A Diagnostic Model for Screening Diabetic Retinopathy Using the Hand-Held Electroretinogram Device RETeval

Abstract

Purpose: To construct a proper model to screen for diabetic retinopathy (DR) with the RETeval.

Method: This was a cross-sectional study. Two hundred thirty-two diabetic patients and seventy controls were recruited. The DR risk assessment protocol was performed to obtain subjects' DR risk score using the RETeval. Afterwards, the receiver operating characteristic (ROC) curve was used to determine the best cutoff for diagnosing DR. Random forest and decision tree models were constructed.

Results: With increasing DR severity, the DR score gradually increased. When the DR score was used to diagnose DR, the ROC curve had an area under the curve of 0.881 (95% confidence interval: 0.836-0.927, P < 0.001), with a best cutoff value of 22.95, a sensitivity of 74.3% (95 CI: 66.0%~82.6%), and a specificity of 90.6% (95 CI: 83.7% ~94.8%). The top four risk factors selected by the random forest were used to construct the decision tree for diagnosing DR, which had a sensitivity of 93.3% (95% CI: 86.3%~97.0%) and a specificity of 80.3% (95% CI: 72.1% ~86.6%).

Conclusions: The DR risk assessment protocol combined with the decision tree model was innovatively used to evaluate the risk of DR, improving the sensitivity of diagnosis, which makes this method more suitable than the current protocol for DR screening.

Keywords: decision tree; diabetic retinopathy; diagnostic model; electroretinogram; risk factor.