key: cord-0839355-gqn5z7dr authors: MacFadden, Derek R; Brown, Kevin; Buchan, Sarah A; Chung, Hannah; Kozak, Rob; Kwong, Jeffrey C; Manuel, Doug; Mubareka, Samira; Daneman, Nick title: Screening Large Population Health Databases for Potential COVID-19 Therapeutics: A Pharmacopeia-Wide Association Study (PWAS) of Commonly Prescribed Medications date: 2022-03-29 journal: Open Forum Infect Dis DOI: 10.1093/ofid/ofac156 sha: c8641425777f0a7f8ad1e962b68e4a3f19399154 doc_id: 839355 cord_uid: gqn5z7dr BACKGROUND: For both the current and future pandemics, there is a need for high-throughput drug screening methods to identify existing drugs with potential preventative and/or therapeutic activity. Epidemiologic studies could complement lab-focused efforts to identify possible therapeutic agents. METHODS: We performed a pharmacopeia-wide association study (PWAS) to identify commonly prescribed medications and medication classes that are associated with the detection of SARS-CoV-2 in older individuals (>65 years) in long-term care homes (LTCH) and the community, between January 15 (th), 2020 and December 31 (st), 2020, across the province of Ontario, Canada. RESULTS: 26,121 cases and 2,369,020 controls from LTCH and the community were included in this analysis. Many of the drugs and drug classes evaluated did not yield significant associations with SARS-CoV-2 detection. However, some drugs and drug classes appeared significantly associated with reduced SARS-CoV-2 detection, including cardioprotective drug classes such as statins (weighted OR 0.91, standard p-value <0.01, adjusted p-value <0.01) and beta-blockers (weighted OR 0.87, standard p-value <0.01, adjusted p-value 0.01), along with individual agents ranging from levetiracetam (weighted OR 0.70, standard p-value <0.01, adjusted p-value <0.01) to fluoxetine (weighted OR 0.86, standard p-value 0.013, adjusted p-value 0.198) to digoxin (weighted OR 0.89, standard p-value <0.01, adjusted p-value 0.02). CONCLUSIONS: Using this epidemiologic approach which can be applied to current and future pandemics we have identified a variety of target drugs and drug classes that could offer therapeutic benefit in COVID-19 and may warrant further validation. Some of these agents (e.g. fluoxetine) have already been identified for their therapeutic potential. M a n u s c r i p t 4 BACKGROUND SARS-CoV-2 (the agent of COVID-19) has caused substantial morbidity and mortality since its recognition in China in December 2019 [1] . Hundreds of millions of COVID-19 cases and millions of attributable deaths have been documented worldwide [2] . Mortality has been particularly high in elderly patients and those with comorbid conditions or residing in long-term care homes (LTCH) [3] . For those patients who develop infection and recover, they remain at risk for longer term sequelae that can impact on their physical and mental health as well as their economic productivity [4] . Significant resources have been invested into developing both effective vaccines for the prevention of disease, and drugs for the treatment of those infected [5] [6] [7] . While vaccines have been shown effective at preventing most severe disease [8] , there are still large populations of individuals including young children, the vaccine-hesitant or ineligible, and those from areas with reduced access to vaccines, that are at particular risk for infection and/or complications of infection [8, 9] . Moreover, even individuals who have been vaccinated can develop severe disease from wild-type strains or emerging variants of concern (VoCs), although to a lesser extent [10] [11] [12] . To date, many of the effective treatments for COVID-19 have been discovered through the 'reuse' of existing drugs that serve other indications [6] . These include drugs with antiviral activity (e.g. remdesivir, molnupiravir), those with local or systemic immunomodulatory properties (e.g. dexamethasone, tocilizumab, baricitinib, budesonide, and sarilumab) [13, 14] , and those with still unclear mechanisms (e.g. fluvoxamine). Many of these therapies have shown impressive impacts at reducing morbidity and mortality in COVID-19, and the success of these agents support the value of identifying active therapeutic agents from existing drugs. Moreover, the regulatory pipeline for A c c e p t e d M a n u s c r i p t 5 trialing and then adopting an existing drug for an off-label indication is simpler than developing new agents for treatment [15] . Despite the significant progress that has been made to date with identifying effective agents, there are still major gaps for improving morbidity and mortality in both outpatient and hospitalized patients; mortality can be as high as 29% in patients receiving mechanical ventilation despite receipt of the most effective therapy to date (dexamethasone) [13] . A major hurdle in the reuse paradigm is efficiently screening candidate agents for prospective study. In vitro studies are used as a basis for justifying the clinical trials of many agents, but this approach is limited by assays that can assess the direct