Wildfire contributions to prenatal PM_2.5 exposure in the contiguous United States, 2003–2019

Abstract

Introduction:

Anthropogenic air pollution has declined substantially across the United States, but wildfire smoke may be offsetting these gains. PM2.5 exposure increases the risk of adverse birth outcomes, and wildfire-related PM2.5 may be more toxic than PM2.5 from other sources. Yet limited data exist on spatiotemporal changes in population-level prenatal wildfire PM2.5 exposure. Existing studies have characterized geographic patterns of wildfire contributions to ambient PM2.5 but have not estimated population-based prenatal exposure at the national scale, a perspective essential for allocating public-health resources to the people most affected.

Methods:

We linked daily county-level PM2.5 data from 2,842 counties (2003–2019) to national birth certificate records covering 64.5 million live births. Using chemical transport modeling, we separated wildfire-related from non-fire PM2.5 and computed period-averaged exposure for baseline (2003–2005) and endline (2017–2019). We assessed trends across nine NOAA climate regions, demographic subgroups, and counties stratified by wildland-urban interface (WUI) status and perinatal healthcare availability, and quantified prenatal exceedance days above 35 µg/m³.

Results:

Prenatal PM2.5 declined from 12.82 to 8.45 µg/m³ (−34.1%), but non-fire PM2.5 declined more steeply (−37.2%), indicating that wildfire offset a portion of these gains. Wildfire's share more than doubled (3.71% to 8.19%), with 0.22 µg/m³ of improvement foregone nationally. Fire-attributable exceedance days were essentially unchanged (∼14%–15% of infants) while non-fire exceedance fell 85%; wildfire now accounts for 65% of remaining events. WUI, low-income, and American Indian/Alaska Native infants bore disproportionate burdens, and maternity hospital shortage counties had wildfire contributions nearly double those of adequate-access counties by 2017–2019.

Discussion:

Increasing wildfire activity is systematically offsetting prenatal air quality gains from anthropogenic emission reductions, disproportionately burdening low-income, rural, and Indigenous communities, with the potential to further erode progress as wildfire activity intensifies.

Introduction

Over the past several decades, the United States achieved substantial improvements in ambient air quality, largely due to effective regulatory policies implemented under the Clean Air Act (1, 2). Between 1970 and 2022, combined emissions of six common pollutants declined by 78 percent, while concentrations of fine particulate matter (PM_2.5) decreased by 39 percent between 1990 and 2010 (3). These improvements have yielded substantial public health benefits, particularly for vulnerable populations including infants, whose health at birth is particularly sensitive to air pollution exposures experienced in the prenatal period (4, 5).

However, emerging evidence suggests that progress in air quality improvements has slowed across much of the country, coinciding with increased wildfire activity across North America (6–8). Wildfires can account for as much as 40% of ambient PM_2.5 concentrations during peak fire years in the Western US (9). Since the mid-2010s, wildfire smoke has influenced annual PM_2.5 patterns in nearly three-quarters of contiguous US (CONUS) states, offsetting up to 25% of air-quality gains achieved in earlier decades (8). This reversal corresponds to approximately four years of lost national air quality progress, consistent with documented increases in wildfire intensity and size attributed to warmer and drier conditions, historical land management practices, and expanding human development at the wildland-urban interface (WUI) (10, 11).

Importantly, emerging evidence suggests that PM_2.5 from wildfire may be more toxic than equivalent doses of ambient PM_2.5 from other sources (12–14), particularly with respect to respiratory health outcomes (15). Toxicological studies have linked the higher toxicity of wildfire-related PM_2.5 to distinct chemical characteristics, including higher carbonaceous content and a greater capacity to induce oxidative stress (16, 17).

Health measured at birth (e.g., birthweight and gestational age) is particularly sensitive to air pollution exposures experienced in the prenatal period (5). Recent studies have linked wildfire smoke exposure during pregnancy to increased risks of preterm birth (18, 19), low birth weight (20–22), and a broader set of perinatal and early-childhood health outcomes (23–25). The underlying mechanisms may involve enhanced maternal-fetal inflammatory responses, oxidative stress pathways, and alterations in placental function (26). Wildfire smoke contains elevated concentrations of polycyclic aromatic hydrocarbons (PAHs) generated by incomplete biomass combustion. PAHs can cross the placental barrier, bind aryl hydrocarbon receptors in placental and fetal tissues, induce oxidative DNA damage, and disrupt placental vascular development — pathways linked to reduced fetal growth, preterm delivery, and altered neurodevelopment (27–30). Because wildfire smoke carries higher PAH and organic-carbon content than typical ambient PM_2.5 from non-fire sources, these mechanisms may be amplified relative to background exposure (26). Adverse birth outcomes are associated with persistent health and cognitive impairment throughout the life course, which underlies the particular public health importance of prenatal wildfire smoke exposure (4).

No studies to date have comprehensively evaluated how prenatal exposure to PM_2.5 among live-born infants has changed over time and space at the national scale, or the relative importance of wildfire in shaping these trends. While previous work has documented spatiotemporal patterns of wildfire smoke and its contribution to PM_2.5 levels, these studies have characterized wildfire pollutants across time and space rather than estimating population-based exposure, which is critical for developing efficient public health mitigation strategies. Our population-based approach to prenatal exposure is particularly useful to policymakers and practitioners that must allocate scarce resources that benefit people rather than land areas; a population-based perspective is especially important in the infant health context where efficient targeting of resources is needed given the high cost and regionalization of intensive neonatal services (31).

Using state-of-the-art chemical transport modeling and national birth certificate data (2003–2019), we provide a national-scale assessment of trends in prenatal PM_2.5 exposure over nearly two decades and document the contribution of increased wildfire activity. We examine these patterns across geographic regions and population subgroups defined by race/ethnicity, county income, urbanicity, and WUI status. We further characterize how these trends differ by perinatal healthcare resource availability, an important dimension given the regionalization of neonatal services. We quantify the population burden of prenatal exceedance days and use shift-share decomposition to distinguish whether trends reflect changes in air quality where people already live or shifts in where births occur.

