Randomizing patients, and changing the care itself — a nationwide trial that hands colorectal surgery risk assessment to AI — a clinical trial (ClinicalTrials.gov)

1Not assisting a reading, but changing the care received

Most trials of medical AI measure how well an image is read or how often a diagnosis lands. This one touches what comes after. Patients scheduled for curative colorectal cancer surgery are randomized to have their risk assessed by a surgeon in the standard way or by a surgeon working with AI assistance, and the level of perioperative care then follows from the assessed risk.

The AI output does not stay in a report; it reaches the care the patient actually receives.

2What changes on the other side of the allocation

1. 1Risk assessmentA surgeon assesses risk by standard methods, or with artificial-intelligence assistance
2. 2StratificationPatients are separated by the assessed level of risk
3. 3Allocation of careThe intensity of perioperative management follows that level
4. 4Outcomes and resourcesCheck whether allocation grew more efficient without worsening complications or death

The primary hypothesis, as recorded, is more efficient and economic care without a deterioration in outcomes. Not improvement, but a change in allocation that does no harm, and that phrasing captures the character of the trial. Intensive perioperative care is finite; hand it to everyone and it runs out. What is being evaluated is a tool for deciding who to concentrate it on.

3How much this rung carries

Roughly a fifth of the clinical trials held here mention randomization. Most of those compare readings, or the way advice is presented; a design in which the allocation changes the treatment a patient receives sits at the heavier end.

That the record states all patients receive standard treatment under national guidelines, with the assessment modality as the only difference between groups, is the other side of that same weight.

4One caution in reading the record

The body of the registration states the number of participating hospitals as seven in one place and eight in another. Registrations are revised over time, and discrepancies between passages can survive. When a figure is quoted, it is safer to confirm which passage it came from.

The record also notes that the preceding pilot suggested reductions in complications, hospital stays and readmissions, but that is the pilot, not a result of this trial.