Locked in Plain Sight: How Clinical Trial Data Hoarding Is Quietly Delaying Cures
The promise of open science has never sounded more credible. Federal agencies reference it in strategic plans. Pharmaceutical executives invoke it at investor conferences. Academic consortia publish earnest white papers about its transformative potential. And yet, for the researchers who actually attempt to access raw clinical trial data—not summary statistics, not curated press releases, but the granular participant-level records that would allow genuine secondary analysis—the experience often resembles navigating a bureaucratic labyrinth designed to discourage entry.
The frustrating reality is that the data exists. Hundreds of thousands of completed clinical trials have generated enormous repositories of biological measurements, adverse event records, genomic profiles, and longitudinal outcomes. The problem is not scarcity. It is access. And the gap between what is technically available and what researchers can realistically use has grown wide enough to constitute a quiet public health failure.
The Architecture of Obstruction
Data sharing in clinical research operates through a patchwork of overlapping requirements, voluntary pledges, and institutional policies that frequently contradict one another. The International Committee of Medical Journal Editors established data availability standards for publications. The National Institutes of Health issued a revised data management and sharing policy that took effect in January 2023. The European Medicines Agency operates its own clinical data portal. The FDA maintains its own submission repositories. Each framework carries different definitions of what constitutes adequate sharing, different timelines for compliance, and different technical standards for formatting.
The result is fragmentation at scale. A researcher attempting to conduct a meta-analysis on, say, cardiovascular outcomes in a particular gene therapy trial cohort may find that one sponsor deposited data in a proprietary software format, another requires a formal data use agreement that takes six months to negotiate, a third posted summary tables but withheld individual participant records, and a fourth claimed an exemption on competitive grounds. None of these sponsors necessarily violated any rule. All of them, collectively, made the analysis impossible.
This is not a hypothetical scenario. A 2022 analysis published in the BMJ found that fewer than half of clinical trials registered on ClinicalTrials.gov had reported results within twelve months of completion. Of those that did report, a significant proportion provided only aggregate outcomes—precisely the data type least useful for detecting subgroup safety signals or identifying unexpected therapeutic responders.
What Gets Lost in the Silence
The downstream consequences of inaccessible trial data are rarely visible to the public, but they accumulate in ways that affect treatment timelines in measurable ways.
Consider the problem of safety signal detection. Adverse events that appear at low frequencies within a single trial may not reach statistical significance in that study's primary analysis. They become clinically meaningful only when data from multiple trials are pooled and examined together. When that pooling is structurally impeded—by incompatible variable naming conventions, missing codebooks, or simply the absence of any mechanism for requesting access—those signals go undetected until the drug reaches a broader population. Post-market surveillance catches some of them. Patients absorb the cost of the ones it misses.
The problem extends to what researchers call the problem of abandoned therapeutic angles. When a trial is designed around a primary endpoint, the sponsor's incentive is to report that endpoint and move on. But clinical trial datasets routinely contain measurements that were never analyzed because they fell outside the study's prespecified scope. A genomics researcher might find meaningful pharmacogenomic patterns in biobanked samples from a cardiovascular trial. A neurologist might identify a cognitive outcome signal buried in a diabetes dataset. These discoveries are only possible if someone can access the underlying records—and access, in the current environment, is rarely straightforward.
The Proprietary Format Problem
Even when data is nominally available, its utility is frequently undermined by technical barriers that receive far less attention than policy-level debates about transparency.
Clinical trial data has historically been collected using a variety of electronic data capture systems, each with its own export conventions. The Clinical Data Interchange Standards Consortium has developed the CDISC standards—SDTM and ADaM in particular—as common frameworks for organizing submission data. Adoption, however, remains inconsistent. Older trial archives, in particular, contain data structured in formats that require substantial reformatting before any analysis is possible. That reformatting work is skilled, time-consuming, and expensive. For academic researchers operating on constrained budgets, it can be prohibitive.
There is also the question of metadata—the descriptive information that gives raw data its meaning. A spreadsheet column labeled "AVAL" is useless without the accompanying codebook explaining that it represents the analyzed value of a specific biomarker measured at a specific timepoint. Metadata completeness varies dramatically across submissions. When it is inadequate, the data itself becomes effectively uninterpretable, regardless of whether it technically satisfies a sharing requirement.
Gatekeeping by Design
Institutional gatekeeping adds another layer of friction. Several major pharmaceutical companies have established controlled access platforms—Project Data Sphere, the Yale Open Data Access Project, and others—that allow researchers to request data under defined conditions. These platforms represent genuine progress. They also, by design, require applicants to submit research proposals, obtain institutional approval, sign data use agreements, and in some cases pay administrative fees. The review process can extend for months.
For researchers pursuing exploratory or hypothesis-generating work—the kind of science most likely to surface unexpected findings—this approval-first model creates a structural disincentive. Proposing a specific research question before accessing data assumes that the researcher already knows what to look for. But some of the most consequential discoveries in medicine have emerged from researchers who were simply allowed to look.
The gatekeeping concern is not without legitimate basis. Patient privacy is a genuine obligation. Competitive intellectual property claims have legal standing. The risk of data misuse or misinterpretation is real. But the current calibration of these protections tilts heavily toward institutional risk management rather than scientific utility—and that calibration carries its own costs, ones that are simply less visible because they manifest as delays and missed discoveries rather than discrete harms.
The Path Forward Requires More Than Pledges
The infrastructure for meaningful clinical data sharing exists in outline, if not yet in practice. Standardized metadata requirements, interoperable repository architectures, streamlined data use agreement templates, and federated analysis platforms that allow computation on data without requiring its physical transfer—all of these technical and policy tools have been proposed, piloted, and in some cases partially implemented.
What has been lacking is the institutional will to treat data accessibility as a scientific priority rather than a compliance checkbox. The NIH's 2023 policy update was a meaningful step. But policy without enforcement, and enforcement without technical support, produces paperwork rather than progress.
For the patients waiting on treatments that existing evidence may already support, the distinction between a curable disease and an untreated one sometimes comes down not to what science knows, but to whether a researcher in Bethesda or Boston can get a clean dataset on a Tuesday afternoon. That is not a scientific limitation. It is an organizational one. And organizational problems, unlike the hardest problems in biology, are ones we already know how to solve.
