Clinical trials generate tons of data that never see the light of day, but an initiative spearheaded by researchers at Yale University is trying to change that. Since 2011, the Yale University Open Data Access (YODA) Project has developed a platform that allows external researchers to gain access to clinical trials data from pharmaceutical companies. Last week (November 27) in Scientific Data, collaborators published their analysis of YODA’s successes and challenges so far.
“It’s nice to have a summary of YODA, which has been one of the leaders of data sharing platforms that have emerged over the past five-plus years,” says Mark Huffman, a cardiologist at Northwestern University who did not participate in the study, but has used the YODA platform to access data from a Johnson & Johnson trial. The authors “help identify some opportunities for how data sharing can move forward using this particular model.”
The YODA project arose from coauthors Joseph Ross and Harlan Krumholz, both physicians at Yale University, whose work showed how much information the medical literature—particularly reports of clinical trials—was missing.
“About half of [clinical] trials that are done in the United States are never published,” Ross tells The Scientist. “Even those that are don’t include all of the information that one might need in order to better come to understand the safety and benefits of a particular therapy.”
Ross says that the drive to have more of these data available was behind the development of YODA. As discussed in the new report, his team began by defining policies for data sharing by corporate partners. Throughout the process, the Yale group has focused on transparency—including being open to public comments for major decisions, such as approval of the final policies put in place for making clinical trial results available—and requires that pharmaceutical companies defer to the academics as independent decision-making authorities when faced with data requests.
We found that we had subtle differences in our replication analysis, which were largely driven by the way in which some of the datasets were defined.—Mark Huffman, Northwestern University
Any investigator who wants clinical data must submit a research proposal, and YODA officials prioritize patient privacy and make sure that the data will not be redistributed. YODA also requires recipients of drug companies’ data to disseminate their results, preferably through the biomedical literature or scientific conference presentations. Data access is currently free.
YODA partnered with Medtronic starting in 2011, to facilitate sharing of individual participant data from both published and unpublished trials for a product, recombinant human bone morphogenetic protein-2, a collaboration that ended in 2015. For this partnership, four teams received de-identified data they analyzed using their own software, and two studies have been published so far. One study showed that certain types of patients—smokers and people younger than 60 years, for instance—had improved clinical outcomes when receiving Medtronic’s drug instead of a bone graft during spinal fusion surgery. The other report indicated that patients who have successful spinal fusions report less pain than those with failed spinal fusions. The external researchers will be required to destroy the data after the use agreement expires—five years in this case.
In 2014, YODA then began an ongoing collaboration with Johnson & Johnson for access to the company’s clinical trials data from pharmaceutical, consumer, and medical device products. As of August 2018, 90 of 100 requests for results from the 270 available clinical trials had been approved, eight had been withdrawn, and two were still under review. The majority of projects based on those requests are still in progress, and there have been 12 resulting publications so far. In one study, for instance, data from trials of the antipsychotic risperidone were pooled with results from trials of other antipsychotics in a meta-analysis to address whether use of these medications leads to an increased risk of death for people with severe mental illness. The authors found no evidence for a link between acute use of antipsychotic medications and death.
The parameters of this data-sharing set-up are different from the Medtronic partnership, in that external researchers must make use of the data via a virtual analytics platform and are not allowed to download results themselves. This presented a set of challenges. As the authors of the report discuss, the arrangement limited the analysis tools available, made uploading complementary data to be analyzed alongside company data challenging, and resulted in a loss of time when learning the platform or when users are logged out unexpectedly and hadn’t saved their work.
Huffman experienced these challenges in accessing data from a cardiology trial from Johnson & Johnson via YODA for a replication analysis his group published in 2017 in JAMA Cardiology. “We found that we had subtle differences in our replication analysis, which were largely driven by the way in which some of the datasets were defined, so when we think about what investigators need to do . . . it’s not just the data that need to be shared but there are things like the statistical code and data dictionaries” that take time to prepare and should also be disseminated, he says. “Investigators who are performing clinical trials and other studies need to think about data sharing on the front end as they’re setting up their data capture systems, not just on the back end.” While he says that it was not the most user-friendly experience, he would use the platform again if there were other cardiology trials available. Right now, there is only one.
“Preparing datasets for sharing is a pain, and it’s nothing [companies] get compensated or rewarded for at all,” says Michelle Mello, a health law scholar at Stanford University who was not involved in the project. She says that the effort and time involved is at least one reason that she has seen a wide spectrum of attitudes toward data sharing from pharmaceutical companies.
Projects like YODA raise other big-picture questions, according to Mello, such as, “Do we need gatekeepers like the Yale group or should companies do this directly?” Another is, if we have independent groups coordinating data sharing, what exactly should they be doing in terms of deciding to release data or not? “My hunch is that things will become more standardized over time,” she says.
At the moment, the biggest dilemma for Ross and his YODA colleagues is paying for the project—and for research done on shared data—long term. “Right now, the companies are investing a lot of resources, and we feel like we’ve developed a process that helps science and helps the scientific community work with secondary data better,” he says. “But now we have to think about how we work with the larger world of funders to . . . make sure that there are actually grants available to support people to work with data like this, as opposed to there only being grants to support people to collect data.”
J.S. Ross et al., “Overview and experience of the YODA Project with clinical trial data sharing after 5 years,” Scientific Data, doi:10.1038/sdata.2018.268, 2018.