A Device Investigation Is Not a Small Drug Trial: Designing Clinical Evidence That Satisfies CDSCO, FDA and MDR

Author: Dr. Ananya Chatterjee
Head of Clinical Affairs
Expertise: Regulatory Strategy
Service: Clinical Research (CRO)
Sector: Diagnostic Devices
Therapeutic Devices
Drug Delivery
Topic: Regulatory Strategy & Compliance
Testing & Analytical Methods
Published: 18 May 2026
7 min read

TL;DR

Device trials differ structurally from drug trials: there is usually no placebo, blinding is often impossible, the "dose" varies with operator skill, and the device itself changes during the study period.
Recruitment is the dominant failure mode across all clinical research. Tufts CSDD analysis of nearly 16,000 investigative sites found that 11% of sites ready to recruit enrolled no one at all, and 48% of all selected sites either enrolled nobody or under-enrolled.
Approximately 80% of clinical trials fail to meet their original enrolment timeline, and a 2011 cohort analysis found 19% of closing or terminating registered trials failed to reach 85% of expected accrual — 481 trials involving more than 48,000 patients that could not answer their own primary question.
In India, clinical investigation of an investigational medical device requires CDSCO permission via Form MD-22 (Form MD-24 for IVD clinical performance evaluation), registered ethics committee approval, and prospective CTRI registration. First participant enrolment must occur within one year of permission.
The single highest-value activity in device clinical research is protocol feasibility done before the protocol is finalised, not after the sites underperform.

Clinical evidence generation for medical devices borrows its vocabulary from pharmaceutical research and almost none of its structure. Teams who plan a device investigation using a drug-trial mental model make a predictable set of errors, and those errors are expensive because they surface late.

Five Structural Differences That Change the Study Design

There is often no meaningful placebo. A sham procedure may be feasible for some interventional devices, but a sham is not a placebo — it carries its own risk, and ethics committees will scrutinise it accordingly. For a wearable monitor or a diagnostic device, the comparator is usually a predicate device or an accepted reference standard, not an inert control.

Blinding is frequently impossible. The patient knows whether a device is attached to them. The clinician knows which device they are using. Where blinding cannot be achieved at the treatment level, it must be pushed to the endpoint level: independent, blinded adjudication of outcomes, and pre-specified, objective endpoint definitions that leave no room for assessor drift.

The operator is part of the intervention. For any procedural or interventional device, outcomes depend on operator technique, and technique improves during the study. This learning curve is a genuine confounder. It is handled by pre-specifying training and credentialing requirements, by requiring a minimum number of roll-in cases per operator before contributing to the primary analysis, and sometimes by explicitly modelling case sequence as a covariate.

The device changes. Iterative improvement is normal in device development and abnormal in drug development. A firmware revision mid-study is a protocol issue: it must be pre-specified, version-controlled, and its impact on the pooled analysis justified. Uncontrolled iteration during a pivotal study can invalidate the dataset.

The endpoint is often a measurement agreement problem, not a survival problem. For diagnostic and monitoring devices, the primary question is frequently "does this device agree with the reference standard closely enough to support the intended clinical decision?" That is a Bland–Altman, sensitivity/specificity, or concordance question — with sample size driven by the required confidence interval width around a performance estimate, not by an event rate.

Recruitment Is Where Trials Actually Fail

The recruitment literature is bleak and consistent.

A Tufts Center for the Study of Drug Development analysis covering nearly 16,000 investigative sites across 151 global Phase II and III trials conducted between 2008 and 2010 found that 11% of sites that were ready to begin recruitment enrolled no patients at all, and that 48% of all sites selected for a trial either enrolled nobody or under-enrolled. Non-activation varied sharply by region: 7% in Western Europe, 8% in Eastern Europe, 9% in Asia-Pacific, 13% in North America, and 20% in Latin America.

More broadly, roughly 80% of clinical trials fail to meet their original enrolment timeline. An analysis of registered trials closing or terminating in 2011 found that 19% either failed to reach 85% of expected accrual or terminated early for insufficient accrual — 481 trials involving more than 48,000 patients whose participation could not meaningfully answer the research question.

Protocol complexity is a contributor rather than a coincidence. Tufts CSDD found the total number of endpoints per protocol rose 86% between the 2001–2005 and 2011–2015 periods. Each additional endpoint adds eligibility constraints, procedures, and site burden — and each of those depresses enrolment.

The operational lesson is that feasibility is a design input, not a site-selection step. Before the protocol is locked, someone should be able to answer, with data rather than opinion:

How many patients meeting these exact inclusion and exclusion criteria present at this site per month?
Of those, what fraction will consent, given this procedure burden and this follow-up schedule?
What is the competing-trial landscape at this site for this population?
Who at the site will actually screen, and what else is on their desk?

Sites will answer optimistically to the first question if asked in the abstract. They answer more accurately if asked to run the eligibility criteria against their last twelve months of records.

The Indian Regulatory Pathway, Specifically

Clinical investigation of medical devices in India is governed by the Medical Device Rules, 2017 (G.S.R. 78(E), notified 31 January 2017, effective 1 January 2018), administered by CDSCO.

