Development and validation of a predictive risk model based on retinal geometry for an early assessment of diabetic retinopathy

Abstract

Aims: This study aimed to develop and validate a risk nomogram prediction model based on the retinal geometry of diabetic retinopathy (DR) in patients with type 2 diabetes mellitus (T2DM) and to investigate its clinical application value.

Methods: In this study, we collected the clinical data of 410 patients with T2DM in the Second Affiliated Hospital of Chongqing Medical University between October 2020 and March 2022. Firstly, the patients were randomly divided into a development cohort and a validation cohort in a ratio of 7:3. Then, the modeling factors were selected using the least absolute shrinkage and selection operator (LASSO). Subsequently, a nomogram prediction model was built with these identified risk factors. Two other models were constructed with only retinal vascular traits or only clinical traits to confirm the performance advantage of this nomogram model. Finally, the model performances were assessed using the area under the receiver operating characteristic curve (AUC), calibration plot, and decision curve analysis (DCA).

Results: Five predictive variables for DR among patients with T2DM were selected by LASSO regression from 33 variables, including fractal dimension, arterial tortuosity, venular caliber, duration of diabetes mellitus (DM), and insulin dosage (P< 0.05). A predictive nomogram model based on these selected clinical and retinal vascular factors presented good discrimination with an AUC of 0.909 in the training cohort and 0.876 in the validation cohort. By comparing the models, the retinal vascular parameters were proven to have a predictive value and could improve diagnostic sensitivity and specificity when combined with clinical characteristics. The calibration curve displayed high consistency between predicted and actual probability in both training and validation cohorts. The DCA demonstrated that this nomogram model led to net benefits in a wide range of threshold probability and could be adapted for clinical decision-making.

Conclusion: This study presented a predictive nomogram that might facilitate the risk stratification and early detection of DR among patients with T2DM.

Keywords: T2DM; diabetic retinopathy; nomogram; prediction model; retinal geometry.

Figure 1

Flow diagram of the analysis.

Figure 2

Least absolute shrinkage and selection operator (LASSO) regression and variable selection. All features were normalized by Z-score to remove the dimension divergence among the observational variables. In the cross-validated error plot, 25 variables with non-zero coefficients were selected at minimum cross-validated error, while five variables with non-zero coefficient remained at minimum cross-validation error within 1 standard error of the minimum.

Figure 3

The nomogram model.

Figure 4

Receiver operating characteristic (ROC) curves of the constructed models. The y-axis means the true-positive rate of the risk prediction. The x-axis means the false-positive rate of the risk prediction. The blue line represents the performance of the nomogram. The cutoff value and area under the receiver operating characteristic curve (AUC) are presented in the picture as well.

Figure 5

Calibration curves of the constructed models. The y-axis means the actual diagnosed diabetic retinopathy (DR). The x-axis meant the predicted risk of DR. The green lines represent the performance of the training cohort and validation cohort, which indicate a closer fit to the red line. This represents a better prediction.

Figure 6

Decision curve analysis (DCA) for the constructed models. The y-axis measured the standardized net benefit. The horizontal line represented that none of the patients has diabetic retinopathy (DR) and all of them were treatment-free. The other oblique line is drawn on the assumption that all of the type 2 diabetes mellitus (T2DM) patients have DR and received clinical intervention.