A machine learning approach reveals features related to clinicians’ diagnosis of clinically relevant knee osteoarthritis
Authors' conclusion
Random forest models presented good performance in explaining clinicians' diagnosis, which helped to reveal typical features of patients recognized as clinically relevant knee OA by clinicians from two different care settings.
Study context
- Population
- Knees from the CHECK cohort, 5–10 years follow-up clinical and radiographic data, evaluated for clinically relevant knee OA
- Participants
- 1,106 knees
- Exposure
- General practitioners (GPs) vs SPs; diagnosis before vs after viewing radiographic data
- Outcome
- Top 10 features related to clinicians' diagnosis of clinically relevant knee OA
Methodological detailsmachine learning, nested 5-fold cross-validation enhanced random forest models, paired clinicians (one GP and one SP) independently evaluating the same knees, 139 clinical and 36 radiographic features evaluated, diagnosis made by GP and SP before and after viewing radiographic data
- Machine learning
- Nested 5-fold cross-validation enhanced random forest models
- Paired clinicians (one GP and one SP) independently evaluating the same knees
- 139 clinical and 36 radiographic features evaluated
- Diagnosis made by GP and SP before and after viewing radiographic data
Main finding
As reported in the abstract
Features related to clinicians' diagnosis of clinically relevant knee OA
Models with top 10 features explaining clinicians' diagnosis, AUC range
0.76-0.83
Observational data: associations, not effects.
Further findings
Knees diagnosed with clinically relevant OA, before and after viewing radiographic data
- GPs, before viewing radiographic data
- 42%
- GPs, after viewing radiographic data
- 43%
- SPs, before viewing radiographic data
- 43%
- SPs, after viewing radiographic data
- 51%
Original publication
A machine learning approach reveals features related to clinicians’ diagnosis of clinically relevant knee osteoarthritis. Rheumatology (Oxford). Published 19 December 2022. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →