Abstract
BACKGROUND A recently published randomized trial (Boulware et al., 2020, NCT04308668) of hydroxychloroquine (HCQ) for post-exposure prophylaxis found a reduction in Covid-19 of 17%. In the context of ambitious powering to detect a 50% reduction, this non-statistically significant finding could translate to a reduction of 22,000/130,828 cases (CDC 8/12/20) among US health care workers (HCW), impacting trajectory and resource utilization models that drive decisions on lockdowns and social distancing.
Data found only in the appendix of Boulware et al. suggested greater differences in the effect HCQ among sub-groups. There were reductions (36%) in younger (<35 years) and increases (110%) in older (>50 years) subjects. Our preliminary analysis revealed a significant negative correlation (slope −0.211, CI −0.328-0.094, p=0.016) between treatment lag and disease reduction, reaching 49% when initiated within one day (RR 0.51, CI 0.176-1.46, p=0.249).
There were also differences in disease reduction by HCQ by type of exposure (HCW − 8% vs. household contacts - 31%; RR 0.691, CI 0.398-1.2). The definitions of exposure severity did not discriminate between the numbers or duration (> 10 minutes) of exposures. Differences between exposure types may result from younger HCW and higher risks in less trained household contacts with little access to advanced PPE. The ex-protocol use of zinc and ascorbic acid were likely confounders, as was the possibly active folate placebo.
Exploratory reanalysis of the raw dataset may inform an age- and stage- nuanced approach to COVID-19 using HCQ testable by prospective studies and may provide insight into the various proposed mechanisms of HCQ.
OBJECTIVES To conduct an exploratory re-analysis of the de-identified raw dataset from a randomized study of the use of HCQ for post-exposure prophylaxis of COVID-19 with view to further defining: a) The time dependent effect of HCQ, b) The age dependent effect of HCQ; c) The sub-stratification of time- and age-dependent effects by exposure type and risk level, as well as by the use of zinc and ascorbic acid. d) The design of future clinical trials to test the hypotheses generated by this study.
METHODS Should granularity of data (by age, time-lag, level and type of exposure) be greater than that originally reported, Fisher Exact test will be used to compare the incidence of COVID-19 in HCQ- and control groups, for each sub-group stratification. Since the degree of loss of data granularity due to de-identification is yet unknown, exploratory analyses involving other demographic characteristics cannot be planned. Where sufficient data granularity exists, univariate regression analyses will be conducted to examine the effect of age- and time lag on any effect of HCQ. The possibility will be explored of conducting multivariate Cox regression analyses with propensity score matching to examine observational data relating to the use of zinc and ascorbic acid.
This analysis will be expanded should a dataset from a similarly designed study (Mitja et al., 2020, NCT04304053), with directionally similar results, become available. This protocol was devised using the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) incorporating the WHO Trial Registration Data Set.
Study Status Protocol version 1.1 (August 19 2020)
Protocol registered at: OSF Registries August 19 2020
Registration doi: https://doi.org/10.17605/OSF.IO/9RPYT
1 STUDY INFORMATION
1.1 Title
Scientific and public title: Preventing disease after exposure to COVID-19 using hydroxychloroquine: exploratory re-analysis of age and time-nuanced effects.
1.2 Authors
Responsible for protocol and analysis
David M. Wiseman PhD, MRPharmS, Synechion, Inc., Dallas, TX.
Responsible for review of protocol, analysis
Mayur S Ramesh MD. Henry Ford Hospital, Detroit, MI
Pierre Kory, MD. Advocate Aurora Critical Care Service, Aurora St. Luke’s Medical Center, Milwaukee, WI
Dan Mazzucco, PhD. ZSX Medical, LLC, Philadelphia, PA. Adjunct Professor, Biomedical Engineering, Rowan University, NJ
No professional writers will be employed.
