Use of Bayesian techniques in clinical trials for rheumatoid arthritis and systemic sclerosis: a scoping review

Abstract

Objective

To gather all relevant literature surrounding the use of Bayesian methods in clinical trials for rheumatoid arthritis and systemic sclerosis; and to assess the use of these methods within said trials.

Methods

Medline and Embase were searched on August 18, 2024. The search strategy and screening process was performed by a single reviewer and verified by a secondary expert. We included studies that presented the primary results of a clinical trial designed to examine a treatment for either rheumatoid arthritis or systemic sclerosis, and that also included the use of a Bayesian technique. From these studies, we extracted the following information: author(s), title, year of publication, study objectives, disease under study, treatment under study, description of study sample, phase of trial, main results, description of Bayesian technique employed, and rationale for use of Bayesian technique (if applicable). The Cochrane risk of bias assessment tool was used to critically appraise each included study. Extracted data were recorded in a spreadsheet and results were synthesized narratively.

Results

A total of 11 studies were included in the final review. Seven of these studies evaluated treatments for rheumatoid arthritis, and four evaluated treatments for systemic sclerosis. A total of five Bayesian techniques were identified. These techniques included the use of posterior probabilities for efficacy analysis, simulation to estimate power and type I error for trials that employed a Bayesian analysis, Bayesian dose-finding algorithms, interim analyses using Bayesian stopping rules, and Bayesian response-adaptive randomization. A variety of rationales for the decision to use Bayesian methods were expressed.

Conclusions

The application of Bayesian methods offers many notable advantages, yet their uptake within rheumatology trials has been slow. Increased awareness of these advantages could greatly benefit the clinical trial world.

Supplementary Information

The online version contains supplementary material available at 10.1186/s41927-025-00563-2.

Background

Within the field of medical research, the research design most capable of providing high-quality evidence is the clinical trial [1]. Historically, clinical trials have been designed within the Frequentist framework [2]. However, more recently there has been a rise in clinical trials that employ Bayesian methods in design or analysis [2]. The two inference frameworks are philosophically different, and this affects the way in which investigators draw conclusions about a treatment effect [3]. Frequentist statistics consider the data the only random component. Inference about the treatment effect, which is assumed fixed albeit unknown, therefore relies on the sampling distribution of the data summaries. This results in probability statements about the data that assume hypothetical fixed values of the treatment effect (p-value). In contrast to this, Bayesian statistics allow specification of probability distributions for the unknown parameter, including the treatment effect. These distributions represent the state of knowledge prior to the study (prior distribution) which are updated by the collected data into a probability distribution that combines all available information (posterior distribution). In addition to philosophical differences, the two frameworks lead to very different methods of clinical trial design and analysis. We argue that the use of Bayesian methods within clinical trials offers a range of advantages. Bayesian inference offers flexibility in sequential analysis in adaptive designs, allowing for incorporation of stopping criteria, as well as accommodating response adaptive randomization [4]. Bayesian methods can also result in a general increase in trial efficiency [5]. Indeed, by using informative priors, researchers can make efficient use of all available information by combining external information with data accumulated within a trial. Many other advantages associated with the application of Bayesian methods to clinical trials have been reviewed in detail elsewhere [6].

Despite the advantages, the uptake of Bayesian methods in clinical trials has been slow. While this has been explicitly studied in other fields, such as oncology [7], there have been no studies to date examining the uptake of Bayesian methods in the field of rheumatology. Understanding if and how Bayesian methods are used in trials for rheumatic diseases is of particular interest, as Bayesian techniques provide notable advantages specific to this field. To highlight one example, in the case of rheumatoid arthritis, which has “a window of opportunity” with earlier control of disease associated with better outcomes, limiting participant exposure to ineffective treatments is critical [8, 9]. Thus, the use of Bayesian adaptive randomization, for example, could be extremely beneficial. The ease with which external information can be incorporated with trial data is also particularly applicable in trials for rheumatic diseases [10]. As many rheumatic diseases are considered rare diseases, sample sizes in rheumatology trials tend to be small. Thus, being able to use all available information maximizes investigators’ abilities to draw clinically relevant conclusions, even with smaller sample sizes.

