Susceptible windows of long-term childhood exposure to air pollution on adult self-reported bronchitic symptoms

Abstract

Background

Childhood exposure to air pollution has long-term effects on adult bronchitic symptoms, but the age windows of susceptibility are understudied.

Methods

We included 1444 participants from the Southern California Children’s Health Study, who were recruited at ages ∼9–10 years in 1992–1993 or 1995–1996 or ages ∼5–7 years in 2002–2003, followed until high-school graduation, and re-contacted again in adulthood (mean age = 33 years) to collect self-reported bronchitic symptoms. Yearly average nitrogen dioxide (NO₂), 8-h maximum ground-level ozone (O₃), and particulate matter of ≤10 µm in diameter (PM₁₀) were estimated by using inverse-distance squared spatial interpolation to participants’ residential history from conception to age 16 years. Log Poisson Distributed Lag Models were fitted to identify susceptible windows of childhood exposure to air pollution on adult bronchitic symptoms adjusted for childhood and adult confounders. We explored sex-specific susceptible windows.

Results

We identified ages 1–2 years as a susceptible window in which NO₂ exposure was associated with a higher risk of adult bronchitic symptoms, with the largest associations observed at age 1 year (risk ratio per 10 ppb = 1.12; 95% confidence interval: 1.01, 1.25). We observed both positive (ages 12–15 years) and inverse (ages 8–11 years) associations with O₃ exposure. Suggestive evidence of increased risk at ages 3–4 years was observed for PM₁₀. There was no evidence of sex differences.

Conclusion

Early childhood might be a particularly susceptible window of exposure to NO₂ (ages 1–2 years) and possibly for PM₁₀ (ages 3–4 years) for increased risk of adult bronchitic symptoms, while early adolescence (ages 12–15 years) might be a susceptible window for O₃ exposure.

Key Messages.

• We assessed susceptible childhood windows for nitrogen dioxide (NO₂), ozone, and particulate matter ≤10 µm in diameter exposure associated with self-reported adult bronchitic symptoms.

• We found evidence that early life (ages 1–2 years) is a susceptible window for NO₂ exposure. There was no evidence of differences by sex.

• Our findings add unique knowledge of susceptible windows for childhood exposure to air pollution for adult respiratory diseases.

Introduction

Lung development begins in utero and continues through adolescence [1], reaching peak volume by age 18 years for females and age 20 years for males [2–6]. Various factors, including body size, behavior (e.g. frequent inhalation and outdoor activities), and immature physiological systems, can influence lung development in childhood, making lungs more vulnerable to environmental exposures [2, 7–9]. Accumulating evidence suggests that chronic exposure to air pollution during childhood can delay lung growth and accelerate functional decline into adulthood [2, 10–12].

Air pollution, such as particulate matter (PM), nitrogen oxides, and ground-level ozone (O₃), can irritate airways through oxidative stress and lead to lung inflammation [13, 14]. Long-term exposure to those air pollutants may result in declining lung function and bronchitic symptoms, including cough, phlegm, and chest congestion [13, 14]. These chronic bronchitic symptoms have been recognized as phenotypes of chronic obstructive pulmonary disease in adults [15, 16] and are correlated with other lung diseases, e.g. lung cancer and asthma [17].

We previously showed that long-term childhood exposure to nitrogen dioxide (NO₂) and PM of ≤10 µm in diameter (PM₁₀) was associated with higher odds of adult bronchitic symptoms, independently of childhood respiratory conditions [18]. This suggests that childhood represents a susceptible period for long-term air-pollution effects. However, existing studies have not identified specific susceptible windows because air-pollution-exposure measurements over long pre-specified time windows (e.g. across several years as in previous studies [18]) are often not sufficiently temporally resolved to identify key specific ages of susceptibility. To address this gap, we leveraged high-temporal-resolution annual average exposure estimates of PM₁₀, NO₂, and O₃ from conception to age 16 years in the Southern California Children’s Health Study (CHS) to examine associations with self-reported adult bronchitic symptoms. We employed data-driven distributed lag models (DLMs) to identify susceptible air-pollution-exposure windows in childhood and investigated effect modification by sex.