1.3 Description
BACKGROUND: A recently published randomized trial found a non-statistically significant reduction in Covid-19 of 17% when hydroxychloroquine (HCQ) was used for postexposure prophylaxis.1 Based on practical considerations2 rather than considerations of clinical meaningfulness, the study was powered detect reduction of development of COVID-19 by 50%. Since this was a “pragmatic” clinical trial, where effect sizes are typically smaller than under tightly controlled “explanatory” trial conditions,3 and with greater heterogeneity, this may have been over-ambitious. Based on CDC (cdc.gov/coronavirus/2019-ncov/cases-updates/cases-in-us.html, 8/12/20) likely underestimates of 130,828 COVID-19 cases among Health Care Workers (HCW) in the USA, a 17% reduction in COVID-19 would have translated to a reduction of about 22,240 cases. Since a similar figure of 17% was considered clinically meaningful for a reduction of mortality using dexamethasone in severe hospitalized cases of COVID-19,4 a 17% reduction of COVID-19 in certain populations may impact modeling of COVID-19 trajectory and health resource estimates that drive decisions on lockdowns and social distancing.
Supplemental data not discussed in the main body of the published paper suggested greater reductions of COVID-19 in a number of sub-groups. There were reductions (36%) in younger (<35 years) and increases (110%) in older (> 50 years) subjects. This is consistent with the findings of an observational prophylaxis study involving mainly younger HCW in India.5 Our preliminary analysis revealed a statistically significant negative correlation (slope −0.211, 95%CI −0.328—0.094, p=0.016, rho CI −1 to −0.42, confirmed independently6) between treatment lag and reduction of COVID-19, reaching 49% when given within one day after exposure (RR 0.51, CI 0.176-1.46, p=0.249). The early use of HCQ is supported by mathematical modeling that considers the peaking of the viral load in the pharyngeal cavity and the effect of the drug to kill infected cells by enhancing cell-mediated immunity.7
Post-Exposure Prophylaxis with HCQ: Time Lag and Age Dependency
There were also differences in development of COVID-19 by type of exposure (HCW vs. household; (OR 0.53, CI 0.299-0.94, p = 0.031) in the placebo group, with a small (8%) reduction in COVID-19 with HCQ in HCW, and a 31% reduction in the household group (RR 0.691, CI 0.398-1.2, p=0.24). The definitions of severity of exposure did not however discriminate between the numbers of exposures or durations longer than 10 minutes. Differences between the different exposure types may have been accounted for by a lower age in the HCW group and higher exposure risk in household contacts, who may have had less access to advanced PPE than HCW as well as to training on hygiene practices. Differences between exposure risk related to type of in hospital-assignment and household risk among UK HCW and likely use of PPE has been suggested to account for differences in rates of COVID-19 in a UK study.8
The study also included observational data relating to use of zinc and Vitamin C. There was a higher incidence of symptoms when Vitamin C was used, both without (20.8% vs 11.2%, p=0.014) and with (14.3 vs. 10.6%, p=0.33) HCQ. For zinc there was a similar relationship in the HCQ group, but not in the placebo group. Since details about timing, dose and reasons for self-medicating with these agents are unknown, these observational data confound the overall findings, as does the use of folate for the placebo. In silico analyses show that folate may interact with SARS- Cov-2. 9,10 In an observational study, blood folic acid levels were significantly lower in severe COVID-19 patients.11 For some other viruses there may be an association between folate deficiency/ supplementation and disease severity/amelioration/prevention - Zika12; HPV13,14; HIV15.
Accordingly, the now publicly released raw dataset may reveal important clues to inform an age- and stage-nuanced approach to COVID-19 using hydroxychloroquine testable by prospective studies. Data from a second16 similarly designed study, with directionally similar results may soon become available for similar analyses.