antiviral effect of agents on target cells or proteins [16] , and do not necessarily capture more complicated actions of drugs including immunomodulation [13, 14] . Additionally, developing in vitro or animal models of emerging pathogens in a timely manner may prove challenging. Rather than using in vitro drug screening, we can use large databases of patients that have exposures to existing drugs, and evaluate their relative risk of COVID-19 as a function of these drug exposures. Using techniques from genomic epidemiology, we can perform a corollary of a genomewide association study (GWAS) but using pharmaceuticals as the exposure. This has been termed a pharmacopeia-wide association study (PWAS) [17] . PWAS and related studies have typically been used for identifying drug harm, such as the evaluation of drug exposure and myocardial infarction risk, but the same approaches could also be used to discover therapeutic agents from existing drugs [17, 18] . PWAS and related studies [18] are not limited to evaluating drugs with direct antiviral activity and can capture immunomodulatory effects along with offering the benefit of assessing populationlevel impacts of particular exposures. These studies are not meant to achieve perfect adjustment/correction of all confounding or selection bias, rather, they are screening tools to A c c e p t e d M a n u s c r i p t 6 identify targets with potential promise that warrant further evaluation. Evaluating agents that reduce an individual's risk of infection can provide a window into potential prophylactic and therapeutic agents. Though these approaches represent a fundamentally different approach to drug screening compared to traditional lab-based methods, PWAS could be a promising and complimentary way of identifying candidate drugs for rapid clinical trials in epidemic and pandemic contexts, including the ongoing COVID-19 pandemic. In this study, we applied a PWAS approach to evaluate potential drug targets for COVID-19 infection that would benefit from further study. We performed a nested case-control study evaluating the associations between all drug exposures and the detection of SARS-CoV-2 in individuals across the province of Ontario, Canada. We used a PWAS with the aim of identifying potential drug targets for further evaluation and validation, but not to achieve perfect adjustment/correction of all confounding or selection bias. This is commensurate with drug screening approaches that seek to identify targets with promise, but not to definitively confirm effectiveness. We obtained study data from linked population-wide administrative datasets housed at ICES (formerly the Institute for Clinical Evaluative Sciences). These datasets were linked using unique encoded identifiers and analyzed at ICES. Population/Cohort. We considered all Ontario residents aged 66-110 years between January 15th, 2020 and December 31st 2020. We excluded residents who had: (1) invalid birth or death dates; (2) non-Ontario postal codes; (3) or were not eligible for Ontario universal health insurance program coverage. We classified individuals into (A) community-dwelling residents and (B) residents of longterm care homes (LTCH). The latter group were identified based on recent assessments recorded in A c c e p t e d M a n u s c r i p t 7 the Continuing Care Reporting System Long-Term Care (CCRS-LTC) database, or physician billings, or prescription drug claims recorded in the Ontario Health Insurance Plan database or the Ontario Drug Benefits (ODB) database, respectively. The ODB database provides near complete information on all prescription drugs for residents 65 years of age or older in Ontario. Cases. We defined a case as laboratory-confirmed molecular detection of SARS-CoV-2 from nasopharyngeal and/or other respiratory specimens recorded in the Ontario Laboratories Information System (OLIS). We selected the index date as the specimen collection date of the earliest testing episode positive for SARS-CoV-2. Controls. We defined a control as any other resident during the study period, this includes individuals that were and were not tested for SARS-CoV-2. We randomly assigned index dates from January 15th 2020 -December 31st 2020. Exposures. We wished to identify chronic drug exposures, and for each oral prescription drug captured in the ODB database, we determined whether an individual was chronically exposed to the drug by looking for a prescription with sufficient days supply that overlapped with the index date and also 30 days prior to the index date. For the analyses, we only considered drugs with an exposure prevalence of =>0.1% in each group of residents. We classified drug exposures by the anatomical therapeutic chemical (ATC) classification levels by linking the Drug Identification Number (DIN) in the ODB database with the Drug Product Database, which is managed by Health Canada. The ATC 4 level typically refers to a group of structurally and/or functionally related chemicals whereas the ATC 5 level refers to individual drugs [19] . We chose to evaluate both drug classes to look for class