Methods

Study population

We focused on three waves of cohort recruitment of the CHS [18–20]: fourth-graders (ages ∼9–10 years) were recruited in 1992–1993 (Cohort ‘C’) and 1995–1996 (Cohort ‘D’) and kindergarteners and first-graders (ages ∼5–7 years) were recruited in 2002–2003 (Cohort ‘E’). A consistent recruitment protocol was used for all three cohorts and all were followed annually, biannually, or biennially with questionnaires through high school. Beginning in 2017, a subset (n = 1444) was re-contacted for adult follow-up. After the exclusion of 79 participants with missing outcomes and one duplicate, 1365 participants were included (Supplementary Fig. S1). Details of the follow-up study are published elsewhere [18].

All parents/guardians of participating children provided written informed consent during the childhood study and child participants provided written assent. During the adult follow-up assessment, participants provided electronic informed consent. The study protocol was approved by the institutional review board of the University of Southern California.

Outcome definition

Adult bronchitic symptoms were identified as self-reported ‘yes’ to any of the following questions during the adult assessment: (i) During the last 12 months, have you had a cough first thing in the morning that lasted for as much as 3 months in a row?; (ii) During the last 12 months, have you had a cough at any other times of the day that lasted for as much as 3 months in a row?; (iii) Other than with colds, do you usually seem congested in the chest or bring up phlegm?; or (iv) During the last 12 months, have you had bronchitis? This approach is consistent with those of previous bronchitic symptoms assessments in the CHS [18, 21–24] and in other studies [25].

Exposure assessment

We assigned annual average ambient PM₁₀, NO₂, and O₃ concentrations to participants’ residential history from pregnancy to age 16 years. We did not include PM of ≤2.5 µm (PM_2.5) due to high missingness from limited monitoring data in the late 1980s and early 1990s in Southern California. Methods for constructing residential histories and exposure assessment were previously published [26]. The monthly average ambient air-pollution concentrations (24-h mean PM₁₀, 24-h mean NO₂, and 8-h maximum O₃) were estimated by using inverse-distance squared spatial interpolation, incorporating the US Environmental Protection Agency’s Air Quality System data [27] and supplemental CHS measurements [20, 28]. Annual averages were estimated from monthly means, requiring ≥75% (9 out of 12 months) of data per age-year.

Covariates

We included several childhood and adult variables as potential confounders based on prior work [18] and knowledge. Childhood variables included race/ethnicity, household income, maternal smoking during pregnancy, presence of smoker in the home, and presence of mold/mildew in the home. Sensitivity analyses included childhood bronchitic symptoms and asthma. Adult variables included age, highest educational attainment, body mass index (BMI), and current smoking status. We additionally considered community at study recruitment and CHS cohort (C, D, or E) to capture spatial and temporal variability in ambient air pollution. To assess sex-specific susceptibility, we stratified analyses by biological sex.

Statistical methods

We first examined temporal correlations between air pollutants by using Spearman’s correlation. To explore susceptible windows, we fitted log Poisson DLMs with robust standard error estimators with a cross-basis of a log-linear exposure–response association based on prior work [18] and a lag–response association using a natural cubic spline smooth function. This approach identifies susceptible windows of exposure via the simultaneous estimation of air-pollution effects for each age-year while minimizing the potential bias from correlated exposures across the ages [29].

To fit the smooth function, we tested various knots numbers and placements (Supplementary Table S1), guided by biological plausibility, age-specific models (Supplementary Fig. S2), unconstrained DLMs, and quasi-information criterion minimization [30] (expanded in the Supplementary Methods). Susceptible windows were identified when age-specific risk ratios (RRs) (per 10-unit increase in exposure) excluded 1. Sensitivity analyses tested knot choices, excluded potential non-confounders (i.e. adult smoking and adult BMI), and adjusted for childhood bronchitic symptoms and asthma to assess independence from early-life respiratory conditions.

