Tekzavro / Canada
Ask better questions about screening performance
A headline accuracy claim is not enough to judge a clinical tool. Evaluation must relate to its intended purpose, users, population and setting.
Understand what was measured
Identify the reference standard, study population and types of errors reported. Results from one dataset should not be assumed to hold in a different clinical setting.
Consider people and workflow
Ask how qualified users interpret the output, what information they also see and how they handle uncertainty. Review the complete workflow rather than an isolated model score.
Use current primary guidance
Consult Health Canada guidance on machine learning-enabled medical devices when assessing a relevant product. This article does not endorse a device, verify its authorisation or offer a diagnostic threshold.
All guides ยท Contact Tekzavro
Primary reference: Health Canada: good machine learning practice for medical devices.