A clinical algorithm to identify people with the glucose-6-phosphate dehydrogenase p.Val68Met variant at risk for diabetes undertreatment.

Summary

Using routine glucose, HbA1c, and RDW from a single draw, the authors trained and externally validated algorithms to flag individuals with the G6PD p.Val68Met variant whose HbA1c may be systematically underestimated. Among people with diabetes, those predicted as possibly deficient had 1.4-fold higher 20-year diabetic retinopathy rates, supporting the clinical relevance of identifying at-risk individuals for genotype testing and glucose-based monitoring.

Key Findings

• Algorithm trained on 122,307 Black participants using concurrent glucose, HbA1c, and RDW predicted G6PD p.Val68Met deficiency risk.
• In hemizygous males, precision/recall were 31%/81% (possible) and 81%/10% (likely); in homozygous females, 6%/76% and 34%/13%, respectively.
• Predicted possible deficiency in diabetes was associated with a 1.4-fold higher 20-year retinopathy rate (14.3% vs 11.2%).
• External validation across All of Us, BioVU, and MVP supports generalizability.

Clinical Implications

Health systems can deploy this algorithm to identify patients who may benefit from G6PD genotyping and a shift toward glucose-based monitoring and treatment targets, reducing the risk of undertreatment and complications such as retinopathy.

Limitations

• Lower precision in females and trade-offs between possible vs likely deficiency thresholds
• Focus on p.Val68Met and self-identified Black cohorts may limit ancestry generalizability; prospective clinical impact not yet tested

Future Directions

Prospective implementation trials should assess workflow integration, impact on treatment decisions and outcomes, and extension to other G6PD variants and ancestries.