Given the significant exposure and covariate differences between retained and excluded participants (Supplementary Table S3), we assessed robustness to missing exposure data by analysing: (i) full exposure data (pregnancy-age 16 years), (ii) subsets with complete exposure data for selected age ranges, and (iii) imputed exposure (see Supplementary Methods). NO₂ and O₃ analyses included the full (N = 769), truncated (ages 1–12 years, N = 1124), and imputed (N = 1332) samples. Due to the high missingness of exposure from pregnancy to age 16 years (87.84%), PM₁₀ analyses were limited to ages 3–12 years. Supplementary Fig. S1 illustrates the flowchart for analytic samples.

We applied Bayesian distributed lag interaction models (BDLIMs) [31] with the full sample to detect heterogeneity by sex (Supplementary Table S3). BDLIM tests subgroup differences in sensitive windows, effect size, or both [31]. Based on the BDLIM findings, we then performed sex-stratified DLMs with the specified knots (Supplementary Table S4).

Analyses were conducted in R version 4.2.2 [32], with the dlnm [33] and regimes [31] packages.

Results

The characteristics of the study population are presented in Table 1. Among those with full NO₂ and O₃ exposure histories, the average age at adult assessment was 33.7 ± 2.34 years. Participants in the NO₂/O₃ (ages 1–12 years) and PM₁₀ subsets were younger, with mean ages of 32.0 ± 4.72 and 30.7 ± 5.04 years, respectively. Most participants were male (61%) and non-Hispanic White (51%–59%) or Hispanic (28%–36%). Approximately 30% reported childhood household incomes of between US$15 000 and US$50 000. Approximately 70% held a bachelor’s degree and 80% were nonsmokers. Twenty-two percent of participants with full exposure history reported bronchitic symptoms at adult assessment. This percentage was higher among the selected age-range (24%) and PM₁₀ (26%) subsets.

Table 1.

Characteristics of participants included in three main analyses and the overall study population, Southern California Children's Health Study.