effects, and also individual drugs (nested within drug classes) for drug specific effects. A c c e p t e d M a n u s c r i p t 8 Covariates. We identified a number of covariates that could act as potential confounders of the association of medication use and SARS-CoV-2 detection. These include: demographic (age, sex), geographic (census tract or census subdivision for rural areas for community-dwelling residents), facility (for LTC residents), comorbidity (Charlson Comorbidity Index, asthma, cancer, chronic kidney disease including dialysis, chronic obstructive lung disease, coronary artery disease, dementia, depression, diabetes, congestive heart failure, hypertension, history of ischemic stroke, Data Analysis. For each drug we performed conditional logistic regression analyses evaluating the odds ratio of drug exposure amongst cases and controls; we applied separate models for the community and LTCH groups of interest. To account for geographic variability in case rates and drug exposures, we conditioned upon either (i) census tracts for community dwelling individuals, or (ii) facility for LTCH residents. We subsequently combined our effect estimates for community and LTCH using an inverse variance weighted meta-analytic approach (described below). With the exception of the Charlson Comorbidity Index, which was not available for all individuals, all other covariates were included in the model for each of the two populations of interest. We reported adjusted odds ratios, A c c e p t e d M a n u s c r i p t 9 standard and multiple-testing adjusted p-values (see below), and in some instances 95% confidence intervals. Meta-analyzed effect estimates. In order to provide a single effect estimate across the two studied populations, we combined the adjusted odds ratios of drug exposure (when a drug had an exposure prevalence of >=0.1% in both LTCH and community populations) for the community and LTCH resident populations using an inverse variance weighting approach to generate a weighted odds ratio (wOR). Variances were pooled to compute the variance, 95% confidence intervals, and standard p-values of the weighted estimates. Multiple testing. We used the conservative Benjamini-Yekutieli procedure to adjust for multiple testing in a fashion assuming an arbitrary p-value dependence [26] . Recently there has been a concerted effort to present confidence intervals instead of p-values in scientific reports, however we have retained our p-values here as we feel they are an important component of a GWAS-inspired analysis. Visualization. Rainforest plots [27] were constructed to present drug-specific weighted odds ratios of exposure and 95% confidence intervals as well as associated standard and adjusted p-values. Scatterplots were used to compare adjusted odds ratios of exposure between community and LTC residents, by drug (ATC 5) and drug class (ATC 4). Research Ethics and Patient consent. ICES is an independent, non-profit research institute whose legal status under Ontario's health information privacy law allows it to collect and analyze health Table 1 . In LTCH residents, the median age was 86 years in both cases and controls, and the majority of cases and controls were female (67% and 70% respectively). Cases tended to have a higher mean number of hospitalizations, prescriptions, and some specific comorbidities (Table 1 ). In the community, the median age was 74 years for cases and controls, and the majority of cases and controls were female (51% and 54% respectively). There was a trend towards a higher mean number of hospitalizations, physician visits, and prescriptions in cases versus controls ( Table 1) . The weighted odds ratios (wOR) of individual drug exposure (ATC 5) are shown in Figure 1 We performed a sensitivity analysis of the community population, stratifying by age <80 years, and age 80 years or over. These adjusted odds ratios of drug exposure (ATC 5) are shown in Supplemental Figure 7 . These generally show the strongest association measure as consistent across these age strata. We did not perform this analysis for the LTCH group because the median age is 86 years. A c c e p t e d M a n u s c r i p t 12 In this large, nested case-control study across the entire pharmacopeia of commonly prescribed medications we demonstrate PWAS can be used to identify drugs and drug classes that are associated with laboratory-confirmed detection of SARS-CoV-2. We found that the vast majority of commonly used drugs are not associated with either increased or reduced detection of SARS-CoV-2. However, as is the potential power of large drug screens [28] , we have identified some existing agents/classes that warrant further investigation as potential COVID-19 therapeutics. While a number of individual agents showed associations with reduced detection of SARS-CoV-2, we will highlight four agents that demonstrated particularly pronounced reduced SARS-CoV-2 detection and are also commonly used: ezetimibe, fluoxetine, levetiracetam, and diazepam. Ezetimibe inhibits cholesterol absorption from the small intestine, and this