Methods

Air pollution data

We used the U.S. Environmental Protection Agency (EPA) Community Multiscale Air Quality (CMAQ) model (version 5.3.3) to simulate PM_2.5 at 12 km spatial resolution over CONUS for 2003–2019. Simulations were driven by emissions and meteorological inputs from the EPA's Air Quality Time Series Project (EQUATES) (32, 33).

To estimate the contribution of fire-related emissions to ambient PM_2.5, we conducted two CMAQ runs: (1) an all-emissions baseline incorporating all emission sources, and (2) a “no-fire” sensitivity run in which all biomass burning emissions within the modeling domain were zeroed out. The no-fire run excluded U.S. wildfires as well as wildfires in Canada and Mexico, and also excluded domestic agricultural burning and prescribed fires. The difference between the two runs was used as our estimate of fire-attributable PM_2.5 (34–36). Consistent with the literature, we refer to this quantity as “wildfire-attributable” PM_2.5, because wildfire smoke is the predominant component and the dominant driver of the rising trend in fire-attributable PM_2.5 over our study period, although the underlying CMAQ attribution also captures other, smaller sources of biomass burning such as agricultural and prescribed fire (8, 34).

We regridded the 5 km NASA SEDAC Gridded Population of the World (GPW) v4 population data (37) onto the 12 km CMAQ CONUS grid using inverse distance weighting (IDW) implemented in the Python pyresample library. Because SEDAC releases GPW at five-year intervals (2000, 2005, 2010, 2015, 2020), annual population fields for intermediate years were estimated by linear interpolation between adjacent releases. County-level daily PM_2.5 was then computed as the population-weighted mean across CMAQ 12-km grid cells assigned to the county.

We estimated three exposure measures for each pregnancy: total PM_2.5 (PM_2.5-T); non-fire PM_2.5 (PM_2.5-NF), representing exposure that would have resulted without wildfire; and wildfire-related PM_2.5 (PM_2.5-WF), defined as PM_2.5-T minus PM_2.5-NF.

Study population

The study population included all live-born infants recorded in the national birth registry between 2003 and 2019, the full period covered by the EPA EQUATES emissions inventory. These records, compiled from birth certificates by the Centers for Disease Control and Prevention, include nearly all live births in the US, totaling 64.5 million births (3.79 million annually). Birth data were linked with air pollution data based on the mother's county of residence at time of delivery. For each infant, daily PM_2.5 exposure was calculated and averaged over the prenatal period, determined by subtracting gestational age from the delivery date. Gestational age was based on the obstetric/clinical estimate (OE), available on the NCHS natality file for both the 1989 and 2003 birth-certificate revisions and used throughout our 2003–2019 study period. States adopted the 2003 revised birth certificate on a rolling basis through 2016, but the OE measure is available across all states and years with one exception: California in 2004–2006, where the clinical-gestation field was missing. For these California-year cases, OE was imputed via regression on a quadratic in LMP-based gestational age, maternal race, maternal age, and state and year fixed effects, following the data-cleaning protocol described in (31). We did not switch between LMP and OE measures based on calendar year. Consistent with NCHS coding conventions adopted in 2014, gestational age was top- and bottom-coded to 17–47 weeks throughout the study period. Birth records report year-month of birth, and we assumed births occurred on the 15th day of the birth month, consistent with the convention used in prior population-level analyses of NCHS public-use natality data (38), which suppresses exact date of birth to protect confidentiality. Multiple births, births with implausible gestational ages (<20 or >45 weeks), and records with missing geographic information were excluded (approximately 4% of records).

Maternal race/ethnicity was derived from birth certificate records. We harmonized race/ethnicity across the 1989 and 2003 birth certificate revisions to create consistent categories: non-Hispanic White, non-Hispanic Black, Hispanic, Asian/Pacific Islander, and American Indian/Alaska Native (AIAN). The 1989 revision collected race and Hispanic origin as separate items, while the 2003 revision allowed multiple race reporting; we used bridged-race categories provided by the National Center for Health Statistics to ensure comparability over the study period.

This study used de-identified birth certificate data obtained from the National Center for Health Statistics (NCHS) under a restricted-use data use agreement. The University of Maryland Institutional Review Board determined this study to be exempt from full review (45 CFR 46.104(d)(4)) as it involved only secondary analysis of existing de-identified administrative data. The study was conducted in accordance with the Declaration of Helsinki.

Geographic coverage

The analytical domain encompassed 2,842 counties across 48 states within CONUS, organized according to nine climatologically distinct regions established by NOAA: Northeast, Southeast, South, Ohio Valley, Upper Midwest, Northern Rockies and Plains (NRP), Southwest, Northwest, and West (39, 40).

Trend estimation and counterfactual analysis

Because wildfire PM_2.5 varies substantially year-to-year reflecting episodic fire activity, we avoided comparing single-year endpoints and instead averaged exposure over three-year baseline (2003–2005) and endline (2017–2019) periods to smooth year-to-year volatility while preserving a meaningful interval over which to assess change. The 2006–2016 years were not omitted: they were used in the annual linear-trend regressions and in all year-resolved time-series figures (Figures 1, 4, 5). The baseline/endline contrast summarizes cumulative change between two stable periods; the annual trend regressions and time-series figures use the full 17-year series. Within each period, we averaged birth-weighted (i.e., weighted by birth counts so that counties with more births contribute proportionally more) PM_2.5 values across years. We also estimated annual linear trends via regression of birth-weighted PM_2.5 on year for each region and subgroup.

Figure 1

To quantify the impact of increasing wildfire activity, we computed a counterfactual: the endline PM_2.5 that would have prevailed had wildfire contributions remained at baseline levels. Specifically, counterfactual endline PM_2.5 = endline PM_2.5-NF + baseline PM_2.5-WF. The wildfire offset is defined as observed endline PM_2.5-T minus this counterfactual—representing the additional PM_2.5 exposure attributable to the increase in wildfire activity.

Subgroup analysis

To characterize population-level exposure disparities, we computed trends and counterfactuals across demographic and geographic subgroups: maternal race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian/Pacific Islander, AIAN), county income level (terciles of per capita income from the CDC Social Vulnerability Index, defined nationally across all 2,842 counties), urbanicity (NCHS 2013 Urban-Rural Classification: metropolitan vs. non-metropolitan), and WUI status. For each subgroup, births were assigned to all counties where the subgroup had births, and birth-weighted average PM_2.5 exposure was computed using the same period-averaging approach.

Race/ethnicity in our data refers to the race/ethnicity of the mother, recorded on the birth certificate. For race-stratified estimates, county-level PM_2.5 was birth-weighted by the count of births to mothers of each racial/ethnic group in each county, rather than computed by stratifying births into separate subsamples; this birth-weighting approach yields the population-level average prenatal exposure experienced by infants born to mothers of each group while keeping the exposure surface common across groups. Differences across groups therefore reflect the joint distribution of maternal residential geography and ambient air quality and should be interpreted as descriptive characterizations of exposure burden rather than as causal effects of maternal race/ethnicity on exposure.

Perinatal healthcare resource assessment

We augmented our analysis with county-level data on perinatal healthcare resources to characterize wildfire PM_2.5 exposure trends in areas with limited healthcare access. Drawing on methods described in Boudreaux et al. (31), we classified counties using two binary indicators of perinatal healthcare shortage: (1) whether the nearest maternity care hospital to the population-weighted county centroid was low volume (fewer than 500 annual births) or geographically isolated (more than 30 miles from the nearest alternative hospital), and (2) whether the nearest Level 3 or 4 neonatal intensive care unit (NICU) was more than 60 miles from the population-weighted county centroid. These measures capture aspects of healthcare access not fully reflected by standard urban/rural classifications. Because the healthcare resource data represent a single cross-section, we treat these designations as time-invariant over the study period.

Exceedance day analysis

We quantified the number of infants experiencing at least one prenatal day when ambient PM_2.5 exceeded 35 µg/m³, the 24 hour National Ambient Air Quality Standard (NAAQS; this standard was unchanged by the January 2024 EPA revision, which lowered only the annual NAAQS from 12 to 9 µg/m³). For each infant, we determined whether any exceedance day was fire-attributable—defined as a day when PM_2.5 exceeded the threshold with wildfire included but would not have exceeded it without wildfire contributions.

Shift-share (Kitagawa) decomposition

To assess whether changes in prenatal PM_2.5 reflect changes in air quality within the counties where births occur, as opposed to shifts in the geographic distribution of births across counties, we employed a Kitagawa decomposition (41). Nationally and within each climate region, we decomposed the change in birth-weighted average PM_2.5 between baseline and endline into three additive components: (1) a place effect, capturing changes in county-level air quality while holding the distribution of births across counties fixed; (2) a migration effect, capturing changes in the distribution of births across counties while holding county-level air quality fixed; and (3) an interaction term (42, 43). Here “place” refers to the county of maternal residence; the decomposition is computed across counties and then summarized nationally and by climate region.

Statistical analysis

Temporal trends were estimated using ordinary least squares regression of annual birth-weighted PM_2.5 on calendar year. Statistical significance was assessed at α = 0.05. For the subgroup analyses, we calculated the percent change between the baseline and endline periods to estimate the relative change in PM_2.5 and wildfire-related PM_2.5 concentrations over time. Given the descriptive intent of these stratified comparisons, we did not apply multiple comparison corrections; results should be interpreted as hypothesis-generating. We stratified all primary analyses by perinatal healthcare shortage indicators and tested for differences in PM_2.5 trends and wildfire contributions using birth-weighted averages and linear trend regressions.

Results

Descriptive overview

Over the full study period, 3.79 million infants were born annually across 2,842 counties (Table 1). The average prenatal PM_2.5-T was 10.43 µg/m³, of which 0.44 µg/m³ (4.22%) was attributable to wildfire. The Ohio Valley had the highest average PM_2.5-T (12.12 µg/m³) and the Southwest the lowest (6.72 µg/m³). The relative share of PM_2.5-WF was highest in the NRP (7.63%) and Southeast (6.31%). Annual births ranged from 66,361 in the NRP to 667,646 in the Southeast, underscoring the importance of birth-weighted analysis.

Trends in prenatal PM_2.5 and the wildfire offset

Average daily prenatal PM_2.5-T declined from 12.82 µg/m³ in the baseline period (2003–2005) to 8.45 µg/m³ in the endline period (2017–2019), a reduction of 4.37 µg/m³ or 34.1% (Table 2, Figure 1). However, non-fire PM_2.5 declined more steeply—from 12.34 to 7.76 µg/m³ (4.58 µg/m³; 37.2%)—indicating that a portion of the gains from anthropogenic emission reductions were offset by increasing wildfire activity.

Nationally, had wildfire contributions remained at baseline levels, endline PM_2.5 would have been 8.23 µg/m³ rather than 8.45 µg/m³—an additional 0.219 µg/m³ (2.60%). Wildfire's share of total prenatal PM_2.5 more than doubled, from 3.71% at baseline to 8.19% at endline (+4.49 percentage points). While PM_2.5-WF varied substantially year-to-year reflecting episodic wildfire, the upward trend was persistent (Figure 1).

Regional variation was notable (Table 2). The Northwest experienced the largest wildfire offset: 0.674 µg/m³. Without increasing wildfire, prenatal PM_2.5 would have declined by 14.7% (non-fire trend), but the observed decline was only 6.6%—wildfire erased 7.90 percentage points of improvement, more than half the non-fire gains. Wildfire's share in the Northwest surged from 5.11% to 13.34%. The West showed a similar pattern: a wildfire offset of 0.478 µg/m³, with wildfire's share rising from 3.68% to 9.68%. The Northern Rockies and Plains (NRP) was the only region with a negative wildfire offset (−0.052 µg/m³), meaning wildfire PM_2.5 was slightly lower at endline than baseline. This likely reflects the NRP's unusually high baseline wildfire contribution (8.90%, the highest of any region), driven by an active early fire period, combined with high year-to-year volatility in a region with only 66,361 births annually. The NRP result should be interpreted cautiously.

Even in regions less publicly associated with wildfire, contributions grew substantially. The Southeast already had an above-average wildfire share in the study period overall (6.31%, Table 1), partly reflecting agricultural burning and prescribed fire activity, yet its wildfire share still increased from 4.68% to 9.77% (+5.09 percentage points) between baseline and endline, and the South from 4.85% to 9.79% (+4.94 percentage points). The Ohio Valley experienced the largest absolute PM_2.5 decline (6.04 µg/m³) but its wildfire share increased from 3.29% to 7.72% (Supplementary Figure S1). Because our no-fire CMAQ sensitivity run zeroes out all biomass burning (including agricultural and prescribed fire), the elevated baseline fire-attributable share in the Southeast partly reflects these non-wildfire biomass sources, which are most prominent in this region; the temporal increase in the Southeast wildfire share over the study period, however, is more likely driven by increased wildfire activity, given the relative stability of regional agricultural and prescribed burning. County-level spatial analysis demonstrated the greatest improvements in eastern regions (Figure 2), while wildfire-related PM_2.5 increased with considerable geographic heterogeneity.

Figure 2

Subgroup analysis

The subgroup analyses revealed disparities in wildfire exposure across subgroups (Supplementary Table S1, Figure 3). Infants born in WUI counties (680,095 annually) experienced a 7.37 percentage-point increase in wildfire contribution (6.74% to 14.11%), nearly double the 3.84-point increase among non-WUI infants. The wildfire offset in WUI counties was 0.350 µg/m³, compared to 0.190 µg/m³ in non-WUI counties. Rural infants experienced larger wildfire contribution increases than urban infants (+5.78 vs. +4.33 percentage points).

Figure 3

Infants in low-income counties (462,894 annually) experienced the largest wildfire contribution increase (+6.19 percentage points, from 5.31% to 11.51%), the highest wildfire offset (0.311 µg/m³), and the smallest overall PM_2.5-T improvement (−30.2%), consistent with wildfire disproportionately offsetting air quality gains in these areas.

Non-Hispanic Black infants (544,159 annually) had the highest baseline prenatal PM_2.5-T (14.45 µg/m³) but experienced the steepest decline (−37.7%). AIAN infants (35,845 annually) had the smallest PM_2.5-T decline (−24.9%) and the largest wildfire contribution increase (+5.11 percentage points, 6.36% to 11.47%). Hispanic infants (899,449 annually) experienced above-average wildfire contribution increases (+4.36 percentage points) and the second-highest wildfire offset (0.244 µg/m³).

Exceedance day analysis

The proportion of infants experiencing at least one prenatal PM_2.5 exceedance day (≥35 µg/m³) declined from 67.3% during 2003–2005 to 21.9% during 2017–2019—a 67.5% reduction (Supplementary Table S2, Figure 4). This improvement was driven almost entirely by reductions in non-fire exceedance, which fell from 44.2% of infants exposed in 2003 to 10.3% in 2019. In contrast, fire-attributable exceedance was essentially unchanged: 14.9% of infants in the early period and 14.1% in the late period experienced at least one day where wildfire pushed PM_2.5 above the NAAQS threshold.

Figure 4

Wildfire's share of remaining exceedance nearly tripled: fire accounted for 22% of all exceedance exposure in 2003–2005 but approximately 65% by 2017–2019. In the most recent individual years (2017–2018), this share exceeded 70%. By the endline period, wildfire accounted for the majority of remaining acute prenatal PM_2.5 exceedance events.

We further characterized exposure by trimester (Supplementary Table S5, Supplementary Figure S3). The proportion of infants experiencing at least one exceedance day (≥35 µg/m³) within a given trimester declined sharply between baseline and endline in all three trimesters: from 35.1% to 8.7% (T1), 39.2% to 8.5% (T2), and 44.3% to 8.3% (T3). At baseline, wildfire accounted for 8.8–10.0% of PM_2.5 on exceedance days across trimesters; by endline this had risen to 36.2–37.6%, a roughly four-fold increase that was similar across T1, T2, and T3. The shift toward wildfire-dominated exposure thus appears broadly distributed across the pregnancy window rather than concentrated in any single trimester. Formal estimation of trimester-specific health effects is outside our descriptive scope and is the subject of follow-up work.

Shift-share decomposition

The Kitagawa decomposition revealed that changes in prenatal PM_2.5 exposure were driven almost entirely by within-county air quality changes (place effects) rather than by shifts in the geographic distribution of births. Nationally, the place effect accounted for approximately 99% of the decline in total prenatal PM_2.5. This pattern was consistent across all nine climate regions and held for both total PM_2.5 improvements and wildfire PM_2.5 increases. The increase in prenatal wildfire exposure reflects increasing wildfire activity in counties where births already occur, not population movement into fire-prone areas. Full regional decomposition results are reported in Supplementary Table S3. For total prenatal PM_2.5, the place effect accounted for essentially the entire national decline (−4.37 µg/m³ of −4.37 µg/m³ total; 100%); for wildfire-attributable PM_2.5, the place effect accounted for 95.9% of the national increase (+0.210 of +0.219 µg/m³) and migration contributed 2.4%. Across regions, place effects dominated in every case except the Upper Midwest, where the absolute regional change in wildfire PM_2.5 was very small (+0.01 µg/m³) and a positive migration effect (+0.009 µg/m³) accounted for most of it.

Prenatal exposure by perinatal healthcare resources

Among the 2,842 counties in our study, 57% had a maternity hospital that was either low volume or geographically isolated, and 39% were located more than 60 miles from a Level 3/4 NICU. Approximately 630,700 infants per year (16.6%) were born in counties with maternity hospital shortages, and 322,300 infants per year (8.5%) were born in counties more than 60 miles from a NICU (Table 3).

