The associations between gas cooking stoves, indoor NO₂ concentrations, and adverse sleep outcomes in a pediatric sample

Abstract

Study Objectives

To investigate whether gas cooking stove exposure and elevated indoor nitrogen dioxide (NO₂) concentration were associated with adverse sleep outcomes in a pediatric sample.

Methods

Children from urban neighborhoods in Boston, Massachusetts underwent in-home sleep assessments. Indoor NO₂ concentrations were measured continuously over 7 days by devices placed in the participants’ living areas. Primary outcomes were short sleep duration (average 7-day sleep duration <8 h by wrist actigraphy), and sleep-disordered breathing (SDB; ≥5 events/hour with ≥3% desaturation by a home sleep monitor). Associations between gas cooking stove exposure and elevated NO₂ (≥ 69.48 ppb) with each sleep outcome were assessed through logistic regression models, adjusting for demographic and socioeconomic factors and season. Sensitivity analyses further adjusted for health conditions, kitchen ventilation, and various sources of indoor NO₂.

Results

Of the 242 children, 74% (n = 178) were exposed to gas cooking stoves. The median (interquartile range) of the average daily 95th percentile indoor NO₂ was 41.1 (38.4) ppb. Children exposed to elevated indoor NO₂ level were at a 2.88 increased adjusted odds (95% CI: 1.27, 6.55, p = .012) of short sleep duration compared to children exposed to lower levels. A positive but insignificant relationship between indoor NO₂ exposure and SDB was found (odds ratios = 1.23, 0.61, 2.47). Gas cooking stove exposure was unassociated with any sleep outcome.

Conclusions

Exposure to elevated indoor NO₂ was associated with higher odds of short sleep duration in children. Interventions targeting indoor air quality may provide a novel approach for improving sleep health and reducing pediatric sleep disparities.

Statement of Significance Emerging data implicates air pollution as a contributor to pediatric sleep disparities. Prior research mainly focused on particulate matter, with limited data regarding the relationship between nitrogen dioxide (NO₂), and sleep. The present study suggests that children exposed to elevated levels of indoor NO₂, which exceed the environmental protection agency’s annual average standard of 53 ppb, had an almost three times increased odds of actigraphy-measured short sleep duration, even after accounting for various confounders. This finding further supports the role of indoor air quality in children’s sleep health and suggests a novel target for interventions that may reduce pediatric sleep disparities.

Graphical Abstract

Graphical Abstract.

Introduction

Sleep plays a critical role in children’s health and development. Unfortunately, inadequate sleep duration and sleep-disordered breathing (SDB) are prevalent and disproportionately occur in children from minoritized or low-income backgrounds [1, 2]. Exposures to adverse indoor and outdoor environments may contribute to these sleep health disparities, with growing evidence that air pollution adversely affects sleep health [3].

Nitrogen dioxide (NO₂) is one of the most health-threatening gaseous components of air pollution with both indoor (e.g. gas cooking and heating) and outdoor (e.g. traffic and industrial processes) sources [4]. Indoor NO₂ could penetrate deeply into the lung, alter airway caliber and viscoelastic properties, and infiltrate the central nervous system—leading to reduced lung function, impaired gas exchange, and neuroinflammation [5–8]. An adverse impact of indoor NO₂ on sleep may occur secondary to effects on upper airway irritation, inflammation, and neurotoxicity, potentially increasing risk of SDB and/or impaired sleep duration and continuity [8, 9].

Most existing NO₂ research focused on all-cause mortality and cardio-respiratory outcomes [4, 10–12]. Only a limited number of studies reported associations between NO₂ exposure and adverse sleep outcomes, including increased sleep disturbance [13], reduced sleep duration [14], poor sleep quality [15], as well as increased risk of SDB [16–18], though the results were inconsistent [19, 20]. In addition, most studies focused on outdoor NO₂ exposure, but the health effects of indoor NO₂ exposure may vary from outdoor exposures due to different sources, co-exposures, and intensity of exposures. Elevated indoor NO₂ levels may be especially relevant to the sleep health of children since children spend longer periods indoors asleep than adults. Only one study, conducted in a school-based sample of children ages 4–8 years in Australia, found a 4.5 increased odds of habitual snoring among children exposed to increased bedroom levels of NO₂ [21].

