The operator is part of the treatment
Learning curves, implanter technique and center volume all move the outcome. Where clustering by site and operator is not accounted for, confidence intervals can be narrower than the data justify.
Home / Medical Device Biostatistics CRO
Medical device biostatistics CROStatistical strategy, sample-size calculation, SAP development, statistical programming and final analysis for medical device clinical investigations, registries and post-market studies.
Seven sequential stages: study design, sample size, statistical analysis plan, data, statistical analysis, clinical interpretation and regulatory evidence.
Most of these decisions should be resolved before the data are analyzed. By database lock, many of the choices that determine what the study can conclude are already fixed.
The number follows from the effect you expect, the variability you assume and the margin you accept. Move one of them and the study changes size, cost and duration.
An endpoint can be measurable, reproducible and still fail to support the claim the device needs. The two properties are decided separately.
The design shapes what the study can conclude, and what it will not be able to conclude, whatever the results turn out to be.
Event rate, standard deviation, non-inferiority margin, attrition. One usually dominates, and it is worth knowing which before the budget is committed.
The primary method should normally be pre-specified, with sensitivity analyses that test it. A change made once the pattern is visible can still be defensible, but it has to be justified transparently and supported by those analyses.
A blinded review of aggregate data serves a different purpose from a comparative interim analysis. The comparative version needs pre-specification, a data firewall, and an approach to multiplicity where it could stop the study.
Reviewers read the rationale, not only the result. Assumptions that were never written down are the ones that generate questions later.
By the time a study reaches database lock, most of the choices that decide what it can conclude are already fixed in the protocol.
Eclevar biostatisticians work with the clinical, medical, regulatory and data-management teams while these decisions are still open. Where the design is already fixed, the same review identifies what can still be changed and what has to be managed in the analysis.
Medical device investigations introduce statistical structures that are especially important to address: operator effects and learning curves, multiple implants or lesions per participant, device iterations, procedural endpoints and imaging-derived outcomes.
Learning curves, implanter technique and center volume all move the outcome. Where clustering by site and operator is not accounted for, confidence intervals can be narrower than the data justify.
Multiple lesions, several implants, bilateral treatment, paired anatomy. The observations are not independent, and the analysis has to account for that rather than count devices as people.
Design changes, new sizes, software versions and configuration options arrive mid-program. Whether data can be pooled across versions is a statistical decision a reviewer will ask you to justify.
Device and procedural success definitions, adjudicated events and core-laboratory measurements each carry their own variability, and the weighting between components of a composite changes what is really being tested.
Small eligible populations, low event rates, arms with no events at all. Asymptotic methods stop behaving, and more participants are often not the most efficient way to buy power.
Objective performance criteria, performance goals, historical controls, registry comparisons. Each shifts the burden onto the justification of comparability rather than onto randomization.
Repeated measurements over long follow-up, survivorship, revision, device deficiencies and real-world post-market datasets add further structure a general clinical-trial template does not anticipate.
Each can be contracted on its own or as part of a full biostatistics scope. Sponsors regularly engage Eclevar for one and extend later.
Objectives turned into testable hypotheses, endpoint strategy and hierarchy, superiority or non-inferiority framing, randomization and stratification, the analysis populations, and the sensitivity scenarios that show how robust the design is.
Discuss your study designAssumptions on effect, variability and event rate stated explicitly, margins justified rather than inherited, attrition built in, and a scenario table showing what happens if the key assumption is wrong. With the written rationale a reviewer will ask for.
Request a sample-size assessmentSAP authoring or independent review: analysis populations, primary and secondary analyses, multiplicity and the testing hierarchy, covariates and subgroups, missing data, protocol deviations and any interim look, pre-specified and version controlled before lock.
Request a SAP reviewRandomization schedules, block and stratified designs, allocation concealment, and the separation of blinded and unblinded roles, documented so the integrity of the allocation can be demonstrated rather than asserted.
Dataset specifications and analysis datasets, tables, listings and figures, independent quality-control programming, and traceability from the raw record to the number in the report. Reproducible from the specification, not from one person's session.
Pre-specified interim analyses, and blinded review of aggregate data such as the observed event rate, which serves a different purpose from a comparative look. A comparative interim analysis needs the firewall between the unblinded statistician and the study team defined in advance, and an approach to multiplicity where it could stop the study. Where an adaptive or Bayesian element is being considered, Eclevar assesses whether the added complexity is justified by the question; where a data monitoring committee is used, the statistical roles and the scope of Eclevar's involvement are agreed explicitly.