Counties with maternity hospital shortages had lower average total PM_2.5 (8.86 vs. 10.74 µg/m³), reflecting their more predominantly rural location, but substantially higher wildfire contribution (6.26% vs. 3.79%). This pattern was more pronounced for NICU proximity: counties more than 60 miles from a NICU had wildfire contributions of 6.96% compared with 3.95% for counties with closer NICU access (Table 3, Panel A; Figure 5).

Figure 5

Temporal trends revealed diverging patterns. In counties with maternity hospital shortages, wildfire contribution increased from 5.44% (baseline) to 10.57% (2017–2019), a gain of 5.13 percentage points. In adequate-access counties, the increase was 4.13 percentage points (3.49% to 7.62%). For NICU-distant counties, wildfire contribution rose from 6.03% to 11.52% (+5.49 percentage points), compared with 3.61% to 7.79% (+4.18 percentage points) for counties within 60 miles of a NICU (Table 3, Panel B; Figure 5). The gap between shortage and adequate-access areas widened notably after 2015.

Regional stratification showed that the maternity hospital access gap in wildfire contribution was largest in the Northwest (10.0% for shortage counties vs. 3.1% for adequate-access counties), West (10.8% vs. 3.3%), and Northern Rockies and Plains (8.2% vs. 7.0%) (Supplementary Figure S2). In these regions, infants born in healthcare-shortage counties faced wildfire contributions 2–3 times those of infants in adequately served areas. Among counties in the top two quintiles of wildfire contribution increase (Q4–Q5), 65% had maternity hospital shortages. These high-fire, high-shortage counties averaged only 272 births per year, reflecting small, isolated communities facing both rising wildfire exposure and limited access to specialized perinatal care.

The geographic concentration of perinatal healthcare shortage was uneven across NOAA climate regions (Supplementary Table S4). The Northern Rockies and Plains had the highest concentration, with 85.9% of counties classified as maternity-hospital shortages and 80.8% located more than 60 miles from a Level 3/4 NICU; 65.8% of infants born in the region resided in maternity-hospital shortage counties. The Southwest (75.2% maternity-hospital shortage; 72.3% NICU shortage), Upper Midwest (73.0%; 44.0%), and Northwest (69.7%; 52.9%) also showed high concentration. The Northeast had the lowest (34.7%; 15.5%). Three of the four regions with the highest shortage concentration — the Northern Rockies and Plains, Northwest, and Upper Midwest — were also among the regions with the largest absolute increases in wildfire contribution to prenatal PM_2.5 over the study period.

Discussion

The US made important progress in reducing prenatal PM_2.5 exposure during 2003–2019. The average daily prenatal PM_2.5 declined by 4.37 µg/m³ (34.1%), benefiting the vast majority of the 3.79 million infants born annually. However, these gains would have been even larger without increasing wildfire contributions. We estimate that increasing wildfire activity offset 0.219 µg/m³ (2.60%) of the prenatal PM_2.5 reduction nationally, with offsets exceeding 5% in the Northwest and West (Table 2). Our counterfactual analysis estimates that had wildfire contributions remained at baseline levels, endline PM_2.5 would have been 8.23 µg/m³ rather than 8.45 µg/m³. This offset was largest in the Northwest (0.674 µg/m³, representing 7.90% of non-fire improvement foregone) and West (0.478 µg/m³, 5.20%).

Perhaps the most striking finding is the approximate two-thirds reduction in prenatal exposure to at least 1 exceedance day over the past 2 decades—a remarkable public health achievement driven largely by Clean Air Act regulations and related emission reduction policies. Yet this reduction in total exceedance was accompanied by a pronounced compositional shift: while total exceedance fell by roughly two-thirds, fire-attributable exceedance was essentially unchanged, so wildfire accounts for a growing share of the exceedance days that remain. Fire-attributable exceedance held at approximately 14% of infants across the study period, and by the endline period roughly two-thirds of all remaining prenatal exceedance days were attributable to wildfire.

The compositional shift toward wildfire-dominated exposure profiles may have health consequences beyond what PM_2.5 mass concentrations alone would suggest. Wildfire-derived PM_2.5 contains higher concentrations of organic carbon and polycyclic aromatic hydrocarbons (PAHs) associated with greater oxidative stress and inflammatory responses (12, 16, 17). Prior studies that have quantified prenatal PAH dose during wildfire events have reported reduced birth weight in exposed pregnancies, with effects observable at exposure levels comparable to those experienced in recent US fire seasons (22, 44). For pregnant individuals, this enhanced toxicity could translate to greater risks of adverse birth outcomes per unit of exposure than would be predicted from PM_2.5 mass concentration alone.

The wildfire burden is not evenly distributed. Infants in low-income counties experienced wildfire offsets 54% larger than those in high-income counties (0.311 vs. 0.201 µg/m³), and WUI county offsets were 84% larger than non-WUI (0.350 vs. 0.190 µg/m³). Rural infants experienced wildfire contribution increases 34% larger than urban infants (+5.78 vs. +4.33 percentage points), with wildfire's share reaching 11.40% by the endline period (2017–2019). These communities face compounding vulnerabilities: higher wildfire exposure, fewer healthcare resources, and greater distance to specialized neonatal services (31). AIAN infants experienced the smallest overall PM_2.5 improvements and the largest wildfire contribution increases. These patterns suggest that wildfire-driven air quality changes are disproportionately burdening populations that already face elevated health risks (45, 46). Subgroup analyses were descriptive and unadjusted, reflecting our aim to characterize the population-level distribution of exposure rather than to estimate causal effects; characteristics such as income and rurality overlap in important ways and should not be interpreted as independent contributions.

The racial/ethnic patterns reveal how environmental changes intersect with residential geography. Non-Hispanic Black infants had the highest baseline PM_2.5 (14.45 µg/m³) but experienced the largest improvements (−37.7%). This is consistent with Colmer et al. (43), who showed that convergence in absolute racial PM_2.5 disparities was driven largely by improvements in the South, where the majority of non-Hispanic Black births occur. However, the disproportionate wildfire burden on AIAN and Hispanic populations represents an emerging environmental justice concern which compounds historical disparities in birth outcomes, especially in AIAN communities (47). We emphasize that these subgroup comparisons describe the population-level distribution of exposure across maternal racial/ethnic groups, reflecting the joint distribution of maternal residential geography and ambient air quality, and should not be interpreted as causal effects of maternal race/ethnicity on exposure.

