Short-term exposure to air pollution and mental disorders: a case-crossover study in New York City

A meta-analysis of global epidemiological studies found that 17.6% of respondents met criteria for a mental disorder in the preceding 12 months, whereas 29.2% met criteria at some point in their lifetimes (Steel et al 2014). Although substantial advances have been made in research focusing on clinical and pharmacological/psychosocial interventions to prevent and treat common mental disorders, our understanding of risk and protective factors for these conditions remains incomplete. Despite a long-standing interest in the role of psychosocial environmental factors, such as stressful life events and social support (Brown and Harris 1989, Maulik et al 2010), the potential role of the physical environment has gained attention only recently (Basu et al 2018, Chen et al 2018, Liu et al 2021, Yoo et al 2021, Zhang et al 2021).

In terms of the physical environment, there is growing evidence suggesting that exposure to air pollutants is associated with increases in emergency room (ER) visits for mental disorders. For example, Szyszkowicz et al (2009, 2016, 2018), Cho et al (2014), Oudin et al (2018), Brokamp et al (2019) and Thilakaratne et al (2020) found that short-term exposure to air pollution increased the frequency of ER visits related to mental disorders, whereas other studies (Power et al 2015, Kim et al 2016, Pun et al 2017, Vert et al 2017) found that long-term exposure of air pollution is associated with mental illness. However, several studies have failed to find a statistically significant association between fine particulate matter (PM_2.5) and mental disorder related health outcomes (Wang et al 2014b, Chen et al 2018, Li et al 2020). Similarly, inconsistent findings were reported about ozone (O₃) as some studies (Szyszkowicz 2007, Lim et al 2012, Nguyen et al 2021) found positive and significant associations between O₃ and mental disorders and others found no associations (Wang et al 2014b, Oudin et al 2018). Such inconsistent findings across studies on the effects of PM_2.5 and O₃ exposures might be related to these studies relying on city-wide average air pollutant concentration estimates based on fixed monitoring stations. These city-wide averages fail to account for the spatial variation in pollutants across the neighborhoods within cities. As is recognized by many (Chen et al 2018, Song et al 2018, Bai et al 2020, Li et al 2020, Lowe et al 2021), fixed monitoring station-based exposure estimation cannot provide accurate exposure to PM_2.5 or O_3, and thus likely leads to exposure misclassification. Moreover, previous studies mainly used city-level time-series or cross-sectional designs, which suffer from ecological bias and yield findings that are not directly applicable to individual level (Brokamp et al 2019).

Additionally, the mixed results potentially imply the role of other variables that modify the impact of air pollution on mental health outcomes. For example, some studies have found that the effects of air pollution exposure on mental health vary by season, as well as by individuals' age, sex, and race/ethnicity (Chen et al 2018, Li et al 2020, Lu et al 2020, Thilakaratne et al 2020, Lowe et al 2021). A recent cross-sectional study in the Netherlands by Klompmaker et al (2019) has also shown that individuals' surrounding greeneries were associated with decreased psychological distress, whereas a related study by Yoo et al (2022) in New York City (NYC) found that the effect of greenspace exposure on mental health can be moderated by the socioeconomic conditions of the neighborhoods in which individuals reside. Lastly, mixed results from previous studies might be related to the nature of the particular mental disorders represented in these studies. For example, both Gao et al (2017) and Lu et al (2020) found no significant associations between PM_2.5 exposure and total mental disorders, although they found significant associations with specific mental disorders, such as schizophrenia, mood disorders, and anxiety. These previous studies raise the possibility that the effects of short-term air pollution exposure on mental health might vary not only by individuals' demographic characteristics, but also by where they live and the surrounding greenery that they are exposed to and the specific type of mental disorder.

The present study was designed to investigate the impact of short-term exposure to daily ambient PM_2.5 and O₃ assessed at fine spatial resolutions (i.e. 1 km or less) on ER visits related to mental disorders using a case-crossover design. In addition to examining the impact of acute PM_2.5 and O₃ exposure on ER visits related to mental disorders as a whole, we also examined the risk of ER visits for specific mental disorders. We further tested whether the impact of PM_2.5 and O₃ varied by individuals' age, sex, and race/ethnicity, and if the associations are modified by the degree of visibility of greenspace at individuals' residence.

