key: cord-0293066-5v8mlnzr
authors: Holder, M. W.; Heeney, C.; Malden, S.; Perera, U.; Sheikh, A.
title: Deleterious drugs in COVID-19: a rapid systematic review and meta-analysis
date: 2021-09-21
journal: nan
DOI: 10.1101/2021.09.17.21262724
sha: 79f316708e5c948d22fc71a353d09943513844eb
doc_id: 293066
cord_uid: 5v8mlnzr

Background: Concerns have been expressed about a number of drugs that potentially worsen outcomes in patients with COVID-19. We sought to identify all potentially deleterious drug groups in COVID-19 and critically assess the underpinning strength of evidence pertaining to the harmful effects of these drugs. Methods and findings: We performed a rapid systematic review, searching Medline, Embase and two COVID-19 portfolios (WHO COVID-19 database and NIH iSearch COVID-19 portfolio) for papers and preprints related to primary studies investigating drugs identified as potentially deleterious. Primary outcomes were direct measures of susceptibility to infection, disease severity and mortality. Study quality was assessed using the National Heart, Lung, and Blood Institute quality assessment tools. Random-effects meta-analyses were used for data synthesis with further subgroup analyses where possible for specific outcome, study design, statistical adjustment and drug groups when two were combined. Sensitivity analyses were performed by removing any studies at high risk of bias and by publication status. 49 observational studies (15 peer-reviewed papers and 34 preprints) reported primary outcomes for eight drug groups hypothesised to be deleterious. Meta-analysis showed that acute inpatient corticosteroid use was associated with increased mortality (OR 2.22, 95% CI 1.26-3.90), however this result appeared to have been biased by confounding via indication. One subgroup analysis indicated an association between immunosuppressant use and susceptibility to COVID-19 among case control and cross-sectional studies (OR 1.29, 95% CI 1.19-1.40) but this was not found with cohort studies (OR 1.11, 95% CI 0.86-1.43). Studies which adjusted for multiple confounders showed that people taking angiotensin-converting-enzyme inhibitors (ACEIs) or angiotensin-II-receptor blockers (ARBs) required a lower level of care (OR 0.85, 95% CI 0.74-0.98). Furthermore, studies which combined these two drug groups in their analysis demonstrated an association with a lower mortality (OR 0.68, 95% CI 0.55-0.85). Conclusions: We found minimal high quality or consistent evidence that any drug groups increase susceptibility, severity or mortality in COVID-19. Converse to initial hypotheses, we found some evidence that regular use of ACEIs and ARBs prior to infection may be effective in reducing the level of care required, such as requiring intensive care, in patients with COVID-19.

been reported in the scientific literature and media, leading to considerable 1 confusion for both patients and healthcare professionals. 2

Especially during the early stages of the pandemic, most of the advice stemmed 3 from outcomes in other respiratory infections including severe acute respiratory 4 syndrome (SARS) (20, 21) and Middle East respiratory syndrome (MERS) (14) or 5 hypotheses based on the mechanism of viral entry into cells (19, 22, 23) and general 6 effects on immunity (17, 18) . In addition, many scientific papers have been 7 published prior to peer review to speed up information sharing, which has led to 8 questions surrounding the scientific validity of some of these findings (24). 9

We aimed to identify, critically appraise and synthesise the evidence on drugs 10 which may be deleterious in Methods

This study is reported as per PRISMA guidelines on the reporting of systematic 13 reviews (25). In response to the exponential rate at which new research on this topic 14 is being published, a rapid review methodology was adopted following the 15 Cochrane rapid review protocol (26) with the aim of producing reliable results in a 16 timely fashion. 17

Search strategy 18 As the availability of studies increased, our search strategy evolved iteratively to 19 capture as much as possible of the available evidence base. This study consisted of a 20 three-stage search strategy involving literature from four databases using the 21 differentiate severe or critical disease from mild or moderate, or diagnostic criteria 1 for ARDS. Level of care was defined as hospitalisation (if compared with 2 community care of COVID-19 cases), ventilation (mechanical or non-invasive 3 ventilation), ICU care or mortality in combination with other outcomes including 4 remaining as an inpatient. Mortality was defined as any measure of mortality 5 outside of or within hospital, either within a set time frame or at any point during 6 data collection. 7

