Decision letter: Identification of drugs associated with reduced severity of COVID-19 – a case-control study in a large population

Abstract

Article Figures and data Abstract Introduction Materials and methods Results Discussion Data availability References Decision letter Author response Article and author information Metrics Abstract Background: Until coronavirus disease 2019 (COVID-19) drugs specifically developed to treat COVID-19 become more widely accessible, it is crucial to identify whether existing medications have a protective effect against severe disease. Toward this objective, we conducted a large population study in Clalit Health Services (CHS), the largest healthcare provider in Israel, insuring over 4.7 million members. Methods: Two case-control matched cohorts were assembled to assess which medications, acquired in the last month, decreased the risk of COVID-19 hospitalization. Case patients were adults aged 18 to 95 hospitalized for COVID-19. In the first cohort, five control patients, from the general population, were matched to each case (n=6202); in the second cohort, two non-hospitalized SARS-CoV-2 positive control patients were matched to each case (n=6919). The outcome measures for a medication were: odds ratio (OR) for hospitalization, 95% confidence interval (CI), and the p-value, using Fisher’s exact test. False discovery rate was used to adjust for multiple testing. Results: Medications associated with most significantly reduced odds for COVID-19 hospitalization include: ubiquinone (OR=0.185, 95% CI [0.058 to 0.458], p<0.001), ezetimibe (OR=0.488, 95% CI [0.377 to 0.622], p<0.001), rosuvastatin (OR=0.673, 95% CI [0.596 to 0.758], p<0.001), flecainide (OR=0.301, 95% CI [0.118 to 0.641], p<0.001), and vitamin D (OR=0.869, 95% CI [0.792 to 0.954], p<0.003). Remarkably, acquisition of artificial tears, eye care wipes, and several ophthalmological products were also associated with decreased risk for hospitalization. Conclusions: Ubiquinone, ezetimibe, and rosuvastatin, all related to the cholesterol synthesis pathway were associated with reduced hospitalization risk. These findings point to a promising protective effect which should be further investigated in controlled, prospective studies. Funding: This research was supported in part by the Intramural Research Program of the National Institutes of Health, NCI. Introduction SARS-CoV-2 is a new single-stranded RNA virus, which was first identified in December 2019, and has rapidly spread into a global pandemic of primarily respiratory illness designated as coronavirus disease 2019 (COVID-19). This disease is associated with significant mortality, particularly among elderly or overweight individuals, raising considerable concerns for public health. Until a vaccine or specifically designed therapies are available, it is urgent to identify whether existing medications have protective effects against COVID-19 complications using available real-world data. With this aim, we performed a case-control study on electronic health records (EHRs) from Clalit Health Services (CHS), the largest healthcare provider in Israel. Materials and methods Participants and data sources Request a detailed protocol We collected data from the CHS data warehouse on adult patients aged 18 to 95 years, who tested positive for SARS-CoV-2 from the beginning of the pandemic through November 30, 2020, and were admitted for hospitalization through December 31, 2020. Each patient was assigned an index date, which is the first date at which a positive RT-PCR test for SARS-CoV-2 was collected for the patient. Patients’ demographic characteristics were extracted, along with existing comorbidities, clinical characteristics including body mass index (BMI), and estimated glomerular filtration rate (eGFR) at the baseline, defined as of February 2020. In addition, the list of drugs or products acquired by each patient in CHS pharmacies was collected for the month preceding the index date, defined as the 35 days prior to this date. Reliable identification of medications procured for a given month is enabled by the fact that in CHS, distinct prescriptions are issued for each calendar month. When medications are provided in advance for multiple