2.1. Study area and participant

2.2. Environmental exposure assessment

2.3. Statistical analysis

Case-crossover analysis is widely used to estimate acute associations of time-varying exposures, including ambient air pollutant concentrations and extreme temperature, with daily mortality or morbidity in a particular geographic region (Janes et al 2005, Armstrong et al 2014). Given a sample of subjects who experienced the event ('case'), exposure associated with the event is compared with exposure at 'control' (or 'referent') times. Because the case-crossover design makes within-subject comparisons, time-independent confounders (e.g. socio-economic conditions, pre-existing health conditions) are well controlled, but also the effects of time-dependent confounders can be effectively controlled by design if the referent times are matched with index time with respect to time-dependent confounders (e.g., the referent time periods are restricted to the same season).

In the present study, we used a case-crossover design with a conditional logistic regression model to estimate the association between environmental exposure and mental disorder-related ER visits. For each individual, the degree of PM_2.5 and O₃ exposures on the day when the individual visited the ER for mental disorders ('case' day) was compared with the exposure estimates on the other days before and after the visit (the same day of week in one and two weeks prior to and after the ER visit), referred to as 'control' days hereafter. By selecting control days from the same day of week before and after the case day, we assessed the associations between changes in individuals' exposure to air pollutants ( PM_2.5 and O₃) and their mental health outcomes while effectively controlling for long-term temporal trends, seasonality, day of the week, and potential time-invariant confounders, such as the individual's age, sex, and other fixed participant characteristics (Janes et al 2005, Carracedo-Martínez et al 2010).

In the conditional logistic regression model (see appendix C for the model specification), we investigated PM_2.5 and O₃ at both single day lags from the day of ER visit (lag 0) up to 6 d (lag 6) and moving average lags for the period up to 6 lagged days (lag 0–6). The four sets of measurements estimated for control days (one and two weeks before and after the ER visit) were also included. We further adjusted for potential time-varying confounders by using natural cubic splines with 6 degrees of freedom (df) for ambient daily temperature and 3 df for dew point temperature. To account for potential cumulative effects of ambient temperature and dew point temperature, we used 6 (lag 0–6) and 3 (lag 0–3) day moving averages, respectively.

Given that the greenspace exposure metric is time invariant, we did not adjust for GVI but rather assessed it as a potential effect modifier for the PM_2.5 and O₃ effect on mental health outcomes. Specifically, we assessed whether the association between PM_2.5 and O₃ exposure and mental disorder-related ER visits differ by varying levels of greenness at the participant's residence using stratified analyses by modeling the PM_2.5- and O₃-ER visits for mental disorder for each of the four quartiles of GVI in the study area. The four quartiles (0–25th, 25–50th, 50–75th, 75–100th percentile of GVI) were chosen based on a previous study (Qiu et al 2021), although we also assessed effect modification by high ($\gt$50th percentile) and low ($\leqslant$50th percentile) levels of greenspace exposure to ensure the robustness of our results.

We also conducted stratified analyses for PM_2.5 and O₃ exposure effects on ER visits for mental disorders to examine effect modification by sex (female and male), age groups (below 18 years old, 19–34 years old, 35–49 years old, 50–64 years old, and above 65 years old), and race/ethnicity (non-Hispanic Black (Black), non-Hispanic White (White), non-Hispanic Asian (Asian), and Hispanic).

The statistical significance of effect modification was tested by calculating the 95% confidence interval (95% CI) of the difference between the effect estimate within each pair of the subgroup as (Zeka et al 2006, Wang et al 2018b, Chen et al 2019):

$$\begin{align} 95\%\,\,\textrm{CI} = ({\hat Q}_1 -{\hat Q}_2) \pm 1.96 \sqrt{({\hat \sigma}_1^2 + {\hat \sigma}_2^2)}, \end{align} \tag{ 1 }$$

where ${\hat Q}_1, {\hat Q}_2$ denote the model coefficients in the conditional logistic regression model for sex, age group, and race/ethnicity, as well as GVI. The corresponding standard errors are denoted as ${\hat \sigma}_1^2, {\hat \sigma}_2^2$, respectively.

