Associations between acute exposures to PM_2.5 and carbon dioxide indoors and cognitive function in office workers: a multicountry longitudinal prospective observational study

A growing body of evidence is showing associations between air pollution and impaired cognitive function. Outdoors, evidence is starting to emerge that chronic exposure to fine particulate matter (PM_2.5) may also impact cognitive performance through impacts to the central nervous system (CNS) (Block and Calderón-Garcidueñas 2009). Ambient air pollution can penetrate indoors, where most exposures can occur due to the high percentage of time spent indoors. The extent to which ambient pollution penetrates and accumulates indoors is influenced by ventilation, infiltration, and filtration (Ben-David and Waring 2016). Consequently, building design and operations can affect indoor exposures to CO₂ and PM_2.5, impacting building occupants' cognition.

Researchers have recently identified associations between exposure to PM_2.5 and various effects on the CNS and neurodegeneration (Block and Calderón-Garcidueñas 2009). A hypothesized mechanistic pathway is that inhalation of PM_2.5 may lead to a level of chronic, systemic inflammation that impacts the CNS. This hypothesis is supported by animal studies that have detected inflammatory biomarkers in the brain after exposure to PM_2.5 and ultrafine PM, as they may be capable of crossing the blood–brain barrier (Campbell et al 2005). Epidemiological studies among older adults have found that chronic exposure to outdoor PM_2.5 is associated with a higher prevalence of neurocognitive development outcomes (Freire et al 2010) and neurodegenerative diseases in adults (Jayaraj et al 2017), the decline of cognitive function among older adults (Power et al 2011, Weuve 2012). More recently a short-term association of PM_2.5 and cognitive function among older men in the Normative Aging Study was reported (Gao et al 2021). Effects of PM_2.5 on neurocognitive performance have also been found in younger populations, including fetuses (Guxens et al 2014), school-age children (Saenen et al 2016), and adolescents (Kicinski et al 2015).

Inside buildings, other pollutants beyond PM_2.5 are often found in higher concentrations than outdoors due to the limited dilution capacity of enclosed spaces. Controlled environmental exposure studies among office workers have found independent and monotonic effects of same-day CO₂ concentrations, as well as effects of CO₂ as a subrogate of ventilation rates, on several domains of cognitive function (Satish et al 2012, Allen et al 2016). The mechanism explaining the effects of CO₂ as an independent pollutant or as a proxy for ventilation rates on cognitive function at relatively low concentrations (<3000 ppm) is still not fully understood. It is hypothesized that elevated CO₂ in the bloodstream resulting from ambient exposure (Vehvilainen et al 2016) results in sympathetic activation of the autonomic nervous system (MacNaughton et al 2016). Activation of the sympathetic nervous system influences cognitive performance, including strategy and working memory (Starcke and Brand 2012).

Despite the growing evidence of the effects of air pollution on cognitive function, most studies have focused on chronic exposures, exposures during early life or among seniors, and outdoor exposures rather than indoor exposures, where people spend 90% of their time (Klepeis et al 2001). There is scarce information about the acute effects of indoor air pollution on cognitive function among the working populations or whether the effects are observed even at low exposure levels. These questions are especially relevant for the working-age population since even small effects in cognitive function could translate into substantial short- and long-term economic and productivity losses. To investigate whether cognitive function was associated with real-time indoor concentrations of PM_2.5 and CO₂, we measured the cognitive performance and indoor air quality among office workers in six countries during a 12 month prospective longitudinal cohort study.

2.1. Study design

2.2. Study population

Companies with office buildings were eligible to participate in the Global CogFx study if they had ten or more employees and agreed to install a network of indoor environmental sensors. These companies either owned or leased a portion of the building or in some cases the entire building. Companies in knowledge work fields, such as architecture, technology, real estate investment, coworking, and engineering, expressed interest in participating. In total, 43 buildings were recruited; this analysis includes 42 buildings since one building in the US was excluded because the sensor package installed did not include a PM_2.5 sensor. Table 1 shows details on the participation levels per country.

