 |
|
|
Resources Services |
|
|
|
|
Resources Services |
 |
Using AI to Support Clinical Reasoning and Diagnosis: Opportunities, Limits, and Best Practices |
Artificial intelligence tools are increasingly being used to support clinical learning, diagnostic reasoning, and point-of-care information seeking. These tools may help users organize clinical findings, generate possible differential diagnoses, suggest related conditions, or identify questions to ask during patient care. However, AI tools do not replace clinical judgment, patient examination, medical training, or consultation with appropriate clinical experts.
Searching for information and making a diagnosis are related but different tasks. A literature search seeks published evidence. A diagnosis requires collecting patient information, interpreting findings, applying clinical reasoning, and considering the patient's full context.
Clinicians do not simply match one symptom to one disease. They synthesize the patient's history, physical examination, risk factors, test results, and clinical context to develop and refine a differential diagnosis.
Patients often describe symptoms in everyday language rather than medical terminology. A key part of clinical reasoning is translating the patient's story into accurate clinical concepts.
Patient statement: "I get all swollen up."
The clinician asks follow-up questions: Where is the swelling? When does it happen? What were you doing before it started? Does it improve with rest or elevation?
If the patient explains that the swelling is in the feet, occurs at the end of the day, and follows prolonged standing, the clinician may describe the finding as dependent pedal edema.
Using accurate medical terminology may help an AI tool provide more relevant information about possible causes, such as chronic venous insufficiency, prolonged standing or sitting, medication effects, heart failure, kidney disease, or liver disease.
AI tools may help clinicians and students think through a differential diagnosis, especially when findings are unusual or when a rare condition should be considered. The quality of the response depends heavily on the quality of the clinical description entered into the tool.
A patient reports sudden facial swelling that has occurred several times a year since childhood. The patient also reports that a brother and an aunt have had similar episodes.
The clinician identifies several important medical concepts:
Entering a phrase such as facial edema that is episodic and familial into a medical AI or clinical information tool may suggest conditions to consider, including hereditary angioedema. The tool may also provide a structured differential diagnosis and possible workup.
This does not mean the AI tool has made the diagnosis. It means the tool may help identify a possible diagnosis for the clinician to evaluate.
Different AI tools are designed for different purposes. Some are general purpose tools, while others are clinical decision support or evidence-based medical information tools. The source of the information matters.
| Tool or Resource | Access | Best Used For | Important Cautions |
|---|---|---|---|
| ChatGPT | Free and paid versions available | Explaining concepts, organizing findings, generating study questions, translating complex information into plain language, and practicing clinical reasoning. | General purpose AI. Do not use as the final authority for diagnosis or treatment. Do not enter identifiable patient information. |
| OpenEvidence | Free access for eligible medical users | Asking clinical questions in natural language and retrieving medical context answers with supporting sources. | Works best with accurate medical terminology. Review cited sources and confirm recommendations. |
| ClinicalKey AI | Mercer subscribed resource | Point-of-care clinical questions, evidence based summaries, textbook based medical information, and clinical decision support. | Does not replace clinician evaluation, diagnosis, or treatment planning. Review linked evidence. |
| DynaMedex / Dyna AI | Mercer subscribed resource | Evidence based clinical summaries, guideline based information, and concise answers to clinical questions. | Best for focused clinical questions. Confirm recommendations with the full topic and clinical context. |
| PubMed customized for MUSM full text | Free resource; Mercer links connect users to subscribed full text when available | Finding biomedical literature, case reports, reviews, clinical studies, and background evidence. | PubMed does not diagnose. Search results require interpretation and appraisal. |
| VisualDX | Mercer subscribed resource | Visual differential diagnosis, especially for dermatology, rashes, lesions, and morphology based findings. | Image based tools are aids only. Diagnosis depends on the full clinical picture. |
| AMBOSS | Mercer subscribed resource | Medical student learning, clinical knowledge review, differential diagnosis practice, and exam preparation. | Excellent for learning and reinforcement, but should not be used alone for patient care decisions. |
| AccessMedicine / Medical Textbooks | Mercer subscribed Textbooks | Authoritative background reading, disease overviews, pathophysiology, diagnosis, and treatment foundations. | Check dates and supplement with current guidelines or evidence when needed. |
| Web of Science / Embase / PsycINFO | Mercer subscribed library databases | Advanced literature searching, evidence discovery, citation tracking, and research on diagnostic methods or clinical reasoning. | These databases support evidence discovery, not diagnosis. Consult a librarian for comprehensive searches. |
Prompt 1:
A patient presents with bilateral foot swelling that worsens after standing all day and improves with leg elevation. What are common causes of dependent pedal edema, and what history questions would help narrow the differential diagnosis?
Prompt 2:
Create a differential diagnosis for episodic facial edema with a family history of similar symptoms. For each diagnosis, list key supporting findings, red flags, and initial tests to consider.
Prompt 3:
Explain the difference between pattern recognition and analytic reasoning in clinical diagnosis. Give an example of when each approach might be useful.
The following readings provide additional information on clinical reasoning, diagnostic decision making, artificial intelligence in healthcare, and the responsible use of AI in clinical practice.
Suggested Learning Path: New to the topic? Start with Bowen and Croskerry to learn how clinicians develop diagnoses, then read Schwartzstein and Omiye to explore how artificial intelligence may support, but not replace, clinical reasoning.
|
```