The operative mechanics:

Application is made to the Central Licensing Authority in Form MD-22 for a pilot or pivotal clinical investigation of an investigational medical device, accompanied by the information specified in the Seventh Schedule. Permission is granted in Form MD-23.
In vitro diagnostics follow a separate route: clinical performance evaluation is applied for in Form MD-24.
Fees are ₹1,00,000 for a pilot or pivotal clinical investigation and ₹25,000 for clinical performance evaluation. Institutions run or funded by Central or State Government are exempt.
Ethics committee approval from a registered ethics committee is required, and in practice the EC approval letter should be in hand before the CDSCO application.
CTRI registration must be prospective. Retrospective registration compromises publishability and creates regulatory exposure.
First participant enrolment must occur within one year of the grant of permission. Missing this window requires fresh prior approval from the CLA.
Compensation obligations for study-related injury or death are prescribed under MDR 2017 and must be provided for.

A device claiming substantial equivalence to a predicate device may still require CDSCO permission for clinical investigation; equivalence does not automatically exempt.

Aligning One Study Across Three Jurisdictions

Running separate studies for FDA, EU MDR, and CDSCO is usually unaffordable for a company of any size below large-cap. Running one study designed to serve all three is achievable, but only if the constraints are imposed at protocol design rather than reconciled afterwards.

The design decisions that make a study portable:

GCP baseline. Design to ISO 14155:2020 (clinical investigation of medical devices for human subjects — good clinical practice). It is the international standard, it is recognised by notified bodies, and it is compatible with CDSCO's GCP guidelines and with FDA expectations for non-IDE and IDE studies alike.

Endpoint definitions that satisfy the strictest reviewer. If EU MDR clinical evaluation under Annex XIV requires demonstration of clinical benefit, and the FDA pathway would accept a performance endpoint, design to the clinical benefit endpoint. Downgrading is easy; upgrading requires a new study.

Statistical analysis plan finalised and version-controlled before database lock. Post-hoc analyses are treated as hypothesis-generating by every regulator, without exception.

Data integrity infrastructure. ALCOA+ principles, audit trails, and a validated EDC system. A study with immaculate science and poor data provenance will not survive a sponsor inspection.

Population representativeness. A notified body assessing an EU MDR clinical evaluation will ask whether the investigated population is representative of the intended EU population. An investigation conducted entirely in one Indian centre may be scientifically sound and still generate an equivalence question. Multi-centre, and where feasible multi-country, design pre-empts this.

Where Device Clinical Programmes Break Down

The study is designed to the regulatory minimum and cannot support reimbursement. Clearance and payment are different evidentiary bars. A study powered to demonstrate substantial equivalence may say nothing about comparative effectiveness, health economics, or workflow impact — which is what a payer or a hospital procurement committee will ask about. Adding a health-economic endpoint at design costs little; generating one afterwards costs a second study.

Sample size calculated from an optimistic effect estimate. Effect sizes from small feasibility studies are systematically inflated. Powering a pivotal study on a pilot point estimate, without discounting, is one of the most common causes of a well-run trial producing a non-significant result.

No pre-specified handling of missing data. In wearable and remote monitoring studies, data loss is not random — it correlates with the patients who are least adherent, least well, or have the most difficult skin interface. Complete-case analysis in that setting is biased in a predictable direction.

Ethics committee and CDSCO timelines treated as parallel when they are sequential. EC approval typically precedes the CDSCO application. Planning them concurrently compresses nothing and creates a rework loop.

Device supply and logistics underestimated. Devices must be labelled for investigational use, tracked by serial number, calibrated, maintained, and returned. Consumables expire. In a monitoring study, the reconciliation burden is real and is usually assigned to no one.

The Framing That Helps

The purpose of a device clinical investigation is not to generate a positive result. It is to generate a defensible answer to a pre-specified question, in a population that resembles the intended users, with data that a reviewer who has never met you can verify.

Studies designed with that as the objective tend to enrol, tend to complete, and tend to survive review. Studies designed to produce a marketing claim tend to do none of those things.

RhythmRx supports clinical investigation design and execution for cardiac monitoring and diagnostic devices, including CDSCO submission pathways, ethics committee coordination, and multi-jurisdiction protocol alignment.

Dr. Ananya Chatterjee is Head of Clinical Affairs at RhythmRx, working on clinical investigation design, ethics committee coordination and CDSCO submission strategy for cardiac diagnostic and monitoring devices.

Sources

Getz K, Tufts CSDD — analysis of ~16,000 investigative sites across 151 global Phase II/III trials (2008–2010), reported in Applied Clinical Trials, 2012.
Tufts CSDD Impact Report — 86% increase in total protocol endpoints between 2001–2005 and 2011–2015.
Analysis of registered trials closed or terminated in 2011 — 19% failed accrual goals or terminated early; 481 trials, >48,000 participants. Reported in Applied Clinical Trials.
Medical Device Rules, 2017 (India), G.S.R. 78(E) — Rules 51, 52, 59; Seventh Schedule; Second Schedule (fees); Forms MD-22, MD-23, MD-24.
ISO 14155:2020 — Clinical investigation of medical devices for human subjects: Good clinical practice.
EU MDR 2017/745, Annex XIV and Annex XV; ICMR National Ethical Guidelines for Biomedical and Health Research Involving Human Participants, 2017.