Table 1. Building and participant characteristics by country.

Variable	All countries	China	India	Mexico	Thailand	UK	USA
Personal demographics
Participants	302	34	68	8	10	49	133
Age n (%)
18–29	110 (36.4)	16 (47.1)	29 (42.6)	3 (37.5)	4 (40)	23 (46.9)	35 (26.3)
30–39	125 (41.4)	16 (47.1)	29 (42.6)	4 (50)	6 (60)	14 (28.6)	56 (42.1)
40–49	43 (14.2)	2 (5.9)	7 (10.3)	1 (12.5)	0 (0)	11 (22.4)	22 (16.5)
50–59	22 (7.3)	0 (0)	3 (0)	0 (0)	0 (0)	1 (0)	18 (0)
60+	2 (0.7)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	2 (0)
Job category n (%)
Managerial	62 (20.5)	5 (14.8)	22 (32.4)	2 (25)	—	8 (16.3)	26 (19.4)
Professional	162 (53.6)	9 (26.5)	6 (38.2)	6 (75)	5 (50)	28 (57.1)	91 (68.7)
Technical	48 (15.9)	9 (26.5)	14 (20.6)	—	3 (33.3)	10 (20.4)	7 (5.2)
Secretarial or clerical	22 (7.3)	10 (29.4)	4 (5.9)	—	2 (16.7)	1 (2)	7 (5.2)
Other	8 (2.6)	1 (3)	2 (3)	—	—	2 (4.1)	2 (1.5)
Educational level n (%)
Incomplete college	14 (4.7)	4 (11.8)	4 (5.9)	—	—	4 (8.1)	2 (1.5)
2 year degree	9 (3)	1 (3)	3 (4.4)	—	—	2 (4.1)	3 (2.2)
4 year degree or professional degree	149 (49.2)	14 (41.2)	30 (44.1)	8 (100)	10 (100)	19 (38.7)	72 (54.4)
Graduate degree	130 (43.1)	15 (44.2)	31 (45).6	—	—	24 (48.9)	56 (41.8)
Buildings characteristics
n	42	8	10	1	1	6	16
Cities (# buildings)	30	Chengdu (3), Shanghai (3), Zhuhai (2)	Bengaluru (5), Chennai (1), Gurugram (1), Hyderabad (1), Mumbai (1), Pune (1)	Culiacan (1)	Bangkok (1)	Croydon (2), Birmingham (1), Cambridge (1), London (1), Sheffield (1)	LA (2), San Francisco (2), Boston (1), Clearwater (1), Chicago (1), Cleveland (1), Denver (1), Minneapolis (1), NYC (1), Omaha (1), Overland Park (1), Phoenix (1), Seattle (1), Washington DC (1)

In each building, recruitment was conducted on a rolling basis until the building recruitment target of ten participants was met. The research team tried to ensure that participants within a building were balanced in terms of age and sex. Participant eligibility criteria included being between 18 and 65 years of age, speaking one of the languages in the study (English, Chinese, or Spanish), having a smartphone compatible with the study's research app, having a full-time position at the company, have a permanent workstation in the office and working there for at least 3 days a week, not being colorblind, and not being a current smoker. The analysis was limited to participants with two or more tests to eliminate any 'first test' effects where participants were getting familiar with the app and test format. Limiting the analysis to those with two or more tests allows us to consider the first test as training and reduce a potential confounding effect of test learning with the exposure (Goldberg et al 2015). Figure 1 shows the number of participants that met the eligibility criteria for this analysis. Our study protocols were General Data Protection Regulation (GDPR)-compliant and were reviewed and approved by the Institutional Review Board at the Harvard T.H. Chan School of Public Health.

