Liberating the data from clinical trials

On behalf of the DCCT/EDIC Research Group, we are grateful for the citation by Henry and Fitzpatrick (1) who praised our “bold move” in sharing our data over many years with outside investigators. We did so by making DCCT/EDIC data and biosamples available through the National Institute of Diabetes and Digestive and Kidney Diseases Central Repository (2). While we share the authors’ enthusiasm with regard to the importance of data transparency and sharing, we think it is important to correct several inaccuracies in their editorial.

In supporting their thesis that “liberating” clinical trial data will facilitate re-analyses and new analyses, Henry and Fitzpatrick state that there have been “over 220 ancillary studies using DCCT data”. We suspect that they misinterpreted the publications from EDIC to be “ancillary studies”. The Diabetes Control and Complications Trial (DCCT) was a randomized clinical trial conducted during 1983-1993. The Epidemiology of Diabetes Interventions and Complications (EDIC) is the long-term epidemiologic study of the DCCT cohort. EDIC is not an ancillary study of the DCCT, but rather a continuation of the original project. In fact, the total publication record by the DCCT/EDIC Research Group is about 250 publications to date (3). Henry and Fitzpatrick also mistakenly note that several of the papers derived from data sharing “have had an impact on the management of diabetes”, but then reference our main study paper (3).

The shared data (and biologic samples) available through the NIDDK Central Repository are meant to allow investigators outside of the Research Group to perform completely independent analyses of DCCT/EDIC data. The number of publications that have come from such analyses is relatively small, perhaps 25 in total. To the best of our knowledge and in our judgment, none of the publications by outside investigators using data accessed through the Central Repository has materially influenced diabetes care to date.

We agree that data sharing is desirable but the benefits should not be exaggerated. Large and comprehensive studies such as DCCT/EDIC usually publish the most important analyses, often leaving only re-analyses, that may be difficult to publish, or less interesting clinical and scientific topics for other investigators to pursue.

David M. Nathan, MD
Saul Genuth, MD
Bernard Zinman, MD
John M. Lachin, ScD

1) Henry D, Fitzpatrick T. Liberating the data from clinical trials. BMJ 2015;351:h4601

2) https://www.niddkrepository.org/ (accessed September 30, 2015)

3) https://edic.bsc.gwu.edu/web/edic/publications

Henry and Fitzpatrick make a powerful case for more effective sharing of trial data through linkage to administrative data. Consented linkage is not controversial and is widely used for cohort studies and, increasingly, for trials. However, the initiative in Ontario to reactivate ‘dormant’ trials to capture long-term outcomes and to answer new questions through linking trial data to administrative data pushes new boundaries, which could generate a wealth of new evidence to unresolved questions, quickly and cheaply.(1)

Could the UK do something similar? The UK holds a treasure trove of dormant trials that could generate new science. Thanks to incentives from funders, more than three decades of high-quality, publicly-funded trials have been retained and archived for re-use. Combined with a wealth of population-level administrative data from healthcare and other sectors such as education, welfare, and employment, the UK has an un-paralleled resource. The jewel in the crown is primary care data.

Many services and sectors could benefit from reactivating dormant trials in the UK. For example, medicines regulators could create incentives to ensure that trials for licensing of medicines or devices can be linked to administrative data for long-term monitoring of adverse outcomes.(2) Children stand to benefit too. A wealth of trials in early childhood lie dormant that could produce important evidence on long-term wellbeing from early interventions if linked to administrative data from health and other sectors. For example, a series of MRC-funded trials of >4000 children in the 1980s and 1990s showed benefits of early enriched nutrition for preterm infants for cognition at school age but potential harms due to increased cardiovascular risks.(3,4) Enriched nutrition is now standard practice across the world. No new studies are likely to be funded and dwindling response rates and biased attrition mean that valid answers from consented follow up of past trials is not achievable. Follow-up of these trials through linkage with administrative data in adolescence and adulthood could help determine whether benefits for the head (based on education achievement data) outweigh the harms for the heart (based on indicators of cardiovascular disease from primary care and hospital data). Both these outcomes will impact on long-term economic success, which could be directly assessed if data is linked to economic administrative data.

So what are the barriers? Firstly, cost. Primary care data costs six-figure sums that few researchers or academic institutions can afford. Patients in 10% of GP practices in the UK contribute data to the Clinical Practice Research Datalink but few analysts, apart from those in the commercial sector, can afford to re-use the data. The solution is to prohibit profits from selling of administrative data from public services and allow only cost recovery for data processing.

Second is capacity. Hospital data for England is more affordable than data from primary care but the Health and Social Care Information Centre currently have a backlog of >200 applications.(5) Each week they receive more applications than they process. The solution is adequate government investment in data services for all government agencies that provide administrative data (not just health). Investment needs to support research and development within data providers to make more effective use of administrative data, improve consistency and transparency across government administrations, and to forge closer partnership with the research community. Funding needs to be conditional on outputs that benefit services and society rather than lump sums for infrastructure, which can disappear.

The third barrier is decisions about data access that are based on narrow interpretation of rules rather than weighing of societal benefits and risks. Trials from the 80s and 90s did not specifically ask for consent to link to administrative data 30 years later. However, it may be unethical to prevent benefits to society of new evidence from being realised due to lack of consent (given stringent safeguards to avoid identification of individuals).(6) Ontario is reviewing their policies to make better use of past investments by trial participants and research funders. The UK should do the same. Connecting rigorous science to data from public services will reap benefits far beyond the original trial questions.

(1) Henry D, Fitzpatrick T. Liberating the data from clinical trials. BMJ 2015;351:h4601
(2) Harron K, Gamble C, Gilbert R. E-health data to support and enhance randomised controlled trials in the United Kingdom. Clinical Trials 2014;12(2).
(3) Isaacs EB, Ross S, Kennedy D, Weaver LT, Lucas A, Fewtrell M. Pediatrics 2011;128(4):e890-e98.
(4) Kennedy K, Ross S, Isaacs EB, Weaver LT, Singhal A, Lucas A, Fewtrell M. Arch Dis Child 2010;95(8):588-95.
(5) http://www.hscic.gov.uk/dars-dashboard
(6) Allen J, Holman CD, Meslin EM,Stanley F. Privacy protectionism and health information: is there any redress for harms to health? J Law Med. 2013 Dec;21(2):473-85.