Further examination of these data may also provide insight into the various proposed mechanisms of HCQ. At early stages, consistent with incubation period estimates of 3-8 days17, HCQ alone may be effective in pre-emption because it interferes with viral attachment and initial infectivity. Using zinc (or Vitamin C) may be futile, ineffective or counterproductive in otherwise healthy individuals with no zinc dysregulation. HCQ may be sufficient in young people, however once infection has occurred, HCQ’s actions as an ionophore18 may be more important, and require the presence of zinc for viral inhibition.19
OBJECTIVES: To conduct more detailed analysis of the raw dataset relating the use of HCQ for post-exposure prophylaxis of COVID-19 with view to further defining:
The time dependent effect of HCQ, on post exposure prophylaxis for COVID-19.
The age dependent effect of HCQ, on post exposure prophylaxis for COVID-19.
The sub-stratification of time- and age-dependent effects by exposure type and risk level, as well as by the use of zinc and ascorbic acid.
The design of future clinical trials designed to test the hypotheses generated by this study.
METHODS: A post hoc exploratory re-analysis will be conducted of a dataset obtained from a randomized, controlled studies on the use of hydroxychloroquine to prevent COVID-19 after exposure, This analysis will be expanded should a dataset from a second similar study become available.
The re-analysis will explore apparent exposure risk, age and time-nuanced effects of HCQ with three main hypotheses:
That HCQ exerts a negative age-dependent effect in reducing COVID-19 when given prophylactically after exposure.
The HCQ exerts a negative time-dependent effect in reducing COVID-19 when given prophylactically after exposure.
That any age- and time-dependent effects of HCQ will be further dependent on type and risk level of exposure and the use of zinc or ascorbic acid.
This protocol was devised using the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) 20,21 incorporating the WHO Trial Registration Data Set22 which it references.
1.4 Hypotheses
That HCQ exerts a negative age-dependent effect in reducing COVID-19 when given prophylactically after exposure.
The HCQ exerts a negative time-dependent effect in reducing COVID-19 when given prophylactically after exposure.
That any age- and time-dependent effects of HCQ will be further dependent on type and risk level of exposure and the use of zinc or ascorbic acid.
2 DESIGN PLAN
2.1 Study type
This study is a post hoc exploratory analysis of a dataset obtained from an original study1 that was a randomized, controlled, blinded clinical trial with a pragmatic design with participants from the USA and Canada. Part of the data related to the use of zinc and ascorbic acid which were observational in nature. Data from a similarly designed but open-label, cluster randomized study16 conducted in Spain, if available, will be subjected to similar analyses.
2.2 Blinding
The original study1 was a double-blinded placebo-controlled study. The second study16 was an open-label, cluster-randomized trial.
2.3 Is there any additional blinding in this study?
Details of blinding, implementation of allocation, enrollment, assignment of interventions and unblinding may be found in the publicly-released documentation for the original studies. No further blinding will be employed for the purposes of this exploratory re-analysis.
2.4 Study design
The original stud(y/ies) were interventional studies that used COVID-19 related endpoints to compare the performance of two groups - HCQ treatment vs. or placebo-1 or no-16 treatment. The current protocol will further stratify data by age-, treatment-lag, exposure type and risk level and zinc and ascorbic acid use.
2.5 Randomization
Details of randomization are found in the publicly-released documentation for the original studies. No further randomization will be employed for the purposes of this exploratory re-analysis. Zinc and ascorbic acid were used in one study,1 but this use was not part of the protocol and was uncontrolled. Accordingly, data concerning zinc and ascorbic acid is considered as retrospective and observational.
3 SAMPLING PLAN
3.1 Existing data
The two studies have been published and examined along with their supplementary summary statistics released at the time of publication.1,16 The raw dataset for the first1 of these has been released and has been transmitted in email form to us by the original authors, but this has not yet been accessed. The raw dataset for the second study16 has not been released. Accordingly, this protocol is being registered prior to accessing the data: As of the date of submission, the data exist, but have not been accessed by the sponsor or principal investigator.
