June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
An EHR-Based Predictive Model to Understand Lost-to-Follow-Up Risk in Patients with Diabetic Retinopathy
Author Affiliations & Notes
  • Durga S Borkar
    Duke University Eye Center, North Carolina, United States
  • Chun Xu
    Duke University, Durham, North Carolina, United States
  • Majda Hadziahmetovic
    Duke University Eye Center, North Carolina, United States
  • Benjamin Goldstein
    Duke University, Durham, North Carolina, United States
  • Footnotes
    Commercial Relationships   Durga Borkar Verana Health, Code C (Consultant/Contractor); Chun Xu None; Majda Hadziahmetovic None; Benjamin Goldstein None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2820 – A0150. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Durga S Borkar, Chun Xu, Majda Hadziahmetovic, Benjamin Goldstein; An EHR-Based Predictive Model to Understand Lost-to-Follow-Up Risk in Patients with Diabetic Retinopathy. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2820 – A0150.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : Prior studies have shown that patients with diabetic retinopathy, particularly proliferative diabetic retinopathy, have high rates of becoming lost-to-follow-up (LTFU), as well as worse clinical outcomes after inadvertent treatment lapses. The purpose of this study was to create a predictive model for risk of becoming LTFU after a clinical encounter for diabetic retinopathy using electronic health record (EHR) data.

Methods : EHR data were extracted for all diabetic retinopathy clinical encounters at an academic medical center in Durham, NC from January 1, 2014 to March 15, 2020. Specifically, data on healthcare utilization, demographic factors, and ophthalmic clinical parameters were collected. Information on imaging and therapeutic procedures (i.e. anti-VEGF injection or panretinal photocoagulation) was included. LTFU was determined by diabetic retinopathy stage based on ICD coding and the recommended follow up interval in the AAO Preferred Practice Pattern for diabetic retinopathy. LASSO and random forest models were constructed using LTFU as the outcome using (1) only data available prior to the clinical encounter and (2) data collected during the ophthalmic clinical encounter.

Results : Data for 25,576 clinical encounters for 7,948 patients were extracted. Of these, 12,971 encounters (50.7%) were for proliferative diabetic retinopathy. LASSO and random forest models were created using pre-encounter data, including healthcare utilization and demographic data. Both models had an area under the curve (AUC) of 0.72. A LASSO model utilizing these parameters, as well as ophthalmic factors, had an AUC of 0.74 while the random forest model had an AUC of 0.75. Additional analyses were performed to construct a LASSO model for LTFU by race. While there was some variation, the AUC for all models for racial subgroups were between 0.70 and 0.80.

Conclusions : High performing EHR models for health services utilization typically have an AUC of 0.70 to 0.80. This EHR-based study using structured data fields presents a high performing predictive model to understand risk of becoming lost-to-follow-up for diabetic retinopathy patients. Models involving ophthalmic clinical factors perform slightly better than models considering healthcare utilization and demographic data alone. This study provides further opportunities for direct implementation into clinical care to understand LTFU risk in real time.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×