A New Method for Partial Correction of Residual Confounding in Time-Series and other Observational Studies

TL;DR

This paper introduces a novel regression-based method using future exposure indicators to reduce residual confounding in time-series observational studies, demonstrated through simulations on ozone effects on asthma visits.

Contribution

The paper presents a new approach that employs an exposure-based indicator with specific properties to correct residual confounding, supported by causal theory and simulation evidence.

Findings

Including future ozone levels reduces residual confounding.

The method achieves modest bias reduction in effect estimates.

Simulations validate the theoretical advantages of the approach.

Abstract

Introduction: Methods now exist to detect residual confounding. One requires an "indicator" with two key properties: conditional independence of the outcome (given exposure and measured covariates) absent confounding and other model miss-specification; and an association with unmeasured confounders (like the exposure). We now present a new method for correcting for residual confounding in time-series and other epidemiological studies. We argue that estimators from models that include an indicator with these key properties should have less bias than those from models without the indicator. Methods: Using causal reasoning and basic regression theory we present theoretical arguments to support our claims. In simulations, we empirically evaluate our approach using a time-series study of ozone effects on emergency department visits for asthma (AV). We base simulations on observed data for ozone, meteorological factors and asthma. Results: In simulations, results from models that included ozone concentrations one day after the AV yielded effect estimators with slightly or modestly less residual confounding. Conclusion: Theory and simulations show that including the indicator based on future air pollution levels can reduce residual confounding. Our method differs from available methods because it uses a regression approach involving an exposure-based indicator rather than a negative outcome control.