Since the application of Bayesian techniques can be particularly advantageous to rheumatology trials, it is important to understand if and how they are currently used within this field. To obtain a better understanding of this, we undertook a scoping review with the purpose of gathering all relevant literature surrounding the use of Bayesian methods in clinical trials for two rheumatic diseases of interest, namely rheumatoid arthritis and systemic sclerosis. These diseases were selected as they are exemplary of the type of rheumatology trial that would most benefit from Bayesian methods. For instance, both rheumatoid arthritis and systemic sclerosis are rare diseases and thus, would greatly benefit from trial design that maximizes inference capacity with small sample sizes. We also sought to assess the use of these Bayesian methods within said trials.

Methods

The methods of this scoping review follow PRISMA reporting guidelines [11]. A checklist outlining how specific guidelines were met is included in the Supplemental Material. We created a systematic search strategy in consultation with a health sciences librarian using a combination of medical subject headings and key words for clinical trials, rheumatoid arthritis, systemic sclerosis, and Bayesian methods. We searched two electronic databases: Medline and Embase. We restricted the search to articles published in English, but put no limits on the place of publication or date of publication. The specific search terms used for each database can be found in the Supplemental Material. The scoping review was registered in Open Science Framework (10.17605/OSF.IO/P6YH4); a protocol for this review can also be found at this location.

We included studies that presented the primary results of a clinical trial designed to examine a treatment for either rheumatoid arthritis or systemic sclerosis, and that also included the use of a Bayesian technique. We excluded conference abstracts, systematic reviews and meta-analyses, commentaries, communications, books, and letters to the editor. A single reviewer (MMC) performed the database searches and saved all records in a reference manager. This same reviewer performed a de-duplication process before moving all identified records to Rayyan Systematic Review Software [12]. MMC then screened the title/abstracts of each identified record based on the aforementioned eligibility criteria. Full texts were retrieved for each record deemed eligible after title/abstract screening. MMC then screened the full texts using the same eligibility criteria. Note that all screening processes were only performed by a single reviewer given that the objective of this research was primarily descriptive. We therefore sought to increase efficiency regarding the screening processes by only using one reviewer. Moreover, we believe the implicated bias associated with the use of only one reviewer, in this context, was not sufficiently threatening to warrant the addition of a second independent reviewer. That being said, after the completion of the screening process, MMC presented the list of articles included after full text screening to a second reviewer (MH), who used subject expertise to add records missed through the search strategy. Only articles digitized as of August 18, 2024 were included. The Cochrane risk of bias tool for randomized trials was used to conduct a critical appraisal of each included article [13]. This process was also performed by a single reviewer (MMC).

The following data were extracted from each included article: author(s), title, year of publication, study objectives, disease under study, treatment under study, description of study sample, phase of clinical trial, main results, description of Bayesian technique employed, rationale for use of Bayesian technique (if applicable). This process was also performed initially by a single reviewer (MMC) and then reviewed by a secondary reviewer (MH). Extracted data were recorded in a spreadsheet and results were synthesized narratively.

Results

Our search strategy returned 1,259 records, of which 977 were screened for eligibility after excluding duplicates (Fig. 1). By screening titles and/or abstracts, 134 articles were selected for full text review, and 9 studies met the eligibility criteria. An additional 2 studies were included after the revision of a second reviewer with subject expertise.

Fig. 1.

Details of screening procedures

In total, we identified eleven studies that presented the primary results of a clinical trial designed to evaluate the treatment for either systemic sclerosis or rheumatoid arthritis, and also included a Bayesian technique. Nine of these studies were identified through the original screening process [14–22] and two others were added by the secondary expert [23, 24]. Specific characteristics of the included studies are detailed in Table 1. All studies were published after 2000, with the majority (nine) having been published after 2010 (Table 1). Five studies scored as having a “low” risk of bias [14, 16–18, 20], two scored as having “some concerns” in regard to bias [19, 22, 23] and three scored as having a “high” risk of bias [15, 21, 24]. Specific results of the risk of bias assessment for each study can be found in the Supplemental Material. Four out of the eleven studies evaluated treatments for systemic sclerosis [14, 15, 23, 24], while the other seven evaluated treatments for rheumatoid arthritis [16–22]. All included studies presented the results of either a phase 1 or phase 2 trial, or a combination of phase 1 and 2.

Table 1.

