Clinically grounded
Start from a real clinical problem and real evidence, not from a model looking for a use. Medicine first, technology second.
Responsible AI in medicine
AI in medicine, built to deserve trust
A physician-scientist's perspective on building artificial intelligence for healthcare that is safe, validated, transparent, and genuinely useful to clinicians and patients.
- EASY-1Registered RCT
- Medtech4HealthInnovation Award
- KTHMedical Device Regulations
- FDAClinical Investigator
Principles
Four commitments for clinical AI.
Validated in the real world
Prove value where care happens, including in randomized trials, not only on benchmarks. EASY-1 is the standard, not the exception.
Transparent and accountable
Clinicians should understand why a tool recommends what it does, and who is responsible when it is used.
Respectful of the workflow
A tool that does not fit the clinician's day will not be used, however clever it is. Fit the work.
Perspective
Why this matters.
Artificial intelligence will change medicine, but only the parts that earn clinical trust will last. The hard work is not a clever model. It is the evidence, the transparency, the regulation, and the design that let a clinician rely on a tool with a patient in front of them.
Dr. Damon Tojjar has spent his career on that bridge: diabetes genetics, global clinical development at Novo Nordisk, medical device regulation, and the clinical decision support of EASY Diabetes. The throughline is simple. Build technology that serves the patient, and prove it.
Research integrity
Getting the evidence right, and keeping it right.
Integrity and reproducibility
A result is only useful if it is real and repeatable. Careful methods, honest reporting, and reproducible analysis are the foundation of trustworthy medicine.
How science self-corrects
Peer review, replication, correction, and retraction are how science fixes its own mistakes. Understanding these mechanisms is part of reading the evidence well.
Evidence-based medicine
Judging what a study can and cannot support: the design, the effect size, the biases, and whether it applies to a given patient. The discipline behind every sound recommendation.
Transparency and conflicts
Disclosed interests, shared data, and clear authorship let readers weigh a claim on its merits. Openness is a feature of good science, not an afterthought.
These are the standards Dr. Tojjar writes to, and writes about, on the evidence blog: clear, sourced, and fair.
Frequently asked
About responsible clinical AI.
When should a clinical AI tool be trusted?
When it starts from a real clinical problem and real evidence, is evaluated where care happens (ideally in a randomized trial, as EASY Diabetes was in EASY-1), is transparent about how it reaches a recommendation, and fits the clinician's workflow.
What does validation in the real world mean here?
Proving value in the setting where the tool is used, not only on a held-out benchmark. EASY-1 compared the EASY Diabetes decision-support system against standard of care in a randomized controlled trial.
How does regulation fit in?
Medical-device and data regulation (EU MDR, IVDR, FDA pathways, and Good Clinical Practice) is treated as a design input from the start, not a box to check at the end, so a tool is safe and accountable by construction.
Contact
Get in touch.
For advisory work, collaboration, or speaking on responsible medical AI, reach out directly.