📝✅ We suggest several improvements:
✓ Report demographic composition of training datasets
✓ Validate algorithms across diverse populations
✓ Include performance metrics by demographic groups
✓ Develop more inclusive validation processes /5

🏥📋 These findings have important clinical implications:
🔸 Brain age algorithms are increasingly used as health biomarkers
🔸 Systematic prediction errors could impact diagnosis accuracy
🔸 May contribute to existing healthcare disparities /4

🔍💡 What contributes to these algorithmic differences?
Training data composition appears to be a key factor:
🏥 UK Biobank: 94% White participants
📊 Many datasets lack demographic diversity
🧬 Algorithms may not generalize well across populations
#DataDiversity #AIResearch /3