Three-Stage Method to Estimate Distributed Lag Effects for a Time-Series Analysis of Noncontinuous Exposure Data

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Time-Series Analysis

Background:

• Distributed lag models (DLMs) are crucial for assessing delayed health impacts of environmental exposures but require complete data.
• Noncontinuous exposure data limits DLM application, leading to omitted variable bias (OVB) in simpler models.

Purpose of the Study:

• To develop and validate a novel three-stage method for accurately estimating distributed lag effects from noncontinuous exposure data.
• To correct for OVB inherent in analyzing sparse environmental exposure time-series data.

Main Methods:

• A three-stage approach was developed: fitting lag-specific models, correcting bias using exposure autocorrelation, and smoothing estimates.
• Statistical simulations were used to validate the method's accuracy and robustness.
• The method was applied to analyze PM2.5 and daily mortality data in Beijing (2013-2015).

Main Results:

• Simulations demonstrated the method's ability to accurately recover true lag-response functions and cumulative effects, even with weekly data points.
• The three-stage method utilized nearly 90% more observations than traditional DLMs in the Beijing case study.
• A 10 μg/m³ increase in PM2.5 was associated with a 0.37% rise in all-cause mortality (lag 0-13 days).

Conclusions:

• The proposed three-stage method effectively overcomes data limitations in time-series environmental epidemiology.
• It corrects for OVB, maximizes the use of sparse data, and enhances statistical power.
• This robust statistical tool offers a practical solution for analyzing environmental exposures and health outcomes with incomplete data.