June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Mathematical Modeling of Geographic Disparities in Vision Impairment and Machine Learning Predictive Capabilities Using National Health Surveillance Data
Author Affiliations & Notes
  • Dean A VanNasdale
    College of Optometry, The Ohio State University, Columbus, Ohio, United States
  • Christopher Anderson Clark
    School of Optometry, Indiana University, Bloomington, Indiana, United States
  • Erica Shelton
    College of Optometry, The Ohio State University, Columbus, Ohio, United States
  • John Crews
    Independent Contractor, Georgia, United States
  • Footnotes
    Commercial Relationships   Dean VanNasdale None; Christopher Clark None; Erica Shelton None; John Crews None
  • Footnotes
    Support  CDC/NACDD Grant VI2020A
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 3394 – A0181. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Dean A VanNasdale, Christopher Anderson Clark, Erica Shelton, John Crews; Mathematical Modeling of Geographic Disparities in Vision Impairment and Machine Learning Predictive Capabilities Using National Health Surveillance Data. Invest. Ophthalmol. Vis. Sci. 2022;63(7):3394 – A0181.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To mathematically model the frequency distribution of county-level vision impairment (VI) prevalence estimates between 2015 and 2019 and assess for geographic disparities. To use machine learning algorithms to test the feasibility of predictive modeling of VI prevalence estimates.

Methods : American Community Survey (ACS) data were used to estimate the county-level prevalence of adult VI in 3220 counties and county-equivalents in the United States for each year between 2015 and 2019. The VI prevalence frequency distribution was assessed for normalcy using the chi-square goodness of fit test. Kurtosis of the frequency distribution was calculated for each year. Machine learning decision tree software was developed in Matlab (IBM, Natick, MA) with a training data set of VI prevalence estimates stratified by age for 2,720 counties from 2014. The training set was used to develop an algorithm to predict the change in vision impairment from 2014-2019 and tested against data from 500 randomized counties held in reserve.

Results : The frequency distribution of vision impairment was not normally distributed for any of the 5 years assessed (p<0.001 for each year). Kurtosis ranged from 15.2 to 20.4. Linear regression showed the machine learning algorithm predicted change in VI compared to the actual VI prevalence change with an R2 coefficient of 0.244. The biggest predictors of change in VI based on the model were the prevalence of VI in the 65 and older age cohort from 2014 and the prevalence of VI in the 18-64 age cohort.

Conclusions : National health surveillance data demonstrate skewed frequency distributions for VI prevalence, indicating vision impairment disparities at the county-level in the United States. This analysis allows identification of locations with disproportionately high VI prevalence and capabilities to monitor changes to that distribution over time. Preliminary machine learning algorithms suggest that predictive capabilities of VI with limited training data are possible.

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.

×