Why your PMCF sample size is probably wrong

Written by HATEM RABEH, MD, MSc Ing

Your Clinical Evaluation Expert And Partner

in
S

The sample size calculation appears in almost every PMCF plan I review. It sits there, formatted nicely, often citing statistical formulas and confidence intervals. But when I ask what clinical question this sample size is designed to answer, the conversation stops.

The problem is not statistical ignorance. Most manufacturers have access to statisticians. The problem is strategic confusion. They calculate sample sizes before defining what needs to be proven.

This matters more under MDR because Notified Bodies are trained to identify statistical window dressing. A sample size without clinical purpose is a red flag, not a strength.

The Question Comes First

MDCG 2020-7 does not require statistical power calculations for every PMCF study. It requires clinical data collection that addresses residual uncertainties and confirms safety and performance claims.

The starting point is always the clinical evaluation. What uncertainties remain after you have reviewed the literature and analyzed your existing data? What performance claims need continuous verification? What safety concerns require ongoing monitoring?

Only after these questions are clear can you decide whether a sample size calculation is meaningful.

Three Categories of PMCF Data Collection

Not all PMCF activities require the same statistical approach. Confusing these categories leads to inappropriate sample size calculations and deficiency letters.

Category 1: Safety Surveillance

Most devices need ongoing safety monitoring. This is not hypothesis testing. This is signal detection.

You are not trying to prove that your adverse event rate is below a specific threshold with 95% confidence. You are watching for patterns, clusters, and emerging risks.

For safety surveillance, sample size calculations often do not apply. What matters is systematic data collection from a representative population over a defined period. The MDR requires continuous monitoring, not statistically powered studies to confirm safety assumptions.

I see manufacturers calculate sample sizes for safety surveillance based on expected adverse event rates. Then they get deficiencies because the approach does not match the objective. Safety surveillance is about detection capability, not statistical proof of absence.

Category 2: Performance Confirmation

Some PMCF plans aim to confirm that the device performs as claimed in routine clinical use. This might require sample size considerations, but only if you are testing a specific performance hypothesis.

For example, if your IFU claims a diagnostic sensitivity of 90% and your pre-market data came from a controlled study, you might conduct PMCF to verify this performance in real-world conditions.

Here, a sample size calculation makes sense. You define the expected performance, the acceptable lower confidence limit, and the precision you need. Then you calculate the minimum sample size that delivers this precision.

But notice the sequence. The clinical question defines the statistical question. Not the other way around.

Category 3: Hypothesis Testing

Some PMCF activities are designed to answer specific clinical questions that emerged from the clinical evaluation. For instance, whether a new version of your device performs equivalently to the previous version. Or whether a specific patient subgroup experiences different outcomes.

These are hypothesis-driven studies. They require formal sample size calculations because you are making statistical inferences.

This is the only category where classical power calculations are expected. And even here, the hypothesis must come from the clinical evaluation. If your CER identified no residual uncertainties requiring hypothesis testing, forcing a powered study into your PMCF plan creates regulatory friction.

What Reviewers Actually Check

When I review a PMCF plan with a sample size calculation, I look for alignment. Does the sample size connect to a clinical question identified in the CER? Does the statistical approach match the study objective?

Most deficiencies arise from misalignment. The CER lists routine safety monitoring as the primary PMCF objective, but the plan includes a sample size calculation designed for hypothesis testing. Or the plan states it will confirm performance, but the sample size calculation assumes a specific adverse event rate.

These disconnects signal that the manufacturer does not understand their own PMCF strategy.

Another common issue is citing sample sizes from literature without adaptation. A manufacturer references a clinical study that used 150 patients and adopts the same target without explaining why that number is appropriate for their device, their claims, and their residual uncertainties.

Notified Bodies notice this immediately. It suggests the sample size is decorative, not functional.

How to Approach Sample Size Justification

Start with the clinical evaluation report. Identify the specific uncertainties or claims that require PMCF data.

For each objective, ask whether you are monitoring, confirming, or testing.

If you are monitoring safety signals, describe your data collection approach and justify the duration and scope based on device risk, volume, and patient exposure. Sample size calculations usually do not apply.

If you are confirming performance, define the performance parameter, the expected value, and the precision you need. Then calculate the sample size required to achieve that precision. Justify why that precision level is clinically meaningful.

If you are testing a hypothesis, state the hypothesis explicitly. Define your null and alternative hypotheses, your acceptable error rates, and your minimum clinically important difference. Calculate the sample size accordingly.

In all cases, document your reasoning. Do not present a sample size as if it emerged from statistical necessity. It emerged from clinical strategy.

When Small Numbers Are Acceptable

Manufacturers worry that small sample sizes signal weak evidence. This is not always true.

For rare conditions or niche devices, small numbers may be the only realistic option. What matters is whether the data collection is systematic and whether you can demonstrate that the sample represents your target population.

I have seen PMCF plans with 30 patients that were accepted because the manufacturer clearly explained the epidemiology, the market size, and the data collection strategy. The Notified Body understood that 30 patients represented a significant proportion of the total user base.

Contrast this with PMCF plans targeting 200 patients with no justification beyond

For the complete framework on writing PMCF documentation, read our guide on PMCF plans and reports under MDR.