months, the date at which the prescription for each month of treatment begins is recorded. This study has been approved by the CHS Institutional Review Board (IRB) with a waiver of informed consent, approval number: COM-0046–20. Patient data that could identify participants were removed prior to the statistical analyses in accordance with the protocol approved by the CHS IRB. Software Patients’ data were extracted and processed from CHS data warehouse using programs developed in-house in Python and SQL. Case-control design and matching Request a detailed protocol Hospitalized COVID-19 patients were assigned to two distinct case-control cohorts, which differ in the way control individuals were selected. In cohort 1, control patients were chosen among the general population of CHS members. Since controls can be selected from among millions of individuals, five controls were selected to match each case (5:1), with comprehensively matched baseline attributes, including age, sex, BMI category, socio-economic and smoking status, chronic kidney disease (CKD) stage for patients with renal impairment, and main comorbidities diagnoses (hypertension, diabetes, CKD, congestive heart failure [CHF], chronic obstructive pulmonary disease [COPD], malignancy, ischemic heart disease). For the matching procedure, patients with undocumented BMI were considered as having a normal BMI, unless an obesity diagnosis was present. Each control was assigned the same index date as the matched case, provided that the patient was still alive and a member of CHS at this date. EHR data were collected for controls using the same procedure described for cases. Cohort 1 is designed to identify drugs that affect the overall risk for hospitalization for COVID-19, where the effect could combine a decreased risk of detectable infection, and a decreased risk for hospitalization once infected. In cohort 2, control patients were chosen among patients who had a positive test for SARS-CoV-2 but had not been hospitalized as of December 31, 2020. Given the smaller size of the pool from which controls can be drawn, only two controls were matched for each case patient. Attributes that were matched were the age, sex, smoking status, Adjusted Clinical Groups (ACG) measure of comorbidity (Shadmi et al., 2011) and presence/absence of an obesity diagnosis. The index date taken was the date of the first positive SARS-CoV-2 PCR test both for cases and for controls. Cohort 2 is more specifically suited to identify drugs that are associated with a decreased risk for COVID-19 hospitalization in patients who had a proven infection with the virus. In both cohorts, there were a minority of case individuals for which enough matching controls could not be found; these cases were not included in their respective cohorts. Patients who were pregnant since February 2020 were also excluded. Outcome measures Request a detailed protocol In each cohort, and for each medication anatomical therapeutic chemical (ATC) class, the odds ratio (OR) for hospitalization was computed, comparing the number of patients who acquired a medication belonging to the class in the 35 days preceding the index date, in the case and the control groups. Statistical analysis Request a detailed protocol OR for hospitalization for drugs acquired in the case versus control groups and statistical significance were assessed by Fisher’s exact test. Correction for multiple testing was performed using the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995), which gives an estimation of the false discovery rate (FDR) in the list. To assess the effects of being in one of two high-risk subgroups, Ultra-Orthodox Jews and Arabs, we used multivariable conditional logistic regression analyses performed in each of the cohorts. In each cohort, we modelized the OR for hospitalization, using subgroup membership and purchased medications as explanatory factors. To assess for possible associations between the protective effect of a medication and BMI, we partition the matched subjects into four BMI ranges: <25, 25 to 30, 30 to 35, >35. Then we redid our association analyses in each range. Statistical analyses were performed in R statistical software version 3.6 (R Foundation for statistical