All statistical analyses were conducted in R software (version 4.0.2) using survival and spline packages to analyze conditional logistic regression models and gstat and dismo to reconstruct GVI surface.

3.1. Summary of environmental variables and ER visits

A total of 457 755 ER visits were made due to mental health-related illnesses in NYC between 1 January 2010 and 31 December 2016. The characteristics of ER visits are summarized in Table 1. Most patients were 19–34 years of age (141 573, 31.03%), 35–49 years old (122 157, 26.80%) and 50–64 years old (97 761, 21.40%). Relatively smaller percentages of cases were aged 18 or younger (66 089, 14.56%) or 65 or above (28 303, 6.21%). The majority of visits (287 909, 63.06%) were by males. We also found that Black patients made 31.09% (141 498) of mental disorder visits, while 30.84% (140 715) were by White patients, 3.21% (14 614) by Asian patients, 18.37% (83 800) by Hispanic patients, and 16.49% (75 256) by 'Others'. Regarding the specific type of mental disorders, ER visits related to psychoactive substance use were the most frequent (43.08%), followed by anxiety disorders (19.79%), mood disorders (17.64%), psychotic disorders (10.64%) and dementia (0.83%).

Table 1. Characteristics of emergency room visits (New York City, 2010–2016).

Characteristics	Total mental disorders	Substance abuse	Anxiety	Mood disorders	Psychotic disorders	Dementia
Total counts
N	457 755	196 507	90 642	81 086	46 243	3816
Age group (%)
0–18	14.56	3.72	17.73	22.57	4.15	0.08
19–34	31.03	29.78	36.55	32.32	35.34	0.31
35–49	26.80	32.47	24.07	24.30	30.43	1.21
50–64	21.40	29.33	14.91	16.31	23.54	9.30
above 65	6.21	4.69	6.73	4.51	6.53	89.10
Sex (%)
Female	36.94	20.89	59.10	52.62	34.66	56.60
Male	63.06	79.11	40.90	47.38	65.34	43.40
Race/ethnicity (%)
White	30.84	34.28	34.28	28.95	20.66	38.00
Black	31.09	26.02	27.67	31.35	49.06	32.63
Asian	3.21	2.75	3.97	3.53	3.37	3.77
Hispanic	18.37	18.65	18.42	19.91	13.35	13.84
Others	16.49	18.30	15.66	16.26	13.56	11.77

Air pollutant concentrations over the study area are summarized in Table 2. During the study period, the daily average of PM_2.5 was 8.85 $\mu\textrm{g}\,\textrm{m}^{-3}$ with a standard deviation of 8.65 $\mu\textrm{g}\,\textrm{m}^{-3}$ and the daily mean of O₃ was 37.10 ppb with a standard deviation of 10.62 ppb. The daily average temperature in the study region varied from −10.31 ^∘C to 30.65 ^∘C with a mean of 13.27 ^∘C and a standard deviation of 8.67 ^∘C. The spatial distribution of GVI ranged between 0.00 and 92.45 with the mean of 18.25 and a standard deviation of 15.37. The spatial distributions of air pollutants (PM_2.5 and O₃) and green view index are presented in Figure 1.

Zoom In Zoom Out Reset image size

Table 2. Summary statistics of daily ambient PM_2.5, temperature and green view index (New York City, 2010–2016).

	Mean	SD	Min	Q25	Median	Q75	Max	IQR
PM2.5 ($\mu\textrm{g}\,\textrm{m}^{-3}$)	8.85	8.65	0.00	4.90	7.42	11.32	99.06	6.42
O3 (ppb)	37.10	10.62	13.50	29.14	35.44	44.05	78.34	14.91
Temperature (∘C)	13.27	8.67	−10.31	5.23	13.39	20.63	30.65	15.40
Dew point temperature (∘C)	6.46	10.31	−28.14	−1.13	7.77	15.33	24.62	16.46
Green view index	18.25	15.37	0.00	6.08	14.05	26.76	92.45	20.69

SD: standard deviation.Q₂₅, Q₇₅: the first and the third quartiles.IQR: inter quartile range ($\textrm{Q}_{75}$–$\textrm{Q}_{25}$).

