Empiric Demonstration of the Trend-in-Trend Study Design for Causal Inference

• Activity Number: 407
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Xinyao Ji*+
• Keywords: Trend In Trend ; Difference In Difference ; Propensity Score

Abstract

The central impediment in inferring causation from epidemiologic studies is the possibility that exposed and unexposed individuals differ with respect to unmeasured characteristics that might influence the rate of the outcome. Such impediment is known as non-exchangeability, selection bias, or unmeasured confounding. To address this problem, we introduce a novel methodology called Trend-in-Trend. Intuitively, if the frequency of a causal exposure increases, the frequency of the outcomes should increase. If the frequency of a causal exposure declines, then the frequency of the outcomes should also decline. Therefore, given a population dataset, if we can identify subgroups who experienced different time trends in exposure frequency and if the magnitude of these time trends are associated with time trends in outcome, we get the evidence for causation. To find the existence of trends in exposure prevalence, we use the propensity score model, regressing the exposure status on covariates associated with each individual. We then divide the whole population into subgroups according to their propensity score quantiles and apply the Trend-in-Trend methodology to the group-level data.