Why a Medical Device Clinical Investigation Is Not a Drug Trial Template

Key points#

• The ethical core is shared across human research: protect participants, ask a worthwhile question, obtain valid consent, use qualified teams, and preserve credible records.
• A device investigation must represent the intended patient, user, environment, procedure, accessories, and device configuration.
• Learning curves, center skill, visible treatment assignment, and co-interventions can affect outcomes in ways that need explicit design and analysis.
• Device accountability includes versions, serial or lot identifiers, software, accessories, calibration, maintenance, returns, and disposition where relevant.
• Device deficiencies and product changes can matter even when no participant has yet experienced a serious adverse event.

Start with the device-use system, not a borrowed protocol#

A medicinal product trial often defines the intervention through active ingredient, dose, route, schedule, and formulation. A device asks for all of that study discipline and then adds a larger system around it: the clinician who selects the patient, the person who operates the product, how it is placed and configured, the accessories and the setting, who reads the output, and which version of the thing is sitting in the room.

That system can determine the result. An implant may perform differently with placement technique. Imaging software may depend on acquisition settings and scanner compatibility. A home monitor may fail because alarms, charging, connectivity, or instructions do not fit real use. A surgical instrument may improve as operators move along a learning curve. Treat any of that as background noise and the study can end up precise about the wrong intervention.

ISO 14155:2026 is the current international standard for good clinical practice in medical device investigations. It replaced the 2020 edition in March 2026 and covers design, conduct, recording, and reporting to assess device clinical performance or effectiveness and safety. The EU MDR adds binding regional requirements for investigations within its scope. ICH E6(R3) is written for medicinal-product trials, but its principles of participant protection, quality by design, proportionate controls, and reliable results remain a useful comparison. Device studies still need their device-specific standard and applicable law.

Let intended purpose define the evidence target#

Before you select endpoints, write the intended purpose in operational terms:

• target condition and patient population;
• intended user and required competencies;
• use environment;
• anatomical site or specimen where relevant;
• indications, contraindications, and limitations;
• clinical benefit or performance claim;
• expected duration and frequency of use; and
• the device, accessories, and configuration being tested.

The clinical investigation plan should then connect each important claim to an endpoint and each important risk to appropriate safety information. A technical success endpoint may show that an implant was placed, but not that it remains effective or improves a patient-important outcome. A diagnostic accuracy result may establish discrimination, but not show that use changes clinical decisions or health outcomes. Endpoint duration must match the claim. A 30-day study cannot settle a five-year durability assertion.

Annex XV of the MDR requires research methods appropriate to the device, clinically relevant endpoints, a sufficient number of observations, intended users, and an environment representative of normal conditions of use. Those requirements make representativeness part of study validity, not a detail to address after the analysis.

Separate product effect from user, procedure, and center effects#

Operator dependence introduces variation that should be planned rather than explained away later. Relevant measures can include prior experience, training received, and procedure volume. They can include adherence to technique, setup or calibration steps, and protocol deviations. Standardization helps interpretation, but training should not create a level of support that will be absent after launch.

A learning curve can produce time trends within a site. Early participants may have different results from later participants, and high-volume centers may differ from new centers. Design options include training thresholds, run-in cases, and stratified randomization. They include limits on operator participation, collection of procedure-level variables, and prespecified analyses by experience or study phase. The right choice depends on whether the claim assumes expert use or ordinary intended use.

Causal attribution is often mixed. A harm may involve a device malfunction, difficult anatomy, a use error encouraged by the interface, and a procedural response. Forcing every event into a single cause can hide the system failure. Record enough information to reconstruct the sequence, then assess device, user, procedure, and environment contributions without assuming they are mutually exclusive.

Design controls and masking around what is feasible#

Randomization remains valuable when it can answer the question and is ethically acceptable. Masking may be harder. A visible implant, distinctive interface, audible alarm, or different procedure can reveal assignment to participants or clinicians. A sham procedure may control expectation and assessment bias, but it introduces its own risk and requires careful ethical justification.

When full masking is not feasible, bias can still be reduced through:

• blinded central reading or outcome assessment;
• separate unmasked treatment and masked assessment roles;
• objective and clinically meaningful endpoint definitions;
• independent event adjudication;
• standardized follow-up and co-intervention rules;
• prespecified handling of crossover and rescue treatment; and
• analyses that acknowledge operator and center clustering.

Single-arm studies and performance goals may be appropriate for some questions, but the comparison must be clinically relevant and based on a defensible source. Historical performance can become stale when practice, patient selection, adjunctive care, or endpoint definitions change. Operational inconvenience alone is not a scientific reason to choose a weak comparator.

Track the evidence object at version level#

Device accountability extends beyond counting dispensed units. Depending on the product, your study record may need model, serial or lot number, implant identifier, software and firmware, algorithm version, accessories, consumables, calibration, maintenance, storage, return, and final disposition.

These details answer a basic question: what exactly did each participant receive or use? Without them, you cannot link a result reliably to the product later submitted or marketed. Configuration records also make safety investigations possible when a problem affects one component, production lot, or software release.

The comparator deserves similar care. A study that compares against an active device must define its configuration, maintenance, operator training, and permitted settings. Otherwise, apparent superiority can result from uneven setup rather than the feature under study.

Record deficiencies before they become injuries#

The EU MDR definition of a device deficiency covers inadequacy in identity, quality, or durability. It covers inadequacy in reliability, safety, or performance. That includes malfunction, use error, or inadequate manufacturer information. A deficiency can be important even when circumstances prevented harm.

For example, an alarm may fail but a clinician may notice the problem, or a component may disconnect before use. These events can reveal a hazard and threaten study integrity. Article 80 requires records for specified adverse events, serious adverse events, certain device deficiencies that might have led to a serious adverse event, and new findings related to them. Reporting duties depend on seriousness, causal relationship, and applicable requirements.

The protocol and supporting procedures should define detection, documentation, and assessment. They should define escalation, reporting, and device quarantine. They should define technical investigation and preservation of evidence. Safety review should not wait for the same fault to injure someone.

Control changes without losing interpretability#

Devices often evolve while clinical development is underway. A change in material, component, or algorithm may reduce risk or improve performance. So may a change in interface, manufacturing process, sterile barrier, accessory, or placement technique. It may also mean that participants no longer received the same intervention.

Change control should document the reason, design inputs, and risk assessment. It should document verification and validation, regulatory handling, and implementation date. It should document affected units and likely effect on the clinical evidence. Under the MDR, modifications likely to have a substantial impact on participant rights, safety, health, or the robustness or reliability of the data follow a formal notification and assessment process.

The statistical plan may need to stratify by configuration, test for interaction, or restrict the primary analysis to a final version. Pooling versions should be justified by evidence that the change does not alter the relevant clinical behavior. The same principle applies to major training or procedural revisions.

Place the investigation inside a lifecycle evidence plan#

A clinical investigation is one source inside a broader clinical evaluation. Nonclinical testing, risk management, and usability engineering together set what the study still has to answer. So do prior clinical data, the literature, and the state-of-the-art alternatives. Post-market surveillance and post-market clinical follow-up cover what is left once it closes.

Your final report should include unfavorable results, missing data, and deviations. It should include device deficiencies, configuration history, and limitations, not only the primary endpoint. A successful short-term result may leave questions about rare harms, durability, new users, or later versions. A negative study may identify a design or use condition that needs to change.

The practical distinction is not that device investigations use weaker trial methods. They use the strongest ethical and scientific methods that fit the technology. Their validity depends on treating the user, procedure, setting, and version as parts of the intervention whenever those parts can change safety or performance.