If you’re interested in contributing a short What Am I Reading post, we’d love to hear from you! Email us at cache@colorado.edu.

Written by Elizabeth Sorensen Montoya, Ph.D. University of Colorado Boulder www.elizabethsorensenmontoya.com

The relationship between air pollution and negative health outcomes (including mortality), has been well-documented across many disciplines. A bit more recently, researchers have begun to explore how air pollution may also affect cognitive decline. In fact, the Lancet Commission (2020) has identified air pollution as a risk factor for dementia.   

As I’ve been reading recent studies in this space, I’ve been struck not just by the findings, but also by the different approaches researchers use to measure something as invisible and transient as air pollution, and something as complex as cognition. 

Below, I provide a quick tour of the different approaches used in the studies I’ve been reading: Shaffer et al. (2021), Shi et al. (2022), and Bishop et al. (2023). 

How do we measure exposure? 

Although it would make researchers’ lives easier, air pollution does not respect geographic boundaries and can vary substantially even over small spatial scales.  Given this mobile nature of air pollution, alongside the likely importance of long-term exposure, assigning accurate exposure is difficult. Choosing one measurement strategy over another, of course, comes with tradeoffs.  

While the focus of this post will be on measures of outdoor air pollution, it is important to note the limitations of relying solely on outdoor measures.  People spend the majority of their time indoors, where pollutant levels are estimated to be two to five times higher than outdoor levels (Wallace et al., 1986) due to factors like poor ventilation and indoor sources like cooking and cleaning products. Observing only outdoor exposure—even if at the individual-level—misses an important channel of exposure, resulting in measures that may significantly underestimate true exposure. I explore the growing body of work on indoor air pollution in a future post.  

That said, much of our current understanding of the health impacts of air pollution at scale comes from the use of outdoor, fixed-site pollution monitors. These monitors provide highly accurate and frequent (in many cases, hourly) measurements, but only in certain places and often concentrated in urban and traffic-heavy areas. In low- and middle-income countries, limited monitoring infrastructure makes it difficult to assess air quality consistently, often requiring support from international partnerships. In these settings, air quality monitors are commonly placed at U.S. embassies to help fill gaps in monitoring coverage, though recently the U.S. government has discontinued the public sharing of these data. 

While using fixed-site monitor data can improve accuracy, it can also limit the study population and introduce endogeneity. Monitors may be disproportionately located in areas where air pollution has historically been a problem, or in areas with specific demographic characteristics. These characteristics, like income, race, or education, can also influence cognitive outcomes throughout one’s life course, making it harder to separate the true effect of pollution from these other confounding factors.  

Some studies attempt to address this by augmenting monitor data with more individualized measures of exposure. Shaffer et al. (2021), for instance, leverage the Adult Changes in Thought cohort study, a longitudinal study of adults aged 65 and older in Seattle, which placed low-cost air pollution sensors in participants’ homes. Using these individual-level measures, they find that increases in long-term exposure to PM2.5 are associated with a substantial increase in dementia diagnoses. While this approach enhances spatial resolution and better captures individual-level exposure, it is typically constrained by geography and sample selectivity, potentially reducing generalizability.  
 
Of course, few studies will have the resources to place monitors in individual homes, but there are other ways to leverage the more publicly available fixed-site monitor data. Bishop et al. (2023) take a clever approach that trades some precision for scale and causal identification by coupling monitor data with a quasi-experimental design exploiting the U.S. Clean Air Act’s attainment and nonattainment classifications. Counties that just exceed the federal PM2.5 standard are required to implement pollution controls, while similar counties just below the threshold are not, creating a regulatory discontinuity that serves as a natural experiment.  The authors find that increases in long-term exposure increase the probability of a dementia diagnosis, with effect sizes similar to those reported in Shaffer et al. (2021). While this approach lacks individualized exposure, likely resulting in some measurement error, it offers a strong identification strategy and broad generalizability.  

Another common strategy is to use modeled or satellite-based estimates of exposure to increase spatial coverage, even in areas lacking ground monitors.  Shi et al. (2022) use satellite-derived models to estimate long-term PM2.5 exposure across the entire United States, producing complete spatial coverage and allowing them to link pollution estimates to cognitive outcomes in large, nationally representative samples. Like the other studies, they find that increases in long term exposure increase the likelihood of a dementia diagnosis.  

The use of these satellite data makes it possible to study massive populations—even at a global scale. However, like other strategies, this approach comes with tradeoffs: modeled data are less precise than observed pollution levels from monitors and depend on assumptions about how pollution moves through time and space. These assumptions, though grounded in environmental science, can still introduce error. Further, resolution varies by product, typically ranging from 1 to 10 kilometers. In general, there is a trade-off between spatial detail and global or temporal coverage: finer-resolution data require more satellite observations and greater computing power.  

