ILLUSTRATIVE EXAMPLE
Clinical Trial Registry Analysis for a
Results Disclosure Study
Time from primary completion to results posting and changes to primary outcomes across completed drug trials, analyzed in R from public ClinicalTrials.gov records with a registered protocol and a dated snapshot.
This example does not describe a real client or project. Every engagement is confidential, during the project and after it ends.
RESEARCH QUESTION
What the Meta-Research Set Out to Measure
This trial registry analysis example follows a drug policy research team that wants to know how many completed drug trials post summary results within a year of primary completion, whether that differs by sponsor type and phase, and how often a primary outcome changes between registered versions of a record. Each answer is a proportion or an odds ratio with its 95% confidence interval.
The records come from ClinicalTrials.gov, a public registry of trial registrations and summary results, not of participant-level data. The team registers its protocol before any data are extracted, which fixes the definition of the trial cohort, the outcomes and the comparisons in advance.
STUDY DESIGN AND DATA
About 5,000 Registry Records at One Snapshot
A cross-sectional meta-research cohort of about 5,000 completed interventional drug trial records, taken from a dated snapshot of a public relational copy of the registry and queried from R. Each record holds sponsor class, phase, enrollment, primary completion date and the dates results were first submitted and first posted. Earlier registered versions, needed for the outcome coding, are taken from the registry's public record history. Whether a trial was required to report results cannot be settled from registry fields alone.
- Design: cross-sectional meta-research cohort, fixed snapshot date
- Outcome: time from primary completion to first posting of results, and change in primary outcome between versions
- Software requested: R, with extraction and analysis scripts and the snapshot date
METHOD AND ASSUMPTIONS
Why Unposted Trials Are Censored, Not Counted as Failures
A trial without posted results at the snapshot date may still post later, so it is censored at that date and not counted as one that never posts. Time to posting is described with the product-limit estimate, and the proportion posted by 12 months is read from it. Posting within 12 months is also modeled with a logistic model on sponsor type, phase and enrollment, restricted to trials whose primary completion is at least 12 months before the snapshot, with standard errors that allow for sponsors running many trials. The first posted date, when results become public, is the primary date. Because the one-year requirement refers to submission, and posting also depends on the registry's quality control review, the first submitted date is used in a sensitivity analysis.
Under US federal rules, applicable clinical trials must submit results within one year of primary completion, with delays and extensions permitted, and applicability cannot be fully determined from registry fields. The analysis therefore treats this as a limitation, never calls a trial noncompliant, and repeats the estimates in trials whose fields point to an obligation to report results. Primary outcome changes are coded from the registered versions, and two coders independently code a random sample of about 300 records so that agreement is reported as Cohen's kappa with its interval.
OUTPUT AND INTERPRETATION
How the Results Are Read
The output has a flow diagram from registry records to the analyzed trials, the cumulative proportion with results posted over time overall and by sponsor type and phase, and the odds ratios with 95% confidence intervals beside the adjusted proportions posting within 12 months, since odds ratios alone exaggerate differences when posting is common. The proportion with a changed primary outcome is reported with its interval, split by whether the change came before or after the study start.
The findings describe trials in the registry on the snapshot date. A record can lag the true status of a trial, and results may be published elsewhere, such as in a journal, so the report states what a registry-only measure can and cannot say. Differences between sponsor types are presented as descriptive patterns, with no motive assigned, and the policy conclusions are for the team to draw.
REPORTING
What the Researcher Received
A written report with the registered protocol, the extraction code and a methods section that gives the snapshot date, structured to the STROBE items that apply to a cross-sectional cohort of records. The research question, the interpretation and the conclusions remain with the research team.