Isabel is designed to help clinicians expand a differential; iatroX supports clinical-reference questions and structured learning. In either setting, a list of possible diagnoses is the beginning of reasoning, not its conclusion. The useful next steps are to check the information supplied, identify competing explanations and decide what evidence would distinguish them.
This comparison is published by iatroX and includes iatroX. Public product descriptions were reviewed on 20 September 2026. The worked examples are original teaching exercises, not observed output from either product or evidence of comparative diagnostic accuracy.
Compare the right Isabel product
Isabel Healthcare currently distinguishes its clinician-facing DDx Companion, patient-facing self-triage product and Clinical Educator. The existence of one function should not be used to describe every product under the brand. This article concerns clinical differential support and relevant learning, not an evaluation of patient self-triage.
The Clinical Educator also matters to a fair comparison. Isabel should not be characterised as offering a diagnosis list but no educational activity. Its published product family includes case-based reasoning education. The practical question is which exact product and task a reader wants to evaluate.
For iatroX, distinguish Ask-iatroX and Brainstorm from examination questions, Tutor and simulations. A general case discussion, an answered clinical question and a formal examination-style encounter have different purposes. Their presence within one platform does not make their outputs interchangeable.
Preserve the facts before expanding the possibilities
Consider an original fictional case involving fever, rash and recent travel, the type of presentation used in the earlier comparison. The first useful step is not to request the longest possible differential. It is to establish what those three phrases actually mean in the case.
Which observations are documented? What is known about the timeline? Does the travel history describe a relevant exposure or merely a destination? What has been examined, and what is simply unrecorded? The exercise does not supply enough information to diagnose or prescribe. It is designed to show why compressed case labels can conceal material uncertainty.
A learner should write a short problem representation that preserves the distinction between established facts and questions still to ask. A differential tool cannot recover omitted facts reliably by inferring them from the eventual diagnosis. If the input says less than the clinician knows, the output may address the wrong problem.
A longer list can make the decision harder
Adding candidate diagnoses can be useful when it challenges a premature conclusion. It becomes less useful when the learner treats every possibility as equally relevant, or responds by investigating everything without a reasoned sequence.
Ask what each candidate contributes. Does it explain the time course? Does it fit the important positive findings? Does it require an assumption that has not been established? What evidence would make it less plausible? These are original teaching questions, not a proprietary assessment algorithm.
The point is not to dismiss rare diagnoses. It is to distinguish a useful alternative from a merely possible label. The strongest contribution of a differential aid may be one neglected line of enquiry, not a large number of additions.
Build an evidence map rather than copy the list
| Candidate explanation | Supporting facts in the case | Facts that do not fit | Information still needed |
|---|---|---|---|
| Initial working explanation | Only findings explicitly supplied | Contradictions or unusual features | A question or examination finding that would change confidence |
| Important alternative | The particular feature making it relevant | Features requiring another explanation | The most useful discriminator |
| Less likely possibility | Its limited connection to the case | Why it should not dominate the plan | What would justify reconsidering it |
This is an original worksheet for discussion. It deliberately avoids naming a diagnosis for the fictional fever-and-rash case, because the omitted details are the learning problem. A supervisor can supply a clinically reviewed case and use the same structure to assess the learner's reasoning.
After completing the table, remove any statement that originated as an assumption but now reads like a fact. That simple editorial check can expose a surprisingly important change in the case representation.
What changes when the diagnosis is already known?
The earlier comparison also used an established asthma diagnosis to distinguish differential generation from management guidance. That remains useful, provided "known diagnosis" does not mean "no further uncertainty".
A clinician may need a recommendation, a rationale, clarification of a changed presentation or a check that another process has not been overlooked. The appropriate resource depends on that unresolved question. A guideline-oriented reference may suit a management enquiry, while reconsidering the differential may be sensible when the case no longer fits the original explanation.
Do not claim that one platform becomes categorically useless in either setting. Instead, ask what output is required now and whether the relevant product provides evidence supporting it.
Evaluate a differential tool without manufacturing a winner
For a documented comparison, use original or appropriately licensed cases with the same information available to both systems. Record the exact product, access tier, date and any version information. Agree in advance what the evaluation concerns: relevant alternatives, misleading suggestions, missing questions or the usefulness of explanations.
A case review should distinguish presence of the eventual diagnosis somewhere in a list from useful support for the clinician's next decision. It should also record whether the output encouraged unsupported assumptions. No test results, sensitivity figures or response-time advantage are claimed in this article.
Where users add follow-up information, preserve the chronology. A system that receives a decisive new finding is not being evaluated under the same conditions as one that only received the initial description.
The learning opportunity after the differential
Under iatroX's September 2026 specification, Brainstorm and Ask-iatroX can support case organisation and clinical clarification. A relevant question and Socratic Tutor discussion can then investigate why the learner favoured one explanation. These are educational uses, not confirmation of a real patient's diagnosis.
A useful follow-up changes one material detail in the fictional case. Ask the learner whether the initial ordering still makes sense and require an explanation. This is more informative than asking them to repeat the same list from memory. A later simulation may rehearse information gathering or explanation where a suitable track exists, while bedside and procedural competence require separate practice.
A scenario-based conclusion
Choose a dedicated differential product when expanding and examining candidate explanations is the unresolved task. Use an appropriate clinical reference when the need is a bounded evidence question. Consider structured learning when the recurring difficulty is how the clinician reasons through a case. Complementary use is entirely reasonable, but every output must remain connected to the facts and the sources behind it.
Frequently asked questions
Does Isabel only generate diagnosis lists?
Its current product family also includes self-triage and Clinical Educator offerings. Evaluate the exact product rather than assigning one description to the entire brand.
Is a diagnosis appearing in an AI differential evidence that it is correct?
No. It is a candidate explanation whose relevance must be assessed against the actual clinical information and appropriate evidence.
What should a learner do after generating a differential?
Identify supporting and conflicting findings, state the missing information and explain which discriminator matters next. The aim is a better reasoning process, not a longer copied list.