Select characteristics of included studies

Authors	Year	Research objectives	Treatment under study	Description of study sample	Phase of clinical trial	Main results	Bayesian methods employed	Advantages of Bayesian framework outlined in text
Trials for Rheumatoid Arthritis (RA)
Choy et al.	2013 [19]	To investigate the safety, tolerability, pharmacokinetics and pharmaco-dynamics of GSK315234 in individuals with RA.	Intravenous GSK315234; a humanised anti-Oncostatin M immunoglobin (IgG1) monoclonal antibody.	Individuals between 18 and 75 years of age, who fulfilled the 1987 American College of Rheumatology (ACR) criteria for RA.	1/2	No evidence of efficacy of GSK315234 observed in individuals with RA.	Bayesian adaptive dose-finding algorithm; simulations to estimate power associated with given sample sizes.	Authors noted that a Bayesian adaptive dose-finding algorithm allowed for sequential accumulation of knowledge throughout the trial, which, in turn, maximizes trial efficiency.
Conaghan et al.	2023 [21]	To examine the efficacy, safety and pharmacokinetics of BMS-986,142 versus placebo on a background of methotrexate in individuals with RA.	BMS-986,142; an oral, reversible kinase inhibitor.	Individuals 18 years of age or older, who fulfilled the 2010 ACR criteria for RA, with moderate to severe disease demonstrated by an inadequate response to methotrexate.	2	No evidence of efficacy of BMS-986,142 observed in individuals with RA.	Bayesian adaptive dose-finding algorithm; Bayesian interim analyses.	Authors noted that a Bayesian adaptive dose-finding algorithm and interim analyses resulted in a maximisation of the patient risk-benefit ratio.
Genovese et al.	2013 [20]	To assess the dose-response relationship, safety and efficacy of tabalumab in individuals with RA.	Tabalumab; a human monoclonal antibody that neutralises soluble and membrane B-cell activating factor.	Individuals between 18 and 75 years of age, who fulfilled the 1987 ACR criteria for RA and were taking methotrexate for 16 weeks or more.	2	Results provide evidence for the existence of a dose-response relationship and efficacy of tabalumab in individuals with RA.	Bayesian adaptive randomization; simulations to estimate power associated with given sample sizes.	Authors noted that Bayesian adaptive randomization was used with the intention to estimate the dose-response relationship more precisely and to allocate patients to more effective dosing.
Glatt et al.	2019 [16]	To evaluate the efficacy and safety of certolizumab pegol in combination with bimekizumab in individuals with RA.	Certolizumab pegol; a Fc-free PEGylated tumour necrosis factor inhibitor, in combination with bimekizumab; a monoclonal IgG1 antibody.	Individuals between 18 and 69 years of age, who fulfilled the 2010 ACR criteria RA with moderate to severe disease.	2	Results provide evidence of efficacy of certolizumab in combination with bimekizumab in individuals with RA.	Bayesian efficacy analysis using posterior probabilities; simulations to estimate power associated with given sample sizes.	No discussion of Bayesian advantages within the main text.
Pratt et al.	2021 [22]	To determine the maximum tolerated dose of seliciclib in individuals with RA.	Seliciclib; an orally available cyclin-dependent kinase inhibitor.	Individuals 18 years of age or older, who fulfilled the 1987 ACR criteria for RA with moderate to severe disease and stable treatment with anti-TNF therapy.	1	Results provide evidence that the maximum tolerated dose of seliciclib in individuals with RA is 400 mg.	Bayesian dose-finding algorithm; Bayesian interim analyses using posterior probabilities.	Authors noted that the Bayesian dose-finding algorithm is more accurate and potentially more efficient in identifying maximum tolerated doses in comparison to traditional rule-based approaches. Furthermore, the authors noted that use of Bayesian adaptive designs offer significant ethical and financial advantages such as when participants can be offered expedited enrolment into following trials.
Robinson et al.	2020 [17]	To evaluate the efficacy and safety of ritlecitinib in individuals with RA.	Ritlecitinib; an irreversible inhibitor of JAK3.	Individuals between 18 and 75 years of age, who fulfilled the 2010 ACR criteria for RA.	2	Results provide evidence of efficacy of ritlecitinib in individuals with RA.	Bayesian efficacy analysis using posterior probabilities.	Authors noted that Bayesian analysis, which allowed for historic use of placebo data, allowed for a decreased placebo sample size, thereby limiting the exposure of patients to placebo.
Visvanathan et al.	2019 [18]	To evaluate the safety, efficacy and therapeutic mechanism of BI 655,064 in patients with RA.	BI 655,064; a humanised, antagonistic anti-CD40 monoclonal antibody.	Individuals between 18 and 70 years of age, who fulfilled the 1987 ACR criteria of RA and were unresponsive to methotrexate.	2	No evidence of efficacy of BI 655,064 in patients with RA.	Bayesian efficacy analysis using posterior probabilities; simulations to estimate power associated with given sample sizes.	No discussion of Bayesian advantages within the main text.
Trials for Systemic Sclerosis (SSc)
Denton et al.	2017 [14]	To determine the effect of selexipag on the frequency of Reynaud’s phenomenon (RP) attacks in patients with RP secondary to SSc.	Selexipag; an oral, selective IP prostacyclin receptor agonist.	Individuals 18 years of age or older, who fulfilled the 2013 ACR criteria for SSc and had a score of 3 or more for the RP item.	2	No evidence of efficacy of Selexipag observed in individuals with RP secondary to SSc.	Bayesian efficacy analysis using posterior probabilities; simulations to estimate power associated with given sample sizes.	No discussion of Bayesian advantages within the main text.
Farge et al.	2022 [23]	To assess the safety of mesenchymal stromal cells in individuals with SSc.	A single infusion of allogeneic bone marrow-derived mesenchymal stromal cells.	Individuals between 18 and 70 years of age, who fulfilled the 2013 ACR criteria for SSc with a modified Rodnan skin score of at least 15, and had at least one of the following organ involvement: lung, heart, or renal.	1/2	Results provide evidence that a single infusion of mesenchymal stromal cells in individuals with severe diffuse SSc is safe.	Bayesian estimation of intolerance using posterior probabilities.	No discussion of Bayesian advantages within the main text.
Farge et al.	2002 [15]	To determine the feasibility, the tolerance and the efficacy of autologous hematopoietic stem cell transplant in individuals with SSc.	Autologous hematopoietic stem cell transplant.	Individuals with early, rapidly progressive diffuse SSc and either heart and/or lung involvement.	1/2	Results provide evidence that haematopoietic stem cell transplant in severe SSc appears feasible with low toxicity and short term clinical benefits.	Bayesian interim and efficacy analyses using posterior probabilities.	Authors noted that a Bayesian approach provides a natural, intuitive, appealing way to monitor patient outcomes and appears more efficient in protecting patients from overly toxic or unfeasible treatment.
Johnson et al.	2009 [24]	To demonstrate how the use of Bayesian analysis makes efficient use of available data and provides results in a format that is more clinically relevant to consumers of the medical literature using data from a trial of SSc.	Methotrexate.	Individuals 18 years of age or older, who fulfilled the 1980 ACR criteria for SSc with diffuse disease.	N/A	Reanalysis provides evidence against methotrexate being ineffective in individuals with SSc.	Bayesian efficacy analysis using posterior probabilities (reanalysis of original data).	Authors noted that the Bayesian approach allows for more efficient use of the data at hand, allows for more clinically relevant interpretation of the data, allows for the determination of the probability of simultaneous benefit in multiple outcomes, and allows for the inclusion of previous knowledge into study analysis.