Although hospitalisation rates would normally imply more severe disease than 8 simply testing positive, in many countries testing was largely only available in 9 hospitals (32-34). If a study compared those admitted to hospital due to COVID-19 10 with untested individuals, this was considered analogous with testing positive for 11 the purpose of analysis. However, if the control group contained positive cases 12 tested in the community, then this was considered as requiring a higher level of 13 care. 14 Due to the common pathway affected by both ACEIs and ARBs , and the 15 recommendation against co-prescription (35), these drug groups were combined in 16 analyses. Henceforth, when combined, ACEIs and ARBs will be referred to as renin-17 angiotensin-aldosterone blockers (RAASBs). 18

For the purposes of this review, we considered immunosuppressants as a class of 19 drugs used primarily to suppress the human immune system. Most included 20 studies investigated immunosuppressants in this manner, although some studies 21 analysed individual immunosuppressants. Where corticosteroids were grouped 22 together with other immunosuppressants, these studies were included here. 23 Narrative synthesis 24 For groups of outcomes for a particular drug group that had fewer than four effect 25 estimates with calculable and relevant odds ratios (ORs), these were synthesised 26 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 narratively. Study description tables were used to summarise study characteristics 1 and compare findings. 2

3 Any drug groups with a sufficient number of comparable studies were synthesised 4 using meta-analyses. Considering the implications of random-effects models on 5 statistical power (36), outcomes were deemed eligible for meta-analysis if more than 6 four studies were identified that investigated the same drug group. When sufficient 7 numbers of results were available, these were then run with sub-group analyses to 8 attempt to address any heterogeneity in the results and investigate which 9 confounding factors may have affected findings. The subgroup analyses included 10 study design, level of adjustment for confounders, acute doses or regular users, the 11 two subsets of disease severity defined above and whether ACEIs and ARBs were 12 analysed together or separately. Additionally, preprint studies that had not yet been 13 formally peer-reviewed were also removed during sensitivity analyses. 14 We included studies that either reported an OR, relative risk (RR) or the raw data 15 that allowed the calculation of OR in the meta-analyses. If available, we used an 16 adjusted OR, otherwise we used an unadjusted OR, either reported in the paper or 17 manually calculated. If the calculated OR exhibited a different result from the 18 conclusion reached by the paper, this was not included in the results. 19

If more than one outcome was reported, these were all included if they were in 20 different categories (i.e. severe disease and mortality). If multiple outcome subsets 21 were reported within the same categories, the outcome with the highest number of 22 patients in the exposed group, and therefore greatest weight, was used. The only 23 deviation from this was within the severity of disease category, where the 24 diagnostic index would take precedence to reduce any potential selection bias for 25 increased level of care. 26 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 We used MetaXL (Version 5.3; EpiGear International Pty Ltd) to run random-effects 1 meta-analyses and create forest plots and funnel plots. Results were reported with a 2 95% confidence interval (CI). We used Cochrane's Q and I 2 test to assess for 3

heterogeneity. An I 2 <30% was considered low heterogeneity, ≥30% but <50% was 4 considered moderate and ≥50% was considered high heterogeneity (37). Sensitivity 5 analyses were performed on all models by removing each study to assess their risk 6 of bias on the pooled effect estimates and heterogeneity. Publication bias was 7 assessed for any meta-analyses that contained 10 or more studies by visually 8 inspecting funnel plots for asymmetry. 9 10 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021.

2 From the first two searches, we identified eight drug groups that were hypothesized 3 to be deleterious in COVID-19 from 178 peer-reviewed publications comprising 4 ACEIs, ARBs, corticosteroids, Immunosuppressants, mineralocorticoid receptor 5 antagonists (MCRAs), NSAIDs, Statins and Thiazolidinediones (TZDs). 6

From the combined total of papers from all three searches, we identified 63 papers 7 with primary data related to these drugs, with 49 measuring at least one primary 8 outcome included in the study (Fig 1) . 9

. CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 ( Table 1 ). The majority of papers were cohort studies (n=29) and the remainder were 1 either cross-sectional studies (n=9), case-control studies (n=8), case series (n=2) or a 2 hybrid design (n=1). There were no experimental or quasi-experimental studies 3 included in this review. Thirteen studies investigated more than one relevant drug, 4 with one study (38) presenting data for all eight drug groups. 5 Exposure status and outcome measures: 8 Most studies analysed outcomes for patients taking drugs regularly prior to 9 contracting COVID-19. Studies that analysed ACEI, ARB, MCRA, statin and TZD 10 usage measured outcomes only from those taking the drugs regularly, although one 11 study compared outcomes of continued use of ACEIs and ARBs after admission 12 with withdrawal of treatment (39). One study analysed acute treatment with the 13 immunosuppressant tocilizumab (40), and one analysed acute treatment with the 14 NSAID celecoxib (41). All the remaining studies, which looked at the use of drugs 15 for treating COVID-19 focussed on corticosteroids. There was a mix of studies 16 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 assessing acute treatment doses of corticosteroids and those taking long-term 1 corticosteroids on severity of disease. All studies analysing susceptibility to 2 infection with corticosteroids assessed those taking long-term steroids and all 3 studies analysing mortality assessed those receiving acute doses. 4

Quality assessment:

The majority of the studies were assessed as being of fair quality (45%), followed by 6 good quality (35%); the remaining studies were assessed to be of high risk of bias 7 and rated as poor quality (20%). Generally, studies were downgraded due to limited 8 justifications of sample sizes used and insufficient controlling for potential 9 confounders. 10 Sensitivity analysis by publication status:

As a considerable portion of the included studies were published as preprints that 12 had not formally been peer-reviewed, we undertook sensitivity analysis on any of 13 these studies eligible for meta-analysis to determine their influence on the overall 14 effect estimates when included in random-effects models. None of the included 15 preprints were found to significantly influence any of these results, described 16 below, and they were therefore included in the analyses. Results of the sensitivity 17 analyses can be viewed in Appendix 7. 18

For four of the drug groups (MCRAs, NSAIDs, statins and TZDs), meta-analysis 20 revealed no evidence of protection or harm for any of the groups of outcomes. None 21 of these drug groups had any studies analysing mortality, and only NSAIDs had 22 enough studies analysing severity of disease to perform a meta-analysis (Appendix 23 6 - Fig 14) . No pooled effect estimates showed any statistically significant evidence 24 of harm in patients with COVID-19 ( is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Results of studies looking into the effect of drug groups on the susceptibility to 4 COVID-19 infection are summarised in Appendix 3 - Table 3 . 5 RAASB 6 Ten studies were used in the meta-analysis of susceptibility to 38, [42] [43] [44] [45] [46] [47] [48] [49] . Seven stratified their results into those taking ACEIs and 8 those taking ARBs, and three combined the two drug groups (Fig 2) . 9

The total pooled effect estimate from all the studies showed no evidence that 10 RAASBs affect the risk of contracting COVID-19 (OR 0.94; 95% CI 0.82-1.07). Visual 11 inspection of the funnel plot showed no evidence of publication bias (Appendix 4 -12 Fig 12) . When split into subgroups for drug group, study design and statistical 13 adjustment, this result was unchanged (Appendix 6 -Figs 18-20). 14 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Corticosteroids 5 Four studies investigated the effects of corticosteroids on susceptibility to 44, 48, 49) . One study, rated as good quality, measured only those taking 7 inhaled corticosteroids (38), therefore meta-analysis was not conducted. This study 8

was also the only one to report an increased risk of contracting COVID-19 amongst 9 patients taking corticosteroids, with the remaining three studies reporting null 10 effects (Appendix 3 - Table 3 ) . 11 Immunosuppressants 12 Five studies (32, 38, 44, 46, 48) analysing the effect of immunosuppressants on 13 susceptibility to COVID-19 were included in the meta-analysis (Fig 3) . 14 The initial analysis showed no statistically significant evidence of harm or benefit 15 (OR 1.11; is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 found the same result for case-control and cross-sectional studies (OR 1.29; 95% CI 1 1.19 -1.40) (Fig 3) . 2

