Abstract
Purpose :
To develop and validate a risk calculator to identify individuals at high-risk for diabetic retinopathy (DR).
Methods :
This study includes patients with diabetes from an urban, academic ophthalmology center. Patient demographic characteristics were generated using descriptive statistics. Univariable and multivariable cox proportional hazard model identified risk factors of high-risk diabetic retinopathy. Stepwise akaike’s information criteria (AIC) was used to select the best multivariable model. Bootstrap estimate of calibration accuracy for 3-year survival probability showed the existence of model overfitting. The model was then validated using 200 bootstrap resamples to obtain the slope shrinkage factor which could be used to update the model to have better survival prediction. Nomogram and R Shiny app were developed to predict 3-year and 5-year survival probabilities. Statistical analyses were performed using R (The R Foundation, Vienna, Austria) and SAS 9.4 (Cary, NC).
Results :
A total of 1363 patients (female, 51.7%) were included in this study. The mean age [SD] was 61.2 [13.5]. The mean hemoglobin A1C [SD] was 8.7 [2.5]. The ethnic and racial breakdowns were 42.9% white, 33.1% Black/African American, 27.5% Hispanic/Latino, and 3.65% Asian. Most patients had a current primary care provider (96.4%), primarily spoke English (80.3%), and had public insurance (63.7%). Patients had a history of the following diabetic and chronic health conditions: nephropathy (10.0%), neuropathy (17.8%), nonhealing ulcer (11.2%), dyslipidemia (31.3%), uncomplicated hypertension (47.5%), and complicated hypertension (2.1%). The validated predictive model included having a primary care provider (HR 2.61; Table 1), insurance provider (private [HR 0.90]; uninsured [HR 2.05]), hemoglobin A1C (HR 1.08), dyslipidemia (HR 0.70), uncomplicated hypertension (HR 1.31), and complicated hypertension (HR 1.66; Table 1). A nomogram was developed to predict 3-year and 5-year probability of developing DR (Figure 1).
Conclusions :
Using a predictive model, we show that adult patients with diabetes can be stratified according to their risk of DR. Risk stratification methods may help to improve clinical and surgical decision-making for patients with DR.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.