There were five different Bayesian methods that we identified within the included studies (Table 1). The technique most commonly employed was the use of posterior probabilities for efficacy analysis in seven of the included studies [14–18, 23, 24]. Five studies used simulation techniques to estimate power and type I error rate for trials that employed a Bayesian analysis [14, 16, 18–20]. A Bayesian dose-finding algorithm was used in three of the included studies [19, 21, 22]. Only two studies used Bayesian interim analysis during their trial [21, 22]. Notably, only Genovese et al. [20] used a Bayesian adaptive randomization procedure during their trial.

In terms of justification for the use of Bayesian methods, seven of the eleven studies provided a rationale for the use of Bayesian methods within the main text (Table 1). Five of these seven studies specified that the Bayesian approach provided some additional benefit to the participants of their trial [15, 17, 20–22]. These advantages included maximizing the participants’ risk-benefit ratio, protecting participants from overly toxic or unfeasible treatments, allocating participants to more effective dosing regimens, being able to offer patients expedited enrolment into following trials, and limiting the number of participants exposed to placebo. The second most cited rationale for the use of Bayesian methods was an increase in trial efficiency, in one way or another. This increase in efficiency was noted by three of the included studies, however, was considered differently between studies. Choy et al. [19] and Johnson et al. [24] specified trial efficiency in terms of data accumulation and use. In contrast, Farge et al. [15] noted that the use of Bayesian methods enabled their trial to protect participants more efficiently. Both Pratt et al. [22] and Genovese et al. [20] indicated that they thought Bayesian dose-finding methods were able to more accurately estimate the dose-response relationship than traditional rule-based methods. Finally, Johnson et al. [24] identified other advantages of the Bayesian approach. Those being, a more clinically relevant interpretation of the data, allowing for the determination of the probability of simultaneous benefit in multiple outcomes, and allowing for the inclusion of previous knowledge into study analysis.