Zoom In Zoom Out Reset image size

2.3. Air quality assessment

2.4. Cognitive function assessment

2.5. Statistical methods

Table 1 summarizes the participant information at the building level. In general, the study had balanced participation by gender (Females = 58%), despite some countries where females were overrepresented (e.g. China and Thailand). Most of the participants (92.3%) had a high educational level, either with a 4 year college degree or graduate education. Table 2 provides summary statistics of the indoor environmental data. Between countries, we found more similarities in the levels of CO₂ than PM_2.5. There was little variance in exposure (i.e. UK) and no overlap between the countries' IQR in some cases. Regarding temperature and relative humidity, the values are largely within previously observed values in indoor environmental quality investigations of office buildings (Apte et al 2000, MacNaughton et al 2017).

Five cognitive performance metrics of office workers were significantly associated with real-time indoor concentrations of PM_2.5 and CO₂ (figure 2). For Stroop, response time was significantly associated with CO₂ and PM_2.5 concentrations. An interquartile (IQR) increase in CO₂ (IQR = 315 ppm) was associated with a 1.92% (95% CI: 1.39, 2.46) and a 0.85% (95% CI: 0.32, 1.39) increase in response time for the unadjusted and adjusted Stroop models, respectively. For PM_2.5, an IQR increase (IQR = 8.8 $\,\mu$g m⁻³) was associated with an increase of 0.78% (95% CI: 0.39, 1.17) and 0.82% (95% CI: 0.42, 1.21) in the unadjusted and adjusted models of Stroop response time. The interference effect, measured as the difference between congruent and incongruent response times, was associated with both air quality parameters. For each IQR increase in CO₂ and PM_2.5, the adjusted model estimates showed a 7.88% (95% CI: 2.08, 13.86) and a 6.18% (95% CI: 2.08, 10.3) increase in interference time, respectively. Regarding Stroop throughput, an IQR increase in CO₂ was associated with a 1.32% (95% CI: −2.3, −0.38) decrease in the adjusted model; an IQR increase in PM_2.5 had a decrease of 0.7% (95% CI: −1.38, −0.01) in Stroop throughput. When evaluating non-linearity in the association between exposure and cognitive performance metrics, we found a significant non-linear association between CO₂ and Stroop response time and Stroop throughput (figure 3). We observed a linear association between 400 and 1200 ppm. The association is not well-defined after 1200 ppm due to the low number of measurements collected above that concentration (CO₂ 95th percentile = 1363.0 ppm).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For the ADD test, increases in CO₂, but not in PM_2.5, were significantly associated with slower response times. For an IQR increase in PM_2.5, there was a 0.66% (95% CI: −0.16, 1.49) non-significant increase in the adjusted model ADD response time. For each IQR increase in CO₂, there was a 1.13% (95% CI: 0.18, 2.11) increase in the adjusted ADD response time. For ADD throughput, an IQR increase in PM_2.5 was significantly associated with lower ADD throughput (−1.51%, 95% CI: −2.65, −0.37); an IQR increase in CO₂ resulted in a non-significant decrease 0.72% decrease in ADD throughput (95% CI: −2.11, 0.63).

Table 3 reports the stratified associations between low (<12 $\,\mu$g m⁻³) and high (⩾12 μg m⁻³) PM_2.5 levels and Stroop and ADD test performance. At PM_2.5 levels equal to or above 12 μg m⁻³, an increase in PM_2.5 concentrations was significantly associated with lower performance on four out of five cognitive test outcomes. Overall, the effects at low PM_2.5 levels were similar directionally and magnitude to those at high concentrations, although none of the associations were statistically significant.

The standard deviation, intraclass correlation of the multi-level random effects, and marginal and conditional R-squared values from the five main adjusted models are presented in table 4. The between-participant variance is higher than the between-building variance in all models. Also, the ICC at the participant level is consistently higher than at the building level, suggesting that the clustering between participants accounts for a larger proportion of the variance than the between-building clustering. The fixed effects explained a larger proportion of the outcome variability in the Stroop test than in the ADD test. The random-effects had more explanatory power for all models than the fixed effects, which is supported by the higher variability observed in the between-participants random intercept.

