Using Propensity Scores in Convenience Samples

• Activity Number: 386 - SPEED: Statistics in Epidemiology Part 1
• Type: Contributed
• Date/Time: Wednesday, August 10, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Olivia M. Bernstein Morgan* and Brian G. Vegetabile and Joshua D. Grill and Daniel L Gillen
• Companies: University of California, Irvine and RAND Corporation and University of California, Irvine and University of California Irvine
• Keywords: propensity scores; convenience samples; Alzheimer's Disease; NACC

Abstract

Convenience samples are abundant in the age of Big Data. They are often used for causal inference but can be subject to confounding combined with selection bias. For example, the National Alzheimer's Coordinating Center (NACC) Uniform Data Set is often used to identify risk factors of Alzheimer’s disease but likely over-represents highly educated and non-Hispanic White participants, relative to the US population. We compare approaches for implementing sampling weights for convenience samples estimated with an auxiliary dataset into propensity score adjusted analyses. Additionally, we derive an analytic variance estimate for the outcome model using simultaneous estimating equations that accounts for the uncertainty in estimating the sampling weights and propensity scores. We conducted a simulation study to compare the impact of incorporating sampling weights into the propensity score and outcome models or ignoring them. We then applied this approach to NACC data to estimate the effect of vitamin E supplementation on functional activities using NACC data with estimated sampling weights. NACC data are contributed by the NIA-funded ADCs and are funded by NIA/NIH Grant U24 AG072122.