Discussion

In this study, we sought to assess the literature surrounding the use of Bayesian methods in clinical trials for rheumatoid arthritis or systemic sclerosis. Our search strategy yielded eleven studies that presented the primary results of a clinical trial for either rheumatoid arthritis or systemic sclerosis, and that also employed a Bayesian technique. A range of Bayesian methods were used within these trials, the most common being the use of posterior probabilities for efficacy analysis. Moreover, we found that the authors of these studies cited a variety of reasons for their choice to use Bayesian methods. Again, there was variation in authors’ rationales, the most common being that Bayesian methods result in significant advantages to participants of clinical trials.

While we have already noted the authors’ views on the employment of Bayesian methods within the cited trials (if expressed), we will now highlight our view of how Bayesian methods shaped these trials. We will focus this discussion on three areas of interest, namely trial flexibility, ability to maximize information with small sample sizes, and the interpretations and conclusions of said trials. As previously stated, the sequential nature of the Bayesian framework is ideal for accommodating trial flexibility. This is well exemplified in the trial conducted by Genovese et al. [20] in which Bayesian adaptive randomization was used to make adjustments to dose assignment groups as the trial progressed. Moreover, the Bayesian framework employed in the trial conducted by Pratt et al. [22] allowed for expedited enrolment of certain patients into the phase 2 trial set to follow their own trial. It is features like these that are particularly beneficial to rheumatology trials as they limit participant exposure to either ineffective treatment (in the case of the former) or transition time in which no treatment is received (in the case of the latter). As previously stated, since the progression of certain rheumatic diseases depends on early control, design features that best enable participants’ capacity to undergo a treatment that is most likely to benefit them are crucial.

Another trial feature that is well highlighted within the articles identified in this scoping review is the way in which select trials maximized information with small sample sizes. For instance, the trials conducted by Robinson et al. [17] and Glatt et al. [16] both used informative priors in their analyses. Glatt et al. [16] noted that the informative prior used in their trial was approximately equivalent to the amount of information that would have been generated with thirteen participants. That is, by using an informative prior Glatt et al. [16] required a sample size of thirteen fewer participants than if that same trial were designed without an informative prior. While the recruitment of thirteen extra participants may seem trivial, in the case of rare diseases, such as many rheumatic diseases, sometimes finding even one additional participant can be challenging. In the case of Robinson et al. [17] informative priors were used to supplement information in the placebo group. Once again, this limited participant exposure to ineffective treatments. These are benefits that would have been fundamentally challenging under the frequentist framework which does not allow for formal inclusion of prior information in analyses.

The final feature we will touch upon in our discussion of the implications of using Bayesian methods in rheumatology trials surrounds the interpretations and conclusions drawn from each trial. In the case of the study conducted by Johnson et al. [24], which used Bayesian methods to reanalyze the data from a trial originally conducted within the frequentist framework, the main conclusions drawn from the two analysis methods were entirely different. When the trial was analyzed using frequentist methods, the authors concluded that the treatment under study (methotrexate) was not an effective treatment method for symptoms associated with systemic sclerosis. The reverse conclusion was made when Johnson et al. [24] reanalyzed the data using Bayesian methods. It is impossible to know whether differences in conclusions drawn between the two inference frameworks would be as extreme as in this example for each other trial included in this review. However, there are a couple additional details regarding trial conclusions that we will highlight here. In the case of the trial conducted by Conaghan et al. [21], which used Bayesian predictive probabilities in their interim analyses, we can say that the decisions drawn from these analyses were better informed than they would have been by conditional power, which ignores the uncertainty about parameter estimates. Bayesian predictive probabilities accounts for the uncertainty associated with the estimated parameter used to generate future data, as well as the variability of the future data, resulting in more robust decisions [3]. In terms of interpretations, it is commonly known that effects estimated through Bayesian processes are summarized with probability distributions; this also applies to effects estimated in the trials included in this review. The primary advantage of summarizing estimated effects in this manner is that one can directly estimate the probability that an effect lies with any interval of interest. This is of course informative to researchers, but also facilitates easy communication of trial results to participants and the public at large. This type of inference is not possible within the frequentist framework. While these benefits associated with the Bayesian framework are not specific to rheumatology trials, we believe they were well highlighted in the included trials and important enough to warrant a brief discussion.

