Short-term PM_2.5 air pollution and “deaths of despair” in Michigan, 2006-2016: a case-crossover study

Abstract

Introduction

Particulate matter is a known risk factor for cardiovascular and respiratory illness. Ecological analyses have indicated associations for mental health outcomes, but there has been only limited work on the impacts of individual exposure to air pollutants and mental health-related mortality. This research uses a novel case-crossover design to assess associations between short-term lag exposure to particulate matter 2.5 (PM_2.5) and mortality from suicide, alcohol-related liver failure and opioid drug overdose.

Methods

Records for all deaths from suicide, alcohol-related liver failure, and overdose were obtained from the Michigan Department of Health and Human Services (MDHHS) for the years 2006 to 2020. Residential addresses were geocoded, and latitude/longitude locations were added to the database. Gridded raster data of daily PM concentrations at 1km resolution were obtained from the CAFE Climate and Health Research Coordinating Center Collection. PM exposure for all 14 days preceding the recorded date of death was extracted for all decedents at each point. A case-crossover design was used, where everyone served as their own control, staggering exposure days on the same weekday in the same month of the date of death. Lag associations of PM and mortality were then tested using regression models stratified by race and sex within a distributed lag non-linear modelling framework, including the cross-basis for PM exposure.

Results

There were 30,930 mental health-related deaths included in the dataset between 2006-2016. Among these, there were 13,336 suicides, 6,875 deaths from alcohol-related liver failure and 10,719 deaths from overdose. The mean PM_2.5 exposure was 9.65 micrograms per cubic meter. Patterns of association between time and exposure to PM differed between the types of mortality. While no association was statistically significant, exposure to extremely high levels of PM had a positive association with mortality from all three types of mortality at and around lag day 7, but not after. Seasonal analyses indicated that increasing exposure to high levels of PM was significantly associated with suicide in the Fall and Winter months. In the Summer, exposure to any amount of PM over the WHO recommended safe level up to two weeks was associated with increased risk for suicide mortality. Tests for effect modification by Black/white and binary sex groups did not indicate significant differences in mortality risk given exposure.

Conclusions

While we did not find overall statistically significant associations of PM exposure with mental health-related mortality, our results suggest that there might be possible associations of cumulative extreme levels of PM exposure over one week. As suicide, alcohol-related mortality and overdose are the end of a long cascade of events and factors, future work might test lag associations with non-fatal mental health-related episodes or assess individuals’ exposure directly.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-26183-0.

Introduction

Fine ambient particulate matter PM is an air pollutant ubiquitous in both urban and rural environments worldwide [1]. Ambient PM concentrations are expected to increase with climate change related temperature increases [2–4] and urbanization [5]. Exposure to PM is considered to be a major risk to human physical health by the World Health Organization [6] and has been associated with a number of human health impacts including respiratory disease [7, 8], cardiovascular disease [9], lung cancer [10] and diabetes [11]. However, associations between ambient exposure to PM and mental health outcomes are less well understood. This question becomes salient as economic development and changes in weather patterns exacerbate poor air quality and as the overall worldwide burden of mental health increases [12].

There have been several studies of associations of PM with prevalent mental health outcomes such as depression, anxiety and suicidal ideation [13]. One study in California found a significant relationship between PM exposure and mental health-related emergency department visits [14]. Other studies have shown a significant relationship between PM exposure and depression [15] when evaluating behavior [16], controlling for age [17] and stratifying by gender [18]. A study in Denmark suggested that same or previous day exposure to increased levels of PM is associated with increased numbers of psychiatric emergency room visits using a case-crossover design [19]. However, other studies investigating these relationships have found insufficient evidence linking air pollution to depression [16].

Individual biological and psychological factors likely significantly contribute to suicide risk, and ecological factors greatly impact the emergence and development of these individual attributes [20]. One of the primary biological linkages between PM exposure and mental health-related mortality is likely via the activation of the toll-like receptor 3 (TLR3) [21], a pattern recognition receptor within the immune system.