This 1 year longitudinal study investigated the acute effects of low-level indoor PM_2.5 and CO₂ exposures on the cognitive function of 302 office workers in six countries around the world. We found significant associations of selective attention (Stroop) and processing speed (ADD) with exposure to these pollutants indoors, at concentrations commonly observed indoors.

PM_2.5 levels indoors result from a combination of indoor and outdoor sources. Indoor sources of PM_2.5 include resuspension of settled dust, indoor smoking, cooking, and secondary generation from cleaning products; however, many indoor sources are less prevalent in office buildings than other building types (Patel et al 2020). PM_2.5 of outdoor origin also penetrates indoors depending on several building-related factors, such as ventilation, infiltration, and filtration. In cities and countries with high levels of outdoor PM_2.5, levels indoors often exceed health-based limits (Cohen et al 2017, Song et al 2017, Balakrishnan et al 2019). In our study, China, India, Thailand, and Mexico, the annual outdoor PM_2.5 mean concentrations exceed the NAAQS annual mean standard of 12 $\,\mu$g m⁻³ and WHO air quality guideline value of 10 $\,\mu$g m⁻³ (Shaddick et al 2018). For comparison, in our study, the median of hourly indoor PM_2.5 concentrations during work hours in India and China were 8.0 and 18.0 $\mu$g m⁻³, respectively. In contrast, the median of hourly concentrations during work hours in the US and the UK was 1.7 and 1.0 $\,\mu$g m⁻³, respectively.

We found that higher indoor PM_2.5 levels were significantly associated with decreased performance in Stroop response time, interference time, and throughput, and lower ADD throughput. Our results are consistent with previously published work. In an experimental study, Shehab and Pope found an association between selective attention from the exposure to candle-generated indoor PM and exposure to traffic pollution (Shehab and Pope 2019). The authors reported no significant effects on the Stroop color-word test, but the study may have been underpowered given their small sample size (n = 33). A study in Belgian classrooms investigating the acute effects of indoor PM_2.5 on cognitive function found a 3% increase in Stroop response times associated with an increase in 8 μg m⁻³ PM_2.5 (Saenen et al 2016). For the same increase in PM_2.5 concentration, we found an increase of 0.8% in Stroop response times. Observing these effects in working-age adults was novel, as research often focuses on children and older adults. It is often hypothesized that these populations are more susceptible to the neurocognitive effects of air pollution due to their developing or declining CNS functions. The higher magnitude of the association in younger children compared to adults might support such an argument.

We found that higher CO₂ levels in office buildings were associated with decreased performance in all Stroop test metrics, and increased ADD response times. Our findings are consistent with and expand upon other studies of the effects of CO₂ and ventilation on cognitive function. Several studies have examined the impacts of ventilation on cognitive performance, using CO₂ as a proxy (Haverinen-Shaughnessy et al 2015, Maddalena et al 2015, Zhang et al 2017), as we did in our study. Haverinen-Shaughnessy and Shaughnessy found significant improvement in arithmetic standardized test scores associated with higher ventilation rates among elementary school students (Haverinen-Shaughnessy et al 2015). Others have investigated the direct effects of CO₂, independent of the ventilation rate (Allen et al 2016, 2019). Allen et al found that short-term exposures to ventilation rates at 20 cubic feet per minute/person and CO₂ above 900 ppm were associated with significantly lower test scores using the strategic management simulation tool (Allen et al 2016). Satish et al also found effects of CO₂, independent of ventilation, when concentrations exceeded 1000 and 2500 ppm relative to a 600 ppm baseline. The effects of PM_2.5 on ADD response time were not statistically significant, suggesting that the exposure had a larger effect on test accuracy than on test speed. Conversely, the association between CO₂ and ADD throughput was not significant. A plausible explanation for these results is that these two environmental exposures have differential effects on attention and processing speed versus working memory, which could also explain the larger effects we report of CO₂ on Stroop interference time.