In terms of frequency of trials identified, it was not particularly surprising to find so few trials that incorporated a Bayesian technique in design or analysis. The slow uptake of Bayesian methods has been noted by previous authors and exists within other medical fields as well [5]. For instance, within oncology trials, a recent systematic review identified only 225 trials that incorporated a Bayesian component [7]. While this number may seem much larger than the eleven identified within this review, there are many more oncology trials than trials for either rheumatoid arthritis or systemic sclerosis. In fact, in clinicaltrials.gov [25], there are over 100,000 oncology trials registered, compared to the 4,100 trials registered for either rheumatoid arthritis or systemic sclerosis. These numbers suggest a similar proportion of the application of Bayesian methods in trials between the two disease areas. Furthermore, in both fields Bayesian methods are more likely to be applied to early phase clinical trials. Fors & González [7] noted that Bayesian methods are typically used in phase 1 and 2 oncology trials. As previously noted, the same was found for trials for rheumatoid arthritis or systemic sclerosis. Moreover, the models used within these trials are also similar between the two fields.

While many note the considerable advantages associated with the Bayesian approach, it is clear that the uptake of Bayesian methods is slow, regardless of medical discipline. Many have hypothesized as to why this might be. A survey conducted by Clark et al. [5] identified that the main perceived barrier stopping investigators from incorporating Bayesian methods into their trials is insufficient knowledge of these methods. This same survey identified that the second most commonly perceived barrier is a lack of regulatory guidance. These barriers are echoed by Ruberg et al. [26]. Importantly, Ruberg et al. [26] noted that in addition to these perceived barriers, the uptake in Bayesian methods stems from a reluctance of certain investigators who believe the incorporation of prior information results in subjectivity in decision making, and outweighs any benefits provided by the Bayesian framework.

This review is not without limitations. Mainly, screening, data extraction, and risk of bias assessment were all performed by only one reviewer. While all processes were verified and supplemented, if necessary, by an additional secondary reviewer, it is still possible that there is bias in the selection of trials included in this review. We encourage interested readers to perform their own search strategies if they wish to verify the articles included for selection. In terms of the results of the bias assessment, we note that in other circumstances having included select trials that scored as having either a “high” risk of bias or “some concerns” in regard to bias may be concerning. However, since the objectives of the current research were primarily descriptive these scores have a lesser impact on the broad conclusions drawn from this review.

Conclusions

In this manuscript, we have summarized the literature surrounding the use of Bayesian methods within trials for rheumatoid arthritis and systemic sclerosis. These trials were used to speak about the use of Bayesian methods within the broader field of rheumatology trials. We highlighted that certain advantages enabled by the Bayesian paradigm greatly benefit rheumatology trials specifically. These advantages include limiting participant exposure to ineffective treatments which assists early disease control, and maximizing information with small sample sizes. Despite these benefits, we note that the uptake of Bayesian methods within the field of rheumatology is slow. We believe that if awareness of these advantages continues to spread throughout the medical research community, and investigators strive to become increasingly familiar with the implementation of Bayesian methods, the Bayesian paradigm has the potential to greatly benefit the world of clinical trials in rheumatology.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

M.M.C. developed and executed the search strategy and screening process, synthesized the information of all included articles, prepared Figure 1; Table 1, and wrote the manuscript. M.H. contributed to concept development, provided supervision, and revised and edited the manuscript. S.G. provided supervision and revised and edited the manuscript. S.H. contributed to concept development and revised and edited the manuscript. All authors read and approved the final manuscript.

Funding

Maureen Churipuy is supported by a trainee award from the Canadian Network for Statistical Training in Trials (CANSTAT) funded by Canadian Institutes of Health Research (CIHR) grant #262556. Other portions of this work were funded by a CIHR grant #183643 awarded to Dr. Sabrina Hoa and a McGill Interdisciplinary Initiative in Infection and Immunity grant awarded to Dr. Marie Hudson.

Data availability

This research is exclusively based on published literature therefore, data sharing is not applicable.

Declarations

Ethical approval

This research did not involve human participants therefore, did not require acquisition of ethics approval or patient consent.

Consent for publication

This research does not involve human participants therefore, consent for publication is not applicable.

Competing interests

The authors declare no competing interests.