A few papers have tested associations of air pollution with mental health-related mortality other than suicide. When measuring associations between PM exposure and suicide, accounting for seasonal conditions [22] as well as lag adjustment [23], significant results have been reported. Other studies were not able to show links between PM exposure and suicide [18, 24–26]. We could find no research on PM exposure and drug overdose or alcohol-related liver failure even though there is considerable cross-over between overdose, alcohol-related liver failure and suicide. Deaths ruled as drug overdoses may be misclassified suicide attempts or may be the result of suicidal-like thinking but not outright suicidal intent [27–29]. Furthermore, alcohol abuse has been reported as a risk factor for suicide [30, 31]. Understanding how PM exposure may influence these interactions is therefore important given that often PM exposure is highest in vulnerable urban populations. We also found no literature on how exposures to PM and their associations with suicide or substance use related mortality might differ between racial and gender groups.

Research goals

We tested three main hypotheses related to PM exposure and mental health-related mortality from suicide, overdose, and alcohol-related liver failure using data from the State of Michigan:

• Increased mean PM exposures will increase risk for one or more types of mental health-related mortality.

• Associations of PM will accumulate over time. We expect that prolonged exposures to extreme levels of PM over days or weeks will have greater impacts on mortality risk than simple acute exposures.

• Patterns of exposure and risk for mental health-related mortality will differ between demographic groups (e.g. sex, race, age). As risk for mortality from self-harm or substance use differs between demographic groups, so will associations of PM exposure with mortality.

Methods

Data

The Michigan Department of Health and Human Services (MDHHS) maintains records on all deaths from all causes which occur within the State of Michigan. Information recorded includes sex, age, race/ethnic group, and marital status. Records also include the residential address of the decedent, even for those whose primary residence is outside the state of Michigan. This study focused on three major mental health-related causes of death as recorded in Michigan death records: suicide, drug overdoses, and alcohol-related liver failure. We included records that had ICD codes associated with the three types of mortality as a listed principal or related cause of death. For suicides these codes were X60-X84 (intentional self-harm), and Y870. For alcohol-related liver failure this included codes K70x. For overdose, we used codes T40x. We also included records that included terms associated with any of the three types of mortality in the narrative description of the death (e.g. “hanging”). The three categories, suicide, alcohol and overdose were treated as mutually exclusive categories. Overdoses where there was clear intent were included in the suicide category. Records without clear dates of death or locational information were excluded from the analysis.

Since this is a study which tests same day and lag associations of PM exposure with mortality, it was important to ensure that date of death properly reflected short-term exposure windows. Thus, we only included records where there was clear evidence that the person died on the same day or the day after the suicide or overdose event. All records which noted that the death occurred two days or more after the event were excluded. This study was approved by the Institutional Review Boards of the University of Michigan and the MDHHS.

Particulate matter (PM) data

Daily PM concentrations for the contiguous United States for the years 2005–2016 were obtained from the CAFE Climate and Health Research Coordinating Center Collection hosted by Harvard University [32]. The dataset includes model-based predictions of PM concentrations (micrograms per cubic meter) in grid cells at a resolution of 1 km for the years 2005 to 2016. All data were extracted and added to the mortality dataset using the listed location of death. Daily PM concentrations were extracted for the death date listed on the death record and all 14 days preceding.

Statistical methods

We assumed exposure locations to be the decedent’s residential address and excluded residences outside of Michigan. We used a time-stratified case-crossover design to assess associations between environmental exposures and mortality. The case-crossover design is effective for evaluation of acute-onset outcomes in response to short-term exposures [33, 34]. In this design, each individual serves as their own control, with case days, i.e., the date of death, and control days. We defined control-days as same day-of-the-week within the same month and year as the case date of death and assigned environmental exposures and lags accordingly. This approach allowed us to account for seasonality and possible temporal autocorrelation by matching case and control days in small time windows [35].

To model potentially non-linear and delayed associations, we used distributed lag non-linear models (DLNMs) [36] with natural cubic splines for both the exposure and lag dimensions. Alternative spline parameterizations were tested, and the optimal specifications were selected using Akaike’s Information Criterion. We then performed a validation step to see how estimates changed based on different parameterizations and using different exposure percentiles.

