Pharma has more drugs to test. But the system built to test them is under strain.
The number of drugs in phase 2 and 3 has climbed about 9% year over year, according to a new analysis by Bain & Company. In an April survey of 120 sponsors, CROs and trial-site professionals, respondents estimated that 53% of current phase 2 and 3 trials faced delays versus 44% before 2020. And more than three-quarters of respondents expect trial complexity to increase over the next three years. Just 2% expect it to ease.
That raises a bigger question for pharma R&D: As discovery tools improve and pipelines expand, can clinical development keep pace?
Some of that strain is inseparable from the science. Targeted therapies, biomarker-defined populations and complex modalities can legitimately make trials harder to run.
However, other burdens come from choices sponsors make, including protocols crowded with nonessential procedures, fragmented site systems and technology pilots that never change the work.
“Demands on clinical development are growing faster than the industry’s ability to execute,” Michael Nelson, an associate partner in Bain’s Healthcare and Life Sciences practice focused on pharmaceuticals, and Brittany Rodriguez, a practice senior manager at Bain, said in written responses to PharmaVoice.
More programs, fewer eligible patients
The most obvious pressure point is recruitment.
In Bain’s survey, 86% of sponsors and CROs ranked patient recruitment among their top three causes of trial delays, up from 64% in the firm’s 2022 survey. Protocol amendments also rose as a reported source of delay, from 28% to 37%.
Part of the problem is structural. Precision medicine has made it possible to define smaller groups of patients who may be more likely to benefit from a therapy. But narrower eligibility criteria also mean more studies can end up competing for the same limited pools of patients, investigators and sites.
Bain also reports that average U.S. clinical-trial enrollment has fallen about 28% over the past decade. In the survey, 45% of sponsors and CROs expected further declines, citing costs, access to patients abroad and competition from international trial hubs.
A shrinking U.S. trial footprint also risks producing less evidence reflecting U.S. populations and treatment practices, while limiting investigators’ experience with drugs before launch, Bain warned.
Recruitment pressure is only part of the story. Trial-site respondents highlighted operational barriers that sponsors and CROs have more power to fix. Of 40 trial-site respondents, 45% ranked slow contracting and activation among their biggest challenges, while 35% cited managing progress across multiple software platforms — both ahead of patient recruitment itself.
“Sponsors can’t create more eligible patients, but they can ensure that investigators and patients have a positive trial experience,” Nelson and Rodriguez said.
Sponsors can decide how much work each trial asks of patients and sites.
“Necessary complexity is whatever the science and evidence actually require,” the Bain experts noted. “Avoidable complexity is everything layered on top that doesn’t change the decision a trial is meant to inform.”
The complexities sponsors can control
Laura Galuchie of Merck and Ken Getz of the Tufts Center for the Study of Drug Development wrote this year that phase 3 trials now collect nearly 6 million data points per protocol on average. Their analysis, based on 105 phase 2 and 3 protocols from 14 biopharma companies, found trial data volume has grown more than 280% over the past decade and about 11% a year since 2020.
Getz, in written comments to PharmaVoice, said roughly one-third of data collected in phase 2 and 3 protocols either does not support primary or key secondary endpoints or is gathered more often than needed. That leaves “a significant opportunity to simplify our protocol designs and avoid collecting so much data unnecessarily,” he said.
The impact goes beyond data volume: The researchers estimated that noncore procedures and nonessential data accounted for 25% to 30% of patient and site burden.
A separate peer-reviewed study led by AstraZeneca researchers tested that idea with a 26-question tool covering trial design, regulatory demands and burdens on patients and sites.
After researchers used the tool to reassess 16 challenging phase 2 through phase 4 trial protocols, total complexity scores fell for 12. In a separate analysis of 26 studies, higher scores were correlated with longer times to site activation and early patient enrollment.
Complexity is not just an inconvenience for trial teams. It can slow execution — and, as the research underscores, some of it can be reduced at the design stage.
AI can remove friction. It can also bloat the queue.
AI is often pitched as one way out of the clinical-development crunch. But Bain’s survey suggests AI adoption is outpacing changes to how trials actually run.
Forty-nine percent of sponsors and CROs said they are piloting AI tools without restructuring roles, while 16% said they are redesigning roles and workflows around AI. Just 8% cited site activation among AI’s areas of biggest expected impact.
The underlying issue? Automating a task is not the same as rebuilding the system around it.
Getz and co-authors made a similar argument in a September analysis of AI adoption in drug development. Few sponsors and CROs have scaled AI beyond isolated pilots, they wrote, citing gaps in data readiness, governance and process redesign.
Yet Getz also points to limited successes. “We’ve seen a few limited cases where an AI-enabled solution materially impacted clinical development timelines,” he told PharmaVoice, citing a Tufts analysis of Medable’s AI trial-monitoring agent. That analysis paired actual-use data with oncology benchmarks and modeled assumptions. A detailed paper has been submitted for peer review, and the estimates do not establish widespread end-to-end acceleration.
Success upstream could create a different problem. Faster discovery and more assets could congest clinical development if site capacity, recruitment and operational systems do not expand with the pipeline.
Therein lies the dilemma. AI can help clean data, identify patients or sharpen protocol design. It can’t manufacture eligible patients, experienced investigators or hours in a site coordinator’s day.
Bain’s Nelson and Rodriguez cautioned that AI remains too early in its deployment across development workflows to deliver materially shorter timelines, while noting that localized efficiencies are promising.
Fix the system, not just the tools
The burden of unnecessary procedures and data collection has drawn regulatory attention, too.
FDA guidance on integrating randomized trials into routine clinical practice, issued as a draft in 2024, recommends streamlined protocols and procedures focused on essential data collection. It also says using existing healthcare institutions and clinical expertise can reduce startup times and speed enrollment.
A separate revised FDA draft guidance outlines how master protocols can streamline evidence generation. The agency’s demonstration project on selective safety-data collection explores reducing data collection in appropriate late-stage trials involving drugs with well-characterized safety profiles.
The point is not to lower the evidentiary bar. It is to design feasible protocols, collect data that answers the central research question and make trials less burdensome for patients and sites. Technology is useful when it relieves those pressures, not when it adds another layer.
Bain pointed to three priorities: earlier cross-functional decisions, trials designed around the evidence that matters and an easier experience for patients and sites. Those improvements, they argued, must form part of a broader overhaul of how clinical development operates.
But if discovery keeps getting faster without a corresponding redesign of clinical development, pharma bottlenecks may simply be shifting from one developmental silo to another.