alteration of the cholesterol synthesis pathway (as occurs also with statins and fibrates) may be a mechanism for beneficial effect in COVID-19 [32] . A large case-control study from Israel identified that drugs related to the cholesterol synthesis pathway, including ezetimibe, ubiquinone, and rosuvastatin were associated with reduced risk of hospitalization with COVID-19 and support our findings [18] . The association between fluoxetine and reduced COVID-19 diagnosis may have a foundation in a unique immunomodulatory effect found with SSRIs, due to σ-1 receptor (S1R) agonism [33] , which may act to reduce proinflammatory cytokine production. The SSRI fluvoxamine has shown promise for treating outpatients with COVID-19 infection [34] , and thus fluoxetine which shares similar mechanistic properties including potent S1R agonism could reduce symptoms of COVID-19 infection and thus yield reduced diagnosis. Levetiracetam (an anti-epileptic agent) and diazepam (an anxiolytic) both have no known antiviral activity against SARS-CoV-2 or clear immunomodulatory effects and may benefit from A c c e p t e d M a n u s c r i p t 13 further evaluation (along with the many other agents demonstrating reduced association with SARS- We also identified a number of drug classes that showed strong associations with reduced SARS-CoV-2 identification, and the majority of these appeared to be cardioprotective agents, including lipid modifying agents such as statins, inhibitors/blockers of the renin-angiotensin system (e.g. ACEinhibitors and angiotensin receptor blockers), and beta-blockers. There are multiple reasons these agents may have shown reduced associations with SARS-CoV-2 detection. Firstly, they may play a role in protecting individuals from acquiring COVID-19 [29] . In particular, ACE-inhibitors and angiotensin receptor blockers have been the focus of much speculation in the literature given that SARS-CoV-2 binds to the ACE2 receptor, and there are a number of ongoing and completed prospective clinical trials to evaluate the clinical impact of these drugs [30] . Secondly, they may reduce the severity of illness, and thus reduce the likelihood of case identification (diagnosis). Other cardioactive classes that showed a reduced association with SARS-CoV-2 identification were the digitalis glycosides, specifically digoxin. This may be due to reduced case finding through a stabilizing effect on cardiac function and thus reduced symptomatology. Interestingly, this could also be due to a direct antiviral effect of digoxin, which has been noted in prior in vitro studies [31] . We also identified classes of drugs associated with increased SARS-CoV-2 detection. Some of these classes include immunomodulatory drugs (e.g. sulfasalazine and leflunomide) that may alter the predisposition of patients for developing severe disease. Other classes associated with increased SARS-CoV-2 detection include antibiotics (e.g. macrolides and sulfa agents), proton pump inhibitors, and iron. The association with antibiotics may be reflective of residual confounding among populations with chronic lung disease or other chronic diseases that require frequent or continuous A c c e p t e d M a n u s c r i p t 14 antibiotic use, and may be at risk of more symptomatic/severe disease. It is unlikely that these findings represent reverse causation, as our exposure definition requires the use of the agent at 1 month prior to the index diagnosis. The harmful association with iron may be due to worse outcomes associated with iron deficiency (with iron use a proxy for iron deficiency) [35, 36] , and the harmful association with proton pump inhibitors could relate to suspected detrimental effects of hypochlorhydria [37, 38] . When comparing all drugs or drug classes, we did not find an association between adjusted odds ratios of exposure for LTCH and community dwelling individuals. This is expected, because only a minority of estimates should be truly concordant between the populations (representing signal) and the rest should be random (representing noise). As with any epidemiologic study, associations that have been outlined here could be due to residual confounding or selection bias. However, we did adjust for a large array of potential confounders in our analysis. Moreover, we used an approach which offered the least risk of collider stratification bias [39, 40] . Nevertheless, we expect residual confounding or selection bias to be driving many of the protective and harmful associations in this PWAS analysis, and the goal of PWAS is as a high throughput screening test to highlight potential targets for further epidemiologic, in vitro or in vivo validation. Another potential limitation of our study was the selection of the outcome, namely SARS-CoV-2 detection, which is not a direct measure of symptomatic COVID-19 infection nor the impacts on outcomes in patients who are diagnosed with COVID-19. However, we chose this outcome to identify drug exposures that might act to prevent acquisition or detection of acquisition, for which there would be an underlying important