Cox proportional hazards models with strata defined by matched identifiers were fit, providing estimates of relative risks (RR) that are inherently adjusted for time-invariant confounders. Model predictions were obtained from cross-basis functions using crosspred, yielding lag-specific and cumulative RRs with 95% confidence intervals.

One of the goals of this research was to see if the impacts of PM exposure on mental health-related mortality differed between Black/white racial groups and binary sex groups. Some research has suggested differences in health outcomes between male/female sex groups [37] and racial groups [38]. To test for possible effect modification, we stratified models by subsetting the full dataset on these demographic variables. We then compared the magnitude and direction of estimates and tested for interactions using those variables.

Results

There were 30,930 deaths between 2006-01-01 and 2016-12-31 in the analysis sample when excluding those with missing address information. There were 13,336 deaths from suicide, 6,875 deaths from alcohol-related liver failure, and 10,719 deaths from drug overdoses. Men represented 80% of suicide deaths, 70% of alcohol-related liver failure deaths, and 65% of overdose deaths. The mean age for suicides was 46.6 years, 55.4 years for alcohol-related deaths, and 41.8 years for overdose deaths. For people under the age of 18, there were 515 suicides, 75 cases of overdose and one case of alcohol-related liver failure. As these were relatively small numbers compared with the total number of deaths, we opted to leave them in the dataset.

The distributions of racial groups within cause of death varied. While Black decedents were 14% of all deaths, roughly reflecting overall demographics in Michigan, only 8% of suicides and only 10% of alcohol-related deaths were of Black people. However, more than 18% of deaths from overdoses were in Black decedents. See Table 1 for full results.

Table 1.

Descriptive statistics of the dataset

	[ALL]	Suicide	Alcohol	Overdose
	N = 30930	N = 13336	N = 6875	N = 10719
Sex:
Male	22690 (73.4%)	10753 (80.6%)	4871 (70.9%)	7066 (65.9%)
Female	8240 (26.6%)	2583 (19.4%)	2004 (29.1%)	3653 (34.1%)
Race/Ethnicity:
White	26463 (85.6%)	11895 (89.2%)	5892 (85.7%)	8676 (80.9%)
Black	3498 (11.3%)	1039 (7.79%)	727 (10.6%)	1732 (16.2%)
Other	969 (3.13%)	402 (3.01%)	256 (3.72%)	311 (2.90%)
Age (years)	46.6 (15.8)	46.0 (18.1)	55.4 (10.7)	41.8 (12.9)

The mean PM exposure on the day of death was approximately 9.6 for all types of death. Black suicide decedents had the highest exposure at 10.52 . Males and females had similar PM exposures. See Table 2.

Table 2.

PMExposure at Lag 0: Mean (95% CI) by mortality category and demographic subgroup

Category	All	Black	White	Female	Male
All deaths	9.64 (SD=5.85)
Suicide	9.65 (SD=5.87)	10.52 (SD=6.60)	9.61 (SD=5.81)	9.65 (SD=6.01)	9.65 (SD=5.84)
Overdose	9.66 (SD=5.79)	9.97 (SD=5.90)	9.63 (SD=5.80)	9.70 (SD=5.75)	9.63 (SD=5.82)
Alcohol	9.52 (SD=5.84)	9.88 (SD=5.51)	9.41 (SD=5.88)	9.89 (SD=6.41)	9.39 (SD=5.64)

Model results

We parameterized the crossbases of PM using 4 and 5 degrees of freedom for the exposure and lag spaces. Sensitivity analyses of different parameterizations and exposure percentiles can be seen in Fig. S1.

Figure 1 shows plots of cumulative relative risk estimates for all decedents, suicide, alcohol-related liver failure and overdose along the range of exposures at same day, from the day previous, and one and two weeks previous. There was evidence to suggest positive, though non-significant, associations of PM exposure with all deaths, suicide, alcohol-related liver failure and overdose at 7 days. At two weeks of cumulative exposure, this association seems to be null.

