Reducing Ophthalmic Health Disparities Through Transfer Learning: A Novel Application to Overcome Data Inequality

Area of Science:

• Ophthalmic healthcare disparities
• Machine learning applications in medicine
• Health equity research

Background:

• Healthcare systems exhibit race disparities, leading to unequal clinical data and biased prediction outcomes.
• Data imbalance across racial groups hinders equitable healthcare analysis and intervention development.

Purpose of the Study:

• To reduce healthcare disparities caused by data imbalance using advanced transfer learning (TL) methods.
• To improve prediction accuracy for underrepresented racial groups in ophthalmic care.

Main Methods:

• Population-level analysis of ophthalmic healthcare disparities using electronic medical records (N = 785).
• Application of regression-based TL models to transfer information from White to African American (AA) and Asian groups.
• Comparison of TL model performance against conventional prediction approaches.

Main Results:

• Observed disparities in socioeconomic status and disease severity among AA and Asian groups.
• TL models demonstrated marked to comparable improvements in prediction accuracy (lower MAE/MSE) versus conventional methods.
• TL identified unique features of visual field mean deviation (MD) rate of change specific to each racial group.

Conclusions:

• Novel application of TL enhances analytical reliability, particularly for small sample sizes.
• TL offers an equitable and efficient strategy to mitigate healthcare disparities by improving predictions for data-disadvantaged populations.