Our stratification by perinatal healthcare resources reveals a compounding vulnerability: communities with the largest wildfire contribution increases are often the same ones with the least access to specialized perinatal care. Counties with maternity hospital shortages experienced wildfire contributions nearly double those of adequate-access counties by 2017–2019 (10.57% vs. 7.62%), and counties far from NICUs saw contributions reaching 11.52%. These communities—often small, rural, and in the western US—face rising air pollution burdens from wildfire smoke coupled with limited access to the specialized obstetric and neonatal care needed to manage pollution-related pregnancy complications. Prior work has documented that these healthcare-shortage areas tend to have fewer OB-GYN providers, greater distances to maternity hospitals, and greater distances to NICUs (31), suggesting limited capacity for clinical response to wildfire-related health risks. This finding extends our population perspective beyond demographic subgroups to identify the healthcare infrastructure dimension of wildfire exposure disparities.

Our shift-share decomposition clarifies that the dominant channel for changing prenatal exposure is within-county air quality change, not population redistribution. WUI expansion does not appear to be a primary driver of changing prenatal PM_2.5 exposure at the regional scale. The wildfire emissions themselves must be addressed through fuel management, prescribed burns, and climate adaptation (48).

Our study is the first to take a comprehensive population-based perspective on the long-term evolution of prenatal air pollution exposure, documenting how wildfire contributions have changed across all regions of CONUS. Our birth-weighted approach emphasizes human impact rather than geographic coverage, providing estimates directly relevant to public health resource allocation. The counterfactual framework quantifies the specific cost of increasing wildfire activity in terms of foregone air quality improvement. Our stratification by perinatal healthcare resources adds policy-relevant context about which communities face the greatest combined burden of rising wildfire exposure and limited healthcare access.

Several limitations warrant consideration. County-level residence at birth may not capture residential mobility during pregnancy (49, 50). Our chemical transport modeling approach for wildfire attribution involves uncertainties related to emissions inventories and model parameterizations (34–36). Alternative approaches have different limitations (51, 52), and no consensus exists on optimal wildfire attribution methods. We could not account for behavioral adaptations that may modify actual exposures, and these protective behaviors are likely unequally distributed across socioeconomic groups. Our perinatal healthcare resource measures are time-invariant and may not capture facility openings, closures, or volume changes during the study period; however, this approach describes current resource availability relevant to ongoing clinical and policy decisions. For gestational age, OE was missing for California in 2004–2006 and was imputed via regression on LMP-based gestational age and other covariates, following the data-cleaning protocol described in (31); affected births represent less than 3% of the study population, and sensitivity to this imputation is expected to be small given that exposure is averaged over a 250–300 day pregnancy.

Conclusions

While the US achieved important progress in reducing prenatal PM_2.5 exposure during 2003–2019, increasing wildfire activity systematically offset a portion of these gains, with wildfire's share of prenatal PM_2.5 more than doubling nationally. Approximately two-thirds of remaining prenatal PM_2.5 exceedance days are now attributable to wildfire, and the burden falls disproportionately on low-income, rural, and Indigenous communities. Approximately 630,700 infants per year are born in counties with maternity hospital shortages where wildfire contributions exceeded 10% by the end of the study period.

Traditional emission reduction policies, based on the regulation of anthropogenic sources, will not be sufficient to address wildfire-related exposures and it is unclear whether US policymakers will prioritize strategies that will reduce the occurrence of mega-fires that have widespread population impacts. Healthcare providers in fire-prone regions should incorporate wildfire risk assessment and education into prenatal care protocols, with particular attention to communities with limited perinatal healthcare infrastructure. Enhanced air quality monitoring, early warning systems, and the distribution of climate adaptation technologies (air filters, masks, and clean air shelters) may help mitigate wildfire-related PM_2.5 burdens.

References

• 1.

  CurrieJWalkerR. What do economists have to say about the clean air act 50 years after the establishment of the environmental protection agency?J Econ Perspect. (2019) 33(4):3–26. 10.1257/jep.33.4.3

  • CrossRef
  • Google Scholar
  • View reference in article

• 2.

  AldyJEAuffhammerMCropperMFraasAMorgensternR. Looking back at 50 years of the clean air act. J Econ Lit. (2022) 60(1):179–232. 10.1257/jel.20201626

  • CrossRef
  • Google Scholar
  • View reference in article

• 3.

  U.S. Environmental Protection Agency, Office of Air and Radiation. The Benefits and Costs of the Clean Air Act from 1990 to 2020: Second Prospective Study. Washington, DC: EPA (2011).

  • Google Scholar
  • View reference in article

• 4.

  CurrieJZivinJGMullinsJNeidellM. What do we know about short- and long-term effects of early-life exposure to pollution?Annu Rev Resour Econ. (2014) 6(1):217–47. 10.1146/annurev-resource-100913-012610

  • CrossRef
  • Google Scholar
  • View reference in article

• 5.

  BellMLBelangerKEbisuKGentJFLeeHJKoutrakisPet al. Prenatal exposure to fine particulate matter and birth weight: variations by particulate matter composition and sources. Epidemiology. (2010) 21(6):884–91. 10.1097/EDE.0b013e3181f0a860

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 6.

  McClureCDJaffeDA. US Particulate matter air quality improves except in wildfire-prone areas. Proc Natl Acad Sci USA. (2018) 115(31):7901–6. 10.1073/pnas.1800351115

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 7.

  ClayKMullerNZWangX. Recent increases in air pollution: evidence and implications for mortality. Rev Environ Econ Policy. (2021) 15(1):154–62. 10.1086/711999

  • CrossRef
  • Google Scholar
  • View reference in article

• 8.

  BurkeMChildsMLde la CuestaBQiuMLiJGouldCFet al. The contribution of wildfire to PM_2.5 trends in the USA. Nature. (2023) 622:761–6. 10.1038/s41586-023-06522-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 9.