3.2. Statistical analyses

We examined the associations between two daily air pollutants of PM_2.5 and O₃ exposures and ER visits related to mental disorders using 457 755 ER visits that occurred in NYC, USA, between 2010 and 2016. To our knowledge, there is only one prior study to examine the adverse effects of the short-term air pollution exposure on broadly defined mental disorders, as well as specific mental disorders, through a case-crossover design using spatially and temporally resolved PM_2.5 concentration and temperature estimates (Brokamp et al 2019). However, this prior study focused only on a single air pollutant (PM_2.5) and children/adolescents, whereas our results suggest that the effects of both PM_2.5 and O₃ vary by age groups. Specifically, our results indicated that the mental-disorder related ER visits increased by 2.78% (95% CI: 1.82%–3.76%) with a 10 $\mu\textrm{g}\,\textrm{m}^{-3}$ increase in ambient PM_2.5 concentration on a moving average lag 0–3 d and 0.71% (95% CI: 0.28%–1.15%) with 10 ppb increase in ambient O₃. Furthermore, this effect was modified by age and race/ethnicity, but not by sex or surrounding greenery.

We first examined the effect of daily PM_2.5 and O₃ exposures on ER visits for any of the mental disorders examined in the present study. A significant positive association was found between ER visits for mental disorders and daily air pollutants of both PM_2.5 and O₃ exposures. The association of increased treatment services for mental disease with high levels of ambient PM_2.5 concentration is consistent with previous studies on both short-term (Gao et al 2017, Chen et al 2018, Song et al 2018, Lee et al 2019, Li et al 2020) and long-term (Power et al 2015, Kim et al 2016, Lin et al 2017, Pun et al 2017, Ran et al 2021) associations. Similarly, the positive and significant effect of O₃ on total mental disorders is generally consistent with findings from previous studies (Szyszkowicz 2007, Szyszkowicz et al 2016, Kioumourtzoglou et al 2017).

Our findings of significant effects of PM_2.5 exposure on psychotic disorders, substance abuse, and anxiety disorders agree with previous studies in China (Gao et al 2017, Bai et al 2020) and Canada (Szyszkowicz et al 2018). However, our results did not demonstrate significant associations between short-term exposure to PM_2.5 and ER visits related to mood disorders or dementia, while significant associations were reported in previous studies (Gao et al 2017, Wang et al 2018a, Peters et al 2019, Qiu et al 2019, Lu et al 2020, Shi et al 2020, Wei et al 2020). Similarly, we found significant effects of O₃ on both mood disorders and psychotic disorders but not on substance abuse, anxiety disorders, or dementia. Although several past studies have also reported positive associations between O₃ and mood disorders (Szyszkowicz 2007, Szyszkowicz et al 2016, Nguyen et al 2021), null results have also been reported (Oudin et al. 2018, Wang et al., 2014b). One potential source of these inconsistent findings lies in differences in the source of health outcomes of previous studies, as some studies focused on hospital admissions (Qiu et al 2019), surveys (Shi et al 2020), or daily hospital visits (Wei et al 2020). Given that there is very limited research on the effects of air pollution exposure on specific mental disorders, further investigation is needed to corroborate our findings and to explain why air pollution was associated with exacerbations of symptoms leading to ER visits for some types of mental disorders, but not for others.

The underlying mechanisms by which air pollutants impact emergency room visits for psychiatric disorders are mostly unknown. Some previous studies reported that air pollutants may affect the mental status through their impact on inflammatory processes, the immune system, oxidative stress, and cerebral neurotransmitter concentrations (Sirivelu et al 2006, Van Berlo et al 2010, Thomson 2019). Despite growing evidence that air pollution is significantly associated with psychiatric diseases (Szyszkowicz et al 2009, Gao et al 2017, Kim et al 2019), it is unclear if the processes underlying the relationship between exposure to air pollutants and mental illness reflects a shared process that cuts across mental disorders or that there are processes that are unique to particular disorders. Similarly, the associations between different air pollutants and particular mental disorders varies per study. For example, Thilakaratne et al (2020) found a negative association between ER visits for substance abuse and NO₂ in California, whereas the present study showed a positive association between PM_2.5 exposure and substance abuse in NYC. Increased understanding of key mechanisms will be needed to identify risk factors that contribute to susceptibility.