The existing literature both supports indoor NO₂ exposure as an independent risk factor for adverse health conditions [12], as well as an increased vulnerability of children to elevated NO₂ levels given their developing immune, respiratory, and nervous systems and the time they often spend indoors [22, 23]. Given the potential impact of indoor NO₂ on sleep health, we studied the association between exposure to gas cooking stoves which is the primary source of indoor NO₂, and elevated indoor NO₂ concentrations with several sleep outcomes in a sample of children living in an urban area from predominantly low-income households. We hypothesized that use of gas cooking stoves and increased indoor NO₂ level would be associated with short sleep duration and SDB. In addition, given the limited data available and interest in sleep as a multi-dimensional construct, we evaluated associations with sleep efficiency and fragmentation, sleep disturbance and sleep-related impairment, and SDB-related quality of life.

Methods

Study design

The Environmental Assessment of Sleep Youth (EASY) study aimed to study indoor environmental risk factors for sleep problems in children living in urban communities in or near Boston, Massachusetts. The study recruited 330 children living from the 41 urban public elementary schools participating in the School Inner-City Asthma Study (SICAS) [24], which administered school-wide health surveys to the students’ caregivers; or from the discharge registers of the Boston Children’s Hospital (BCH) Emergency Department, which primarily serves families covered by public health insurance. Eligible participants were aged 6–12 who had lived in their current residence for at least one month or for 50% of the time prior to enrollment and who had a caregiver who spoke English or Spanish. The study excluded children with severe chronic diseases or conditions that would impede the ability to complete the study procedures. The study was approved by the Boston Children’s Hospital Institutional Review Board. Before enrollment, written informed consent from caregivers and assent obtained from the children. Details of the study design and procedure were described in a previous publication [25].

Home visits were conducted by research staff for the purposes of characterizing the home physical environment (e.g. kitchen ventilation characteristics and combustion sources), to collect dust for allergens, and to set up sleep and environmental monitors to collect in-home assessments of sleep and air quality [26]. After the implementation of restrictions related to the COVID-19 pandemic (March 2020), home visits were modified such that caregivers were instructed on setting up of monitors and the home examination occurred via video-call. Children enrolled before the start of the COVID-19 pandemic (approximately 50% of the sample) also participated in a clinic-based research visit at BCH Clinical Research Center where skin allergy testing and pulmonary function testing were performed. Caregivers also completed questionnaires during study visits or over the phone.

Primary exposure: Gas cooking stoves and elevated indoor NO₂

Exposure to gas cooking stoves was assessed based on caregiver-reported responses to a question regarding the type of cooking fuel used in the home using a standardized home environmental assessment [26]. Possible answers included natural gas, electric, and others. The use of gas cooking stoves was defined as the use of natural gas as their cooking fuel, or propane (n = 1) when other types of cooking fuel were specified. Those who reported using electric or other types of fuel were classified as not using gas cooking stoves.

Environmental Multipollutant Monitoring Assembly (EMMA) devices were used to collect the indoor gas samples in each participant’s main living area for at least 7 days [27]. The EMMA device is a portable air pollution sensor platform equipped with Alphasense NO2B43F for monitoring NO₂ with high sensitivity and precision, showing an R² of 0.93 and an RMSE of approximately 3.20 ppb in paired-sensor field tests [27]. NO₂ concentrations were sampled and recorded every one minute over the period that overlapped with actigraphy monitoring. The data were scaled and calibrated with standard methods described previously [27]. The data were cleaned according to visual inspection of the NO₂ concentration trajectory graph by EMMA engineers, trimming or removing records with signs of clear malfunction, such as instances where gas levels exhibited an instantaneous rise and rapid dispersion or an extreme value was surrounded by normal values on both the preceding and subsequent minutes. The all-day (i.e. 24 h) 95th percentiles of the indoor NO₂ levels were calculated for each monitoring day, and averaged over the monitoring period. Elevated indoor NO₂ was defined as an average 95th percentile indoor NO₂ measurements ≥80th percentile value in our sample (≥ 69.48 ppb), which is modestly higher than the NO₂ annual average safety level proposed by the US Environmental Protective Agency under the National Ambient Air Quality Standards (53 ppb) [28].

Outcomes

The primary outcomes were short sleep duration and SDB. Sleep duration was assessed through seven-night wrist actigraphy (ActiGraph GT3X+ Activity Monitor; ActiGraph, Pensacola, FL). Devices were placed on the child’s nondominant wrist 30–60 min before bedtime and removed 30–60 min after waking up each day. Caregivers completed a sleep diary during the monitoring period to indicate the child’s bed and wake times [29]. Data were reviewed and annotated by a research technician utilizing activity counts, light level data, and bed and wake times recorded in the reported sleep diary [30]. The main sleep period was identified based on visible changes in activity counts that were congruent with the diary-reported bed and wake times and changes in light levels. The Sadeh algorithm, appropriate for children [31], was used to classify each 60-s epoch as sleep using ActiLife version 6.13 software. Sleep duration was defined as the average time spent asleep for each main sleep period, averaged across all measurement days. Short sleep duration was defined as an average daily sleep duration less than 8 h, which is about 1 SD less than the pooled actigraphy based sleep duration of children ages 6–8 and 9–11 years [32].