  BurkeMDriscollAHeft-NealSXueJBurneyJWaraM. The changing risk and burden of wildfire in the United States. Proc Natl Acad Sci USA. (2021) 118(2):e2011048118. 10.1073/pnas.2011048118

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 10.

  WesterlingAL. Increasing western US forest wildfire activity: sensitivity to changes in the timing of spring. Philos Trans R Soc B Biol Sci. (2016) 371:20150178. 10.1098/rstb.2015.0178

  • CrossRef
  • Google Scholar
  • View reference in article

• 11.

  AbatzoglouJTWilliamsAP. Impact of anthropogenic climate change on wildfire across western US forests. Proc Natl Acad Sci USA. (2016) 113(42):11770–5. 10.1073/pnas.1607171113

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 12.

  AguileraRCorringhamTGershunovABenmarhniaT. Wildfire smoke impacts respiratory health more than fine particles from other sources: observational evidence from southern California. Nat Commun. (2021) 12:1493. 10.1038/s41467-021-21708-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 13.

  DeFlorio-BarkerSCrooksJReyesJRappoldAG. Cardiopulmonary effects of fine particulate matter exposure among older adults, during wildfire and non-wildfire periods, in the United States 2008–2010. Environ Health Perspect. (2019) 127(3):037006. 10.1289/EHP3860

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 14.

  LiuJCPereiraGUhlSABravoMABellML. A systematic review of the physical health impacts from non-occupational exposure to wildfire smoke. Environ Res. (2015) 136:120–32. 10.1016/j.envres.2014.10.015

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 15.

  ReidCEBrauerMJohnstonFHJerrettMBalmesJRElliottCT. Critical review of health impacts of wildfire smoke exposure. Environ Health Perspect. (2016) 124(9):1334–43. 10.1289/ehp.1409277

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 16.

  WilliamsKMFranziLMLastJA. Cell-specific oxidative stress and cytotoxicity after wildfire coarse particulate matter instillation into mouse lung. Toxicol Appl Pharmacol. (2013) 266(1):48–55. 10.1016/j.taap.2012.12.017

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 17.

  WegesserTCPinkertonKELastJA. California Wildfires of 2008: coarse and fine particulate matter toxicity. Environ Health Perspect. (2009) 117(6):893–7. 10.1289/ehp.0800889

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 18.

  Heft-NealSDriscollAYangWShawGBurkeM. Associations between wildfire smoke exposure during pregnancy and risk of preterm birth in California. Environ Res. (2022) 203:111872. 10.1016/j.envres.2021.111872

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 19.

  PicciottoSHuangSLurmannFPavlovicNChangSYMukherjeeAet al. Pregnancy exposure to PM_2.5 from wildland fire smoke and preterm birth in California. Environ Int. (2024) 186:108583. 10.1016/j.envint.2024.108583

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 20.

  HolstiusDMReidCEJesdaleBMMorello-FroschR. Birth weight following pregnancy during the 2003 southern California wildfires. Environ Health Perspect. (2012) 120(9):1340–5. 10.1289/ehp.1104515

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 21.

  BrewBKGibsonPGCollisonAKarmausWSzwecSHollidayEet al. Prenatal wildfire smoke exposure and adverse neonatal outcomes in a high-risk cohort of pregnant women with asthma. Am J Obstet Gynecol. (2026) 18. 10.1016/j.ajog.2026.02.022

  • CrossRef
  • Google Scholar
  • View reference in article

• 22.

  ArdiyaniVKellyFPhillipsDWoosterMGrosvenorM. Reflection of prenatal exposure to wildfire smoke on reduced birth weight in 2015–2019. E3S Web Conf. (2024) 485:07010. 10.1051/e3sconf/202448507010

  • CrossRef
  • Google Scholar
  • View reference in article

• 23.

  AbdoMWardIO’DellKFordBPierceJRFischerEVet al. Impact of wildfire smoke on adverse pregnancy outcomes in Colorado, 2007–2015. Int J Environ Res Public Health. (2019) 16(19):3720. 10.3390/ijerph16193720

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 24.

  FooDStewartRHeoSDhamraitGChoiHMSongYet al. Wildfire smoke exposure during pregnancy and perinatal, obstetric, and early childhood health outcomes: a systematic review and meta-analysis. Environ Res. (2024) 241:117527. 10.1016/j.envres.2023.117527

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 25.

  ZhangYYeTYuPXuRChenGHuangCet al. Preterm birth and term low birth weight associated with wildfire-specific PM_2.5: a cohort study in New South Wales, Australia during 2016–2019. Environ Int. (2022) 174:107879. 10.1016/j.envint.2022.107879

  • CrossRef
  • Google Scholar
  • View reference in article

• 26.

  BasilioEChenRFernandezACPadulaAMRobinsonJFGawSL. Wildfire smoke exposure during pregnancy: a review of potential mechanisms of placental toxicity, impact on obstetric outcomes, and strategies to reduce exposure. Int J Environ Res Public Health. (2022) 19(21):13727. 10.3390/ijerph192113727

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 27.

  WhyattRMBellDAJedrychowskiWSantellaRMGarteSJCosmaGet al. Polycyclic aromatic hydrocarbon-DNA adducts in human placenta and modulation by CYP1A1 induction and genotype. Carcinogenesis. (1998) 19(8):1389–92. 10.1093/carcin/19.8.1389

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 28.

  PereraFTangDWhyattRLedermanSAJedrychowskiW. DNA Damage from polycyclic aromatic hydrocarbons measured by benzo[a]pyrene-DNA adducts in mothers and newborns from northern manhattan, the world trade center area, Poland, and China. Cancer Epidemiol Biomarkers Prev. (2005) 14(3):709–14. 10.1158/1055-9965.EPI-04-0457

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 29.

  ChoiHWangLLinXSpenglerJDPereraFP. Fetal window of vulnerability to airborne polycyclic aromatic hydrocarbons on proportional intrauterine growth restriction. PLoS One. (2012) 7(4):e35464. 10.1371/journal.pone.0035464

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 30.