In the age-specific analyses, we found that associations between both air pollutants (PM_2.5 and O₃) and ER visits were more pronounced among the age group 19–34 years old relative to other age groups. For PM_2.5 only, the youth (18 years old or below) and 55–64 years were more susceptible than other age groups. In contrast, previous studies reported that older adults, particularly those aged 64 years and older, were more vulnerable in studies based in China (Li et al 2020) and South Korea (Cho et al 2014, Kim et al 2019). Although Thilakaratne et al (2020) observed that youth (below 18 years old) were vulnerable to homicide/inflicted injury related emergency department visits following exposure to high levels of carbon monoxide, this association was not significantly different from other age groups. To our best knowledge, there is only one other study (Brokamp et al 2019) reporting that young people were particularly vulnerable to the effects of ambient PM_2.5 exposure on risk for mental disorder outcomes.

In terms of greenspace exposure, we found no evidence that the degree of greenspace visibility surrounding individuals' residence made individuals more or less susceptible to the effects of exposure to air pollution. That is, the percent change in ER visits related to total mental disorders associated with PM_2.5 exposure was not significantly different for varying levels of greenspace exposure. However, it is possible that unique aspects of NYC may have had an impact on these results. While it seems likely that in many urban environments, greenery is positively correlated with social advantages, this is not the case in NYC. Neighborhoods in the highly affluent areas of lower and mid-town Manhattan tend to be surrounded by skyscrapers and concrete, whereas most neighborhoods in the less affluent boroughs of the Bronx and Staten Island are surrounded by thick trees at the street level. These mixed conditions in greenspace of NYC may have contributed to the elevated susceptibility to air pollution exposure in some areas, and the opposite effect in others.

The main strengths of this study are as follows: first, the use of spatially resolved daily PM_2.5 and O₃ exposure estimates enabled us to overcome the limitations of existing studies where city- or region-wide coarse air pollution estimates were used as a proxy measurement. In addition, the use of the large sample of ER visits in NYC also enabled us to investigate specific mental disorders, and assess differentiated risk for specific disease, such as substance abuse and psychotic disorders. Third, our individual-level case-crossover design automatically adjusted for time-invariant confounding and time-varying meteorological factors (ambient temperature) were controlled for in the models, providing stronger inference in potential causality than ecological study designs such as time-series and cross-sectional studies.

Our study is not short of limitations. First, spatial patterns of ER visits and their associations with socio-economic conditions have not been taken into account, and these potentially could impact individuals' reaction to PM_2.5 exposure. Second, the present study only examined PM_2.5 and O₃, and it is possible that other pollutants such nitrogen dioxide and sulfur dioxide have different associations with ER visits for mental disorders. Given the potential unique aspects of New York City, further studies in other regions are needed to confirm our findings. Perhaps of most importance, it will be critical for future studies to investigate the underlying mechanisms by which exposures to PM_2.5 and O₃ is linked to ER visits for mental disorders.

This study suggests that there is a positive association between short-term exposure to ambient fine particulate matter and increased ER visits for total mental disorders, as well as for substance abuse and psychotic disorders in particular. Our findings of higher vulnerability to air pollution among the specific age groups (i.e. 19–34 years old and youth) and non-Whites warrant follow-up research to better understand why and how these individuals are particularly susceptible.

The data generated and/or analysed during the current study are not publicly available for legal/ethical reasons but data are available from the corresponding author on reasonable request.

We investigated delayed effects beyond lag 0 to lag 6 by calculating the percent changes in the risk of visiting ER for total mental disorders over different days for both single day and accumulated day (moving average) lag. To account for the delayed effects of the two air pollutants simultaneously, we considered 13 different lags {lag 0, lag 1$, \ldots,$ lag 6, lag 0–1, lag 0–2$, \ldots,$ lag 0–6} for each pollutant, which resulted in a total of 169 different combinations of lag specifications for PM_2.5 and O₃.