SDB was assessed by objective monitoring using the WatchPAT 200 U (Itamar Medical, Caesarea, Israel) (n = 137) or the Nonin WristOx 3150 oximeter (Plymouth, MN) (n = 85), conducted for one night at the beginning of the EMMA monitoring protocol. Both devices have been used to assess the severity of SDB in pediatric populations [33, 34]. The WatchPAT device is a type 3 home sleep apnea testing device that records finger arterial tonometry, pulse oximetry, snoring (microphone), and movement (activity). After automatic analysis of respiratory events, a trained technician edited events to only retain those associated with a minimum of 3% oxygen desaturation. If the finger probe did not fit, children used the WristOX instead. During the COVID-19 pandemic, the WristOX was also preferentially used in all participants as a means of simplifying the protocol. SDB was defined as ≥5 events/h for both devices, each with >3% desaturation, which is a level indicating moderate SDB in pediatric population [35]. The prevalence of SDB was similar between the two devices (WatchPAT: 29.2%, WristOX: 29.4%).

In total, 49 participants underwent a second (n = 44) or third (n = 5) concurrent assessment of sleep with EMMA monitoring due to technical issues with initial recordings from the EMMA device (n = 15) or sleep devices (n = 34) that impacted the concurrent collection of environmental and sleep data. In these cases, only the last set of valid concurrent values were analyzed. If no overlapping measurements were available, we averaged all available sleep and EMMA data.

Secondary outcomes

Secondary outcomes include caregiver reported habitual loud snoring, Obstructive Sleep Apnea (OSA)-related quality of life (OSA-18 score [36]), sleep disturbance (pediatric Patient-Reported Outcomes Measurement Information System [PROMIS] sleep disturbance scale [37]), sleep-related impairment (pediatric PROMIS sleep impairment scale [37]), and actigraphy-based sleep efficiency. Habitual loud snoring was defined as caregivers’ responses to a question from the OSA-18 indicating that their child had loud snoring most or all of the time during the past four weeks [38]. The total OSA-18 score is the summed score of all items from the caregiver-completed OSA-18 questionnaire, ranging from 18 to 126, with higher scores indicating more severe symptoms related to SDB. Symptom severity includes five domains: sleep disturbances, physical symptoms, emotional distress, daytime problems, and caregiver concerns quantified using a point scale ranging from 1 (none of the time) to 7 (all the time) [36, 39, 40]. OSA-18 score was dichotomized with 40, indicating mild sleep impairment [36]. Sleep maintenance efficiency was defined as the percentage of sleep time after estimated sleep onset until final wakening divided by total time from sleep onset through final awakening, with low sleep maintenance efficiency defined as sleep maintenance efficiency <85% [41]. PROMIS sleep disturbance and sleep impairment T scores were converted to percentile ranks using T-distribution tables from the 8-item sleep disturbance parent proxy short form and the eight-item sleep-related impairment parent proxy short form, respectively [37]. T scores ≥55 (i.e. ≥0.5 SD above the population mean) were defined as elevated PROMIS sleep disturbance or impairment T scores [37, 42].

Covariates

Socio-demographic characteristics reported by the caregiver included the child’s age (years), sex (male or female), race and ethnicity (Hispanic, Non-Hispanic Black, Non-Hispanic White, and Non-Hispanic Other), and the caregiver’s annual household income (<$25 000, $25 000 and <$75 000, $75 000). Physical activity (hours/day) was assessed as the total reported daily average time spent walking, jogging or running, bicycling, jumping, swimming, jumping rope, gym or physical education class, and other physical activities in the last three months. Health-related factors included body mass index (BMI) percentile and asthma history. Obesity was defined by a BMI ≥95th percentile for their age, sex, and height. Asthma was defined by a score of ≥5 on the International Study of Asthma and Allergies in Childhood eight-item global wheezing questionnaire [43]. Neighborhood socio-economic status (SES) was characterized using the Child Opportunity Index (COI) 2.0, ranging from 0 to 100, with a higher COI suggesting higher neighborhood SES [44, 45]. Indoor PM_2.5 level was quantified as the mean all-day concentrations measured by the EMMA devices over the same monitoring period as indoor NO₂, with an accuracy of approximately 68% relative to integrated filter measurements and a median R² of 0.82 in field colocation tests. Average indoor daily temperature was measured and quantified in the same way as indoor PM_2.5, using the EMMA devices. Kitchen ventilation was defined as caregiver-report presence of a working exhaust fan venting to the outside, a fan in the wall, or windows used to exhaust air from the kitchen. Season when assessments were performed were categorized as spring, summer, autumn, and winter. The daily average outdoor NO₂ concentration (ppb) during each child’s monitoring period was obtained from the Environmental Protection Agency’s (EPA) Air Quality System (AQS) API by the epair package in R [46].