From database lock through the analysis and the tables, listings and figures, to the statistical sections of the Clinical Investigation Report or Clinical Study Report. Because biostatistics and medical writing can operate within the same delivery team, statistical outputs can move directly into clinical interpretation and the report with fewer handoffs.
Statistical input to submissions and to questions raised during review: the supplementary analyses that answer a specific query, clarification of methodology already applied, and the rationale behind a design decision taken years earlier.
Discuss regulatory statistical supportBiometrics is the combined discipline of biostatistics, statistical programming and clinical data management. Sponsors outsource it as one scope so that the analysis, the datasets and the people who defend them stay under a single accountable lead.
A sample size calculation for a medical device trial follows from four inputs: the effect you expect, the variability you assume, the margin you accept and the attrition you plan for. We state each one explicitly, show what happens to the number when the dominant assumption is wrong, and write the rationale a reviewer will ask for rather than a single figure.
A responder analysis converts a continuous outcome into the proportion of participants who reach a defined threshold. It is often the most clinically readable endpoint a device can carry, and the most contested: the threshold has to be justified before the data are seen, and the handling of participants who cannot be classified has to be pre-specified alongside it.
For implants, device survivorship is rarely a simple time-to-event question. Death and removal for an unrelated reason are competing risks, and a Kaplan-Meier estimate that treats them as censoring will overstate the revision rate. We use cumulative incidence functions and report the competing events alongside the primary estimate, which is what registry comparisons require.
Real world data from registries, claims databases and post-market follow-up carry selection and confounding that a randomized design controls by construction. The analysis has to earn what the design cannot assume: a pre-specified target estimand, an explicit approach to confounding, and sensitivity analyses that show how far the conclusion can be pushed before it breaks.
Each of these can be contracted on its own or as a full biometrics scope. Sponsors regularly engage Eclevar for a sample size assessment or an independent Statistical Analysis Plan review, then extend to programming and final analysis once the design is fixed.
See EDC and clinical data management for the data side of the same scope.
Statistical programming after database lock is the last twelfth of the work. The strategy that decides whether the analysis can answer the clinical question is set at stage one.
Before the protocol is fixed
Before and during enrollment
After lock
A statistician brought in at stage nine can describe the study you built. A statistician involved at stage one helps you build a study that can answer the question.
Most sponsors do not contact a biostatistics CRO at the start of a program. They contact one when something in an active study stops behaving the way the protocol assumed.
Does the study still have adequate power?
Slow recruitment is an operational problem with statistical consequences. The impact depends on whether the shortfall affects participant numbers, event accumulation, follow-up duration or the assumptions underlying power, and the options differ accordingly: extended follow-up, a revised endpoint definition, or a reasoned decision to continue as planned.
Review the assumptionsIs the sample size still appropriate?
A blinded review of the aggregate event rate can often be carried out without compromising the study, provided it is pre-specified or properly documented and the review stays blinded to treatment allocation.
How should they be handled without undermining the analysis?
The mechanism matters as much as the proportion. We look at whether missingness is related to outcome, define the primary handling method and the sensitivity analyses that test it, and write both down before the data are unblinded.
Which analyses remain scientifically defensible?
There is usually a narrow set of changes that preserve the integrity of the study and a wider set that would compromise it. The value of an external statistician here is in drawing that line explicitly and in writing.
Are the populations, endpoints and methods appropriate before database lock?
An independent read of a SAP written elsewhere, against the protocol and the intended claim, before the point at which changes stop being pre-specification and start being post hoc.
Request a SAP reviewHow should the sponsor respond, and with which supplementary analyses?
Some questions are answered by explaining what was already done, some by a supplementary analysis, and some by stating a limitation clearly. Choosing wrongly between the three is what turns one question into three.
Discuss regulatory statistical supportCan the proposed comparison be justified scientifically and statistically?
The question is comparability: how the historical population was defined and measured, which covariates are available, what residual bias remains, and whether the design can carry the weight of the intended claim.
Can the analysis be picked up mid-study without losing traceability?
It starts with an assessment: SAP status, dataset specifications, what has been programmed, what has been validated, and what would have to be rebuilt. Worth having before the transfer is agreed.
Request an independent reviewThe statistical problem is set by the device category and the endpoint framework it uses.
Composite safety endpoints, device and procedural success, reintervention, time-to-event analysis with competing risks, and imaging or hemodynamic measurements read by a core laboratory. Endpoint definitions come from the framework for the specific procedure: VARC-3 for transcatheter aortic valve replacement, MVARC and TVARC for mitral and tricuspid.
