Abstract
Purpose: :
To compare an abbreviated method for determining ETDRS diabetic retinopathy severity level with the standard method used in the ETDRS.
Methods: :
In the standard ETDRS method, a grade for each abnormality considered in the diabetic retinopathy severity scale is recorded separately for each relevant field of the 7–standard fields. A computer algorithm then combines these grades to determine the severity level for the eye. In the abbreviated method, the grader (who must have a detailed knowledge of the definitions of the steps in the scale) examines all 7 fields and assigns the appropriate level directly. This method is more rapid because it does not require that every abnormality be graded in every relevant field. In this comparison 52 eyes (previously graded using the standard system) were selected to include all steps in the scale and were graded by each of 20 graders using the abbreviated system. Agreement was assessed by percent agreement (exact and within 1 step) and by kappa statistic, unweighted and weighted as follows: 1.0 for agreement, 0.75 for 1 step disagreement and zero for all other disagreements.
Results: :
For 11 of 20 graders exact agreement was >=75%, agreement within 1 step >=90%, unweighted kappa >=0.72 and weighted kappa >=0.85. For 4 graders, agreement within 1 step was <=87% (80, 83, 86.5, 86.5) and weighted kappa was <=0.75 (0.68, 0.70, 0.72 and 0.75). The remaining 5 graders were intermediate between these groups. Averages for all 20 graders were: exact agreement 74%, agreement within 1 step 91.7%, unweighted kappa 0.71, weighted kappa 0.84. In another group of 91 eyes in which duplicate grading with the standard method was carried out, corresponding values were: 74%, 91.2%, 0.68 and 0.78.
Conclusions: :
Reproducibility with the abbreviated method is similar to that with the standard method.
Keywords: clinical research methodology • diabetic retinopathy