Identifying the severity of diabetic retinopathy by visual function measures using both traditional statistical methods and interpretable machine learning: a cross-sectional study
Authors' conclusion
Ensemble ML models predicted status of diabetic eye disease with high accuracy using just age, sex and measures of visual function. Interpretable ML methods enabled us to identify profiles of visual function associated with different stages of diabetic eye disease, and to disentangle associations from artefacts of the data collection process. Together, these two techniques have great potential for developing prediction models using untidy real-world clinical data.
Study context
- Population
- Participants of the Northern Ireland Sensory Ageing Study grouped as no DM, DM no DR, DR no DMO and DR with DMO; 12 variables from nine visual function tests1,901 eyes
- Participants
- 1,032 participants
- Exposure
- Ensemble machine learning (ML) models vs multiple logistic regression for classifying DR severity stage
- Outcome
- Accuracy of classification of group membership for tasks A (DM no DR vs no DM), B (DR no DMO vs DM no DR) and C (DR with DMO vs DR no DMO)
Methodological detailsensemble machine learning, interpretable machine learning, multiple logistic regression comparison, Missing values imputed using chained equations
- Ensemble machine learning
- Interpretable machine learning
- Multiple logistic regression comparison
- Missing values imputed using chained equations
Main finding
As reported in the abstract
Classification accuracy, tasks A–C
| Group | Accuracy task A | Accuracy task B | Accuracy task C |
|---|---|---|---|
| Ensemble ML | 0.92 | 1.00 | 0.84 |
| Logistic regression | 0.84 | 0.61 | 0.80 |
Accuracy task A
Ensemble ML0.92
Logistic regression0.84
Accuracy task B
Ensemble ML1.00
Logistic regression0.61
Accuracy task C
Ensemble ML0.84
Logistic regression0.80
Observational data: associations, not effects.
Further findings
Variable importance (interpretable ML)
- Reading index highly ranked for tasks A and B
- Near visual acuity and Moorfields chart acuity important for task C
- Microperimetry variables ranked highly for all three tasks, partly due to a data artefact (a large proportion of missing values)
Original publication
Identifying the severity of diabetic retinopathy by visual function measures using both traditional statistical methods and interpretable machine learning: a cross-sectional study. Diabetologia. Published 19 September 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →