Materials and methods
This study was performed between February and August 2019 at two tertiary hospitals in Western Sydney. Western Sydney is a diabetes hotspot, with an ethnically diverse, low-socioeconomic population.15 The treating endocrinologist offered study participation to consecutive patients with diabetes over the age of 18 years, who attended routine appointments at foot ulcer or integrated care diabetes clinics.16 Informed, written consent was obtained from all participants. Patients with photosensitive epilepsy were excluded from participation, and patients with a family history of angle-closure glaucoma were not offered mydriasis.
Data acquisition and screening investigations were performed by a trained medical student (LSW). Best-corrected visual acuity (VA) was tested with a 3 metre handheld Snellen chart for portability, with pinhole correction tested if 6/6 vision was not initially achieved. The patient was seated in a darkened room for 10 min while a validated questionnaire was conducted regarding ethnicity, spoken languages, type and duration of DM, time and site of previous diabetic eye screening, known DR status, and systemic diseases.17
Bilateral two-field 45° non-stereoscopic, color, non-mydriatic photographs were taken using a portable, non-mydriatic retinal camera (RetinaVue 100; Welch Allyn, Macquarie Park, Australia) (online supplemental file 1). Both macula and optical-disc centered photos were taken of each eye. After 2–3 min, combined pupillometry and electroretinogram (ERG) was performed bilaterally (RETeval, Welch Allyn). Skin-adhesive electrodes were placed beneath both eyes and the Ganzfeld Dome was held over the patient’s eye. Flicker ERG flashed light of varying intensity into each eye for approximately 45 seconds, and patterns of pupillary constriction were recorded. A single numerical output was generated, indicating the patient’s risk of DR.
Patients were subsequently offered pupillary dilation. This was initially with Minims Tropicamide 1% (Bausch & Lomb, Laval, Canada); however, uptake was poor due to concerns over induced visual disturbance. Subsequently, a less concentrated formulation, Minims Tropicamide 0.5% (Bausch & Lomb), was offered. Following mydriasis, bilateral two-field fundus photographs were retaken with the RetinaVue 100, and these dilated images were used as the clinical reference standard.18 19 Overall, 53.1% of the study population (n=145) consented to mydriasis.
The patients’ most recent (within the last 3 months) HbA1c, estimated glomerular filtration rate (eGFR), and body mass index (BMI) were obtained from their electronic medical record. All results were reviewed by a consultant ophthalmologist (HPD) with access to clinical data, who assessed image quality, graded DR severity, and determined follow-up plans with either an optometrist or an ophthalmologist, within a clinically relevant timeframe. Letters detailing these results were sent to the patient and their nominated general practitioner. To confer evaluation reliability, a second trained grader (HN), masked to clinical data and previous gradings, reassessed all photographs for DR severity and image quality. Computerized image reordering was used to avoid recall bias. Grading conflicts were arbitrated by a medical retina fellowship-trained ophthalmologist (ATF), masked to clinical data and previous gradings.
DR severity was graded according to the International Classification of Diabetic Retinopathy (ICDR) and Diabetic Macular Oedema (DMO) grading scale.20 DR was defined as the presence of mild, moderate, or severe non-proliferative diabetic retinopathy (NPDR), proliferative diabetic retinopathy (PDR) and/or DMO on ophthalmology evaluation. VTDR was defined, in accordance with the Practice Guidelines for Ocular Telehealth, as an eye having severe NPDR, PDR or DMO.5 DMO was defined as any hard exudate or obvious retinal thickening within the macula region.5 20 Image quality was rated using a 5-point Likert scale (1, inadequate for any diagnostic purpose; 2, unable to exclude all emergent findings; 3, only able to exclude emergent findings; 4, not ideal but still able to exclude subtle findings; and 5, ideal quality).21
Statistical analysis
To determine the sensitivity and specificity of the RETeval ERG device, an optimal cut-off score defining DR risk had to first be determined. The receiver operating characteristic (ROC) curve was used to identify the optimal DR cut-off score using Euclidean distances.22 The score with the lowest Euclidean distance represented the optimal DR cut-off score. Accuracy was measured by the area under the ROC curve, which measures the test’s ability to correctly classify those with and without disease. The sensitivity and specificity of the non-mydriatic camera and ERG were then determined using point estimates with 95% CI, using the mydriatic photographs as the clinical reference standard. Inter-rater grading agreement was assessed using a Kappa test. Patients who declined pupil dilation were excluded from diagnostic accuracy analyses. If no gradable mydriatic photographs were obtained, this was recorded as a screening failure, and these patients’ results were also excluded from diagnostic accuracy analyses.
Patient characteristics associated with DR were determined using a univariate analysis. Covariate status was determined as of the day of screening, and dark-adapted horizontal pupil size was used.23 Continuous variables were presented with means and standard deviations (mean±SD). Associations between DR and continuous variables were determined using two-tailed Student’s t-tests. Skewed data were presented with medians and interquartile ranges. The associations between DR and skewed data were assessed using Mood’s median tests. Changes in discrete variables were analyzed with χ2 tests.
Predictors of DR were determined using a logistic regression model. DM duration, eGFR and RETeval score were treated as categorical variables, while HbA1c and pupil size were treated as dichotomous variables. Appropriate control groups were determined from similar studies. DM duration was grouped into ≤4, 5–14, 15–24, and ≥25 years, with ≤4 years as the control.24 eGFR was grouped into <30, 30–59, 60–89, and ≥90 mL/min/1.73 m2, with ≥90 mL/min/1.73 m2 as the control.25 Finally, RETeval score was grouped into not measurable, <22 and ≥22, with <22 as the control. Among the dichotomous variables, HbA1c of ≥7.0%26 and pupil size of <4 mm were coded as 1.
Covariates with a p value of ≤0.1 were entered into a multivariate logistic regression model where variables were then chosen using stepwise, backward elimination. The logistic regression model was validated with c-statistics and Hosmer-Lemeshow statistics. Significance was defined as a p value of <0.05. Statistical analyses were conducted with SAS software V.9.4.