Patient-reported outcomes, implant survivorship and time to revision, radiographic outcomes, and non-inferiority against an established implant with the margin anchored on a minimal clinically important difference. Repeated measures over multi-year follow-up mean attrition is planned for from the start.
Complete closure as a binary endpoint against continuous area reduction, time to healing with censoring, recurrence, and longitudinal measurement of a wound that changes shape between visits. Etiology stratifies the analysis; country variation in standard of care is a covariate.
Diagnostic performance rather than treatment effect: sensitivity and specificity against a performance goal, agreement against a reference standard, algorithm discrimination and calibration, and datasets where one participant contributes hundreds of clustered observations.
Learning curves that make the first cases systematically different from the last, surgeon and center effects modeled rather than averaged away, and procedural endpoints with skewed distributions.
Small eligible populations, long follow-up, responder definitions built on scale thresholds, and within-participant or crossover designs used where a parallel-group study of the required size is not feasible.
The same discipline in all three settings. What changes is who reads the rationale and what they expect it to justify.
Under EU MDR 2017/745 and ISO 14155:2026, the statistical section of the clinical investigation plan carries the justification for the design, the sample size and the analysis.
Early feasibility, first-in-human and pivotal development for US sponsors, where the statistical rationale is normally discussed with the agency before the study runs.
Post-market datasets behave differently: no randomization, variable follow-up, and a structure designed for surveillance rather than hypothesis testing.
The analysis inherits every decision taken in the data chain before it. When the statistician sees the dataset for the first time at database lock, the inheritance is a surprise.
Nine sequential stages from case report form design, electronic data capture, data cleaning and query resolution, through database freeze and lock, statistical programming, tables listings and figures, clinical interpretation, and the clinical report.
Statistical problems are cheaper to solve before database lock than after it.
The biostatistician reviews the case report form against the analysis before the database is built, so every variable the statistical analysis plan needs is collected in the form it needs, with the derivations defined rather than reconstructed later. The same team runs clinical data management and biostatistics, and works alongside clinical monitoring so central review and site visits look at the same signals. Where a sponsor already has a data management provider, Eclevar works to their specifications instead.
Every item below can be contracted individually.
Four ways sponsors normally begin. The first is often a single meeting; the others are scoped as proposals.
Study design, endpoint strategy, sample size
For sponsors preparing a protocol, or reconsidering one before it is submitted. The output is a written recommendation on design, endpoints and participant numbers, with the assumptions exposed.
Discuss study designProtocol, SAP, programming, final analysis
For sponsors who need an outsourced statistics partner for one study, working alongside their own team or another CRO. A full biostatistics workstream from protocol to statistical report.
Request a proposalClinical operations, data management, biostatistics, medical writing
For sponsors outsourcing several functions, where the value is in the connections between them. One accountable team from protocol to final report.
Discuss full-service deliveryStatistical review, gap assessment, remediation strategy
For sponsors with an existing study, protocol, SAP or dataset that needs an independent read, or a transfer of biostatistics from another provider. Starts with an assessment, not a commitment.
Request an independent reviewOne program where Eclevar's statistical contribution is documented, and two where the published design shows the kind of statistical problem such programs pose. Each panel states which it is.
A ceramic hip resurfacing system needs early safety and performance evidence and, in a second connected part, long-term survivorship and radiographic assessment across four planned evidence domains.
Clinical investigation strategy, study synopsis and endpoint architecture, the long-term follow-up plan, integration of patient-reported outcome measures, digital data capture and the statistical methodology behind the two connected parts.
Endpoint architecture across two parts fixes what can be pooled and what stays separate, how attrition over multi-year follow-up is handled, and whether survivorship can be interpreted alongside the early safety data.
Status and contribution are as recorded in the Eclevar program register and published on the client programs page. The investigation is in progress. No result or regulatory outcome is claimed, and no enrollment, follow-up or delivery is stated as complete.
Comparative clinical, quality-of-life and health-economic evidence across three valve platforms within one prospective randomized UK study rather than three, covering four evidence domains.
Clinical program delivery and evidence coordination, including UK site delivery, data management and reporting. The program is ongoing and currently enrolling.
Three parallel arms across four evidence domains raise the questions of which comparisons carry the conclusion, how the testing hierarchy is ordered across those domains, and how operator and center effects are handled in an interventional procedure.
Program context, not biostatistics proof. Manufacturer names and trademarks identify the platforms evaluated and do not imply endorsement of, or affiliation with, Eclevar.
Two clinically distinct wound populations, diabetic foot ulcer and venous leg ulcer, across 160 participants and 14 clinical sites, with endpoints intended to serve both Notified Body scrutiny and reimbursement discussions.