	Analytic samples			Full study population (N = 1365)
	NO2 and O3 conception to age 16 years (N = 817)	NO2 and O3 ages 1–12 years (N = 1178)	PM10 ages 3–12 years (N = 819)
Adult age (years)
Mean (SD)	33.7 (2.34)	32.0 (4.72)	30.7 (5.04)	32.1 (4.76)
Median (minimum, maximum)	34.0 (23.0, 67.0)	33.0 (20.0, 67.0)	32.0 (20.0, 67.0)	33.0 (20.0, 70.0)
Sex [n, (%)]
Male	497 (61)	714 (61)	497 (61)	820 (60)
Female	320 (39)	464 (39)	322 (39)	545 (40)
Adult body mass index
Mean (SD)	27.7 (6.54)	27.8 (6.60)	27.8 (6.80)	27.8 (6.61)
Median (minimum, maximum)	26.5 (4.78, 61.9)	26.6 (4.78, 61.9)	26.5 (4.78, 61.9)	26.5 (0.345, 61.9)
Missing	0 (0%)	0 (0%)	0 (0%)	33 (2.4%)
Cohort [n (%)]
C	396 (48)	494 (42)	191 (23)	578 (42)
D	415 (51)	492 (42)	437 (53)	572 (42)
E	6 (1)	192 (16)	191 (23)	215 (16)
Childhood town code [n (%)]
Alpine	65 (8)	93 (8)	47 (6)	112 (8)
Lake Elsinore	61 (7)	76 (6)	61 (7)	88 (6)
Lage Gregory	81 (10)	101 (9)	46 (6)	109 (8)
Lancaster	59 (7)	79 (7)	15 (2)	94 (7)
Lompoc	45 (6)	58 (5)	28 (3)	71 (5)
Long beach	86 (11)	116 (10)	117 (14)	137 (10)
Mira Loma	76 (9)	101 (9)	64 (8)	128 (9)
Riverside	65 (8)	96 (8)	81 (10)	104 (8)
San Dimas	70 (9)	115 (10)	113 (14)	127 (9)
Atascadero	72 (9)	84 (7)	29 (4)	97 (7)
Santa Maria	47 (6)	66 (6)	47 (6)	80 (6)
Upland	88 (11)	133 (11)	111 (14)	148 (11)
Glendora	2 (0)	25 (2)	25 (3)	29 (2)
Anaheim	0 (0)	16 (1)	16 (2)	20 (1)
San Bernadino	0 (0)	4 (0)	4 (0)	5 (0)
Santa Barbara	0 (0)	15 (1)	15 (2)	16 (1)
Race/ethnicity [n (%)]
White	482 (59)	660 (56)	418 (51)	773 (57)
Black	30 (4)	40 (3)	29 (4)	45 (3)
Hispanic	232 (28)	379 (32)	295 (36)	429 (31)
Other	73 (9)	99 (8)	77 (9)	118 (9)
Income (US$) [n (%)]
<7500 to 14 999	64 (8)	109 (9)	76 (9)	130 (10)
15 000 to 49 999	299 (37)	398 (34)	263 (32)	465 (34)
50 000 to 99 999	289 (35)	426 (36)	287 (35)	487 (36)
≥100 000	67 (8)	116 (10)	98 (12)	134 (10)
Don’t know/missing	98 (12)	129 (11)	95 (12)	149 (11)
Adulthood latest school [n (%)]
Up to high school	69 (8)	120 (10)	88 (11)	140 (10)
Undergraduate	573 (70)	847 (72)	595 (73)	979 (72)
Postgraduate	175 (21)	211 (18)	136 (17)	246 (18)
Biological mother smoking during pregnancy [n (%)]
No	703 (86)	1027 (87)	724 (88)	1171 (86)
Yes	106 (13)	138 (12)	88 (11)	179 (13)
Don’t know/missing	8 (1)	13 (1)	7 (1)	15 (1)
Any daily smoking in subject’s childhood residence [n (%)]
No	692 (85)	1007 (85)	703 (86)	1161 (85)
Yes	112 (14)	150 (13)	98 (12)	182 (13)
Don’t know/missing	13 (2)	21 (2)	18 (2)	22 (2)
Current smoking status [n (%)]
No	675 (83)	950 (81)	659 (80)	1097 (80)
Yes	132 (16)	208 (18)	147 (18)	246 (18)
Don’t know/missing	10 (1)	20 (2)	13 (2)	22 (2)
Mold/mildew ever in subject’s childhood residence [n (%)]
No	557 (68)	816 (69)	582 (71)	954 (70)
Yes	238 (29)	336 (29)	222 (27)	382 (28)
Don’t know/missing	22 (3)	26 (2)	15 (2)	29 (2)
Self-reported bronchitic symptoms [n (%)]
No bronchitic symptoms	638 (78)	892 (76)	609 (74)	1027 (75)
Bronchitic symptoms	179 (22)	286 (24)	210 (26)	338 (25)
Missing NO2 [n (%)]	48 (5.88%)	54 (4.58%)	17 (2.08%)	596 (43.66%)
Missing O3 [n (%)]	0 (0.00%)	0 (0.00%)	2 (0.24%)	525 (40.15%)
Missing PM10 [n (%)]	658 (80.54%)	633 (53.74%)	0 (0%)	1199 (88.35%)

Current smoking status, based on reporting smoking >0 cigarettes in last 12 months. Body mass index, calculated by dividing self-reported weight by self-reported height squared, kg/m².

The median annual exposure to air-pollutant concentrations decreased at older ages, from a mean of 38.23 (pregnancy) to 23.64 (age 16 years) ppb for NO₂, from 49.78 to 43.30 ppb for O₃, and from 50.56 (age 3 years) to 39.06 (age 12 years) µg/m³ for PM₁₀ (Fig. 1). Correlations across different air-pollutant concentrations were higher at closer ages (Supplementary Fig. S3). Within each age, NO₂ and PM₁₀ had the strongest correlations (Spearman’s correlation coefficient range 0.49–0.82). NO₂ and O₃ were weakly correlated (–0.30 to 0.32), positive until age 11 years and negative from 12 to 16 years, whereas NO₂ and PM₁₀ were highly positively correlated, especially for ages 5–12 years (≥0.74) (Supplementary Fig. S4).

