Accuracy of Primary Care Clinicians in Screening for Diabetic Retinopathy Using Single-Image Retinal Photography

Tillman F. Farley; Naresh Mandava; F. Ryan Prall; Cece Carsky

doi:10.1370/afm.857

Article Figures & Data

Tables

Additional Files

View popup

Table 1.

Criteria for Referral to Eye Care Specialist

Uninterpretable (inadequate) photographs

Severe diabetic retinopathy as evidenced by any of the following:

Diffuse intraretinal hemorrhage in all 4 quadrants

Any hard lipid exudate present within 1/3 disc diameter of fovea

An optic-disc-sized patch of hard lipid exudate, any part of which is within 1 disc diameter of the fovea

Venous beading present in 2 quadrants

Any intraretinal microvascular abnormalities

Any neovascularization

Any vitreous hemorrhage

Other abnormalities, including retinal scarring or past laser surgery

Any unknown abnormalities

View popup

Table 2.

Results of Photograph Interpretations (n = 1,040) by Primary Care Clinician and Ophthalmologist

Reading Outcome, All Patients	Ophthalmologist	Primary Care Clinician	Sensitivity	Specificity	PPV	NPV
NPV = negative predictive value; PPV = positive predictive value.
^a For identification of any diabetic retinopathy, regardless of whether it needed referral.
^b Other abnormalities include laser and other retinal scarring.
^c Some patients were referred for more than 1 finding, but each patient referred was counted only once.
Diabetic retinopathy, overall^a	113	156	0.85	0.94	0.62	0.98
Vitreous hemorrhage	5	7	1.0	0.997	0.71	1.0
Intraretinal microvascular abnormality	5	3	0.2	0.998	0.33	0.996
Venous beading	6	11	0.50	0.995	0.27	0.997
2 quadrants	4	9	0.50	0.99	0.22	0.99
Intraretinal hemorrhage	98	90	0.73	0.98	0.80	0.97
Diffuse	4	6	0.75	0.99	0.50	0.99
Neovascularization	5	4	0.4	0.997	0.50	0.998
Lipid exudates	56	108	0.96	0.95	0.50	0.99
1/3 disc diameter	39	52	0.82	0.98	0.62	0.99
Disc diameter	16	29	0.94	0.99	0.52	0.99
Referred for other abnormality^b	39	26	0.66	0.97	0.38	0.96
Referred for diabetic retinopathy	46	69	0.89	0.97	0.59	0.97
Referred for inadequate photograph	302	296	0.87	0.96	0.89	0.95
Referred, total^c	344	356	0.90	0.93	0.87	0.95

View popup

Table 3.

Ophthalmologist Referrals Missed by Primary Care Clinician

Abnormality Missed	No./Denominator^a(%)
^a Denominator is the number of referrals for each category made by the retinal specialist.
^b Clinicians missed severe diabetic retinopathy in 5 cases; 1 patient was referred for inadequate photograph and does not count as missed referral.
Inadequate photograph	18/302 (6.0)
Other abnormality needing referral (total)	13/39 (33.3)
Abnormal cup-to-disc ratio	5
Severe arterial attenuation and tortuosity	1
Unknown abnormality	3
Choroidal nevus	2
Small optic nerve	1
Arterial plaque	1
Diabetic retinopathy^b	4/46 (8.7)
Total missed referrals	35/344 (10.1)

View popup

Table 4.

Patient Screening Rates by Clinic Access to Retinal Camera

Clinic (Access Model)	No. Screened	No. of Patients With Diabetes Mellitus	Screening Rate	PValue^a
^a For 95% confidence interval of rates, compared with C1, using χ² analyses.
C1 (permanent)	314	673	47	NA
C2 (twice yearly)	261	587	44	.436
C3 (twice yearly)	114	297	38	.017
C4 (twice yearly)	130	519	25	<.001
C5 (refer in)	95	368	26	<.001
C6 (refer in)	30	142	21	<.001
Total	944	2,586	37	<.001

Additional Files

Tables

Supplemental Figures

Supplemental Figure 1. Retinal photograph showing venous beading (vb) and intraretinal microvascular abnormality (IRMA), missed by primary care clinician reader; Supplemental Figure 2. Retinal photograph showing lipid exudates within 1/3 disc diameter of fovea missed by primary care clinician reader.

Files in this Data Supplement:

Supplemental data: Figure 1 - PDF file, 1 page, 174 KB
Supplemental data: Figure 2 - PDF file, 1 page, 98 KB
The Article in Brief

Accuracy of Primary Care Clinicians in Screening for Diabetic Retinopathy Using Single-Image Retinal Photography

Tillman F. Farley , and colleagues

Background Early detection and treatment of diabetic retinopathy, an eye disease common in people with diabetes, can slow or prevent the development of blindness. Screening for diabetic retinopathy usually involves an ophthalmologist (eye specialist), but low-income patients may have limited access to such specialists. Researchers have tested a screening program in which primary care doctors are trained to read a single image taken with a retinal camera, a tool that can identify diabetic retinopathy. This study looks at the doctors' accuracy in reading the photographic images and correctly identifying patients needing referral to ophthalmologists.

What This Study Found Primary care doctors trained to read single images from a retinal camera have acceptable accuracy in screening for diabetic retinopathy. They screened 1,040 diabetic patients at a low-income health center. One hundred thirteen (10.9%) were found to have diabetic retinopathy, 46 severe enough to need referral to an ophthalmologist. Participating doctors missed diabetic retinopathy in 8.7% of the patients who had it (4 out of 46), but discovered it in 41 others. Of the 344 patients the ophthalmologist believed needed referral, primary care doctors failed to refer 35 (10.2%), usually because the doctor failed to recognize an inadequate photograph or abnormalities other than diabetic retinopathy.

Implications

Using single-image retinal photography in primary care offices, with primary care doctors reading the images, may be a cost-effective way to help reduce vision loss in diabetes patients who have limited access to ophthalmologists.
Further training may be needed to help primary care doctors recognize other common eye problems.