The only study looking at susceptibility that was not included in the meta-analysis 3 (50) looked at individual immunosuppressive drugs compared against each other, 4 finding janus kinase inhibitor usage was more prevalent among rheumatology 5 patients with COVID-19, whereas no evidence of a protective or harmful effect of 6 methotrexate was observed. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Effects of drug groups on severity 1 Results of studies looking into the effect of drug groups on the severity of COVID-2 19 infection are summarised in Appendix 3 - Table 4 . 3 RAASB 4 Among the total pool of studies, there was no evidence of benefit or harm with 5 RAASBs (OR 0.93; 95% CI 0.77 -1.12) (Fig 4) . No individual studies affected either 6 the pooled effect estimate or the heterogeneity. Visual inspection of the funnel plot 7 showed no evidence of publication bias (Appendix 4 - Fig 13) . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ;  (adjustment for at least age, sex and one or more co-morbidities) was performed. 1 This showed statistically significant evidence of a protective effect from ACEIs and 2 ARBs for adjusted results (OR 0.86; 95% CI 0.77 -0.96) (Fig 5) . Sensitivity analysis 3 showed that the removal of the result from Bean, DM et al (51) removed the 4 statistically significant pooled effect estimate (OR 0.90; 95% CI 0.80 -1.02). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Creating diagnostic severity index or higher level of care subgroups showed 1 statistically non-significant and heterogeneous results (OR 0.84; 95% CI 0.56 -1.27; I 2 2 45% and OR 0.94; 95% CI 0.77 -1.15; I 2 62%, respectively). When these two groups 3 were analysed separately and subgroup analysis was run for level of adjustment, 4 this showed a statistically non-significant result for adjusted studies using the 5 severity index (OR 0.77; 95% CI 0.39 -1.49). However, further statistically significant 6 evidence of benefit for RAASBs among adjusted studies in regard to the highest 7 level of care needed was observed (OR 0.85; 95% CI 0.74 -0.98) (Fig 6) is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101/2021.09.17.21262724 doi: medRxiv preprint 1 Eight studies with verifiable data assessed the risk of corticosteroids increasing the 2 risk of developing severe disease (49, (52) (53) (54) (55) (56) (57) (58) . The majority were cohort studies, 3 which varied by study quality and other methodological factors (Appendix 3 -4 Table 4) . 5

Four cohort studies measured outcomes from acute treatment doses of 6 corticosteroids, reaching different conclusions; however, one study did not have a 7 calculable OR and so a meta-analysis was not possible. 8

The four studies which used long-term steroids as their exposure were analysed 9 together in a meta-analysis. One study ran different analyses for oral and inhaled 10 steroids. As these can be co-prescribed, this analysis was run twice with each model 11 including either the inhaled or oral corticosteroid measurement. Neither model 12 showed statistically significant evidence of harm or benefit for severity of COVID-19 13 for those taking long-term steroids (inhaled corticosteroids = OR 1.40; 95% CI 0.81 -14 2.39; I 2 36%; oral corticosteroids = OR 1.36; 95% CI 0.76 -2.45; I 2 44% (Fig 7) ). 

Immunosuppressants 20 Meta-analysis was conducted on four studies (46, 54, 56, 59) is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021.

poorly on aspects of controlling for confounding during quality assessment 1 (Appendix 3 - Table 4 ). 2

The pooled effect estimate showed no statistically significant evidence of benefit or 3 harm (OR 0.68; 0.27 -1.71) (Fig 8) . 

The studies not included in this meta-analysis reached opposing conclusions, but in 8 very different scenarios. The only study to look at acute usage of tocilizumab (40) 9 was a case control study, finding evidence of benefit in terms of requiring 10 ventilation (OR 0.42: 95% CI 0.2 -0.89) or ICU care (OR 0.17; 95% CI 0.06-0.48). An 11 additional case control study (60) , found that those on immunosuppressants had 12 more severe COVID-19 when compared with their family members who tested 13 positive for SARS-CoV-2, but no OR was available nor calculable. 14

Results of studies looking into the effect of drug groups on mortality in infection are summarised in Appendix 3 - Table 5 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. no evidence of harm or benefit against hypertensive controls (OR 0.91; 95% CI 0.62 -1 1.32), the general population (OR 1.11; 95% CI 0.67 -1.85) or for the pooled effect 2 overall (OR 0.97; 95% CI 0.74 -1.29) (Fig 9) . Álvarez, JE et al were removed. These two studies were also the only two in this 12 analysis that were rated as poor during the quality assessment. 13 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. 