Figure 1.

Median (solid line), 25th–75th percentiles (smaller shaded area), and 5th–95th percentiles (larger shaded area) of annual average exposure concentration by age in the study population, Southern California CHS, California, USA. Age of –1 indicates pregnancy. Note: n = 791 for children with complete NO₂ exposure history from pregnancy to age 16 years; n = 840 for children with complete O₃ exposure history from pregnancy to age 16 years; n = 839 for children with complete PM₁₀ exposure history from ages 3–12 years.

The DLM results (Fig. 2) showed that NO₂ exposure at ages 1–2 years was associated with a higher risk of adult bronchitic symptoms, with the largest associations observed at age 1 year [RR per 10 ppb = 1.12; 95% confidence interval (CI): 1.01, 1.25]. O₃ exposure at ages 4–5 and 12–15 years was associated with increased risk and was the highest at age 14 years (RR = 1.51; 95% CI: 1.23, 1.86). Statistically significant inverse associations were observed for O₃ at ages 8–11 years and were the strongest at age 10 years (RR = 0.80; 95% CI: 0.71, 0.90). The single-pollutant model showed similar patterns, though NO₂ was non-significant in the full-exposure analysis (Supplementary Fig. S5).

Figure 2.

Predicted RRs and 95% confidence intervals (CIs) for associations between 10-ppb increase in annual 24-h average NO₂ and 8-h maximum O₃ exposure and adult bronchitic symptoms from mutually adjusted two-pollutant DLMs among three samples, Southern California CHS, California, USA. Age of –1 indicates pregnancy. Note: Models were adjusted for race and ethnicity, childhood household income, maternal smoking during pregnancy, presence of smoker in childhood home, presence of mold/mildew in childhood home, town at study recruitment, CHS cohort, age, latest education, current smoking status, and BMI at adult assessment, as well as NO₂ or O₃ (mutually adjusted). Solid dots represent the estimated RRs from the fitted DLMs and red dots represent significant findings in which the 95% CIs did not cross the null. Dash horizontal line represents the null value (RR = 1). The three different samples are: (i) children with complete 18 years of exposure history and covariates (complete exposure); (ii) children with complete 12 years of exposure history (ages 1–12 years) and covariates; (iii) full cohort with imputed exposure (18 years) and complete covariates.

In secondary analysis (ages 1–12 years, larger sample), NO₂ showed similar patterns but attenuated associations, with susceptible windows extending to age 3 years (Fig. 2). O₃ associations at ages 8–12 years became non-significant, while inverse associations at ages 1–2 years were significant (Fig. 2). DLM using imputed exposure yielded similar trends but all CIs contained the null (Fig. 2). Results remained consistent after excluding adult covariates or adjusting for childhood respiratory conditions (Supplementary Figs S6 and S7).

BDLIM analysis suggested heterogeneity by sex in the effect size but not in the window for NO₂ and in the window only for O₃ (Supplementary Table S3). However, sex-stratified analyses revealed overlapping lag–response patterns and CIs across ages (Fig. 3). Results were robust to covariate changes (Supplementary Fig. S8).

Figure 3.

Predicted RRs and 95% CIs for associations between 10-ppb increase in annual 24-h average NO₂ and 8-h maximum O₃ exposure and adult bronchitic symptoms from DLMs, stratified by males (n = 306) and females (n = 463), Southern California CHS, California, USA. Age of –1 indicates pregnancy. Note: Female model was adjusted for race and ethnicity, childhood household income, maternal smoking during pregnancy, presence of smoker in childhood home, presence of mold/mildew in childhood home, town at study recruitment, CHS cohort, age, latest education, current smoking status, and BMI at adult assessment, as well as NO₂ or O₃ (mutually adjusted). Male model included the same covariates except maternal smoking during pregnancy due to collinearity. Solid dots represent null estimates. Asterisk symbols represent significant findings in which the 95% CI did not cross the null. Dash horizontal line represents the null value (RR = 1).