3.2 Explanation of existing data
Once this protocol has been registered, the raw dataset will be downloaded and inspected. Other than the analyses already published by the original authors, further data patterns are unknown. Moreover, since there are a number of methods that may be used to de-identify data23 and the level of data granularity is unknown, the type of analyses possible cannot yet be determined.
3.3 Data collection procedures
Details of data collection procedures including inclusion and exclusion criteria are found in the publicly-released documentation for the original studies. No further data collection will be employed for the purposes of this exploratory re-analysis.
3.4 Sample size
Details of sample size are found in the publicly-released documentation for the original studies. The first study1 had 414 and 407 subjects in the HCQ-treated and placebo arms respectively. The second study16 had 1116 and 1198 subjects in the HCQ and no-treatment arms respectively.
3.5 Sample size rationale
Sample size rationales are provided in the original studies. This exploratory study will examine smaller subgroups and without control over their size. The statistical challenges of performing sub-group analyses are well-known.24,25
3.6 Stopping rule
Not applicable
4 VARIABLES
4.1 Manipulated variables
The manipulated variable (treatment arm) employed in the original studies is:
Treatment for post-exposure prophylaxis of COVID-19 with levels as:
Treatment with HCQ
No (or placebo) treatment
Data will be stratified and sub- stratified by:
Subject age
Time between exposure to COVID-19 and treatment
Type of exposure (household vs. health worker contact)
Level of exposure risk
Treatment with zinc or ascorbic acid
4.2 Measured variables
The primary outcomes defined in the original studies were:
Presence of COVID-19, based on symptom-based criteria with expert review.1
Presence of PCR-confirmed, symptomatic Covid-1916
Secondary endpoints were:
Prevention of transmission of COVID-1916
4.3 Indices
The presence of symptom- or test- based COVID-19 will be expressed a percentage of the number of subjects for each subgroup and Odds or Risk Ratios computed with 95% confidence intervals. Risk ratios may also be expressed as its complement, the reduction of risk.
5 ANALYSIS PLAN
5.1 Statistical models
Fisher’s Exact test will be used to compare the incidence of COVID-19 in HCQ- and control groups, for each sub-group stratification described above. Since it is as yet unknown how the raw data have been de-identified23 and the level of data granularity is unknown, other possible analyses cannot yet be determined. Where sufficient data granularity exists, univariate regression analyses will be conducted to examine the effect of age- and time lag on any effect of HCQ. The possibility will be explored of conducting multivariate Cox regression analyses with propensity score matching to examine observational data relating to the use of zinc and ascorbic acid. See also comments received after version 1.0 of this protocol was registered from Dr. Boulware (section 5.6).
Should the dataset become available for the second study, it may be possible to aggregate data with those from the first study. As this is an exploratory analysis, no adjustment will be made for the clustering used in the second study.
5.2 Transformations
Since it is as yet unknown how the raw data have been de-identified23 and the level of data granularity is unknown, other possible analyses and the need for data transformation cannot yet be determined.
5.3 Inference criteria
Since all analyses performed in this protocol are exploratory in nature, p- values from two-tailed tests will be reported if ≤ 0.1 without adjustment for multiple comparisons.25
5.4 Data exclusion
The ability to conduct the planned analyses will depend on the methods used to de-identify data and the level of granularity present in the raw data set: For example:
Subject age: age in years vs. age in year categories (e.g. 18-35)
Time lag between exposure and treatment: days/hours vs. time categories (<1 day, 2, days, <3 days etc.)
Exposure level – distance and time vs. exposure categories (> 10 minutes)
Data that are no more granular than is provided in the published studies will not be eligible for further analysis.
5.5 Missing data
As per the original studies, missing or incomplete data will not be included in the analyses. As this protocol is exploratory in nature, sensitivity analyses will not be conducted.
5.6 Exploratory analysis
Since it is as yet unknown how the raw data have been de-identified23 and the level of data granularity is unknown, further exploratory analyses cannot yet be determined. However, the relationships between the primary outcomes and demographic variables, as well as exposure type and level will be explored.