computing). Role of the funding source Request a detailed protocol The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. AI, IF, and AT had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Through December 31, 2020, 10,295 adult patients between the ages of 18 and 95 had a recorded COVID-19 related hospitalization in the CHS database. The matching procedure was able to identify control individuals from the general population in ratio 5:1 for 6530 patients in the first cohort, and control patients in ratio 2:1 for 6953 SARS-CoV-2 positive individuals in the second cohort. The characteristics of the matched populations are shown in Table 1. Table 1 Demographics and clinical characteristics of the two matched cohorts of patients (hospitalized versus non-hospitalized). Cohort 1Cohort 2COVID-19 hospitalized (cases)Not hospitalized (controls)COVID-19 hospitalized (cases)Not hospitalized (controls)n653032,650695313,906Age (mean, SD)64.6 (16.1)64.8 (15.8)65.7 (16.0)65.7 (15.8)Sex, female (%)3259 (49.9)16,295 (49.9)3381 (48.6)6762 (48.6)Hospitalization severity (n, %)Mild condition3008 (46.1)2676 (38.5)Serious condition851 (13.0)1043 (15.0)Severe condition1621 (24.8)1903 (27.4)Deceased1050 (16.1)1331 (19.1)Smoking status (%)Never smoker5012 (76.8)24,218 (74.2)5156 (74.2)10,312 (74.2)Past smoker1115 (17.1)5808 (17.8)1338 (19.2)2676 (19.2)Current smoker403 (6.2)2624 (8.0)459 (6.6)918 (6.6)Nb visits at primary doctor in last year (mean, SD)8.1 (7.9)7.9 (7.3)8.2 (8.4)7.5 (7.1)Comorbidity (%)Arrhythmia887 (13.6)4242 (13.0)1278 (18.4)2221 (16.0)Asthma527 (8.1)2941 (9.0)650 (9.3)1376 (9.9)Congestive heart failure (CHF)228 (3.5)1140 (3.5)784 (11.3)851 (6.1)Chronic obstructive pulmonary disease (COPD)148 (2.3)740 (2.3)603 (8.7)776 (5.6)Diabetes2976 (45.6)14,880 (45.6)3425 (49.3)5549 (39.9)Hypertension3850 (59.0)19,062 (58.4)4396 (63.2)8102 (58.3)Ischemic heart disease (IHD)1464 (22.4)7320 (22.4)1838 (26.4)3113 (22.4)Malignancy1087 (16.6)5435 (16.6)1280 (18.4)2766 (19.9)Chronic kidney disease (CKD)102 (1.6)510 (1.6)1086 (15.6)1117 (8.0)Obesity (documented diagnosis)3761 (57.6)17,837 (54.6)3975 (57.2)7950 (57.2)Body mass index (BMI) (mean, SD)28.7 (5.7)28.6 (6.5)29.1 (6.3)28.5 (5.7)BMI group (%)<18.517 (0.3)85 (0.3)51 (0.7)93 (0.7)18.5 to 251070 (16.4)5350 (16.4)1244 (17.9)2471 (17.8)25 to 302295 (35.1)11,475 (35.1)2264 (32.6)4870 (35.0)30 to 352053 (31.4)10,265 (31.4)2005 (28.8)4267 (30.7)35 to 40761 (11.7)3805 (11.7)886 (12.7)1562 (11.2)>40334 (5.1)1670 (5.1)503 (7.2)643 (4.6)Glomerular filtration rate (GFR) (mean, SD)85.7 (21.6)85.8 (20.3)78.7 (28.2)83.4 (22.4)Chronic kidney disease (CKD) staging (n, %)G13047 (46.7)15,145 (46.4)2837 (40.8)6090 (43.8)G22679 (41.0)13,747 (42.1)2535 (36.5)5722 (41.1)G3a558 (8.5)2817 (8.6)689 (9.9)1257 (9.0)G3b203 (3.1)836 (2.6)391 (5.6)571 (4.1)G441 (0.6)89 (0.3)186 (2.7)160 (1.2)G563 (0.9)28 (0.2)Dialysis2 (0.0)16 (0.0)252 (3.6)78 (0.6) In each of the two cohorts, we counted the number of patients from each group who acquired drugs and other medical products from each ATC class and computed the OR and p-values using Fisher’s exact test. The distribution of OR for drugs for which the p-value was statistically significant (p<0.05) is shown in Figure 1. The OR for most drugs are neutral or associated with an increased risk of COVID-19 hospitalization. Only a small number of items are associated with decreased risk: 1.15% in cohort 1 and 1.75% in cohort 2. Figure 1 Download asset Open asset Histogram showing the distribution of the odd ratios (OR) of medication use with the outcome in cohorts 1 and 2. The overwhelming majority of medications are associated with neutral effect (gray) or increased risk for hospitalization (black, OR>1), only a few are associated with significantly decreased risk (black, OR<1). Table 2 presents the list of drugs and products that were found to be negatively associated with COVID-19 hospitalization in a statistically significant manner in cohort 1 (A) and in cohort 2 (B). We display items for which the p-value is below 0.05, and for which the FDR is less than 0.20, meaning that at least 80% of the items in the displayed list are expected to be true positives. Items are sorted in decreasing order of significance. Table 2 Most