Device start-ups
You are planning bench, usability or pilot studies and need the statistics right.
Medical Devices & Digital Health
Study design and statistics for teams developing medical devices, digital cognitive tests and other digital-health tools: sample size, agreement, reliability and algorithm validation.
Who this is for
Teams we help in this field.
Health-technology companies
You build a tablet, web or phone-based cognitive test and need evidence of how it performs.
Clinical researchers
You are comparing a new device with a reference method.
Audiences we welcome, not a list of clients.
Questions we help answer
What teams in this field ask us.
- How many units or participants do we need for a verification or comparison study?
- Does our device agree with the reference method, and within what limits?
- How reliable is our test when people take it twice?
- Does the test work equally well across age, sex, language and education?
- Is the algorithm in our device accurate on data it has not seen?
- How do we set and test acceptance criteria?
What you receive
- The statistics section of your test plan: methods, acceptance criteria and analysis.
- A sample-size justification: for units, runs or participants.
- A statistical report: ICC, Bland–Altman limits and accuracy, each with confidence intervals.
- An algorithm validation report: performance on data the algorithm has not seen.
- Reproducible code: for every result.
Methods we use
The methods specialists ask about.
- Tolerance intervals
- Equivalence testing (TOST)
- Bland–Altman and Deming regression
- ICC (test–retest)
- Practice-effect correction
- ROC and calibration
- Subgroup fairness checks
- Cross-validation
Evidence
Published work behind this area.
Predicting Brain Age Using Machine Learning Algorithms: A Comprehensive Evaluation
A comprehensive evaluation of machine-learning algorithms on data they had not seen.
Read the paper (external site)The accuracy of T1-weighted voxel-wise and region-wise metrics for brain age estimation
Compared the accuracy of voxel-wise and region-wise image measures.
Read the paper (external site)Patch-wise brain age longitudinal reliability
A longitudinal reliability study: the same logic applies to a digital test taken twice.
Read the paper (external site)The founder holds a PhD in biomedical engineering. We have not published a device or digital-test trial; these papers show the methods we use.
Boundaries
Statistics and study design only. We do not design, build, manufacture or certify devices or tests, run accredited safety testing (for example IEC 60601), make diagnostic claims or prepare regulatory submissions. Device classification and Health Canada licensing remain with the manufacturer.
Packages that fit
Start at the right scale.
The same packages and prices on every page. Each starts with a written scope.
Research Diagnostic
Fixed priceCAD 495
- 60-minute video call
- Two-page risks and next-steps memo
- Optional 30-minute debrief
Paid in full at booking.
Memo within 5 business days of the call.
Ask about thisAnalysis Plan Sprint
Fixed priceCAD 1,250
- Analysis and validation plan
- Scoped sample-size or power calculation
- Methods text or reviewer response
50% to start; 50% on delivery.
Delivery date agreed in the written scope.
Ask about thisData-to-Results Sprint
FromCAD 3,000
- Agreed analysis and quality-control log
- Figures, tables and methods text
- R or Python code and recorded environment
50% to start; 50% on delivery.
Final price and delivery date agreed after scoping.
Ask about thisRelated
Go deeper.
Questions
About working in medical devices & digital health.
Can you test our prototype in a lab?
Not at present. We plan the tests and analyze the results that your team or test lab produces.
Can you say our test is "validated"?
No. We report what the study shows, with its limits. Whether the evidence supports a clinical or regulatory claim is for you and your regulator to decide.
Can you help with design verification and validation documents?
We provide the statistical sections: sample size, acceptance criteria and analysis. Your quality system owns the documents.
Do you work inside an ISO 13485 quality system?
We can follow your procedures, but Cedar is not ISO 13485 certified.
Why is correlation not enough to show agreement?
Two tests can correlate strongly and still disagree by a large, consistent amount. Agreement methods such as Bland–Altman show how big the differences are.
Who owns the analysis and code?
You do, once the invoice is paid, unless the written scope says otherwise.
Discuss a medical devices project.
Start with a short, non-confidential description. We reply with a suggested next step.