Understanding COVID-19 infection among people with intellectual and developmental disabilities using machine learning.

BACKGROUND

People with intellectual and developmental disabilities (IDD) were disproportionately affected by the COVID-19 pandemic. Predicting COVID-19 infection has been difficult.

OBJECTIVE

We sought to address two research questions in this study: 1) to assess the overall utility of a machine learning model to predict COVID-19 diagnosis for people with IDD, and 2) to determine the primary predictors of COVID-19 diagnosis in a random sample of Home and Community Based Services users in one state.

METHODS

We merged three major IDD-specific datasets (National Core Indicators, Supports Intensity Scale, Medicaid HCBS expenditures) from one state to create one combined dataset for analyses that included more than 700 variables. We then built a random forest machine learning algorithm to predict COVID-19 diagnosis and to explore the top predictors of such a diagnosis, when present.

RESULTS

Our algorithm predicted COVID-19 diagnosis in a random sample of HCBS users with IDD with 62.5% accuracy. The top predictors of having a documented case of COVID-19 among our sample were higher age, having high overall, medical, or behavioral support needs, living in a lower-income neighborhood, total Medicaid expenditure, and higher body mass index.

CONCLUSIONS

Results largely followed trends in the general population, and were largely suggestive that increased contact with other people may have exposed a person with IDD to greater COVID-19 risk.