significant associations for medications acquired in the 35 days preceding the index date in two matched cohorts. ATC code and classUse in caseUse in contr.Case %Contr. %Odds ratio (95% conf. int.)p-ValueFDR(A) Cohort 1 (N = 6530 hospitalization cases, N=32,650 controls taken from the general population)C10AA07 Rosuvastatin32823805.027.290.673 (0.596 to 0.758)<0.0001<0.001C10AX09 Ezetimibe737401.122.270.488 (0.377 to 0.622)<0.0001<0.001A16AX30 Ubiquinone (CoQ-10)61650.090.510.181 (0.065 to 0.403)<0.0001<0.001C01BC04 Flecainide71160.110.360.301 (0.118 to 0.641)0.000390.005J07AL02 Pneumococcus vaccine conjugate212200.320.670.476 (0.288 to 0.746)0.000490.006C09BA05 Ramipril-hydrochlorothiazide1278591.952.630.734 (0.603 to 0.887)0.000990.011A10BD07 Sitagliptin-metformin24315013.724.600.802 (0.696 to 0.922)0.001590.017C10AA03 Pravastatin523850.801.180.673 (0.493 to 0.902)0.006590.060N06AB10 Escitalopram21613023.313.990.824 (0.708 to 0.955)0.009300.078M01AC01 Piroxicam242050.370.630.584 (0.365 to 0.894)0.009800.082C09CA06 Candesartan654511.001.380.718 (0.544 to 0.934)0.012370.100M05BA07 Risedronic acid563960.861.210.705 (0.522 to 0.935)0.013190.103G04CB02 Dutasteride302400.460.740.623 (0.411 to 0.914)0.013670.105A11CC05 Cholecalciferol660363410.1111.130.898 (0.821 to 0.980)0.016000.119C09AA08 Cilazapril171530.260.470.554 (0.315 to 0.918)0.017430.124G04BE08 Tadalafil292290.440.700.632 (0.413 to 0.933)0.018620.132S01ED61 Timolol-travoprost8900.120.280.444 (0.186 to 0.913)0.020680.142A10BH01 Sitagliptin332500.510.770.658 (0.443 to 0.950)0.024610.162J07BB02 Influenza vaccine inac39222056.006.750.882 (0.787 to 0.986)0.025480.165N06DX02 Ginkgo folium2420.030.130.238 (0.028 to 0.915)0.025520.165A12CC04 Magnesium citrate312370.480.730.652 (0.433 to 0.952)0.025970.166A10BK01 Dapagliflozin352550.540.780.685 (0.466 to 0.978)0.032830.193(B) Cohort 2 (N = 6953 hospitalization cases, N=13,906 controls taken from patients SARS-CoV-2 positive)C10AA07 Rosuvastatin3549505.096.830.732 (0.643 to 0.831)<0.00010.000C10AX09 Ezetimibe923031.322.180.602 (0.471 to 0.764)0.000010.000J07AL02 Pneumococcus vaccine conjugate20950.290.680.419 (0.245 to 0.685)0.000210.003M05BA07 Risedronic acid471650.681.190.567 (0.400 to 0.789)0.000420.005A16AX30 Ubiquinone (CoQ-10)9560.130.400.321 (0.139 to 0.653)0.000520.006N06AB10 Escitalopram2366103.394.390.766 (0.654 to 0.894)0.000610.007C09BA05 Ramipril-hydrochlorothiazide1213421.742.460.702 (0.565 to 0.869)0.000820.009C01BC04 Flecainide7430.100.310.325 (0.123 to 0.729)0.002530.023S01XA40 Hydroxypropyl-methylcellulose (tears)672030.961.460.657 (0.490 to 0.871)0.002730.025A11CC05 Cholecalciferol737166910.6012.000.869 (0.792 to 0.954)0.002800.025B01AE07 Dabigatran etexilate371240.530.890.595 (0.400 to 0.866)0.005430.042C09AA08 Cilazapril15640.220.460.468 (0.247 to 0.831)0.005790.044N02CC04 Rizatriptan1170.010.120.118 (0.003 to 0.750)0.010650.075A12CC04 Magnesium citrate331080.480.780.609 (0.399 to 0.908)0.011910.080S01KA01 Hyaluronic acid (artificial tears)5310.070.220.322 (0.098 to 0.836)0.012490.083C09DB01 Valsartan-amlodipine2275493.273.950.821 (0.698 to 0.963)0.014450.094A10BD07 Sitagliptin-metformin2335603.354.030.826 (0.704 to 0.967)0.017210.108B03BA51 Vit.B12 combinations311000.450.720.618 (0.399 to 0.934)0.019790.119G03CA03 Estradiol18670.260.480.536 (0.300 to 0.914)0.020470.122C09DA01 Losartan-hydrochlorothiazide1243151.782.270.783 (0.630 to 0.969)0.024240.140S01ED01 Timolol20700.290.500.570 (0.328 to 0.949)0.024920.143G04BD12 Mirabegron22740.320.530.593 (0.351 to 0.967)0.029980.163S01XA02 Retinol (eye ointment)3210.040.150.285 (0.054 to 0.956)0.030150.163Z01CE01 Eye care wipes3210.040.150.285 (0.054 to 0.956)0.030150.163N06AX12 Bupropion6300.090.220.399 (0.136 to 0.976)0.033850.177N06BA04 Methylphenidate8360.120.260.444 (0.178 to 0.972)0.036560.186A12AX05 Calcium-zinc CD0100.000.070.000 (0.000 to 0.892)0.036960.186A11JC02 Multivitamins for ocular use25810.360.580.616 (0.376 to 0.976)0.038270.191 Numbers are of patients from the group who have acquired a medication from the class in the last month before the index date. p-Values are calculated according to Fisher's exact test. Medications are sorted by increasing order of p-values. OR: odds ratio; [95% CI]: 95% confidence interval; FDR: false discovery rate calculated according to Benjamini-Hochberg (BH) procedure. Shown in this table are anatomical therapeutic chemical (ATC) classes for which the p-value is less than 0.05, and for which the FDR is less than 0.20 (about 80% of entries are expected to be true positive). The top ranked medications by significance in cohort 1 were rosuvastatin (OR=0.673, 95% confidence interval [CI] 0.596 to 0.758), ezetimibe (OR=0.488, CI 0.377 to 0.622), and ubiquinone (OR=0.181, CI 0.065 to 0.403); these same three medications were also in the top five by significance of cohort 2: rosuvastatin (OR=0.732, CI 0.643 to 0.83), ezetimibe (OR=0.602; CI 0.471 to 0.764), and ubiquinone (OR=0.181, CI 0.065 to 0.403). It is remarkable that these three drugs act on the cholesterol and ubiquinone synthesis pathways, which both stem from the mevalonate pathway (Buhaescu and Izzedine, 2007); the intermediate product at the branch point is farnesyl polyphosphate (FPP) (Figure 2). Rosuvastatin and other statins specifically inhibit he enzyme HMG-CoA reductase. Ubiquinone is a food supplement available over the counter, which is often recommended to patients prone to muscular pain and receiving a statin treatment (Qu et al., 2018). Risedronate, which also acts on this pathway, and is commonly used to prevent osteoporosis, by blocking the enzyme FPP synthase, is also identified by both cohorts, and is ranked 4th by significance in cohort 2 (OR=0.567; CI 0.400 to 0.789), and 13th in cohort 1 (OR=0.705; CI 0.522 to 0.935). Figure 2 Download asset Open asset Ubiquinone and cholesterol biosynthesis pathway. Ubiquinone and cholesterol biosynthesis pathways originate from a branching of the mevalonate pathway at FPP. Rosuvastatin and other statins can inhibit the HMG-CoA reductase, while risedronic acid and other bisphosphonates can inhibit the FPP synthase. Ac-CoA: acetyl coenzyme A, HMG-CoA: hydroxymethylglutaryl coenzyme A, GPP: geranyl pyrophosphate, FPP: farnesyl pyrophosphate, PPP: polyprenyl pyrophosphate. Other medications that fulfilled the stringent criteria of being identified by both cohorts with an FDR of 80% include the pneumococcal conjugate vaccine (OR=0.476, CI 0.288 to 0.746 in cohort 1; 0.602, CI 0.245 to 0.685 in cohort 2), magnesium citrate (OR=0.652, CI 0.433 to 0.952 in cohort 1; 0.609, CI 0.399 to 0.908 in cohort 2), vitamin D (OR=0.898, CI 0.821 to 0.980 in cohort 1; 0.869, CI 0.792 to 0.954 in cohort 2), flecainide (OR=0.301, CI 0.118 to 0.641 in cohort 1; 0.325, CI 0.123 to 0.729 in cohort 2), escitalopram (OR=0.824, CI 0.708 to 0.955 in cohort 1; 0.766, CI 0.654 to 0.894 in cohort 2), cilazapril (OR=0.554, CI 0.315 to 0.918 in cohort 1; 0.468, CI 0.247 to 0.831 in cohort 2), ramipril combined with hydrochlorothiazide (OR=0.734, CI 0.603 to 0.887 in cohort 1; 0.702, CI 0.565 to 0.869 in cohort 2), and sitagliptin combined with metformin (OR=0.802, CI 0.696 to 0.922 in cohort 1; 0.826, CI 0.704 to 0.967 in cohort 2). Sitagliptin alone is also significant in cohort 1 (OR=0.658, CI 0.443 to 0.950). In addition, we observe interesting patterns in cohort 2, which is designed to identify drugs associated with decreased hospitalization risk in SARS-CoV-2 positive patients: several vitamin or mineral supplementation items appear to have a protective effect, in addition to vitamin D and magnesium citrate, which were identified by both cohorts: vitamin B12 combinations (OR=0.618, CI 0.399 to 0.934), multivitamins for ocular use (OR=0.616, CI 0.376 to 0.976), and calcium-zinc combinations (OR=0.000, CI 0.000 to 0.892). Several ophthalmic items also appear to be associated with significantly decreased odds for hospitalization, including artificial tears, hydroxypropyl-methylcellulose-based (OR=0.657, CI 0.490 to 0.871), or hyaluronic acid based (OR=0.322, CI 0.098 to 0.836); decreased OR are also found for items that may act as a physical barrier to the eye: eye care wipes, which are sterile wipes sold to clean the eyes (OR=0.285, CI 0.054 to 0.956), a retinol-based ointment used to treat cornea abrasion (OR=0.285, CI 0.054 to 0.956), and timolol drops used to treat glaucoma (OR=0.570, CI 0.328 to 0.949). Also associated with decreased odds for hospitalization are several