilized the MedDRA hierarchy for regulatory information of medical products in hypertension, which is grouped based on etiology, manifestation site, or purpose. Here we utilized the 23.0 or earlier version of MedDRA, with the most recent update from April 2020 that includes new COVID-19 related terms and revisions.
Data Mining and Search Strategy
In alignment with our previous multidisciplinary work1,21, we implemented a three-stage approach to curate disparate databases and identified patients with hypertension including pulmonary arterial and intracranial hypertension. First, data mining algorithms were used to identify hypertension datasets and associated post-marketing ADEs for ACEI/ARB drugs that were prevalent among the top reported symptoms in COVID-19 patients. Next, as part of data cleaning, standard libraries were utilized to curate missing information or unify distinct groups within the data. For example, drug names in the FAERS database are reported by a combination of active ingredients, generic names, or brand names. Using PostgreSQL (PostgreSQL Global Development Group), allowed us to map and search all the possible drug names to drug parents in the DrugBank database (Alberta Innovates - Health Solutions, The Metabolomics Innovation Centre) creating a unified dataset22. Additionally, ADEs derived from unstructured data (e.g. text) needed data scrubbing, cleansing,and merging16. For this purpose, deep learning techniques were employed to implement and map the informatic structure of the FAERS database into the international safety reporting guidance coded using terms in MedDRA16. Finally, ADEs associated with medications in the ACEI and ARB classes administered to patients with hypertension were recorded.
Proportional Reporting Ratio
Statistical analysis was performed using SAS (SAS® University Edition version 9.4, North Carolina, U.S). First, data based on the frequency of each ADE related to respiratory, thoracic, and mediastinal disorders/infections were parsed in the MedDRA and FAERS databases. Specific ADEs collected were pulmonary edema, pleural effusion, oropharyngeal pain, dyspnea, dysphonia, cough, sinusitis, pneumonia, nasopharyngitis, bronchitis, pneumonia aspiration, emphysema, and pleurisy (Fig. 1). These ADEs were consistent with globally reported information, which found that pneumonia, pneumonitis, shortness of breath, cough, and sore throat were among the top reported symptoms in COVID-19 positive patients10,11,12,13,14,15. We then employed a method proposed and implemented by the FDA for analyzing ADE disproportionality in pharmacovigilance data by observed-expected ratios16. This method, the proportional reporting ratio (PRR), provides a statistical summary for the commonality of an ADE for a specific drug as compared to the entire database for drugs in the same or other classes16.
We then addressed confounding factors including patient demographics and drugs that are under- reported in voluntary reporting systems, including the FAERS, since conditional slicing and sub- setting can confine the use of quantitative signal detection methods such as PRR. For this purpose, we were able to correct the analysis after applying logistic regression for the known covariates of age, weight, and sex, and combine this approach with PRR to improve analyses of drug effects using the hypertension data sets. As a result, we found that the following identity is chiefly correct in numerous scenarios:
Pr(ADE|drug, age, weight, sex) = Pr (𝐴DE|drug)This helped us to estimate a PRR for a specific drug-ADE combination by calculating the following equation: (see Equation 1 in the Supplementary Files)
where 𝑟𝑖𝑗 gives the total number of a specific ADE 𝑖 ∈ {1,2, . . , 𝐸} for a given drug 𝑗 in {1,2, . . . , 𝐷}. Here 𝐸 and 𝐷 represent the number of all events and drugs in the drug class, respectively. 𝑑𝑟𝑢𝑔∗ denotes the drug class, excluding the specific drug 𝑗. Also, 𝑛𝑗 shows the total events for the given drug 𝑗. As the distribution of PRR samples are all positive, we then applied a log transformation to data and found the confidence interval23 using the following equation: (see Equation 2 in the Supplementary FIles)
Friedman Test Results
Using SAS, sample differences among the four groups—Quinapril, Trandolapril, ACEIs, and ARBs—were assessed for a pairwise analysis with the assumption that data were not normally distributed using the non-parametric Friedman test for two independent unequal-sized data. Friedman test was also applied to perform multiple comparison tests (P values for statistical significance < 0.05). For the non-parametric Friedman test of statistical significance, seven pairwise and multiple comparisons were performed based on the ARBs and ACEIs excluding Quinapril and Trandolapril, hence denoted as ACEIs-2. Tests performed included ACEIs-2 vs. ARBs, ACEIs-2 drugs vs. Quinapril alone, ACEIs vs. Trandolapril alone, Quinapril vs. ARBs, Trandolapril vs. ARBs, and Quinapril and Trandolapril vs. all ACEIs-2 and ARBs.
Principle Component Analysis
Principal components of PRR in pulmonary ADE for ACEIs and ARBs were calculated using the built-in function prcomp in R 3.6 (R Core Team, GNU GPL v2)24. Implementing principal component analysis (PCA) to the drugs with 13 pulmonary ADEs reduced the dimension into a smaller number of PCs, significantly explaining and visualizing variation of ACEIs and ARBs. Biplot was generated using the R package factoextra25.
Data availability
All the data supporting the findings in this study are available in the paper and Supplementary Information. Data related to this paper are available from the corresponding authors upon request.