drug effect that could be repurposed for either prevention or treatment of COVID-19. Lastly, our study occurred largely in pre-VoC time periods, and findings here may be less generalizable to a viral landscape made up of predominantly VoCs. A c c e p t e d M a n u s c r i p t 15 Additional steps can be taken to evaluate these candidate drugs, depending upon the agent and other supporting evidence, and may include in vitro confirmation of anti-viral activity (if suspected), additional observational studies to confirm effects in separate settings/regions, or possibly prospective trial evaluations. In summary, we present an approach for using large epidemiologic cohorts, in a manner akin to genome-wide association studies, to screen for possible drug candidates for the prevention and treatment of COVID-19 that would benefit from further evaluation. These approaches can be used now to search for possible active agents for the ongoing COVID-19 pandemic, as well as in future as we experience the emergence of new pathogens of global concern with epidemic/pandemic spread. M a n u s c r i p t 22 Covid-19 -Navigating the Uncharted WHO Coronavirus (COVID-19) Dashboard. Available at Risk Factors Associated With Mortality Among Residents With Coronavirus Disease 2019 (COVID-19) in Long-term Care Facilities in Ontario, Canada 6-month consequences of COVID-19 in patients discharged from hospital: a cohort study Efficacy and Safety of the mRNA-1273 SARS-CoV-2 Vaccine Repurposed Antiviral Drugs for Covid-19 -Interim WHO Solidarity Trial Results Safety and Efficacy of the BNT162b2 mRNA Covid-19 Vaccine Impact and effectiveness of mRNA BNT162b2 vaccine against SARS-CoV-2 infections and COVID-19 cases, hospitalisations, and deaths following a nationwide vaccination campaign in Israel: an observational study using national surveillance data Covid-19 and Health Equity -Time to Think Big Assessment of protection against reinfection with SARS-CoV-2 among 4 million PCR-tested individuals in Denmark in 2020: a population-level observational study. The Lancet Genomic evidence for reinfection with SARS-CoV-2: a case study Covid-19: Delta variant is now UK's most dominant strain and spreading through schools Dexamethasone in Hospitalized Patients with Covid-19 Tocilizumab in Patients Hospitalized with Covid-19 Pneumonia A white-knuckle ride of open COVID drug discovery High-throughput screening identifies established drugs as SARS-CoV-2 PLpro inhibitors Systematic assessment of prescribed medications and short-term risk of myocardial infarction -a pharmacopeia-wide association study from Norway and Sweden Identification of drugs associated with reduced severity of COVID-19: A case-control study in a large population A multicenter study of the coding accuracy of hospital discharge administrative data for patients admitted to cardiac care units in Ontario Accuracy of Canadian health administrative databases in identifying patients with rheumatoid arthritis: a validation study using the medical records of rheumatologists Identifying cases of congestive heart failure from administrative data: a validation study using primary care patient records. Chronic Diseases and Injuries in Canada Diabetes in Ontario: determination of prevalence and incidence using a validated administrative data algorithm Accuracy of administrative databases in identifying patients with hypertension The increasing burden and complexity of multimorbidity The control of the false discovery rate in multiple testing under dependency Rainforest plots for the presentation of patient-subgroup analysis in clinical trials Coronavirus Treatment Acceleration Program (CTAP) Risk of severe COVID-19 disease with ACE inhibitors and angiotensin receptor blockers: cohort study including 8.3 million people Effect of Discontinuing vs Continuing Angiotensin-Converting Enzyme Inhibitors and Angiotensin II Receptor Blockers on Days Alive and Out of the Hospital in Patients Admitted With COVID-19: A Randomized Clinical Trial Antiviral activity of digoxin and ouabain against SARS-CoV-2 infection and its implication for COVID-19 Cholesterol-modifying drugs in COVID-19 Sigma receptors: Their role in disease and as therapeutic targets Fluvoxamine vs Placebo and Clinical Deterioration in Outpatients With Symptomatic COVID-19 Persisting alterations of iron homeostasis in COVID-19 are associated with non-resolving lung pathologies and poor patients' performance: a observational cohort study Serum Iron Level as a Potential Predictor of Coronavirus Disease 2019 Severity and Mortality: A Retrospective Study Proton Pump Inhibitors are Risk Factors for Viral Infections: Even for COVID-19? Direct COVID-19 infection of enterocytes: The role of hypochlorhydria Collider bias undermines our understanding of COVID-19 disease risk and severity Individual and social determinants of SARS-CoV-2 testing and positivity in Ontario, Canada: a population-wide study M a n u s c r i p t We thank IQVIA Solutions Canada Inc. for use of their Drug Information Database. The authors have no conflicts of interest to report.