Assessment of demographic bias in retinal age prediction machine learning models
Bibliographic record
Abstract
The retinal age gap, defined as the difference between the predicted retinal age and chronological age, is an emerging biomarker for many eye conditions and even non-ocular diseases. Machine learning (ML) models are commonly used for retinal age prediction. However, biases in ML models may lead to unfair predictions for some demographic groups, potentially exacerbating health disparities. This retrospective cross-sectional study evaluated demographic biases related to sex and ethnicity in retinal age prediction models using retinal imaging data (color fundus photography [CFP], optical coherence tomography [OCT], and combined CFP + OCT) from 9,668 healthy individuals (mean age 56.8 years; 52% female) in the UK Biobank. The RETFound foundation model was fine-tuned to predict retinal age, and bias was assessed by comparing mean absolute error (MAE) and retinal age gaps across demographic groups. The combined CFP + OCT model achieved the lowest MAE (3.01 years), outperforming CFP-only (3.40 years) and OCT-only (4.37 years) models. Significant sex differences were observed only in the CFP model ( p < 0.001), while significant ethnicity differences appeared only in the OCT model ( p < 0.001). No significant sex/ethnicity differences were observed in the combined model. These results demonstrate that retinal age prediction models can exhibit biases, and that these biases, along with model accuracy, are influenced by the choice of imaging modality (CFP, OCT, or combined). Identifying and addressing sources of bias is essential for safe and reliable clinical implementation. Our results emphasize the importance of comprehensive bias assessments and prospective validation, ensuring that advances in machine learning and artificial intelligence benefit all patient populations.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".