drugs based on an ACE inhibitor or an angiotensin receptor blocker (ARB), sometimes in combination with another compound. In addition to cilazapril and ramipril-hydrochlorothiazide that were highly ranked in both cohorts, cohort 1 identifies candesartan (OR=0.718, CI 0.544 to 0.934), and cohort 2 identifies valsartan with amlodipine (OR=0.821, CI 0.698 to 0.963), and losartan with hydrochlorothiazide (OR=0.783, CI 0.630 to 0.969). Remarkably, several drugs acting on receptors to neurotransmitters also appear to decrease hospitalization risk: rizatriptan (OR=0.118, CI 0.003 to 0.750), bupropion (OR=0.399, CI 0.136 to 0.976), and methylphenidate (OR=0.444, CI 0.178 to 0.972). In the Israeli population, the two groups that have been reported to be at higher risk are Ultra-Orthodox Jews and Arabs (Muhsen et al., 2021). Therefore, we performed additional analyses with the goal to eliminate membership in either of these groups as a potential confounder and to eliminate possible confounding in concurrently used medications. We performed multivariate conditional logistic regression (Materials and methods) in each of the cohorts. In each cohort, we modelized the OR for hospitalization, using ethnicity and purchased medications as explanatory factors. See Supplementary file 1. Either Ultra-Orthodox or Arab identity indeed appear to be each associated with increased risk for hospitalization. However, even after adjusting for the subgroup membership, most of the medications identified by individual Fisher’s exact tests maintain statistically significant protective effect. Because of the established association between high BMI and COVID-19 severity, it is of interest to know whether any of the protective medications are especially protective in high BMI individuals. Therefore, we performed a subgroup analysis, by partitioning partition BMI into four ranges (see Materials and methods). The results are shown as a forest plot in Supplementary file 1-table 3. In general, the protective effects were seen in most or all BMI ranges and we did not see any striking association between a protective medication and high BMI. Discussion In this large-scale retrospective study, we identified several drugs and products that are significantly associated with reduced odds for COVID-19 hospitalization, both in the general population and in patients with laboratory-proven SARS-CoV-2 infection. Several other research groups have recognized the potential for EHRs to enable large-scale studies in COVID-19 and the challenges of this sort of retrospective research are reviewed in Dagliati et al., 2021; Sudat et al., 2021. To give a few examples, EHRs have also been used to predict: (i) COVID-19 mortality based on pre-existing conditions (Estiri et al., 2021; Osborne et al., 2020), (ii) early diagnosis of COVID-19 based on clinical notes (Wagner et al., 2020), and (iii) eligibility of COVID-19 patients for clinical trials by matching trial criteria with patient records (Kim et al., 2021). Major strengths of our study include: (i) the large sample of hospitalized COVID-19 patients, (ii) the ability to collect comprehensive data about individual demographic and comorbidity characteristics and to build matched case and control populations, (iii) the ability to track hospitalizations and disease severity, owing to a central database established by the Israeli Ministry of Health, and (iv) the capacity to track which drugs and products have been acquired by patients in the period that have preceded SARS-CoV-2 infection, owing to comprehensive digital systems integration in CHS. Another strength is the dual cohort design, with control individuals taken from the general population in the first cohort and from individuals positive for SARS-CoV-2 in the second cohort, with each using different matching criteria, mitigates potential bias that could affect each cohort. The two cohorts allowed us to evaluate the protective effect of drugs that act either by reducing the initial risk of infection or by reducing the risk of hospitalization in those infected. Analyses are based on items procured in the 35 days before the initial positive test. This window was chosen in accordance with the monthly renewal of prescription policy in