Diabetes mellitus (DM)

Diabetic retinopathy (DR)

Compare various machine learning algorithms to construct DR prediction models in patients with type 2 DM and develop a nomogram based on the best model

This cross-sectional study included DM patients undergoing routine DR screening

Patients were randomly divided into training set (244 cases) and validation set (105 cases)

Least absolute shrinkage and selection operator regression for clinical feature selection

Six machine learning algorithms were compared: Decision Tree (DT), K Nearest Neighbors (KNN), Logistic Regression Model (LM), Random Forest (RF), Support Vector Machine (SVM) and XGBoost (XGB)

Model performance was evaluated through receiver operating characteristic curve (ROC), calibration, and decision curve analysis (DCA). A nomogram was then developed based on the best model

Compared with other 5 machine learning algorithms (DT, KNN, RF, SVM, XGB), LM showed the highest area under the ROC curve (AUC of 0.894) and recall rate (0.92) in the validation set

In addition, the calibration curve and DCA results are also relatively good

LM includes disease duration, DPN, insulin dosage, urine protein, and ALB

The nomogram showed robust discrimination (AUC: training set 0.856, validation set 0.868), calibration and clinical applicability after 1000 bootstrap tests in both data sets.

Among six different machine learning algorithms, the LM algorithm showed the best performance

A nomogram based on logistic regression for predicting DR in patients with type 2 DM was established.

This nomogram can serve as a valuable tool for DR detection to facilitate timely treatment

Predictive characteristics of DR by logistic model

Research flow chart

Clinical feature selection via LASSO regression model

Training set, validation set, multi-model ROC, calibration curve, DCA, performance evaluation

Nomogram predicting the likelihood of DR in patients with type 2 DM

Performance of logistic regression model with 1000 bootstraps

Characteristics of patients in training and validation sets

Predicting diabetic retinopathy using different machine learning algorithms has varying importance