Microsoft researchers’ Diagnostic Orchestrator scored 83.2% on one set of complex diagnostic cases and 80.3% on a US emergency-department dataset, according to Distilled Post. The results suggest AI could help weigh which tests to order next, but they do not show that it outperforms specialists or works in real clinical practice.
Microsoft AI Watch analysis
What happened
The system simulates a panel of physicians revising possible diagnoses as new information arrives, while weighing the cost of further tests. Distilled Post says it was tested on 304 complex cases from a medical journal and 325 emergency admissions from a US database. Generalist physicians working under the same text-only conditions scored 36.1% and 46.8%, respectively, at higher estimated cost.
Those comparisons have important limits: the doctors were generalists tackling cases usually handled by specialist teams, without colleagues, reference materials or a patient in front of them. The study does not establish superiority over specialists or performance in real-world care. The cost comparison is estimated, not based on NHS tariffs.
Why it matters
The idea is relevant to the NHS because diagnostic capacity is finite. A tool that helps clinicians decide which investigation is worth ordering could, in principle, reduce low-yield tests and ease pressure on scanners and laboratories. But the evidence described here comes from curated cases and US data, not NHS patients or pathways.
Putting such a system into practice would also raise questions about fragmented records, integration with existing systems and who carries responsibility if software steers a clinician towards ordering fewer tests. A strong score is a reason to investigate further, not a clinical instruction manual.
Our read
This is an interesting test of AI-assisted diagnostic decisions, not proof that the NHS is ready to hand it the waiting-list clipboard. The useful next step is prospective evaluation in real clinical settings, measuring missed diagnoses, unnecessary investigations and time to treatment, alongside costs and patient outcomes.
What to watch
- Whether the researchers publish more detail on the study and its cost assumptions.
- Whether the system is tested prospectively with UK patients and clinical workflows.
- How clinicians and regulators address accountability when software influences test decisions.
Discussion spark: Should the NHS trial AI that recommends which tests to order only after prospective UK studies show it is safe, or is a carefully monitored pilot itself the right way to build that evidence?
Sources and evidence
- Diagnostic AI and the Economics of Ordering Tests in the NHS – Distilled Post (26 October 2026, 07:39 UTC)
not affiliated with or endorsed by Microsoft