Due to the limited range of PM_2.5 exposures in the UK and the USA relative to the rest of the countries in the study, we performed a sensitivity analysis evaluating the effects of PM_2.5 below and above the NAAQS primary annual mean standard of 12 μg m⁻³. We found that the associations between indoor PM_2.5 and lower cognitive performance were stronger in magnitude and significance at concentrations above the NAAQS standard annual mean. This result supports the existence of acute effects of PM_2.5 exposures on cognitive function. Nevertheless, it is difficult to determine whether the effects at lower levels are non-existent or if other country-level covariates associated with cognitive test performance (e.g. quality of educational systems) could have subdued the effects found at higher PM_2.5 levels. We also performed a stratified analysis by time of day, gender, job category, and educational level (results not shown). Overall, the findings are robust across the different strata, maintaining the magnitude and directionality of the associations of PM_2.5 and CO₂ with the cognitive tests. However, some of the strata have a limited number of observations, leading to more uncertain effect estimates within strata.

The model diagnostics show that environmental factors explain a significant portion of the variability in cognitive performance, especially on the Stroop test. The fixed effects on the Stroop test account for 40%–50% of the variance explained by the model. While the variance in cognitive performance between individuals was greater than differences between buildings, these results show that the built environment influences acute cognitive performance of similar magnitude to the aggregate interpersonal differences related to education, sleep, diet, and other factors.

Our study has several limitations. First, it did not measure any exposures outside work. While some of the observed effects may be lagging a home or commute exposure, we consider that the close temporal relationship between exposure and outcome in our study makes this possibility less likely. Second, the random effects analysis shows that between-participant variation is very large compared to the effect size observed from the exposures of interest. Despite participant-level covariates associated with the outcome (e.g. age, gender, educational level), the model has a relatively low fit, particularly in the ADD test with larger confidence intervals (table 3) and lower R-squared values (table 4). Third, there are known limitations to the quality of real-time, commercial-grade environmental sensors. Concerns about the instrument quality have been expressed elsewhere (Clements et al 2017). However, since several sensor packages share the same components (same optical particle counter model), we consider that errors might be similarly distributed in different study locations. Moreover, these commercial-grade packages are improving rapidly (Demanega et al 2021).

Our study advances a growing body of literature on the effects of air pollution on cognitive function in several ways. First, the combination of real-time indoor air quality sensors and a mobile research app to administer in-situ cognitive tests allowed us to study exposure and outcome at a high spatiotemporal resolution. Second, compared to experimental approaches with simulated exposures, the observational study design relied on the natural variability of the exposures of interest across countries representing nearly half of the world's population, adding to the exposures' validity (table 2).

We consider that this study adds to the evidence of the acute impacts on cognitive function associated with poor indoor air quality. Benefits on cognitive function from reducing exposures to PM_2.5 and CO₂ indoors, either by filtration or higher ventilation rates, may positively impact productivity, educational attainment, safety, and many other activities where cognitive performance is important. This study also contributes to the body of literature indicating that building performance is associated with cognitive performance. In addition to the well-established health benefits from lower PM_2.5 levels (e.g. reductions in cardiovascular disease, asthma attacks, premature mortality), and from higher ventilation rates (e.g. reduced infectious disease transmission, fewer sick-building symptoms, and reduced absenteeism), our findings provide further incentive to improve air quality in indoor spaces. Higher ventilation rates and enhanced filtration that exceed current minimum targets are important public health strategies.

This work was supported by the National Institutes of Health Grant T32-ES007069, the National Institute of Environmental Health Sciences Grant P30-ES000002, the National Institute for Occupational Safety and Health Grant T42-OH008416, and a gift from Carrier Global Corporation (Carrier). Additional support was provided by Jones Lang LaSalle, Inc. (JLL) and by several companies that participated in this study, whose identities we cannot disclose because it could inadvertently breach participant confidentiality, who provided funds to cover the costs of study equipment for their building(s). Carrier, JLL, and the companies that participated in the study were not involved in study design, data collection, data analysis, data interpretation, data presentation, or drafting of the manuscript.

The data that support the findings of this study are available upon reasonable request from the authors.