Purchase this article with an account.
Christopher J. Brady, Samantha D'Amico, Jeremy Peavey, Stephen Higgins, Brian Kim; Determinants of Poor Follow-up Adherence for Diabetic Retinopathy. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1086.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Vision loss from diabetic retinopathy (DR) is largely preventable with adequate screening and follow-up care. Unfortunately, adherence to both services is often poor. We sought to characterize adherence to DR follow-up in a rural population, and explore factors associated with poor follow-up.
A retrospective chart review of 320 adult patients with diabetes mellitus seen in the University of Vermont Ophthalmology Division at least once between 10/1/15 and 3/30/16 was done. The records of patients who were told to return for repeat evaluation were assessed for visit completion within double the recommended interval. Demographic factors, stage of DR, and diabetic macular edema (DME) presence were explored for association with poor follow-up. Receiver-operating characteristic (ROC) curves were assessed for identifying individuals unlikely to follow-up.
Of 320 adults seen in the index time period, 192 (60%) were noted in the chart to require a DR follow-up exam. The median age of included patients was 66. Thirty-four (17.7%) had insurance status of “self-pay,” “Medicaid,” or “Medicare/Medicaid,” and were defined as socioeconomically disadvantaged. The majority of patients, 106 (55.2%) had no DR, 22 (11.4%) had mild non-proliferative DR (NPDR), 28 (14.6%) had moderate NPDR, 4 (2.1%) had severe NPDR, and 32 (16.7%) had proliferative DR (PDR). Forty-seven (24.5%) failed to follow-up within the allotted time, and 40 of these never following up at all by the end of study observation (median follow-up time: 195 days, range: 1-728 days). Disadvantaged status, younger age, lower DR severity and absence of DME were all associated with poor follow-up in univariable logistic regression (fig. 1). The area under the ROC curve using a logistic regression model including all covariates was 0.68 (fig. 2).
In our sample, 75% of diabetic patients adhered to follow-up recommendations. Those less likely to follow-up included people of lower socioeconomic status, younger people, and people with no DR or milder DR. People who do not follow-up within double the recommended follow-up interval are unlikely to follow-up at all. While our models and ROC analyses were not currently adequate for clinical use, they are encouraging about the potential for a predictive algorithm to alert clinicians through the EHR that a particular patient might be prone to poor follow-up.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
ROC for predicting poor follow-up
This PDF is available to Subscribers Only