Two studies were not included in the meta-analysis, each reporting conflicting 5 results with regards to mortality (69, 70) (Appendix 3 - Table 5 ). 6 Corticosteroids 7 Seven (62, 67, 71-75) of the 10 studies assessing mortality in those given 8 corticosteroids were included in the meta-analysis. All of these analysed acute doses 9 used to treat COVID-19 admissions in hospital (Appendix 3 - Table 5 ). Two studies 10 (55, 76) reported no evidence of harm or benefit but it was not possible to extract 11 summary data for pooled analysis. The remaining study (54) compared hazard 12 ratios (HRs) for high and low dose corticosteroids, finding a low-dose had no 13 evidence of either harm or benefit (HR 1.26; 95% CI 0.61 -2.58), but a high-dose 14 showed evidence of increased mortality (HR 3.5; 95% CI 1.79 -6.86). 15 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021.

Meta-analysis of the remaining studies showed evidence of harm (OR 2.22; 95% 1.26 1 -3.90) (Fig 11) . The authors of three of the studies highlighted confounding by 2 indication as a likely factor considerably affecting their results, and when these 3 studies were removed, the pooled effect estimate showed there was no clear 4 evidence of an effect (OR 1.48; 95% CI 0.73 -3.00; I 2 67%). Two studies analysed mortality rates in those taking immunosuppressants. (40, 59). 10

Neither found statistically significant evidence of increased or decreased mortality 11 (OR 0.88; 95% CI 0.17 -4.46 and OR 0.90; 95% CI 0.24 -3.42, respectively) 12 (Appendix 3 - Table 5 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint 

2 This rapid systematic review and meta-analysis attempted to identify and assess 3 any potentially deleterious drug groups in covid-19 susceptibility and prognosis. In 4 terms of susceptibility to infection, we found no pooled effect estimates and one 5 subgroup analysis with statistically significant evidence of increased susceptibility 6 (Fig 3) , which would be in keeping with many other infections and a known risk of 7 immunosuppressant use (77). Hence, we found no evidence to suggest that those 8 without COVID-19 should stop taking their medication to reduce their risk of 9 contracting the disease. Stopping immunosuppressants may also result in a flare-up 10 of a person's underlying condition, which can result in increased risk of infection 11 (78). Furthermore, the withdrawal of any necessary medications could cause harm 12 from the disease the drug was being used to treat, as well as from a potentially 13 severe form of COVID-19 for people with hypertension and diabetes who are 14 already at increased risk (79, 80). 15

We found no evidence that long-term use of any of these drugs increases the 16 severity of disease, nor the mortality rates. One subgroup analysis of RAASBs 17 showed a statistically significant relationship between RAASB usage and requiring 18 a higher level of hospital care (Fig 6) . However, this was found in an unadjusted 19 subgroup where the adjusted subgroup showed evidence of protection, highlighting 20 the effect of known confounders, such as age and cardiovascular disease (81), on the 21 results from these observational studies. This evidence of protection against more 22

severe disease was also shown by the pooled effect of all adjusted studies analysing 23 severity (Fig 5) . 24

This rapid systematic review and meta-analysis has synthesised the best available 26 evidence from the first few months of the pandemic for each of these drugs when 27 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101/2021.09.17.21262724 doi: medRxiv preprint taken by those with or at risk of COVID-19. This study has, within the limitations 1 imposed by time and the unparalleled research publication rate, aimed to compile a 2 comprehensive list of drugs that were hypothesized to be deleterious in peer-3 reviewed papers at the time of the searches. 4

This review however has several limitations. The rapidity with which research is 5 being produced and published affected our study in a number of ways. Firstly, we 6 acknowledge that while a rapid review is justifiable on the grounds of swifter 7 completion, it is less robust than a systematic review. Secondly, the studies included 8 in this review were also completed within a short space of time in order to attempt 9

to learn from clinical practice as quickly as possible. As a result of this, many papers 10 that would be relevant to this review, including papers with stronger evidence and 11 more robust methodology than some of those included may now be available. 12

Finally, the research landscape is changing so quickly that the studies included in 13 this review were all published prior to conducting the last search. This will bias our 14 results towards the studies that were produced more quickly and in countries that 15 experienced larger COVID-19 case numbers earlier on in this outbreak. These 16 countries may have large differences in their demographics as well as clinical 17 protocols for both pre-existing co-morbidities and COVID-19, which could impact 18 the generalisability of our findings. 19