Fig. 1.

Cumulative exposure across same day, previous day, one week, and two weeks and relative risk of mortality for all decedents, and deaths from suicide, overdose and alcohol-related liver failure.

Cumulative exposure to the 95th percentile of PM (approximately 31 g/m) was generally associated with elevated risks of mortality outcomes. For all mental health-related mortality, relative risks ranged from RR = 1.02 (95% CI: 0.84, 1.23) at lag 1 to RR = 1.38 (95% CI: 0.70, 2.73) at lag 14, indicating consistent positive associations across lags. For suicide mortality, risks were elevated across all lag periods, with the largest at lag 14 (RR = 1.38, 95% CI: 0.63, 3.03). Overdose-related deaths similarly showed increased relative risks, reaching RR = 1.42 (95% CI: 0.55, 3.65) at lag 7. Associations of PM with alcohol-related mortality exhibited an inconsistent pattern: relative risks were below 1.0 for white, female and male groups at early lags, but estimates for Black decedents were elevated, including RR = 105.7 (95% CI: 1.44, 7751.6) at lag 7. These large risks should be interpreted with caution. Full results are presented in Table 3. There was no evidence to suggest effect modification between any of the demographic subgroups (statistical results not shown, graphical display in Fig. S2.)

Table 3.

Relative risk estimates for selected lag associations of exposure to the 95th percentile of PMexposure and mortality within distributed lag non-linear models. Relative risk estimates and confidence intervals are shown for all decedents, suicides, alcohol-related liver failure and overdose. Results also presented for Black and White racial categories and male and female decedents

Category	Lag (days)	All	Black	White	Female	Male
ALL	0	1 (0.9, 1.12)	0.99 (0.71, 1.36)	1 (0.9, 1.12)	1 (0.85, 1.17)	1.01 (0.9, 1.14)
	1	1.02 (0.84, 1.23)	0.96 (0.54, 1.72)	1.02 (0.83, 1.25)	1.01 (0.76, 1.34)	1.03 (0.83, 1.28)
	3	1.07 (0.79, 1.45)	0.89 (0.35, 2.28)	1.08 (0.78, 1.49)	1.03 (0.65, 1.63)	1.09 (0.77, 1.54)
	7	1.25 (0.81, 1.95)	0.69 (0.17, 2.75)	1.31 (0.82, 2.09)	1.1 (0.56, 2.15)	1.27 (0.77, 2.09)
	14	1.38 (0.7, 2.73)	0.28 (0.03, 2.39)	1.63 (0.8, 3.32)	0.86 (0.3, 2.43)	1.46 (0.68, 3.13)
Suicide	0	1.02 (0.9, 1.16)	1 (0.66, 1.52)	1.02 (0.89, 1.16)	1.02 (0.83, 1.25)	1.02 (0.89, 1.17)
	1	1.05 (0.84, 1.31)	0.97 (0.46, 2.05)	1.04 (0.82, 1.32)	1.03 (0.71, 1.5)	1.04 (0.81, 1.33)
	3	1.11 (0.78, 1.59)	0.82 (0.24, 2.74)	1.12 (0.77, 1.63)	1.07 (0.59, 1.94)	1.1 (0.74, 1.64)
	7	1.29 (0.77, 2.14)	0.44 (0.07, 2.64)	1.36 (0.79, 2.32)	1.14 (0.47, 2.74)	1.29 (0.74, 2.27)
	14	1.38 (0.63, 3.03)	0.24 (0.01, 4.13)	1.54 (0.68, 3.5)	0.82 (0.21, 3.27)	1.43 (0.61, 3.38)
Overdose	0	1.08 (0.86, 1.35)	0.68 (0.36, 1.27)	1.12 (0.88, 1.42)	1.04 (0.79, 1.37)	1.11 (0.84, 1.47)
	1	1.15 (0.77, 1.72)	0.49 (0.16, 1.51)	1.23 (0.8, 1.89)	1.07 (0.65, 1.76)	1.22 (0.74, 2.01)
	3	1.27 (0.67, 2.43)	0.31 (0.05, 1.93)	1.4 (0.7, 2.78)	1.1 (0.49, 2.45)	1.4 (0.63, 3.12)
	7	1.42 (0.55, 3.65)	0.21 (0.01, 2.97)	1.52 (0.56, 4.19)	1.03 (0.32, 3.28)	1.64 (0.49, 5.45)
	14	1.32 (0.3, 5.78)	0.05 (0, 2.79)	1.8 (0.37, 8.83)	0.59 (0.1, 3.43)	2.18 (0.32, 14.73)
Alcohol	0	0.71 (0.48, 1.05)	2.11 (0.78, 5.69)	0.61 (0.4, 0.94)	0.87 (0.49, 1.54)	0.72 (0.46, 1.15)
	1	0.55 (0.27, 1.11)	4.28 (0.71, 25.75)	0.42 (0.2, 0.91)	0.8 (0.29, 2.24)	0.56 (0.25, 1.3)
	3	0.43 (0.14, 1.32)	15.59 (0.84, 290.21)	0.29 (0.08, 0.97)	0.81 (0.15, 4.21)	0.42 (0.11, 1.57)
	7	0.56 (0.11, 2.85)	105.7 (1.44, 7751.62)	0.4 (0.07, 2.27)	1.4 (0.13, 15.67)	0.43 (0.07, 2.81)
	14	1.12 (0.1, 12.8)	69.52 (0.15, 32636.5)	1.14 (0.08, 15.59)	5.13 (0.11, 244.01)	0.56 (0.04, 9.01)

