Longitudinal Screening for Diabetic Retinopathy in a Nationwide Screening Program: Comparing Deep Learning and Human Graders

Table 1

Demographic characteristics of patients with diabetes in the first and second screening, including the prevalence of each diabetic retinopathy severity level and diabetic macular edema.

Characteristics	First screening, DL and HG ()	Second screening, DL ()	Second screening, HG ()
Age, years,
Female, (%)	3,945 (68.8%)	2,874 (69.3%)	2,951 (69.2%)
Hypertension, (%)	3,895 (67.9%)	2,855 (68.8%)	2,921 (68.5%)
FBS, mg/dL in
No NPDR, (%)	4,152 (72.36%)	3,239 (78.09%)	3,256 (76.38%)
Mild NPDR no DME, (%)	589 (10.26%)	448 (10.80%)	449 (10.53%)
Moderate NPDR no DME, (%)	293 (5.11%)	250 (6.03%)	269 (6.31%)
Severe NPDR no DME, (%)	6 (0.10%)	7 (0.17%)	6 (0.14%)
PDR no DME, (%)	47 (0.82%)	11 (0.27%)	17 (0.40%)
DME, (%)	651 (11.35%)	193 (4.65%)	266 (6.24%)

DL: deep learning; HG: trained human graders; FBS: fasting blood sugar; NPDR: nonproliferative diabetic retinopathy; PDR: proliferative diabetic retinopathy; DME: diabetic macular edema. The prevalence of each DR severity level and DME in each cohort was determined by the reference standard.