Another factor heightening the risk of bias, errors and methodological inadequacies, 20 is the lack of peer-reviewed articles and the use of preprints in this review. We 21 conducted a sensitivity analysis in January 2021 to analyse whether the removal of 22 papers that had not made it through the peer-review process at this point affected 23 the results, but there were no significant changes to the pooled outcomes (Appendix 24 7). Nevertheless, we must emphasise that because we have included preprints, the 25 results from this review are only to highlight potential outcomes to evaluate with 26 further studies and should not be used to alter clinical practice. 27

. CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10. 1101 /2021 As all of the evidence in this review is from observational studies, this introduces a 1 risk of bias from confounding factors, despite many studies attempting to 2 statistically adjust for these. Any confounders not adjusted for or as yet unknown 3 will have affected our results. One example of this is the discovery that people with 4 a Black, Asian or minority ethnic background are at an increased risk of severe 5 COVID-19 (82), something that was not apparent at the beginning of the outbreak 6 and was therefore only adjusted for in five of the studies included in the review. 7

Other biases evident in the studies included confounding by indication, most 8 evident among the studies looking at acute treatment of COVID-19 with 9 corticosteroids, and misclassification bias in studies where it was possible the 10 control group had been buying the drug of interest over-the-counter or had stopped 11 taking it immediately prior to admission to hospital. This happened with NSAIDS 12 for example. 13

Using aspects of the GRADE assessment, the quality of evidence presented in this 14 review could be classified as low, primarily due to inconsistency, indirectness, 15 imprecision and risk of bias across each outcome (31). 16

One of the reasons RAASBs, MCRAs, NSAIDs, Statins and TZDs were hypothesized 18 to be harmful is their potentially upregulating effect on ACE2, this being the entry 19 point into cells for SARS-CoV-2. Although there are studies which have shown that 20 all of these drugs may upregulate ACE2, particularly RAASBs, the evidence is 21 largely from animal studies, and some of it is contradictory (83). We found no 22 evidence to suggest any of these drugs worsen outcomes or increase the 23 susceptibility to contracting COVID-19 and some evidence to suggest RAASBs 24 could be protective. Whilst it is unclear if this is related to their effect on ACE2, it 25 appears that the protective nature of ACE2 may in fact provide a target for treating 26 . Studies have also shown similar protective effects of ACEIs (85), 27

ARBs and statins (86) in non-COVID pneumonias. 28

. CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10. 1101 /2021 Another prominent reason for drugs to be hypothesized to worsen outcomes was 1 immunosuppression. Advice emerged early on advising against using steroids in 2 COVID-19 (14) , but this was at odds with the experiences of front-line clinical 3 workers in China who were using it in a large number of cases (87). Guidance 4 surrounding long-term immunosuppressants in those with COVID-19 was also 5 issued (88), however at the same time it appeared that dampening the immune 6 response may have a role in treating acute infections (89). With the discovery that 7 interleukin 6 (IL-6) was elevated in patients with severe COVID-19 (90), IL- 6 8 antagonists such as tocilizumab started to be used therapeutically (91). We only 9

found one study which analysed the outcomes of those treated with tocilizumab 10 and this found a reduction of ICU admissions and need for ventilation when 11 patients were given tocilizumab (40). This was an observational study where the 12 patients were "highly selected" so, although the results were adjusted for 13 confounders, there remains a risk of bias. 14 The majority of studies included in this review measured outcomes in patients 15 taking drugs regularly. We only found one study (39) comparing patients who had 16 the drug withheld during the acute infection with those that continued to take them. 17

This study found no effect on disease severity or mortality but found statistically 18 significant reduction in viral clearance and length of stay in those in whom RAASBs 19 were withheld during their hospital stay. This area requires more research, as any 20 conclusions we draw from this review can only be applied to those who take these 21 medications prior to the onset of COVID-19. In order to guide clinical practice, 22 studies comparing initiated, continued and withheld medications, ideally 23 randomised controlled trials (RCTs), are paramount (92). 24

With most drug groups included in this review, dosage and timing was not usually 25 measured or analysed. One study of corticosteroids compared the effect of the 26 cumulative dose of steroids and the authors concluded that mortality with high-27 dose steroids was higher than with low-dose steroids (54). Another study which 28 found favourable outcomes for corticosteroids as a treatment for COVID-19 29 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10. 1101 /2021 hypothesized that the timing was also important, opting for low dose steroids early 1 on in the disease progression (93). These hypotheses were similarly at risk of 2 confounding by indication, with those going on to develop ARDS more likely to get 3 higher dose steroids at a later point in time. 4