Discussion

Through this research, we have shown evidence to suggest that cumulative exposure to high levels of PM over one week is associated with elevated risks for suicide. We have also shown a positive, though non-significant association of PM exposure and risk for death from alcohol-related liver failure. We did not find an association between PM exposure and substance use mortality.

Other research has found associations of PM exposure and suicide mortality though these studies explored only same day or very short-term lag periods [22, 23]. We could find only one study which explored associations of PM exposure and suicide over longer lag periods [39]. This case-crossover study using a Belgian sample showed only limited associations of exposure of PM10 to suicide risk, and positive associations were only seen in specific age groups. To our knowledge, this study is the only study which has explored links between suicide and PM for lags greater than seven days. Some research suggests links between PM exposure and increased risk for chronic liver disease [40, 41] but we found no other study which looked at alcohol-related liver mortality and PM exposure. We found no other study which explored links between PM exposure and overdose.

Suicide attempts and non-fatal and fatal overdoses have been shown to have different pathologies with similar roots [42, 43]. These differences could explain the striking differences in the patterns of association between PM exposure and mortality found in this research. Future research might focus on what specific clinical aspects of suicidal ideation/attempt and substance use could be impacted by exposure to air pollutants such as PM.

The linkages between PM exposure and suicide are thought to involve epigenetic mechanisms such as DNA methylation and their impact on gene expression, which result in predisposition to mood and behavioral volatility [20]. A major biological pathway linking mental health-related mortality and PM exposure is TLR3 expression and its effects on hippocampal plasticity [44, 45]. In particular, the upregulation of pro-inflammatory cytokines IL-1, TNF-, and IL-6 has been linked with neuroticism and suicide [46–48]. PM exposure also increases hippocampal TNF- and IL-1 cytokine expression resulting in the reduction of hippocampal plasticity which is associated with mood disorders and suicidal completion [49–52]. PM exposure may also drive epigenetic changes by altering both global gene expression [53–57], and gene-specific DNA expression [57–61] via DNA methylation, as well as disrupting standard miRNA functioning [60, 62, 63], and RNA processing [56]. Research suggests links between PM exposure induced epigenetic mechanisms such as DNA methylation and a heightened risk of suicidal behavior [64]. However, it is unclear how the lag windows considered in this study line up with the timing of epigenetic changes and if short-term acute PM exposures can trigger the epigenetic changes. There is, however, evidence that epigenetic changes can be rapidly triggered and can oscillate as part of the circadian rhythm [65], potentially indicating that even acute exposures can trigger epigenetic changes that may increase suicidal behavior.