Statistical analysis

Analyses excluded participants if all sleep outcomes were missing, there was less than 3 days of valid actigraphy data, or more than 50% of days had invalid NO₂ data. In cases where more than one child per family were studied, we only included data from one child (excluding 21 siblings) given the correlated data related to exposure to the same indoor environment. The flow chart of the data cleaning process is shown in the Figure S1.

Descriptive statistics of sleep duration, SDB, and covariates by the gas cooking use and indoor NO₂ level were calculated. Multiple imputation was performed to generate 20 imputed datasets using the predictive mean matching method for numerical variables [47], logistic regression model for binary variables [48], and polytomous regression for categorical variables with more than two levels [48] was conducted on variables with missing values by the mice package in R [49]. Missingness for variables included in the statistical analysis ranged from n = 1 (0.4%) to n = 26 (11%).

Logistic regression analysis was used to analyze the associations of gas cooking use and dichotomized indoor NO₂ level with sleep duration and SDB. Model 0 presented a univariate association between gas cooking stoves and sleep outcomes. Model 1, which is our primary model, adjusted for individual, household, and neighborhood socio-demographic variables (age, sex, race and ethnicity, household income, and COI), and the seasonality of indoor NO₂ measurement. We then developed models to explore the influence of different sets of covariates while mitigating the risk of overfitting. Covariate selection was based on a literature review and the development of directed acyclic graphs (DAG). The DAG illustrating the association between gas cooking and sleep health is shown in Figure S2. Model 2 additionally adjusted for asthma and BMI percentile given that (1) obesity and asthma are risk factors of SDB and might influence other sleep measures (although bidirectional associations are plausible), and (2) pre-existing health conditions might influence the choice of cooking fuels [25, 50, 51]. Model 3 included covariates in Model 1 plus kitchen ventilation given its potential effect on the choice of cooking fuels and sleep health [52]. Model 4 further adjusted for daily average indoor PM_2.5 level based on Model 1 as it can be influenced by gas cooking and previous research has found its association with SDB [53, 54].

For the association between indoor NO₂ and sleep outcomes, the primary model (Model 1) adjusted for indoor temperature in addition to age, sex, race and ethnicity, household income, COI, and seasonality, as it is a key meteorological parameter that can impact the NO₂ concentration (Figure S3) [55]. The same Models 0 and 2 were performed as for gas cooking as the exposure. Model 3 additionally adjusted for physical activity based on Model 1, since it can influence people’s sleep health and the amount of time individuals spend indoors, which may further affect their exposure to indoor NO₂ levels [56, 57]. Models 4 and 5 included model 1 covariates plus gas cooking and daily average outdoor NO₂, respectively, as they represent different sources of indoor NO₂ and may independently influence sleep duration or SDB [3, 18, 52, 54]. Similarly, Model 6 additionally added daily average indoor PM_2.5 based on Model 1 [53].

Other secondary sleep outcomes were similarly analyzed to our primary sleep outcomes.

Results

Sample characteristics are shown in Table 1. A total 242 subjects (43% females, 40% Hispanic, 26% Black, 21% White, and 9% Other) with a mean (SD) age of 9.5 (±1.9) years were analyzed. There were 158 children (65%) from households with annual incomes of less than $75 000, and the mean (SD) COI of 35 (±30.0) was consistent with living on average in neighborhoods with limited educational, economic, and health resources. Overall, 32% of children were classified as obese, and 19% had a history of asthma. The mean (SD) of the average sleep duration was 7.6 (0.9) hours; 152 (63%) were classified with short sleep duration, and 27% were found to have SDB by objective testing.

Table 1.

