COVID-19 Prediction Model

If you were selected to work on a COVID research team, how would you build your prediction model?

1. Identify the dependent variable

• Clinical severity
  • Operationalization:
    • Mild
    • Moderate (hospitalized but not on ventilators)
    • Severe (hospitalized and on ventilators)
    • Deceased (from COVID-19)

2. List all the possible predictors

• Age
• Gender (female/male)
• Race (African American/Asian/White/Other)
• Ethnicity (Hispanic/Non-Hispanic)
• Obesity (Body mass index)
• Preexisting health conditions (e.g. diabetes)
• Insurance (has insurance/does not have insurance)

3. Work up a strategy and plan for your prediction model

• Use the UT Dallas Proactive Testing database
• Split the tested individuals with laboratory-confirmed positive COVID-19 test into randomly selected 70% training and 30% testing sets

3a. Estimation

• Test differences among the four DV groups by using multivariate logistic regression

3b. Resampling methods

• Use bootstrap resampling technique by running multiple simulations on the dataset. In consecutive simulations, parameters will be adjusted to obtain the most precise estimates.