Subsequent to the registration (https://osf.io/4akug) of version 1.0 of this protocol, we received the following unsolicited but welcome suggestions from Dr. Boulware, the principal author of the first1 of the studies considered in this protocol.
“Along with doing the incorrect Watanabe linear regression analysis, consider doing the actual proper analysis for categorical data, which would be a logistic regression model with an interaction term for treatment group * exposure time group.”
Response: The ability to conduct this type of analysis will depend on the level of data granularity which is as of this protocol version (1.1) unknown to us.
“Look at the actual percentages by each of the subgroup days of exposure. There are some minor variation in the event rate between 12.7% to 17.9% between days in the placebo. As these are small subgroups, this creates artifact when looking at the difference. Consider, pooling the placebo event rate across the Day 1-4 exposure groups. If you can say why the Day 2 placebo group incidence rate would be higher than Day 1 or 3 or 4, sure keep it as is. If you pool the placebo together, the nice perfectly linear line is much less linear. For example, the day 2-3 HCQ incidence is 12% vs 12.2%, respectively. The placebo day 2-3 incidence is 16.9% vs. 14.5%, respectively. If you want to say day 2 works great but not day 3, that’s a bit of a stretch. Yes, the difference is 4.9% vs 2.3% -- but this is an over-interpretation of the subgroup analysis – when the incidence of disease at day 2 and 3 is virtually the same with HCQ.”
Response: The suggestion to pool the placebo data may be reasonable as there really should not be any difference by time and all we are seeing there is noise. However, there is still the possibility that folate has some activity [see 1.3], in which case pooling may not be appropriate.
6 OTHER
6.1 Trial identifiers and Registration
This protocol has been registered with the Open Science Framework (OSF, Center for Open Science).
This study has the internal identifier of SYN2001SYN.
6.2 Protocol Version and Revision History
6.3 Sponsor and Contact Information
Contact for public and scientific queries:
Principal Investigator: Dr. David Wiseman
Synechion, Inc., 18208 Preston Road, Suite D9-405, Dallas, 75252
synechion{at}aol.com 972 931 5596
There are no secondary sponsors
6.5 Funding, role of sponsor and funders, declaration of interests
There is no external support for this study. The sponsor is entirely responsible for the design and conduct of this study. The sponsor and principal investigator have no financial or other conflicts of interest in the subject matter of this protocol.
6.6 Ethics, Consent and Confidentiality
Analyses will be performed on a de-identified, publicly released dataset obtained from studies conducted under ethics committee supervision. No further Ethics Committee, IRB approval, informed consent or confidentiality provisions are required.
6.7 Roles of committees
Not applicable
6.8 Harms, ancillary and post-trial-care
Not applicable
6.9 Data management and access
No data entry is required. Microsoft Excel will be used for primary data manipulation. Vassar Stats (vassarstats.net/) will be used for confirmation of calculations. A study report will be complied and submitted for publication. Microsoft Excel files will be made available on request.to other investigators.
6.10 Summary Results
Data will be summarized as part of the study publication.
6.11 Appendix
Consent materials: Not applicable
Biological specimens: Not applicable
6.12 SPIRIT CHECKLIST of PROTOCOL ITEMS
The checklist for the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) 20,21 was used to construct this protocol. As the SPIRIT checklist references (Item 2b) the WHO Trial Registration Data Set22, a cross reference is provided below for the 24-item WHO Data Set. Where there is no corresponding SPIRIT item, it is listed under SPIRIT Item 2b. Section numbers for the required item are indicated. Non-applicable items are marked “NA.”
SPIRIT 2013 Checklist: Recommended items to address in a clinical trial protocol and related documents*
Data Availability
The publicaly released dataset from the study by Boulware et al. is available at: https://covidpep.umn.edu/data