Sample characteristics, overall and by the use of gas cooking and elevated average daily 95th percentile indoor NO₂; the Environmental Assessment of Sleep in Youth study (n = 242)

	Overall n = 242	Gas cooking stoves *		>80th percentile of the average daily 95th percentile indoor NO 2
		No n = 60	Yes n = 178	<80th percentile n = 187	80th percentile n = 55
Socio-demographics
Age (years)	9.5 (1.9)	9.4 (1.9)	9.6 (1.9)	9.5 (1.9)	9.6 (1.9)
Female (%)	103 (43)	23 (38)	80 (45)	77 (41)	26 (47)
Race and ethnicity (%)
Hispanic	97 (40)	24 (40)	70 (39)	70 (37)	27 (49)
Non-Hispanic Black	65 (26)	18 (30)	47 (26)	50 (27)	15 (27)
Non-Hispanic White	51 (21)	8 (13)	42 (24)	43 (23)	8 (15)
Non-Hispanic Other	22 (9)	10 (17)	12 (7)	19 (10)	3 (5)
Household income (%)†
<$25 000	73 (30)	22 (37)	49 (28)	55 (29)	18 (33)
≥$25 000 to <$75 000	85 (35)	28 (47)	56 (31)	67 (36)	18 (33)
≥$75 000	62 (26)	9 (15)	52 (29)	50 (27)	12 (22)
Neighborhood environment
Child opportunity index†,‡	35 (30)	27.4 (28.6)	37.6 (30.1)	36.8 (31.0)	28.4 (25.3)
Use of gas cooking stoves‡	178 (74)	-	-	127 (68)	51 (93)
Kitchen ventilation	179 (74)	42 (76)	136 (76)	140 (75)	39 (64)
Health-related factors
BMI (kg/m2) percentile	70.8 (29.8)	74.4 (29.9)	69.5 (30.1)	69.9 (30.5)	73.8 (27.4)
Obesity (%)	78 (32)	22 (37)	56 (31)	61 (33)	17 (31)
Asthma (%)	46 (19)	15 (25)	29 (16)	39 (21)	7 (13)
Physical activity (hours per day)	2.9 (3.4)	3.0 (2.8)	2.9 (3.6)	3.1 (3.4)	2.4 (3.5)
Sleep disordered breathing
AHI/ODI 3%, median (IQR)	3.4 (3.4)	3.6 (3.2)	3.4 (3.2)	3.3 (3.1)	3.6 (3.8)
≥5 events/h (%)	65 (27)	16 (27)	46 (26)	46 (25)	19 (35)
Sleep duration
Average time spent asleep per main sleep (h)	7.6 (0.9)	7.6 (1.0)	7.6 (0.9)	7.6 (0.9)	7.5 (0.9)
<8 h (%)‡	152 (63)	35 (58)	114 (64)	110 (59)	42 (76)
Seasonality of data collection ‡
Spring (March, April, May)	47 (19)	15 (25)	32 (18)	42 (22)	5 (9)
Summer (June, July, August)	63 (26)	15 (25)	47 (26)	50 (27)	13 (24)
Autumn (September, October, November)	80 (33)	16 (27)	63 (35)	61 (33)	19 (35)
Winter (December, January, February)	51 (21)	13 (22)	36 (20)	33 (18)	18 (33)
Air quality and meteorological factor
NO2 level (ppb)
Average daily 95th percentile indoor NO2, median (IQR)	41.1 (38.4)	24.9 (11.7)	46.6 (35.7)	33.8 (21.9)	90.8 (35.1)
Average daily outdoor NO2	10.3 (4.5)	10.7 (4.9)	10.1 (4.3)	10.2 (4.2)	10.7 (5.4)
PM 2.5  level (μg/m3)
Average daily indoor PM2.5, median (IQR)	6.9 (7.8)	7.8 (7.8)	6.8 (7.9)	6.1 (6.3)	10.6 (14.4)
Temperature (°F)
Average daily indoor temperature	76.7 (4.6)	77.4 (4.1)	76.4 (4.7)	76.7 (4.4)	76.8 (5.1)

SD, standard deviation; BMI, Body Mass Index; ODI, Oxygen Desaturation Index; AHI, Apnea-Hypopnea Index; IQR, Interquartile Range; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

^*Missing data: gas cooking stoves, n = 4 (2%); age, n = 2 (1%); race/ethnicity, n = 7 (3%); household income, n = 22 (9%); Childhood Opportunity Index, n = 14 (6%); kitchen ventilation, n = 2 (1%); BMI percentile, n = 2 (1%); asthma, n = 10 (4%); AHI/ ODI 3%, n = 20 (8%); seasonality of data collection, n = 1 (0.4%); average daily outdoor NO₂, n = 11 (5%); average daily indoor PM_2.5, n = 26 (11%); average daily indoor temperature, n = 8 (3%).

^†Indicates a statistically significant difference (p<.05) in comparisons between participants with and without the presence of gas cooking stoves in their kitchen.

