Clinical Trial Design for Medical Devices: Where Programs Go Wrong

← Back to All Insights

Medical device clinical trials are not small-molecule drug trials. Endpoints, controls, enrollment dynamics and operational constraints all differ, and a trial team that brings a pharmaceutical playbook to a device program can carry assumptions that do not fit.

Design choices are hard to undo once enrollment begins. Good execution can recover a delayed trial; it cannot supply a control group or an endpoint the protocol never included.

Endpoint Selection: Surrogate vs Clinical

Surrogate endpoints can shorten follow-up and reduce sample size, which makes them attractive. FDA's guidance on pivotal device studies says surrogate endpoints may be appropriate when they are validated and directly correlated to clinical benefit, and that early discussion with FDA is critical to determine whether a surrogate has been established as valid. A useful diligence question is whether the chosen endpoint was discussed with FDA, and what the agency said.

Choosing a Control

The same guidance describes several kinds of control: an active control (another device, surgery or sometimes a drug), against which a study may be designed to show superiority or non-inferiority; a placebo or sham control; no-treatment and historical controls; and single-group studies measured against an objective performance criterion or a performance goal, which the guidance treats as a lower level of evidence than an objective performance criterion. The choice follows from the current standard of care and how much is known about the control's effect, and it shapes how persuasive the result will be. It is a separate question from the premarket pathway: the guidance is written primarily for studies supporting PMAs and notes that its recommendations can also apply to 510(k) and De Novo submissions that include clinical data.

Sample Size

Device pivotal studies are often smaller than drug trials, but a sample size is only as sound as its assumptions: the expected effect, the event rate, the non-inferiority margin where there is one, and expected dropout. An assumption carried forward from an older study or a different patient population deserves a direct question.

Multi-Site Enrollment

Device trials enroll differently from drug trials. Investigators may need procedure-specific training, site startup can take longer, and enrollment per site can run below a drug-trial benchmark. Projections built from comparable device studies are more credible than projections borrowed from pharmaceutical analogs. Site selection also matters for interpretation: FDA's guidance notes that selecting only tertiary care centers, where operators may have more specialized experience, could bias the assessment of device performance.

Operator Experience and the Learning Curve

Device performance can depend on operator skill. FDA's guidance recommends that the protocol define which subjects fall within a learning-curve period and that the statistical analysis plan say how those results will be reported, and it notes that a steep learning curve may have ramifications for labeling and user training. A trial that ignores operator effects can produce results that do not carry over to routine use.

Adaptive Designs and Early FDA Feedback

FDA's final guidance on adaptive designs for medical device clinical studies (July 2016) covers prospectively planned modifications based on accumulating data, which, properly implemented, can reduce resource needs, shorten a study or improve its chance of success. The Q-Submission Program (final guidance, May 2025) lets a sponsor request FDA feedback, including through a Pre-Submission, before a pivotal study begins, which is the natural place to test an adaptive design, a control choice or a surrogate endpoint.

Device trial design is its own discipline. For investors and acquirers, the core design questions (endpoint, control, sample size assumptions, operator effects and what FDA said in pre-submission feedback) can often be asked of the protocol, the IDE correspondence and the ClinicalTrials.gov record before the data arrive.