Abstract
Purpose: To develop an algorithm using routine clinical laboratory measurements to identify people at risk for systematic underestimation of glycated hemoglobin because of p.Val68Met glucose-6-phosphate dehydrogenase (G6PD) deficiency.</p>
Methods: We analyzed 122,307 participants of self-identified Black race across 4 large cohorts with blood glucose, glycated hemoglobin, and red cell distribution width measurements from a single blood draw. In UK Biobank, we used recursive partitioning to train 2 models to predict possible and likely G6PD deficiency. We validated the algorithm in NIH AllofUs, Vanderbilt BioVU, and the Million Veterans Program. In the Vanderbilt Synthetic Derivative with clinical but no genetic data, we created a cohort of 48,031 participants with type 2 diabetes and no genetic data to test whether predicted risk for G6PD deficiency was associated with incident diabetic retinopathy.</p>
Results: G6PD deficiency predictions in hemizygous males showed precision/recall of 31%/81% for possible and 81%/10% for likely deficiency. In homozygous females, precision/recall was 6%/76% for possible and 34%/13% for likely deficiency. Patients with diabetes having predicted possible deficiency demonstrated 1.4-fold higher 20-year retinopathy rates (14.3% vs 11.2%, P = .003).</p>
Conclusion: We report a simple clinical algorithm that enables health care systems to identify people who may benefit from G6PD genotyping and glucose-based diabetes monitoring.</p>