After adjustment for other pollutants, PM₁₀ models showed a downward risk trend (Fig. 4). Borderline statistically significant inverse associations were noted at ages 9 and 10 years (RR = 0.93; 95% CI: 0.87, 1.00 and 0.92; 95% CI, 0.85–1.00, respectively) after adjustment for NO₂. When adjusted for O₃, susceptible windows appeared at ages 3–4 years, with the highest RR at age 3 years (RR = 1.13; 95% CI: 1.03, 1.25) (Fig. 4). Additionally, in the sample restricted by PM₁₀ availability, we found susceptible windows at ages 3–4 and 11–12 years for NO₂ in the two-pollutant and three-pollutant models (Fig. 4). Sensitivity analyses excluding adult covariates or including childhood respiratory conditions showed consistent results (Supplementary Figs S9 and S10). However, given the high correlation between PM₁₀ and NO₂, the multi-pollutant model results at older ages should be interpreted cautiously.

Figure 4.

Predicted RRs and 95% CIs for associations between 10-µg/m³ increase in annual 24-h average PM₁₀, 10-ppb increase in annual 24-h average of NO₂, and 8-h maximum O₃, and adult bronchitic symptoms from single-pollutant, two-pollutant (PM₁₀+NO₂; PM₁₀+O₃), and three-pollutant (PM₁₀+NO₂+O₃) DLMs between ages 3 and 12 years, Southern California CHS, California, USA. Note: Models were additionally adjusted for race and ethnicity, childhood household income, maternal smoking during pregnancy, presence of smoker in childhood home, presence of mold/mildew in childhood home, town at study recruitment, CHS cohort, age, latest education, current smoking status, and BMI at adult assessment. Solid dots represent the estimated RRs from the fitted DLMs and red dots represent significant findings in which the 95% CIs did not cross the null. Dash horizontal line represents the null value (RR = 1).

Discussion

With 18 years of exposure history and adult follow-up in the CHS, this study is the first to have explored susceptible windows of childhood ambient air-pollution exposure with adult bronchitic symptoms. Data-driven DLM analyses suggested that early life (ages 1–2 years) might be a particularly susceptible exposure window for NO₂ and early adolescence (ages 12–15 years) for O₃, while O₃ showed inverse associations in mid-childhood (ages 8–11 years). There was some suggestive evidence of an early-life (ages 3–4 years) susceptible window for PM₁₀ exposure.

Studies on the long-term effects of childhood air pollution on adult bronchitic symptoms are limited. Some have investigated related outcomes, such as asthma incidence [34], forced expiratory volume in 1 s (FEV1), and forced vital capacity (FVC) [35, 36]. The Swedish BAMSE study of 2278 children followed from birth until 16 years of age found that exposure to traffic nitrogen oxides and PM₁₀ in the first year of life was associated with reduced FEV1 at age 16 years [35]. The Norway and Sweden RHINESSA study (N = 3428) reported that O₃ exposure from ages 0 to 10 years was associated with lower FEV1 and FVC at age 28 years [36]. PM_2.5 and O₃ during ages 10–18 years were also associated with lower lung function [36]. No association with NO₂ was found, possibly due to the coarse temporal resolution from averaging exposure over 0–10 years, which may have masked the potential early-life effects of NO₂. The Dutch PIAMA study with 3687 participants linked NO₂, PM_2.5, and PM₁₀ at birth to higher asthma incidence until age 20 years [34]. While these studies consistently associated childhood exposure to PM₁₀, NO₂, and O₃ with adverse adult respiratory outcomes, the susceptible exposure windows remain unknown. Few prior studies have focused on prenatal exposures, e.g. [37, 38], limiting the ability to make preventive recommendations across childhood. Limited evidence may stem from methodological challenges, e.g. quantifying time-varying exposures, correlated exposure windows [39], and data limitations, such as insufficient temporal resolution and short follow-up periods. With high temporally resolved air-pollution estimates in the CHS and participants aging into adulthood, we had a unique opportunity to explore these associations.

