Clinical Decision Support for Vestibular Diagnosis: Large-Scale Machine Learning with Lived Experience Coaching (2025, Callejas Pastor et al)
This study looked at whether machine learning could help support the diagnosis of common vestibular disorders. The researchers developed a CatBoost machine learning model to classify six vestibular disorders. The model used 30 clinical features that were selected by computer-based methods and 20 that were added based on vestibular specialist expertise. On unseen test data, the model reached 88.4% accuracy. For Dizzy Care Network, this research supports the value of structured dizziness intake. A well-designed questionnaire may help collect key information before the visit and support better clinical decision-making. The study also reinforces an important point: machine learning should help organize information and support triage, but it should not replace the clinician’s examination, testing, diagnosis, or treatment plan.
Permanent Link: https://doi.org/10.1038/s41746-025-01880-z
Why is this research important for dizziness care?
This research is important because dizziness can be difficult to diagnose. Many vestibular conditions have overlapping symptoms, so clinicians often need detailed information about timing, triggers, hearing symptoms, headache history, motion sensitivity, balance, and near-fainting symptoms. The study shows that machine learning may be most useful when paired with clinical expertise, since vestibular specialists added features they felt were important for real-world diagnosis. Additionally, the model performed best as a decision-support tool, not a stand-alone diagnostic tool. This aligns with the goals of Dizzy Care Network: helping care-seekers reach a more relevant clinician sooner, improving the intake process, and supporting clinicians with organized information about their patients dizziness.
Key takeaways for Care Seekers
- Dizziness can come from many different causes and symptoms often overlap.
- A computer tool or questionnaire cannot diagnose you by itself.
- Machine learning may help organize your symptoms and guide you toward a more relevant type of clinician.
- You still need a trained clinician to review your history, examine you, and decide on the right diagnosis and treatment plan.
Key takeaways for Clinicians
- This study supports the use of machine learning as a clinical decision-support tool for vestibular diagnosis.
- Clinical implementation should be cautious because this was a retrospective, single-center study from Korea, included only six vestibular disorder categories, and still needs broader prospective validation.
- The six disorders were: BPPV, vestibular migraine, Menière’s disease, vestibulopathy, PPPD, and hemodynamic orthostatic dizziness.
- Machine learning may help streamline intake, support triage, and organize complex symptom data, but it should not replace clinician expertise, examination, testing, or individualized treatment planning.