The tradeoffs between precision and coverage are not only limited to the sources of exposure measurement, but also to how it is defined. You may have noticed that that each of the referenced studies considers “long-term” exposure, but each defines and calculates this somewhat differently: Shaffer et al. use a 10-year moving average; Bishop et al. use a decadal cumulative measure; and Shi et al. use up to 17 years of annual averages. While these approaches are well-suited to capturing chronic exposure patterns, they may miss the impact of acute, high-concentration pollution events, which may also present health consequences. For example, extreme events like wildfires or large industrial accidents will be smoothed out in long-term averages, making it difficult to isolate their effects. While the existing body of literature points to long-term exposure as a key risk factor for cognitive decline, short-term, high-intensity events may also play an important role. 

How do we measure cognition? 

Just as there’s no single way to measure exposure, there are multiple ways to measure cognition. Studies differ in how they define cognitive decline, and these differences, just as with differences in exposure measurement, present their own set of tradeoffs. 

From my own reading (which is certainly not exhaustive), it appears that there are two primary ways of measuring cognitive outcomes: administrative data—such as Medicare claims which rely on physician diagnoses recorded using ICD codes, and standardized cognitive assessments (like those conducted in the Health and Retirement Study).  

Shi et al. (2022) and Bishop et al. (2023) both use Medicare claims data, allowing them to study large populations over time. The tradeoff, of course, is that these records may represent only the more severe stages of cognitive decline and may miss early cognitive changes that impact quality of life. These data may also leave out people with lower access to medical care, raising concerns about who is and is not observed.  

Shaffer et al. (2021), on the other hand, rely on longitudinal cognitive assessments conducted by the Adult Changes in Thought study. While the authors use scores from these assessments to assign dementia diagnoses based on a clinical threshold, these or similar data could also be used to explore cognitive changes that occur below the diagnostic threshold, offering a more comprehensive picture of the impact of air pollution on cognition.  However, such cognitive testing is limited to smaller samples.  

Putting it together 
 
Across all these approaches, the trade-off so familiar to environmental and health researchers is clear: increased precision often comes at the cost of generalizability—and a literal cost, requiring funding, personnel, and typically resulting in smaller sample sizes. Conversely, approaches that maximize sample size and generalizability often sacrifice precision and nuance. But each strategy can teach us something different. It shapes what we see and who we see. Studies using fixed monitors or administrative data can cover large populations over long periods but may miss or over-represented people with certain demographic characteristics or fail to consider potential heterogeneous effects. Clinical assessments of cognition offer detailed insights into cognitive change but are limited in sample size and often the length of the period of study. Each approach reveals a different component of the relationship between air pollution and cognition, and the fuller picture emerges when we consider these components together.  

Despite the different measurement strategies and disciplinary approaches of the cited studies, a pattern emerges: air pollution exacerbates cognitive decline. As someone trained in economics but working alongside sociologists and public health scholars, I find this convergence compelling. It also reminds me that how we measure things—what we see and what we miss—shapes the kinds of solutions we imagine. I think this insight itself calls for more interdisciplinary work on this subject—work that CACHE is here to support.  

References:  

Adebayo, T. and Arasu, S. (2025) ‘Scientists raise concerns as the US stops sharing air quality data from embassies worldwide’, AP News, 5 March. Available at: https://apnews.com/article/us-air-quality-monitors-8270927bbd0f166238243ac9d14bce03 

Bishop, K.C., Ketcham, J.D. and Kuminoff, N.V., 2023. Hazed and confused: the effect of air pollution on dementia. Review of Economic Studies, 90(5), pp.2188-2214. 

Livingston, G., Huntley, J., Sommerlad, A., Ames, D., Ballard, C., Banerjee, S., Brayne, C., Burns, A., Cohen-Mansfield, J., Cooper, C. and Costafreda, S.G., 2020. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. The Lancet, 396(10248), pp.413-446. 

Shaffer, R.M., Blanco, M.N., Li, G., Adar, S.D., Carone, M., Szpiro, A.A., Kaufman, J.D., Larson, T.V., Larson, E.B., Crane, P.K. and Sheppard, L., 2021. Fine particulate matter and dementia incidence in the adult changes in thought study. Environmental Health Perspectives, 129(8), p.087001. 

Shi, L., Zhu, Q., Wang, Y., Hao, H., Zhang, H., Schwartz, J., Amini, H., van Donkelaar, A., Martin, R.V., Steenland, K. and Sarnat, J.A., 2023. Incident dementia and long-term exposure to constituents of fine particle air pollution: A national cohort study in the United States. Proceedings of the National Academy of Sciences, 120(1), p.e2211282119. 

University of Michigan. (n.d.) Health and Retirement Study: Cognition Data. Available at: https://hrs.isr.umich.edu/data-products/cognition-data 

Wallace, L.A., Pellizzari, E.D., Hartwell, T.D., Whitmore, R., Sparacino, C. and Zelon, H., 1986. Total Exposure Assessment Methodology (TEAM) Study: personal exposures, indoor-outdoor relationships, and breath levels of volatile organic compounds in New Jersey. Environment International, 12(1-4), pp.369-387.