The association between PM exposure and alcohol-related mortality is less clear. However, evidence points to two possible mechanisms. First, PM exposure causes prefrontal cortex inflammation which is associated with depressive-like responses/behaviors [66] which are in turn associated with an increased vulnerability to developing alcoholism. Second, PM exposure may increase cortisol production which dysregulates the hypothalamic-pituitary-adrenal (HPA) axis. Alterations in HPA functioning have in turn been associated with an increased risk of alcohol dependence and use [67].

Research has shown that there are disparities in exposures to air pollutants between disadvantaged and non-disadvantaged areas of Michigan [68]. Black and Hispanic populations exposed to high levels of air pollutants in urban areas are at a higher risk for emergency hospitalization than White populations [69]. Our results also highlight that associations of PM exposure and suicide are strongest in Black populations. Other research suggests disparities in risk for suicidal ideation and suicide attempt between racial groups [70–72]. While it is unlikely that PM exposure is the sole driver of these disparities, our results suggest that mitigating exposure to air pollutants might be an actionable means to prevent suicide and mental health-related mortality.

At its core, air pollution is an environmental justice issue due to the disproportionate burden of exposure and negative health impact it has on communities of color and people of lower socioeconomic status. Communities of color and residents of lower-income neighborhoods tend to face disproportionately higher exposure to PM, not only within the neighborhoods they live in, but also those they travel to [73]. Research also indicates that PM exposures may differ by racial group. Black and African populations are primarily urban, residing in contexts where housing is on average older, more dense and of lower quality than suburban or rural areas. In Michigan, residents of the City of Detroit, a primarily African American city, were found to have increased cracks and openings in the home which could allow increased exposure [74] as suggested by air exchange rate models. Disparities in exposure to ambient PM have also been found between rental and homeowner units, and between multifamily and single family housing, with rental multifamily units being the most exposed [66]. Future work in this area should strive to approach PM exposure as an environmental justice issue and seek to mitigate current inequities in air pollution exposures that influence physical or mental health.

This study has several limitations. First was the inability to reliably determine the true length of exposure. Higher levels of impulsivity and aggression are known risk factors for suicide among those experiencing depressive symptoms [75]. It is possible that short-term exposure to PM increases impulsivity and subsequently suicide risk [76]. Our inclusion of alcohol-related liver failure might not be appropriate given the long latency period before the development of liver disease. Suicide attempts and completion are often associated with alcohol use [77]. Our data may also be subject to misclassification given that many overdoses may, in fact, be suicides but are misclassified [78, 79]. Finally, we relied on residential addresses included in the death records. These may not accurately reflect decedents’ actual exposure. These cannot account for time-activity patterns including time spent outside the home or even outside of the state [80].

Future work might explore the links between a wide range of air pollutants and suicide risk, with an aim of influencing environmental policy and/or facilitating protective interventions. This work should closely examine disparities in pollution exposures, and might also identify the specific factors that contribute to excess risk for mental health-related mortality within specific demographic subgroups.

Authors' contributions

Conceptualization, P.S.L. and R.B; methodology, P.S.L.; validation, P.S.L.; formal analysis, P.S.L.; investigation, P.S.L., R.B., L.E.; data curation, P.S.L.; writing-original draft preparation, P.S.L., R.B., L.E., V.F., A.T.; writing-review and editing, P.S.L., R.B., L.E., V.F., A.T.; visualization, P.S.L., R.B., L.E.; supervision, P.S.L.; project administration, P.S.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research was conducted with no funding.

Data availability

Data is available from the Michigan Department of Health and Human Services through formal request. Code is available on the Github repository kambanane/PM25despair

Declarations

Ethics approval and consent to participate

This study was approved and considered exempt by the Institutional Review Boards of the University of Michigan and the Michigan Department of Health and Human Services.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.