June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Looking for Low Vision: Deep Learning and Natural Language Processing to Predict Visual Prognosis
Author Affiliations & Notes
  • Sophia Y Wang
    Ophthalmology, Stanford University, Stanford, California, United States
  • Benjamin Tseng
    Ophthalmology, Stanford University, Stanford, California, United States
  • Footnotes
    Commercial Relationships   Sophia Wang, None; Benjamin Tseng, None
  • Footnotes
    Support  NIH NLM T15 LM 007033
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 3502. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Sophia Y Wang, Benjamin Tseng; Looking for Low Vision: Deep Learning and Natural Language Processing to Predict Visual Prognosis. Invest. Ophthalmol. Vis. Sci. 2021;62(8):3502.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Low vision rehabilitation improves quality-of-life for visually impaired patients, but referral rates fall short of national guidelines. Automatically identifying from electronic health records (EHR) patients with poor visual prognosis could allow targeted referrals to low vision services. Since most clinical information is in free-text progress notes, we hypothesize that using information from notes would improve the ability to predict patients’ visual prognosis over using purely structured information such as demographics and billing codes. The purpose of this study was to build deep learning models using natural language processing to integrate EHR data that is both structured and free-text to predict visual prognosis.

Methods : We identified 5612 patients with low vision (best documented visual acuity (VA) <20/40) on ≥ 1 encounter from EHR from 2009-2018, with ≥ 1 year of follow-up from the earliest date of low vision. Patients who did not improve to > 20/40 over 1 year were identified. Ophthalmology notes on or prior to the index date were extracted. Structured data available from the EHR included demographics, billing and procedure codes, medications, and exam findings including VA, intraocular pressure, corneal thickness, and refraction. To predict whether low vision patients would still have low vision a year later, we developed and compared deep learning models that used structured inputs to models that used both structured and free-text progress notes, represented by previously developed ophthalmology domain-specific word embeddings. Standard performance metrics including area under the receiver operating curve (AUROC) and F1 score were evaluated on a held-out test set.

Results : Among the 5612 low vision patients in our cohort, 40.5% (N=2278) never improved to better than 20/40 over one year of follow-up. Deep learning models utilizing structured inputs were able to predict low vision prognosis with AUROC of 79.1% and F1 score of 65.7%. Deep learning models further augmented with free-text inputs were able to achieve AUROC of 81.0% and F1 score of 69.0%.

Conclusions : Free text progress notes within the EHR provide valuable information relevant to predicting patients visual prognosis. Deep learning models utilizing data from EHR to predict ophthalmology outcomes should incorporate information from unstructured free-text progress notes where possible.

This is a 2021 ARVO Annual Meeting abstract.

×
×

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.

×