place in CHS. Limitations of this study are related to it being observational in nature. Best efforts were made to use matching so that patients in case and controls are similar regarding most of the known factors for disease severity, and notably, age, obesity, smoking, and baseline comorbidity. The cases and controls were not matched for ethnicity, which could be a substantial confounding factor. We aimed to get a sensible tradeoff between controlling for confounding factors by rigorous matching and keeping enough patients so that cohorts are representative of the general population. Our analysis is based on medication acquisition in pharmacies and does not ascertain that medications purchased were used. Notably, some of the drugs associated with a protective effect may have been stopped during patient’s hospitalization so that our analysis may have underestimated the full achievable benefits for some of the drugs. Conversely, since drugs tested here were acquired before patients were positive for SARS-CoV-2, the protective effect of some of the drugs may be fully attained only when treatment is started before or early in the infection. The variable behavior of people during the pandemic has been an important factor that can affect the risk of exposure and the severity of infection. We tried to address this cause of variable risk by performing matching in two distinct cohorts and by using only PCR-positive patients in the second cohort. Nevertheless, behavioral factors, which could not measure, can still account for some of the observed differences. Our analyses counted the purchase of each medication, but not the dose or the patient compliance. Therefore, we cannot comment on whether higher doses of the beneficial medications, such as rosuvastatin and ubiquinone, are associated with reduced risk. The medications that are protective are prescribed for a variety of conditions. It is conceivable but unlikely that it is the medical condition, or comorbidity, that provides the protection rather than the medication itself. Three of the comorbidities that have been prominently suggested as relevant to COVID-19 severity and outcome include high BMI, diabetes, and hypertension. Therefore, at the helpful suggestion of the reviewers, we did both subgroup analysis and regression analysis to show that the protective effect of the most protective medications appears not to be associated with BMI (Supplementary file 1). The study design explicitly matched for diabetes and hypertension, so it follows that these two diseases are not associated with the protective effects of the drugs listed in Table 2A and B. However, we recognize the limitation that when the association between the medical condition and the prescription is very specific, such as flecainide for cardiac arrhythmia, we lack suitable data to separate the possible effects of the condition and the medication. Bearing these strengths and potential limitations in mind, our analyses seem to indicate several viral vulnerability points, which can potentially be exploited to effectively reduce disease severity with drugs that are already available. The drugs identified as protective include ubiquinone, which is a food supplement with a very good safety profile that does not even require a prescription in our health system, and rosuvastatin and ezetimibe, two drugs prescribed routinely to reduce cholesterol and that have a very good safety profile. These findings are in line with previous reports that RNA viruses need cholesterol to enter cells, for virion assembly, and to maintain structural stability (Aizaki et al., 2008; Bajimaya et al., 2017; Rossman et al., 2010; Sun and Whittaker, 2003), and that prescribing statins may protect against infection with RNA viruses such as members of family Flaviviridae, including dengue virus, Zika virus, and West Nile virus (Gower and Graham, 2001; Osuna-Ramos et al., 2018; Whitehorn et al., 2016). The involvement of the cholesterol/ubiquinone pathway is further confirmed by the fact that risedronic acid, a drug acting on the enzyme farnesyl pyrophoshate synthase (Tsoumpra et al., 2015; Figure 2) which catalyzes the production of FPP from which the cholesterol and the ubiquinone synthesis pathways