^‡Indicates a statistically significant difference (p<.05) in comparisons between participants exposed to indoor NO₂ levels >80^th percentile of the average daily 95^th percentile and those below the level.

A total of 178 children (74%) lived in households with gas cooking stoves. Those children were more likely to be from homes with higher household incomes, live in more advantaged neighborhoods, and more likely to be exposed to higher levels of indoor NO₂ levels compared to children from households without gas cooking stoves.

The median (IQR) of the average daily 95th percentile indoor NO₂ was 41.1 (38.4) ppb and the mean (SD) of the average daily outdoor NO₂ was 10.3 (4.5) ppb. Children exposed to elevated indoor NO₂ were more likely to reside in less advantaged neighborhoods and have gas cooking stoves in their kitchen.

Associations of gas cooking stove and short sleep duration/SDB

Table 2 shows the odds ratios (OR) and 95% confidence intervals (CI) for the association between use of gas cooking and short sleep duration. In both unadjusted and the adjusted models, an approximate non-significant 1.6-fold increased odds of short sleep duration (Model 1 OR = 1.57, 95% CI: 0.79, 3.10, p = .195) was observed for individuals living in homes with gas cooking stoves compared to other children. No association was found between exposure to gas cooking stoves and SDB (Table 3). Similarly, we observed no association between gas cooking stoves and other secondary sleep outcomes (Tables S1–S5).

Table 2.

Odds ratios (95% Confidence Intervals) for associations of gas cooking stove exposure and short sleep duration^† (n = 242)

	Gas cooking	P-value
Models	OR (95% CI)
Unadjusted	1.27 (0.70, 2.31)	0.433
Model 1 *	1.57 (0.79, 3.10)	0.195
Model 2 (Model 1 + Asthma and BMI percentile)	1.67 (0.83, 3.33)	0.148
Model 3 (Model 1 + Kitchen ventilation)	1.58 (0.80, 3.14)	0.186
Model 4 (Model 1 + Daily average indoor PM2.5)	1.62 (0.81, 3.22)	0.169

OR, Odds Ratio; CI, Confidence Interval; BMI, Body Mass Index; PM_2.5, fine particulate matter.

^*Model 1 adjusted for age, sex, race/ethnicity, household income, Child Opportunity Index, and seasonality.

^†Short sleep duration was defined as having average sleep duration <8 h measured by actigraphy.

Table 3.

Odds ratios (95% Confidence Intervals) for associations of gas cooking stove exposure with SDB^† (n = 242)

	Gas cooking	P-value
Models	OR (95% CI)
Unadjusted	0.74 (0.38, 1.42)	0.361
Model 1 *	0.77 (0.38, 1.55)	0.455
Model 2 (Model 1 + Asthma and BMI percentile)	0.78 (0.39, 1.59)	0.496
Model 3 (Model 1 + Kitchen ventilation)	0.77 (0.38, 1.55)	0.459
Model 4 (Model 1 + Daily average indoor PM2.5)	0.76 (0.38, 1.54)	0.445

SDB, Sleep-Disordered Breathing; AHI, Apnea-Hypopnea Index; ODI, Oxygen Desaturation Index; OR, Odds Ratio; CI, Confidence Interval; BMI, Body Mass Index; PM_2.5, fine particulate matter.

^*Model 1 adjusted for age, sex, race/ethnicity, household income, Child Opportunity Index, and seasonality.

^†SDB was defined as AHI/ODI (≥3% desaturation) ≥5 events/h.

Elevated indoor NO₂ and short sleep duration/SDB

Table 4 illustrates the OR and 95% CI for the association between exposure to higher indoor NO₂ level and short sleep duration. In the primary model adjusted for age, sex, race/ethnicity, household income, COI, seasonality, and indoor temperature, a 2.88 (95% CI: 1.27, 6.55; p = .012) increase in the odds of short sleep duration was found in children exposed to elevated indoor NO₂ exposures. The effect sizes increased slightly after additional adjustment for asthma and BMI percentile (OR = 2.96, 95% CI: 1.30, 6.78; p = .010) and physical activity level (OR = 3.00, 95% CI: 1.31, 6.85; p = .009). The association remained but was attenuated by 6.6% (OR = 2.69, 95% CI: 1.17, 6.22; p = .021) when further adjusted for gas cooking based on the primary model. The associations also persisted after adjusting for daily average outdoor NO₂ or indoor PM_2.5 (Models 5 and 6, respectively).

Table 4.

Odds ratios (95% Confidence Intervals) for associations of elevated indoor NO₂ exposure with short sleep duration^† (n = 242)

	NO 2  (≥80th percentile)	P-value
Models	OR (95% CI)
Unadjusted	2.39 (1.15, 4.99)	0.020
Model 1 *	2.88 (1.27, 6.55)	0.012
Model 2 (Model 1 + Asthma and BMI percentile)	2.96 (1.30, 6.78)	0.010
Model 3 (Model 1 + Physical activity)	3.00 (1.31, 6.85)	0.009
Model 4 (Model 1 + Use of gas cooking)	2.69 (1.17, 6.22)	0.021
Model 5 (Model 1 + Daily average outdoor NO2)	2.88 (1.27, 6.55)	0.012
Model 6 (Model 1 + Daily average indoor PM2.5)	2.84 (1.30, 6.22)	0.009

OR, Odds Ratio; CI, Confidence Interval; BMI, Body Mass Index; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

^*Model 1 adjusted for age, sex, race/ ethnicity, household income, Child Opportunity Index, seasonality, and daily average indoor temperature.

^†Elevated indoor NO₂ exposure was defined as an average 95^th percentile indoor NO₂ measurements ≥80^th percentile value in our sample, and short sleep duration was defined as having average sleep duration <8 h measured by actigraphy.

Table 5 suggests a positive but statistically insignificant association between high indoor NO₂ exposure and SDB. No associations were found between elevated indoor NO₂ level and other sleep outcome variables (Tables S6–S10).

Table 5.

Odds ratios (95% Confidence Intervals) for associations of elevated indoor NO₂ exposure with SDB^† (n = 242)

	NO 2  (≥80th  percentile)	P-value
Models	OR (95% CI)
Unadjusted	1.21 (0.62, 2.35)	0.580
Model 1 *	1.23 (0.61, 2.47)	0.567
Model 2 (Model 1 + Asthma and BMI percentile)	1.33 (0.66, 2.68)	0.430
Model 3 (Model 1 + Physical activity)	1.17 (0.57, 2.39)	0.661
Model 4 (Model 1 + Use of gas cooking)	1.44 (0.69, 2.99)	0.324
Model 5 (Model 1 + Daily average outdoor NO2)	1.21 (0.60, 2.46)	0.586
Model 6 (Model 1 + Daily average indoor PM2.5)	1.15 (0.57, 2.33)	0.687

SDB, Sleep-Disordered Breathing; AHI, Apnea-Hypopnea Index; ODI, Oxygen Desaturation Index; OR, Odds Ratio; CI, Confidence Interval; BMI, Body Mass Index; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter.

^*Model 1 adjusted for age, sex, race/ethnicity, household income, Child Opportunity Index, seasonality, and daily average indoor temperature.

^†Elevated indoor NO₂ exposure was defined as an average 95th percentile indoor NO₂ measurements ≥80th percentile value in our sample, and SDB was defined as AHI/ODI (≥3% desaturation) ≥ 5 events/h.

Discussion

To our knowledge, this is one of the very first studies to assess the associations of both gas cooking stoves and indoor NO₂ exposure with multiple objectively measured and caregiver reported sleep outcomes in a pediatric cohort. Our study revealed that children exposed to elevated indoor NO₂ levels, exceeding EPA’s annual average safety guideline (53 ppb), had a nearly three-fold increased odds of short sleep duration (by actigraphy) when compared to children exposed to lower indoor NO₂ levels; associations persisted across models that adjusted for multiple potential confounders. Our results did not show associations between elevated indoor NO₂ levels and other sleep outcomes, including SDB-related metrics (snoring, OSA-18 score, and apnea-hypopnea index/oxygen desaturation index >5), caregiver-reported sleep-related daytime impairment and sleep disturbance (pediatric PROMIS instruments), and actigraphy-based poor sleep continuity. Additionally, gas cooking stoves, considered a primary source of indoor NO₂, was not statistically associated with adverse sleep outcomes, although similar directional associations were observed for short sleep duration.

Our findings add to the existing literature that highlights the influence of the environment and poor air quality on adverse sleep outcomes, and specifically identified an association between indoor NO₂ with short sleep duration, but not with other sleep outcomes [53]. Our results are consistent with a prior study that reported an association between 5-year levels of outdoor NO₂ exposure with an approximately 30 min reduction in self-reported sleep duration in college students in China [14]. However, the latter population was exposed to high levels of multi-pollutants and the individual effects of NO₂ were unclear.