Full program delivery from protocol design to final study report, with European site coordination and structured data capture. The program is ongoing.
Two populations under one protocol raise the question of stratification against separate analysis, of whether healing is analyzed as a binary endpoint or as time to event, and of how country variation in standard of care enters the model.
Program context, not biostatistics proof.
Panels marked program context illustrate the statistical problem a published design poses. Eclevar's biostatistics scope on those programs is not published, and nothing above states that Eclevar developed the SAP, calculated the sample size, produced the randomization, programmed the analysis or performed the final analysis for them. All program status, size and role wording is as published on the Eclevar client programs page. No result or regulatory outcome is claimed for any program shown.
Statistics should not operate as an isolated downstream function. The endpoint a statistician analyzes is the endpoint a clinician defined, collected in the structure a data manager built, and written up by someone who has to defend it in a report.
Chief Data Officer, Biometry Lead
Owns the frameworks used to collect, standardize and analyze clinical data across Eclevar studies, from the case report form and the dataset specification through the statistical analysis plan to the locked database and the final analysis.
Chief Operating Officer and Head of Cardiovascular
Consultant cardiac surgeon and former TÜV SÜD Team Leader and Senior Clinical Reviewer. Frames cardiovascular and structural-heart endpoints so that what the statistician tests is what the reviewer will read.
Chief Medical Officer, Orthopedics and Spine
NHS orthopedic surgeon and former TÜV SÜD Senior Reviewer for Class III implantable devices. Defines the functional, radiographic and survivorship endpoints behind orthopedic and spine analyses.
Head of Medical Writing
Leads Clinical Evaluation Reports, PMCF documentation and deficiency responses. Carries the statistical result into the clinical report and the clinical evaluation, where it has to survive being read out of context.
Former positions are stated for biographical context only. Eclevar MedTech is independent and is not affiliated with or endorsed by TÜV SÜD, a registered trademark of its respective owner. Notified Body review experience is professional background rather than a medical qualification, and neither it nor a clinical qualification implies that a clinician is a biostatistician.
Eclevar MedTech received the Platinum Award in the xShare and EUCROF Open Call for Clinical Research 2026, co-funded by the European Union, for the clinical-research use case submitted to that call. It is not a ranking of contract research organizations.
Send Eclevar your study synopsis, protocol, sample-size assumptions or existing SAP. We review the statistical strategy before study execution or before database lock, and identify the issues that would otherwise affect interpretation, power or the defensibility of the analysis.
Documents are reviewed confidentially, and a non-disclosure agreement can be put in place before any technical or clinical information is sent.
Cost follows scope rather than a rate card. What moves it most: the design, the number of endpoints and analysis populations, the sample size and site count, the volume of programming, whether there is an interim analysis, and whether biostatistics is standalone or part of an integrated scope. A synopsis and an endpoint list are usually enough for an itemized proposal.
Yes, regularly, alongside a sponsor's own clinical operations team or another CRO. The interfaces that need defining are the data transfer specification, the timing of the analysis relative to database lock, and who signs the SAP.
Yes, and it begins with an assessment rather than a transfer: the current SAP, the dataset specifications, what has been programmed and validated, and what would have to be rebuilt to make the outputs traceable. That tells the sponsor what the transfer costs before it is agreed.
Yes. The review reads the plan against the protocol and the intended claim, and returns findings ranked by their effect on the conclusion. It is most useful before database lock, while changes are still pre-specification rather than post hoc.
Yes, and it is one of the most common first engagements. The deliverable is a calculation with the assumptions written out, the margin justified where one applies, and a scenario table showing how the required number changes if the main assumption is wrong.
Ideally before the protocol is finalized. The endpoint definition, comparator, follow-up schedule and sample size are statistical decisions written in a clinical document, and far cheaper to change while the protocol is a draft. The next best moments are before the case report form is built, and before database lock.
Yes. European clinical investigations under EU MDR 2017/745 and ISO 14155:2026, and support for US sponsors on early feasibility and pivotal development, including the statistical rationale prepared for discussion with the agency. What differs is the expected form of the justification.
A synopsis or protocol, the design and comparator, the endpoints and their definitions, the expected sample size and site count, any existing SAP, the data structure and EDC platform if chosen, the stage of the study, and the deliverables you want quoted. If some of that does not exist yet, the missing pieces are usually the conversation itself.
Share your synopsis, protocol, endpoint strategy or existing statistical analysis plan. We assess the proposed approach and define the biostatistics support required, from study design through to final analysis.