[Paper Review] Air quality and acute deaths in California, 2000-2012

Kenneth K. Lopiano, Richard L. Smith+1 more|arXiv (Cornell University)|Feb 10, 2015

Air Quality and Health Impacts23 references3 citations

TL;DR

This study investigates the association between air quality (PM2.5 and ozone) and acute deaths in California from 2000 to 2012 using a publicly available dataset of daily air quality, weather, and mortality data across eight major air basins. Employing moving median analysis, time series modeling, and leave-one-year-out cross-validation, the authors find little to no evidence linking current levels of PM2.5 or ozone to acute deaths, suggesting air quality may not be a significant driver of acute mortality in California.

ABSTRACT

Many studies have sought to determine if there is an association between air quality and acute deaths. Many consider it plausible that current levels of air quality cause acute deaths. However, several factors call causation and even association into question. Observational data sets are large and complex. Multiple testing and multiple modeling can lead to false positive findings. Publication, confirmation and other biases are also possible problems. Moreover, the fact that most data sets used in studies evaluating the relationships among air quality and public health outcomes are not publicly available makes reproducing the claims nearly impossible. Here we have built and made publicly available a dataset containing daily air quality levels, PM2.5 and ozone, daily temperature levels, minimum and maximum and daily relative humidity levels for the eight most populous California air basins. We analyzed the dataset using a moving median analysis, a standard time series analysis, and a prediction analysis within the following analysis strategy. We examine the eight air basins separately to see if estimates replicate across locations. We use leave one year out cross validation analysis to evaluate predictions. Both the moving medians analysis and the standard time series analysis found little evidence for association between air quality and acute deaths. The prediction analysis process was a run as a large factorial design using different models and holding out one year at a time. Among the variables used to predict acute death, most of the daily death variability was explained by time of year or weather variables. In summary, the empirical evidence is that current levels of air quality, ozone and PM2.5, are not causally related to acute deaths for California. An empirical and logical case can be made air quality is not causally related to acute deaths for the rest of the United States.

Motivation & Objective

• To examine whether current levels of air pollution (PM2.5 and ozone) are causally linked to acute deaths in California.
• To address reproducibility issues in prior air quality-mortality studies by creating and publicly releasing a comprehensive, daily-level dataset of air quality, weather, and mortality.
• To evaluate the robustness of associations across multiple analytical methods and geographic regions in California.
• To test whether weather and temporal variables explain more of the daily death variability than air pollution.
• To challenge the validity of previously reported associations through rigorous cross-validation and model comparison.

Proposed method

• Constructed a publicly available dataset with daily measurements of PM2.5, ozone, temperature (min/max), and relative humidity across eight major California air basins from 2000 to 2012.
• Applied moving median analysis to assess short-term associations between air pollutants and acute deaths.
• Used standard time series modeling to evaluate the relationship between air quality and daily mortality counts.
• Conducted a large factorial prediction analysis using multiple models, with one year held out at a time for validation (leave-one-year-out cross-validation).
• Compared model performance using weather and temporal variables against models including air pollutants to assess predictive contribution.
• Evaluated model fit and significance using statistical criteria and cross-validation to minimize overfitting and false positives.

Experimental results

Research questions

• RQ1 Is there a statistically significant association between daily levels of PM2.5 or ozone and acute deaths in California between 2000 and 2012?
• RQ2 Do the observed associations between air quality and mortality hold across different geographic regions in California?
• RQ3 To what extent do weather and seasonal factors explain daily variation in acute deaths compared to air pollution?
• RQ4 Can models including air quality variables predict acute mortality better than models based only on time of year and weather?
• RQ5 Are the reported associations between air pollution and mortality robust to cross-validation and multiple testing?

Key findings

• The moving median analysis found no consistent or statistically significant associations between PM2.5, ozone, or other pollutants and acute deaths across the eight California air basins.
• Standard time series models also showed little to no evidence of a meaningful association between air quality and acute mortality.
• The prediction analysis revealed that time of year and weather variables (temperature, humidity) explained most of the daily variability in acute deaths, not air pollutants.
• Models including air quality variables did not outperform those based solely on temporal and meteorological factors in cross-validation.
• The study's cross-validation approach, which held out one year at a time, did not support the existence of a robust predictive link between air pollution and acute deaths.
• The authors conclude that current levels of PM2.5 and ozone are not causally related to acute deaths in California, based on empirical evidence and methodological rigor.