The results are summarized in figure A.1 where the shade represents the mean of percent changes of RRs for different combinations of lag specification for PM_2.5 (left panel) and O₃ (right panel) overlapped with circles to denote the lag specifications where both PM_2.5 and O₃ were statistically significant. The results suggest that all the accumulated day (moving average) lags yielded high RR for PM_2.5 (left panel), but the maximum percent change of RR was associated with the single day lag (lag 1) and moving average lag 0-6 for O₃ (right panel). When both PM_2.5 and O₃ were taken into account simultaneously, however, we found only a subset (23 combinations) of lag specifications out of 169 combinations (13 × 13) were significant, and the moving average lag 0–3 d for PM_2.5 and single day lag (lag 1) for O₃ yielded the highest RR value (2.78, 95% CI (1.82, 3.75) for PM_2.5; 0.72, 95 % CI (0.29, 1.15) for O₃). Thus we selected moving average lag 0–3 and single day lag (lag 1) as the main lag for PM_2.5 and O₃, respectively, throughout the analysis.

Zoom In Zoom Out Reset image size

We examined the sensitivity of the risk of visiting ER related to mental disorders associated with the exposure to ambient PM_2.5 and O₃ by varying levels of greenspace surrounding individuals' residence. Specifically, we estimated the percent change in RR of ER visits related to total mental disorders from exposure to ambient PM_2.5 and O₃ by quartile of GVI measured at one's residence. The results summarized in Table B.1 suggested that the changes in risk are similar to each other in the range of 2.06 and 3.08 for PM_2.5 and 0.15 and 1.37 for O₃ regardless of the levels of surrounding greeneries. No statistically significant differences were found between any pair of levels of greenness.

We also conducted a stratified analysis based on 50th percentile of GVI values, and we found that the percent change (2.56, 95% CI (1.2, 3.94)) in risk at high ($\gt$50th percentile) levels of greenness is not significantly different from that (3.01, 95% CI (1.6, 4.38)) at the low ($\leqslant$50th percentile) level of greenness for PM_2.5. Similarly, no significant difference was found between the percent change (0.76, 95% CI (0.15, 1.37)) in risk at high levels of greenness and that (0.66, 95% CI (0.05, 1.27)) at the lower level of greenness for O₃, respectively.

The model formulation for the conditional logistic model with a two-pollutant model of PM_2.5 and O₃ can be written as:

$$\begin{align} Y_{i,j} & \approx Bernoulli (\pi_{i,j}) \nonumber\\[6pt] \pi_{i,j} & = \frac{\exp \{\alpha + \beta_1 PM_{s_i,j} + \beta_2 O_{s_i,j} + ns(Temp_{s_i,j}, \textit{df}_1) + ns(Dew_{s_i,j}, \textit{df}_2) \}}{\sum_{j^{^{\prime}} \in k(j)} \exp \{\alpha + \beta_1 PM_{s_i,j^{^{\prime}}} + \beta_2 O_{s_i,j^{^{\prime}}} + ns(Temp_{s_i,j^{^{\prime}}}, \textit{df}_1) + ns(Dew_{s_i,j^{^{\prime}}}, \textit{df}_2) \}}, \end{align}$$

$Y_{i,j}$ indicates whether the event that the ith individual' ER visit for mental disorder occurs in time interval j in the pre-specified reference window k(j) ($Y_{i,j}$ = 1, event; 0, not). The set of reference periods is denoted as $k(j) = \{j-14, j-7, j, j+7, j+14\}$. The spatially and temporally resolved exposures at the location s_i (i.e. the residence of ith individual) and time j (day) for air pollution (PM_2.5 and O₃) and temperature, and dew point temperature are denoted as $PM_{s_i,j}, O_{s_i,j}, Temp_{s_i,i}, Dew_{s_i,j}$, respectively. The smoothing function is denoted as ns with the corresponding degree of freedom $df_1, df_2$ for temperature and dew point temperature, respectively. To estimate relative risk (RR), we reestimated the model coefficients ($\hat \beta_1, \hat \beta_2$) with the different sets of data $Y_{i,j}$.