Introduction

The UK invests substantial sums in medical research and the international reputation of the sector is vital to the government’s industrial strategy [1]. Unfortunately, systemic problems undermine the rigour of medical research and lead to costly research waste [2].

In this commentary, we propose straightforward measures to reduce waste in medical research that will safeguard investments and ensure the UK remains a productive setting for researchers committed to genuine scientific discovery.

Main text

What are the causes of the reproducibility crisis in medical research?

While fraud is an important problem that leads to non-reproducible research, many of the issues that undermine reproducibility do not involve deliberate misconduct [3, 4]. Here, we concentrate on identifying and suggesting remedies to the systemic threats to research rigour and reproducibility.

Lack of transparency for methods and data

Although most UK medical research is funded through taxation and charitable donations, there are remarkably few requirements placed on researchers to share data, adequately describe methods, or provide their full results. Understanding exactly how studies were carried out requires full transparency of study methods and analytic code (where applicable) to verify results and conclusions, ideally with full access to the data. Without providing this information, the validity of published results can only be taken on trust.

Some journals now have policies requiring authors to make their data, analytic code, and protocols available. In practice, these requirements are routinely ignored [5,6,7]. Considerations of patient privacy or commercial confidentiality are not insurmountable barriers to adequate transparency. Research participants should be offered the opportunity to consent to the sharing of their data, or data can be de-identified and provided via third-party, securely-managed access. The OpenSafely project has provided a successful model throughout the COVID-19 pandemic for safe and secure access to UK patient records for emergency epidemiological research [8]. While individual electronic health records cannot be shared in this instance, the OpenSafely project has made transparency into data access, code, and analysis plans central to the project. The Vivli platform provides another model for securely sharing data collected from clinical research studies and numerous platforms exist for sharing non-sensitive data [9].

Consistent and mandatory sharing of data, code, and protocols via journals is one route to improved transparency. However, there is a strong case that all study documentation for publicly funded research should be reported in the public realm at study completion (regardless of article publication status). The establishment of a single national repository would provide a much more complete and useful record of experiments, as well as reduce duplication in reporting requirements that researchers currently face. The Health Research Authority has recently announced plans for all clinical trials to be automatically registered following ethical approval, in partnership with the ISRCTN registry [10]. This infrastructure could form the nucleus of a more ambitious research repository, hosting comprehensive documentation for all study types.

Proposed measure 1 Journals and funders should ensure authors fulfil their obligation to share study protocols and, as far as possible, analytical code and research data regardless of the study’s ultimate publication status in a journal.

Reporting and publication biases

It is well established that results which are ‘positive’ or novel are much more likely to be published [11] leading to a distorted public record. One study found that 96% of research findings were ‘statistically significant’, a mathematical impossibility [12]. Where unpublished results are taken into account, the evidence of benefit for interventions may diminish or disappear altogether [11, 13].

We can monitor whether clinical trials have been published because of legal and ethical requirements that clinical trials are registered prior to commencement [14]. However, clinical trials are a minority of all medical studies. For other study types, including observational studies that have informed healthcare decision-making throughout the coronavirus pandemic, there is little expectation that researchers pre-register their analysis plans and hypotheses. This makes it impossible to monitor what research is planned, if research plans have been followed, if all results are published, and if interpretation involves unscientific “spin” (Table) [15]. Pre-registering study plans for any research type is now straightforward and free to researchers [16, 17]. Requiring pre-registration for medical research can and should be used to promote transparency and accountability.

Although we believe pre-registration enhances transparency and should be more widely used, even when research is pre-registered, it is common to find unacknowledged deviations between the proposed research and the published manuscript [6]. While registration makes these deviations detectable, detection is onerous and journals often fail to take action when alerted to these issues [6]. Related ‘questionable research practices’, like manipulating analyses to generate statistically significant results (‘p-hacking’) and amending study hypotheses retrospectively to suit the results found (Hypothesising After the Result is Known, or ‘HARKing’) can lead to biased findings when research is not pre-registered or the registrations are not checked against final reports [18].

Registered reports are a publishing format with two methods for preventing these poor research practices. In a registered report, reviewers assess methods before data collection begins. When the researchers and reviewers agree that the design is appropriate, the researchers are given in-principle approval for publication, regardless of the study findings, as long as the proposed methodology is adhered to [19].

Promoting publishing via registered reports addresses both outcome reporting and publication biases for almost all types of research [19]. Since journals decide on publication based on the methods before the results are known, researchers have less incentive to strain to produce eye-catching or positive results. Instead, methodologically sound research enters the record without bias regarding what those results might say. Early evidence suggests that registered reports do improve the quality and rigor of proposed study designs [20, 21]. Unfortunately, medicine lags behind other disciplines like psychology in adopting registered reports, with just over 1% of medical journals offering the format in 2020 [22].

Funders should either incentivise publication via registered report, which will increase their uptake by more journals, or establish their own publication platforms which prioritise the registered report format. The NIHR and Wellcome Trust have both established open and peer-reviewed publication platforms for their funded research [23, 24]. Novel funding pathways are demonstrating that it is possible to integrate peer reviews for registered reports in grant applications which could improve the efficiency of academic discovery and dissemination [25, 26].

Proposed measure 2 Funders and medical journals should incentivise the uptake of registered reports.

Lack of transparency on conflicts of interest

Researchers potential conflicts of interest are difficult to ascertain. Conflict of interest statements in publications are brief and often omit important conflicts, even major financial conflicts like sources of funding or relationships with industry [27]. The RetractionWatch project has logged numerous instances of problematic findings, and eventual retractions of articles, due to important undisclosed conflicts of interest [28]. In the UK, there are currently voluntary registers for those who have received payments from the pharmaceutical industry and for doctors [29, 30]. However, because there is little incentive for individuals to make such declarations, these registers are greatly underutilised. Only 0.002% of doctors’ registered with the General Medical Council (GMC) were listed on the doctors’ voluntary register in 2020 [31]. Patients, the public and other scientists are entitled to be able to check whether a researcher has a conflict of interest and to understand what potential biases might impact a study.

There have been calls for the GMC to curate a registry of doctors’ interests [32, 33]. While such a register would mark an important advance in transparency, it would not cover non-medically qualified researchers. Therefore the UK’s research regulator, the Health Research Authority (HRA), along with the GMC, could be tasked with creating a central register, similar to the US OpenPayments database, to index all medical researchers’ interests using the unique identity numbers (ORCiD) which are already required by institutions and funders [34, 35] Expectations of accurate and up-to-date declarations could be encouraged by employers during researchers’ appraisals, although a legal mandate for regulators to ask for such declarations may require legislation [36].

Proposed measure 3 a mandatory national register of interests for all those who are involved in medical research in the UK should be established, with an expectation that individuals maintain the accuracy of their declarations.

Dysfunctional incentives and research culture

Decisions about who to hire, fund and promote in academia are often informed using reductive, simple metrics such as citations, the journal impact factor of publications or grant income [37, 38]. Such metrics perversely incentivise researchers to generate a high quantity of publications which are perceived to be exciting or newsworthy, rather than prioritising high quality reproducible research that actually benefits patients and the public. Practices that support high quality research by improving transparency and reducing bias, such as registering studies and publishing all results, are not typically used to appraise performance in academia [39]. Initiatives which seek to mitigate dysfunctional incentives and promote practices that are conducive to reproducible research are becoming more widely established throughout the UK. These include the UK Reproducibility Network (UKRN), the San Francisco Declaration on Research Assessment (DORA), Résumé for Researchers, and the Concordat to Support the Career Development of Researchers [40,41,42,43,44]. Such efforts should be supported and expanded.

Proposed measure 4 funders and institutions should stop using metrics such as citations and journal’s impact factor to assess research and instead evaluate based on quality, reproducibility and societal value.

Some clinical professionals are incentivised to undertake research because publication is used as a selection criteria for training programmes and for professional promotion [45]. Such dysfunctional incentives promote research that is undertaken solely for career advancement, by individuals who may lack methodological expertise and commitment to produce high-quality reproducible research. There is no persuasive evidence that authoring scientific publications improves the clinical performance of healthcare professionals. Instead we should aspire to create a system of medical research that produces “less research, better research, and research done for the right reasons” [46].

Proposed measure 5 employers and training programmes for health professionals should remove incentives to publish from their selection procedures.

Outlook

In this commentary we have set out recommendations to improve the transparency and reproducibility of medical research. Not all of these would be easily implemented, and further evidence is needed to articulate their value and best practice. Moreover sincere engagement from funders, government, journals, researchers, and their employing institutions is required. Elsewhere, we highlighted three actions that warrant prioritisation as readily implementable measures with high potential for impact [22, 47]. These are:

  1. (1)

    Mandatory registration of interests for all people and institutions who conduct and publish health research on a single on-line platform accessible to all;

  2. (2)

    Prioritisation by journals and funders of publication of research via Registered Reports; and.

  3. (3)

    Public pre-registration before data collection of the study design of all publicly-funded medical research, along with protocols, analytic code and, where possible, research data and results.

Table 1 outlines how these actions, along with other proposed measures outlined in this commentary could address some of the problems in medical research.

Table 1 Problems in medical research and how they can be mitigated by authors’ proposed strategy
Full size table

There is growing acknowledgement that systemic problems pervade health research. But as long as we fail to take action, medicine’s reproducibility crisis will persist. The recommendations in this commentary aim to advance beyond identification of the problems to meaningful, achievable reform.