Our findings for NO₂ align with those of a previous CHS study that reported NO₂ as a strong predictor for childhood bronchitic symptoms [22]. Our identification of the age of 1–2 years as a susceptible exposure window is consistent with suggestive evidence from PIAMA, BAMSE, and RHINESSA [34–36]. Findings were consistent across multi-pollutant models with varying age ranges and co-pollutant adjustments. The results for the single-pollutant model should be interpreted cautiously due to co-exposure with O₃.

For O₃, increased risks at ages 4–5 and 12–15 years were observed in the full age-range analysis, but these associations were not consistent in secondary or PM₁₀ analyses. Inverse associations at ages 8–11 years appeared only in the complete sample. A previous CHS study also found a statistically non-significant inverse association between O₃ and childhood bronchitic symptoms [22]. The long-term effects of ozone on respiratory health remain inconclusive [40–42]. Some primate studies suggest that O₃ may impair lung function [40], possibly leading to long-term respiratory symptoms. The observed inverse association may reflect unmeasured confounding, such as outdoor activities [43, 44], or exposure misclassification due to indoor–outdoor differences, such as building filtration and air exchanges while people are staying indoors [45, 46]. A study of adult asthma found protective associations with summertime O₃ only in single-pollutant models, underscoring the importance of multi-pollutant approaches [47]. Our age-based exposure metric limits direct parameterization of summertime O₃. Future studies should explore seasonal O₃ measures to evaluate the persistence of this paradoxical effect.

In our study sample, PM₁₀ was highly correlated with NO₂ from age 5 years onward. After adjustment for NO₂, some early-life susceptible windows for PM₁₀ remained, but the CIs widened, with some containing the null. Inverse associations at ages 9–10 years were inconsistent, possibly due to multicollinearity. Our previous study found the average childhood PM₁₀ (birth to 17 years) to be associated with increased odds of bronchitic symptoms in adulthood [18]. It is important to note that the present analysis required more complete age-specific exposure, reducing the sample (n = 819) and ages examined (3–12 years) compared with those of the previous study (n = 1204, birth to 18 years PM₁₀ exposure average). Other evidence also linked childhood PM₁₀ exposure to lung function decline. A CHS study reported non-statistically significant decreased FEV1 growth from ages 10–18 years [48] and the UK-MAAS birth cohort reported a 1.37% (2.52%, 0.23%) decrease in FEV1 with lifetime PM₁₀ exposure (birth to 11 years) [49].

Sex differences in the impact of air pollution on lung development have been reported, but the findings vary by developmental stages [37, 50–53]. Some studies suggest greater male susceptibility in infancy [37, 54]. For example, mid-gestation PM_2.5 exposure is associated with asthma by age 6 years in males but not in females [45]. Males also more frequently exhibit early-life respiratory symptoms, such as chronic cough, respiratory syncytial virus infection, and respiratory distress syndrome at younger ages [50]. Conversely, puberty-related studies suggest that estrogen may exacerbate, while androgens may alleviate, asthma symptoms [50–52]. In our study, we found no evidence of sex differences, possibly due to limited power in the sex-stratified analyses.

Our findings aligned with biological mechanisms. The alveolar stage of lung development (36 weeks’ gestation to age 10 years) is the longest [55], with peak alveolar multiplication (increasing lung size) at ages 1–2 years and completion by 8–10 years [1, 55]. Air-pollution exposure during this stage, overlapping with our identified susceptible windows, may disrupt developmental programming, leading to reduced lung function and adult respiratory symptoms. Two primary hypothesized mechanisms are oxidative stress and inflammation. Air pollutants can induce reactive oxygen species and chronic inflammation [14, 56], contributing to sputum production and respiratory symptoms such as daily cough, congestion, phlegm, and/or bronchitis, collectively known as bronchitic symptoms and phenotypes of respiratory dysfunction [57].

