Medical Device Study for a
Single-Arm Catheter Trial

RESEARCH QUESTION

What the Study Set Out to Ask

This medical device study example follows a clinical research team that plans a single-arm study of a new catheter, with no randomized control group. Results exist from a predecessor device, and the team wants the primary analysis to use them, so that the new study can be smaller than one analyzed on its own.

The question is whether the true success rate of the new catheter, defined as technical success without a major device-related complication within 30 days, exceeds a prespecified performance goal. Borrowing from an earlier device is only defensible when the earlier population and procedure are close enough to the new ones, so the plan states how much borrowing is allowed, what happens if the new data disagree, and what posterior probability counts as success, all before any new patient data are seen.

STUDY DESIGN AND DATA

A Single-Arm Study of About 120 Patients

A prospective single-arm study of about 120 patients treated at several centers, with summary results by center from about 300 patients treated with the predecessor device available as historical data. Each patient has a binary success outcome at 30 days, together with the treating center and baseline risk features recorded for a subgroup check. The historical and planned populations are compared on key characteristics before any prior is built, and the performance goal is written into the plan with its justification from published results for comparable catheters and clinical judgment, so that the goal does not rest on the same data that form the prior.

METHOD AND ASSUMPTIONS

Why a Borrowed Prior, and What Must Hold

A Bayesian model fits this setting because the earlier data enter formally, as a prior with a stated weight, instead of being pooled informally. The prior for the success rate is a mixture of an informative component built from the predecessor data, down-weighted for differences in population and procedure, and a weakly informative component. If the new data conflict with the earlier ones, the weight shifts toward the weak component and borrowing falls. Prior predictive checks show which success rates and counts the prior implies before the new data are fitted, so an implausible prior is found and corrected in the plan, not after the results.

Success is declared when the posterior probability that the success rate exceeds the performance goal passes a threshold fixed in advance. The error rates of such a rule are not fixed by its construction, and borrowing changes them, so the plan simulates its operating characteristics: how often it declares success when the true rate equals the goal or sits just below it, and how often when the true rate is better, with the new data in agreement with the historical data and in conflict with them. Borrowing raises the chance of a wrong success claim when the new population is worse than the old one, and the simulations quantify that. Sampler convergence is checked with R-hat and effective sample size, and the fit with posterior predictive checks.

OUTPUT AND INTERPRETATION

How the Results Are Read

The output reports the prior and its justification, the posterior success rate with its 95% credible interval and the posterior probability of exceeding the performance goal. The observed success proportion with an exact confidence interval sits beside them, so a reader sees what the new data say on their own. One figure plots the prior, the likelihood and the posterior together, and a table lists the operating characteristics of the decision rule under each scenario in the plan.

Sensitivity analyses refit the model with weaker borrowing, with a vague prior, and with a historical set that leaves out the centers least like the new ones, and report how far the posterior probability moves. The interpretation states what the study shows about the success rate of the new catheter in this population and under these assumptions. Decisions about the device itself sit outside the analysis and remain with others.

REPORTING

What the Researcher Received

A written report that documents the prespecified Bayesian plan, the choice of prior and the simulations in a form a methods reviewer can follow, with statistical wording for the methods, matched to the output, for the team to check and incorporate. The study design, clinical interpretation and conclusions remain with the research team.