  DrwalERakAGregoraszczukEL. Review: polycyclic aromatic hydrocarbons (PAHs)-action on placental function and health risks in future life of newborns. Toxicology. (2019) 411:133–42. 10.1016/j.tox.2018.10.003

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 31.

  BoudreauxMYoderMEllicottEPassarellaMLorchS. Perinatal health care resources and wildfire smoke. Med Care. (2025) 63(6):396–404. 10.1097/MLR.0000000000002130

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 32.

  FoleyKPouliotGEythAGillilandAAllenCAppelWet al. EQUATES: EPA’s air QUAlity TimE series project. In: Presented at: 19th Annual CMAS Conference (2020).

  • Google Scholar
  • View reference in article

• 33.

  SimonHAppelWFoleyKBhavePWhitehillASperoTet al. Modeled and observed 2002–2019 trends in ozone day-of-week patterns across US nonattainment areas. Atmos Chem Phys. (2024) 24:1855–71. 10.5194/acp-24-1855-2024

  • CrossRef
  • Google Scholar
  • View reference in article

• 34.

  O’DellKFordBFischerEVPierceJR. Contribution of wildland-fire smoke to US PM_2.5 and its influence on recent trends. Environ Sci Technol. (2019) 53(4):1797–804. 10.1021/acs.est.8b05430

  • CrossRef
  • Google Scholar
  • View reference in article

• 35.

  ChildsMLLiJWenJHeft-NealSDriscollAWangSet al. Daily local-level estimates of ambient wildfire smoke PM_2.5 for the contiguous US. Environ Sci Technol. (2022) 56(19):13607–21. 10.1021/acs.est.2c02934

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 36.

  ReidCEConsidineEMMaestasMMLiG. Daily PM_2.5 concentration estimates by county, ZIP code, and census tract in 11 western states 2008–2018. Sci Data. (2021) 8:112. 10.1038/s41597-021-00891-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 37.

  Center for International Earth Science Information Network (CIESIN), Columbia University. Gridded Population of the World, Version 4 (GPWv4): Population Count. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC) (2018).

  • Google Scholar
  • View reference in article

• 38.

  National Center for Health Statistics. User Guide to the 2019 Natality Public Use File. Hyattsville, MD: NCHS, Division of Vital Statistics (2020).

  • Google Scholar

• 39.

  NCEI. U.S. Climate Regions. National Oceanic and Atmospheric Administration (2023). Available online at:https://www.ncei.noaa.gov/access/monitoring/reference-maps/us-climate-regions. (Accessed July 6, 2026)

  • Google Scholar
  • View reference in article

• 40.

  KarlTKossWJ. Regional and national monthly, seasonal, and annual temperature weighted by area, 1895-1983 (1984). Available online at:https://repository.library.noaa.gov/view/noaa/10238. (Accessed July 6, 2026)

  • Google Scholar

• 41.

  KitagawaEM. Components of a difference between two rates. J Am Stat Assoc. (1955) 50(272):1168–94. 10.1080/01621459.1955.10501299

  • CrossRef
  • Google Scholar
  • View reference in article

• 42.

  CurrieJVoorheisJWalkerR. What caused racial disparities in particulate exposure to fall? New evidence from the clean air act and satellite-based measures of air quality. Am Econ Rev. (2023) 113(1):71–97. 10.1257/aer.20191957

  • CrossRef
  • Google Scholar
  • View reference in article

• 43.

  ColmerJHardmanIShimshackJVoorheisJ. Disparities in PM_2.5 air pollution in the United States. Science. (2020) 369(6503):575–8. 10.1126/science.aaz7497

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 44.

  ArdiyaniVGrosvenorMWoosterMLestariPSuriW. The infiltration of wildfire smoke and its potential dose on pregnant women: lessons learned from Indonesia wildfires in 2019. Heliyon. (2023) 9(8):e18513. 10.1016/j.heliyon.2023.e18513

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 45.

  TessumCWApteJSGoodkindALMullerNZMullinsKAPaolellaDAet al. Inequity in consumption of goods and services adds to racial–ethnic disparities in air pollution exposure. Proc Natl Acad Sci USA. (2019) 116(13):6001–6. 10.1073/pnas.1818859116

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 46.

  MohaiPPellowDRobertsJT. Environmental justice. Annu Rev Environ Resour. (2009) 34:405–30. 10.1146/annurev-environ-082508-094348

  • CrossRef
  • Google Scholar

• 47.

  ElyDMDriscollAK. Infant mortality in the United States, 2018: data from the period linked birth/infant death file. Natl Vital Stat Rep. (2020) 69(7):1–18.

  • Google Scholar

• 48.

  SchoennagelTBalchJKBrenkert-SmithHDennisonPEHarveyBJKrawchukMAet al. Adapt to more wildfire in western North American forests as climate changes. Proc Natl Acad Sci USA. (2017) 114(18):4582–90. 10.1073/pnas.1617464114

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 49.

  ChenLBellEMCatonARDruschelCMRomittiPAShawGMet al. Residential mobility during pregnancy and the potential for ambient air pollution exposure misclassification. Environ Res. (2010) 110(2):162–8. 10.1016/j.envres.2009.11.006

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 50.

  HodgsonSLurzPWShirleyMDBythellMRankinJ. Exposure misclassification due to residential mobility during pregnancy and its effects on the association between PM_2.5 and adverse birth outcomes. Int J Hyg Environ Health. (2015) 218(4):414–21. 10.1016/j.ijheh.2015.01.001

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 51.

  GanRWFordBLassmanWPfisterGVaidyanathanAFischerEet al. Comparison of wildfire smoke estimation methods and associations with cardiopulmonary-related hospital admissions. GeoHealth. (2017) 1(3):122–36. 10.1002/2017GH000073

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 52.

  LarsenAEReichBJRuminskiMRappoldAG. Impacts of fire smoke plumes on regional air quality, 2006–2013. J Expo Sci Environ Epidemiol. (2018) 28(4):319–27. 10.1038/s41370-017-0013-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar