System Maintenance Notice

Please note that maintenance work is scheduled between 4 AM and 5 AM - CET. During this time, you may experience reduced performance. Thanks for your patience.

Publications

Journal Article Epublish

PubMed: 40520481 DOI: 10.1097/EE9.0000000000000403

Assessing bias in measuring power outage exposure with simulations.

Environ Epidemiol 9(4):e403

Authors: Heather McBrien, Daniel Mork, Marianthi-Anna Kioumourtzoglou, Joan A Casey

Affiliations

  • Department of Environmental Health Sciences, Columbia Mailman School of Public Health, New York, New York.
  • Department of Biostatistics, Harvard T. H. Chan School of Public Health, Boston, Massachusetts.
  • Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington.
  • Department of Epidemiology, University of Washington, Seattle, Washington.

Abstract

Background

New national power outage exposure data have become available since 2020, which can support epidemiologic studies of power outage and health outcomes, but exposure assessment challenges remain. Two sources of bias could affect results: available datasets are missing large percentages of observations, and the health-relevant duration of power outages remains unknown. Here, we aimed to determine if existing datasets can produce usable effect estimates in epidemiologic studies despite missing data, and quantify bias introduced by incorrect assumptions about the health-relevant duration of power outages.

Methods

Based on existing data from PowerOutage.us, we conducted simulations representing a county-level study. We simulated and then estimated the effect of daily power outage exposure on hospitalization rates. We measured the magnitude and direction of bias introduced in the presence of incorrect assumptions about the health-relevant power outage duration and when increasing amounts of exposure data were missing.

Results

When the health-relevant power outage duration was underestimated, results were substantially biased towards the null (mean bias: -64.7%, SD: 34.9). Overestimation of the health-relevant power outage duration resulted in smaller bias (mean bias: -6.7%, SD: 30.6). When 80% or more of county-level person-time of power outage data were missing in 80% of study counties, results were severely biased towards the null (mean bias: -54.4%, SD: 39.8).

Conclusions

Our results show that while some bias is likely, sensitivity analyses and careful choices of health-relevant duration can help researchers leverage available power outage data to produce low bias effect estimates in epidemiologic studies of power outages and health outcomes.

MeSH

Bias, Epidemiologic Studies