Advice to the European Medicines Agency (EMA) on data release from clinical trials, published last week, should ultimately help to improve healthcare but reveals widely divided opinion on exactly how data sharing should happen.
The advice documents cover different aspects of clinical data sharing and reuse. Five advisory groups discussed: protecting patient confidentiality, data formats, “rules of engagement” (the process for third parties to obtain the data), good analysis practice, and legal issues. This process followed the EMA’s announcement, in November 2012, to implement a policy on “proactive publication” of data supporting all European drug license applications from 2014.
“We are not here to decide if we publish clinical trial data, but how”, was the impressive opening of the November workshop by Guido Rasi, Executive Director of the Agency. Now, the Agency has much evidence and opinion to review as it seeks to define who should have access to what clinical trial data, when, and how they will be allowed analyse it. Around 200 people representing researchers, drug and device industry representatives, regulators, patients, publishers, editors and medical communicators contributed to the advice documents.
The EMA’s (on schedule) commitment to implementing its policy to establish “trust and confidence in the system” is a significant achievement although we are likely only near the end of the beginning of long process.
Sharing and publication of original research data are important to Faculty of 1000, with F1000Research’s leadership in this domain, as are clinical trials. F1000 (IH) had the privilege of contributing advice to several of the EMA’s documents. The advisory documents cover many issues – too many to comment on point-by-point – but here are a few of the most pertinent.
Whether data are being shared privately between drug companies and independent researchers, or data are being posted publicly online, patient confidentiality is an abiding issue. In the ‘protecting patient confidentiality’ group it was, surprisingly, questioned what value completely de-identified data might have – how could it be useful to research if it is stripped of information? Fortunately there are published peer-reviewed examples of open access clinical datasets which we know have informed future research. And other evidence is growing that many public datasets are reused. There are a number of human- and machine-driven ways of anonymizing patient data but it seems, from the consensus of the advice to the EMA, that the initial focus will be on sharing as much anonymized data as possible without being prescriptive about which anonymization method is used.
Pragmatically, then, it seems unlikely that many of the complete datasets of the first trials the EMA receives under the new policy could be sufficiently anonymous to be posted publicly online. However, if the EMA heeds advice from the advisory group to make metadata about all trials and datasets publicly available this will be progress. Researchers would at least know more about the nature of what data they can request. The 2012 Medtronic-Yale Open Data Access (YODA) policy takes a similar approach. We are moving slowly towards the new default for clinical research being set to ‘open’.
Sharing of data generated commercially raises many legal issues. Several legal directives, such as commercial confidentiality and public interest, will need to be carefully balanced in the EMA’s policy. Copyright was also raised in the legal discussions. One suggestion was that all data submitted to the EMA would be protected by copyright under the EU Database Directive. This seems unlikely, as it assumes all trial data are in a database. Data take many forms within and without databases – from simple tables and spreadsheets to machine-harvestable data points and “facts” from published papers. Whether copyright applies to data is a much debated issue depending on, amongst other things, the legal jurisdiction. Using a waiver or license – a legal tool – specifically designed for data such as Creative Commons CC0 helps overcome this problem. Data repositories such as Dryad and figshare use CC0 for its simplicity and universality, and F1000Research was the first journal to use the CC0 public domain dedication waiver for data it publishes.
Data formats and standards
Standards in research, particularly in data, breed efficiency – efficient reuse, sharing, understanding and computation. The advice on data formats includes some promising recommendations for the open science community. The advisory group was quick to recognise the importance of clinical data standards such as CDSIC – catalogued by F1000Research partner, Biosharing – and file formats that can be read with open source software. But to avoid delays in implementing the policy it seems likely that such standards will not be required and “any format shall be acceptable for all data until the policy is applied by stakeholders”. PDF, a format widely discouraged for data, was even recommended by some as a format for some types of data.
The documents – available with full version history – cover many other areas with equally many opinions, sometimes from within the same stakeholder group. The rules of engagement advice acknowledges a lack of agreement on many of the questions considered. This is a critical issue but perhaps expected. Fear about reanalysis by those without appropriate qualifications is one of the biggest concerns clinical trialists have about sharing their data. It will be fascinating to see what path to openness the EMA now takes in response to all this advice, when their policy is published for public consultation in June.
…but another step back?
Ironically, the same day the advice on publication of trial data was announced, we also learned that the EMA’s existing, less proactive, access to documents policy is being challenged by two pharmaceutical companies on grounds of commercial confidentiality. This news quickly was labelled “a disgrace” by Dr Ben Goldacre the medical doctor, researcher and author who fronts the Alltrials campaign for the registration and reporting of all clinical trials.
The ruling by the European Court, which Alltrials noted puts the EMA in conflict with its own policy, was welcomed by the EMA as an opportunity to clarify the meaning of commercial confidentiality.
This injunction, and advice to the EMA on its data release policy, highlights many differences in opinion on access to data both between different interested parties (such as researchers and drug companies) and also from within the same groups. The BMJ reported last month on support for data access from the drug industry. GSK’s Chief Medical Officer James Shannon reportedly said: “[T]he more eyes that are put on data, the better. That’s why GSK has taken the lead to commit to both patient level data transparency as well as clinical study reports.”
In 2013 there is unprecedented public attention on access to data from clinical research. The UK Government’s Science and Technology Select Committee is reviewing large amounts of oral and written evidence on its recent enquiry on clinical trials; and the Alltrials campaign has attracted more than 50,000 signatures including major pharmaceutical companies. The EMA’s task is now of crafting a policy that satisfies these calls for transparency while respecting patient privacy, commercial confidentiality, and researchers concerns over appropriate secondary analyses. This is nevertheless a real opportunity to change medical evidence for the better. Patients are waiting.
I attended the EMA’s November workshop and was a member of three of their five advisory groups, although I would have liked to have contributed more than time allowed. I was not paid by the EMA for my time. The EMA advice document on protecting patient confidentiality makes several references to a paper I wrote, with co-authors, on anonymizing of clinical data.