Implications for policy, practice and research 5 This review contains exclusively observational studies, many of which have not 6 been peer-reviewed, and so we do not recommend any changes to normal 7 prescribing practice. However, we have found no evidence to suggest medications 8 should be stopped solely due to contracting COVID-19. This goes along with 9 guidelines published for many of these drug groups (94-96) and the findings of 10 other rapid reviews (28, 97). 11

For those taking the drugs highlighted in this review, the binary decision between 12 stopping and continuing their medications, coupled with the contrasting advice 13 from experts, the media and governments, helped fuel the false dichotomy that 14 drug groups were either harmful or not. Due to the large number of factors that can 15 affect outcomes in COVID-19, the harm or benefit derived from starting, stopping or 16 continuing drugs will affect individuals differently, with co-morbidities, 17 demographics, other medications and individual response to the virus playing a 18 role in the progression of the disease. However, this review raises the prospect that 19 drug groups that have been hypothesized to be deleterious in COVID-19, may have 20 the potential to be beneficial in certain circumstances. This is evidenced further by a 21 number of clinical trials, either planned, in progress or completed, involving drugs 22 or drug groups identified during this review (98-103). 23 Furthermore, where trials are unfeasible, as in the case where the aim is to examine 24 the effects of long-term drug use on COVID-19, good quality, large-scale 25 observational studies will be necessary to form the best possible evidence. RCTs will 26 understandably be focussed on finding drugs used to treat COVID-19, the results of 27 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ;  which are beginning to be published (98). However, trials looking at treatments 1 with drugs included in this review may help inform their use in other contexts. 2

We have highlighted the gaps in the research, such as the lack of RCTs and 4 mortality data for NSAIDs, immunosuppressants and long-term steroid use. We 5 have proposed that there should be some focus in future research on these gaps as 6 well as the use of ACE2, or drugs which upregulate it, as a potential target for 7 treating This study found a total of eight drug groups hypothesized to be deleterious in 9 COVID-19. The available data from the first few months of the current pandemic 10 suggest that there is little to no evidence these drug groups increase susceptibility, 11 severity or mortality in COVID-19 and we found some evidence that ACEIs and 12

ARBs may be protective in preventing a more severe disease. 13 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101/2021.09.17.21262724 doi: medRxiv preprint sever* or discontinue? or unsafe or avoid)).mp. [mp=ti, ab, ot, nm, hw, fx, kf, 23 ox, px, rx, ui, an, sy, tn, dm, mf, dv, kw, dq] is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101/2021.09.17.21262724 doi: medRxiv preprint prescribed for its licensed use(s) 23 6. Only looking at medications being used for the treatment of 24 where no medication involved has already been identified as posing a 25 potential risk when prescribed for its licensed use(s) 26

. CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. 8. There is no measure against non-COVID-19 or non-drug exposed control 25 . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101/2021.09.17.21262724 doi: medRxiv preprint 46 Appendix 3: study tables Table 3 . Summary of studies looking into the effect of drug groups on the susceptibility to COVID-19 infection. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10.1101/2021.09.17.21262724 doi: medRxiv preprint

No meta-analysis was performed for susceptibility with corticosteroids.

Severity with acute corticosteroids leaves Wang, D only: OR 2.16 (95% CI 0.49 -9.43).

Susceptibility with immunosuppressants had one study not peer reviewed (Rentsch CT et al.) and sensitivity analysis was run for this already within the study. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10. 1101 /2021 No meta-analysis was performed for mortality related to immunosuppressants.

No meta-analyses were performed for mortality related to either NSAIDs, statins or TZDs.

No meta-analyses were performed for severity related to either statins or TZDs. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10. 1101 /2021 No meta-analyses were performed for MCRAs. . CC-BY-NC 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted September 21, 2021. ; https://doi.org/10. 1101 /2021 

Enzyme Inhibitors

Antihypertensive Agents/ or Angiotensin Receptor Antagonists/ or 15 ARB

((drug? or medication? or prescri*) adj10 (harm or safe* or risk or worse* or 22 Study or Subgroup Yan

Q=8.57, p=0

Case Control & Cross Sectional subgroup

ACEI) Cohort Q=7.84, p=0

I2=85% No Adjustment Q=2.21, p=0