Enrollment cuts and primary-outcome changes, not completion-date delays, predict the termination of industry phase 2–3 drug trials: a landmark analysis of ClinicalTrials.gov version histories

Every new medicine that reaches the pharmacy shelf begins its journey as a clinical trial, a carefully planned experiment where researchers test whether a drug works and is safe for people. These trials are not private affairs; by law, companies must register them in a public database called ClinicalTrials.gov before they begin. This registry acts as a public ledger, recording the trial's goals, how many people it hopes to recruit, and when it plans to finish. Throughout the life of a trial, these records are updated as plans change. For years, industry experts and investors have watched these updates like a weather forecast, believing that repeated delays in a trial's finish date were a clear warning sign that the project was failing and might be cancelled. The assumption was simple: if a company keeps pushing back the deadline, something is wrong.

A recent study challenges this long-held belief by looking not just at the current status of a trial, but at the entire history of changes made to its public record. The researcher, Lars Kullman, examined thousands of drug trials sponsored by pharmaceutical companies to see which specific changes in the registry actually predicted that a trial would be stopped early. By analyzing the version history of these records—essentially reading every single edit made over time—the study found that the warning signs everyone is watching are often misleading. Instead of delays, the true signals of trouble are a company lowering its target number of participants or changing the main measurement used to judge the drug's success.

To reach this conclusion, the researcher gathered a massive dataset of over thirteen thousand industry-sponsored drug trials that had started between 2014 and 2021. From this group, a random sample of nearly four thousand trials was selected for a deep dive. The researcher did not just look at the final outcome; they reconstructed the timeline of each trial as it happened, reading every version of the record from the moment it started until it was either completed or terminated. The study focused on three specific types of changes that could be spotted early: how many times the planned finish date was pushed back, whether the estimated number of people to be recruited was cut by at least one-fifth, and whether the primary way the drug's effectiveness was measured was altered after the trial began.

The analysis used a method called a landmark approach, which means the researcher looked at the state of the trial at specific points in time—twelve, twenty-four, and thirty-six months after it started—and asked whether the changes seen up to that point predicted the trial's eventual fate. The results were striking and overturned the common intuition about delays. The study found that postponing the completion date, even multiple times, had no ability to predict whether a trial would be terminated. In fact, about sixty percent of the trials in the study had delayed their finish date at least once, yet they were just as likely to be completed as those that never delayed. A delay, the data suggests, is simply a record of a plan being revised, not a record of a plan failing.

In contrast, two other signals proved to be powerful predictors of a trial's end. When a sponsor reduced their estimated enrollment target by twenty percent or more, the trial was roughly twice as likely to be terminated compared to those that kept their original target. This signal was consistent across all time points, appearing as a strong warning sign whether it happened in the first year or the third. Similarly, changing the primary outcome measure—the specific metric used to decide if the drug worked—was also linked to a higher chance of termination. While this association was not as strong as the enrollment cuts, it was a reliable indicator that emerged clearly after the first two years of the trial.

The study also looked at how these signals worked together. When a trial showed none of these warning signs, it had an eighteen percent chance of being terminated. However, when a trial displayed all three signals—delayed dates, reduced enrollment, and changed outcomes—the chance of termination jumped to forty percent. This pattern held true even when the researcher looked specifically at the most advanced stage of drug testing, known as phase three, where the risk of termination was highest for trials showing these warning signs. The findings suggest that the public record contains a hidden narrative of a trial's health that is far more accurate than the simple observation of a moving deadline.

The implications of this work are significant for anyone who relies on public data to understand the drug development pipeline. Investors, contract research organizations, and even other scientists planning follow-up studies often use the registry to gauge the stability of a competing or related trial. The study indicates that they have been looking at the wrong indicator. The common practice of treating a delayed completion date as a red flag is not supported by the data. Instead, the true early markers of trouble are a shrinking ambition to recruit patients and a fundamental shift in how success is defined. These changes, visible in the version history of the public record, offer a much clearer picture of a trial's future than the date on the calendar.

This research does not claim to solve the mystery of why trials fail, as many reasons exist, from business decisions to safety concerns. However, it provides a concrete, data-driven method for reading the public record to identify which trials are in genuine danger. By automating the reading of these version histories, it is possible to extract information that the current, static view of a trial record simply does not show. The study concludes that the registry's history is a valuable tool, but only if we learn to interpret the right signals. The delays that practitioners have watched for decades turn out to be noise, while the cuts in enrollment and changes in goals are the clear, early warnings of a trial that will not finish.