Our study has several strengths. First, by leveraging the long-term CHS cohort with detailed residential histories, we estimated the annual average residential air-pollution exposures from conception to age 16 years, capturing yearly exposure dynamics beyond fixed time windows. Second, long follow-up into adulthood allowed us to study the impact of childhood exposure on adult (>18 years old) respiratory outcomes. Only two cohorts (RHINESSA [36] and PIAMA [34]) have explored such an association. Rich follow-up questionnaire data enabled adjustment for various adult and childhood confounders. Third, we used a data-driven DLM approach to identify susceptible windows that have not been studied and applied in this context before. We note, however, that DLM results can be sensitive to parameter input such as knot selection and placement.

Our study also has limitations. First, bronchitic symptoms were self-reported based on recall of the prior 12 months, representing only a snapshot of adult respiratory health and potentially introducing outcome misclassification. However, as participants were unaware of their annual childhood air-pollution exposure when completing the questionnaires, any misclassification is likely non-differential. Symptom-based assessments have been used in multiple CHS investigations and have revealed associations with air pollution, particularly in studies of childhood bronchitis [18, 21–24]. Second, missing exposure data—due to monitoring data gaps or residential histories—posed challenges, particularly for the DLMs. The significant differences in exposure and covariates between included and excluded participants suggest potential selection bias. We evaluated this by analysing additional models with restricted age ranges and imputed exposures. The results moved towards the null in the imputed exposure analysis, possibly from noise in the single imputations; however, we found consistent NO₂ results across the restricted age models. Third, exposures were assessed by using kriging, which has lower spatial resolution than advanced spatiotemporal models, potentially causing exposure misclassification. Fourth, we did not adjust for current air pollution, which could have confounded the results if correlated with childhood exposure and adult outcomes. However, the previous study investigating average childhood exposure and adult bronchitic symptoms in the same CHS data found little impact of current exposure adjustment on the association of childhood exposure with adulthood bronchitic symptoms [18]. Additionally, we adjusted for proxies of childhood and adult socioeconomic status and baseline residential towns, which account for some correlations between current and childhood exposures. Thus, current exposure is unlikely to be a major source of bias. Other unmeasured confounders, such as childhood antioxidant intake, may have introduced bias if correlated with exposures. Last, the limited sample size may have constrained generalizability beyond the study population.

In conclusion, our study provides new insights linking the timing of childhood air-pollution exposure to adult respiratory health. Our results highlighted that adult bronchitic symptoms are susceptible to NO₂ exposure during early life (1–2 years). Another susceptible window might be early adolescence (12–15 years) for O₃ and early childhood (3–4 years) for PM₁₀, but more studies, especially with larger samples, are needed for robust conclusions. Our study emphasizes that reducing air-pollution exposure is critical for children with ongoing lung development.

Ethics approval

The study protocol was approved by the institutional review board of the University of Southern California.

Author contributions

F.C., Z.N., C.V.B., T.B., and E.G. designed the study. C.V.B. and E.G. provided supervision. F.C. and Z.N. conducted statistical analysis. F.C., Z.N., C.V.B., and E.G. interpreted the results and drafted the manuscript. R.H., N.R.P., and F.L. acquired the exposure data. All authors provided critical revision of the manuscript.

Supplementary data

Supplementary data is available at IJE online.

Conflict of interest

None declared.

Funding

This work was supported by the National Institutes of Health (grant# UH3OD023287) and National Institute of Environmental Health Sciences (grant# P30ES007048).

Data availability

Due to the limitations in the original consent forms and HIPAA requirements, the data from the CHS cannot be freely available in the manuscript, supplemental files, or in a public repository. However, we are committed to sharing the data and results acquired as part of this study. The CHS has a process in place for data sharing that involves approval of proposals by a Data Sharing Committee. Investigators who want access to the data will be required to submit a research protocol, which will be reviewed by the CHS Health Data Release Committee and the USC IRB. Please send requests to access this dataset to Dr Frank Gilliland (gillilan@usc.edu).

Use of artificial intelligence (AI) tools

An AI-based tool, ChatGPT, was used to help shorten the introduction and discussion from a prior manuscript version with a higher word count. After using this tool/service, the authors carefully reviewed and revised the shortened text and take full responsibility for the content of the publication. There was no other use of AI tools.