Our study did not observe the hypothesized association between indoor NO₂ and SDB, either measured objectively or by reported snoring. A null association was also reported in a study of outdoor NO₂ exposure and parental-reported snoring or wheezing-related sleep disturbances [19, 20]. In contrast, a study of Australian children ages 4 to 6 years found a 4.5 times increased odds of snoring among children exposed to higher (>60 μg/m³) compared to lower indoor NO₂ (<30 μg/m³) concentrations [21]. Other studies, including one from Taiwan and another study conducted in the Multi-Ethnic Study of Atherosclerosis (MESA) cohort in the USA, found increased annual average ambient NO₂ exposure was associated with polysomnography-based higher AHI and increased odds of sleep apnea among adults [17, 18]. These inconsistent findings may reflect population differences and related population variations in susceptibility; potential effects of multi-pollutants; and differences in outdoor vs indoor exposures and chronicity of exposures. Specifically, we only assessed NO₂ in the immediate one-week time period surrounding SDB assessment; it is possible that assessment of chronic exposure may be more relevant for identifying longer-term effects of NO₂ on naso-pharyngeal inflammation and other pathogenic pathways. SDB was measured using a single overnight assessment with focus on respiratory events that resulted in oxygen desaturation, potentially underestimating obstructive events that resulted in arousal without desaturation. Also, unlike the MESA polysomnography-based study, we did not observe an effect of NO₂ on sleep efficiency (or fragmentation) as measured by actigraphy. While actigraphy potentially provides more representative sleep–wake data over multiple nights, it also is limited by misclassification, and in adolescents may under-estimate sleep periods [58].

Some previous research observed associations between outdoor NO₂ exposure and other sleep quality-related outcomes, though our study did not find those associations with indoor NO₂ level [13, 15, 16]. Indoor NO₂ is primarily from gas cooking stoves and poorly vented cooking practices, tobacco smoking, combustion heating, and air exchange from outdoor NO₂. In contrast, the outdoor or ambient NO₂ sources are often attributable to industrial combustion and vehicles [28]. Nonetheless, indoor NO₂ exposure may be particularly salient to the sleep health of children given that children spend longer times sleeping (indoor) than youth and adults. For example, it is estimated that children under 12 spend 75.6% of their time indoors, while youth and adults ≥12 spend 66.8% of their time indoors [22]. It is possible that the null associations with sleep quality outcomes in our study were due to misclassification of chronic exposures related to assessment of air quality over a single week, or due to under-estimation of children’s sleep disturbances by caregivers.

Few studies have investigated the association between gas cooking and sleep health [59]. One study conducted among older adults in India found those who used solid fuel for cooking had 1.25 times higher odds of experiencing sleep problems compared to those who used clean fuel [59]. Instead, most existing research on gas cooking focused on lower airway respiratory diseases such as asthma in children [6]. A meta-analysis indicated a positive association between asthma with gas cooking but not with indoor NO₂; in contrast indoor NO₂ was reported to be associated with wheeze but not asthma [6]. Differences in the reported health effects of gas stoves compared to directly measured indoor NO₂ levels may reflect variations in home ventilation and gas cooking stove characteristics and usage that influence NO₂ levels associated with gas cooking. In addition to NO₂, pollution emitted from gas cooking stoves also includes PM_2.5, ultrafine particles, benzene, carbon monoxide (CO), and formaldehyde (HCHO) [60, 61]. Our previous study found that higher indoor PM_2.5 exposure was associated with appropriately three times higher odds of SDB in the same study cohort [53], although no association was observed between indoor PM_2.5 and short sleep duration (Table S11). Another simulation study also found indoor PM_2.5 concentration was associated with more severe OSA [62]. Moreover, prior research has suggested that being exposed to elevated indoor HCHO concentration was associated with higher risk of OSA among males [63]. These findings imply that pollutants other than NO₂, which are emitted from gas cooking, may have unique effects on sleep health that are not captured by direct measurement of NO₂. Therefore, there is value to studying gas cooking exposure as well as directly measuring NO_2.

Though children living in households with gas cooking stoves were exposed to higher indoor NO₂ levels than children in homes without gas stoves (with and without gas cooking stoves: 46.6 ppb vs 24.9 ppb), we found weaker associations for gas cooking stoves compared to directly measured indoor NO₂. In our sample, gas cooking stoves were more common in children from higher socio-economic households while indoor NO₂ levels were higher in association with low socioeconomic markers. The higher indoor levels in the latter households may be explained by poor stove ventilation and higher outdoor NO₂ levels, stressing the variability in NO₂ levels not captured by the presence or absence of a gas cooking stove and the value of direct NO₂ measurements. Also, among the participants using gas cooking in our sample, the majority reported using natural gas for cooking, which may explain our varying results. Previous reports have suggested that respiratory symptoms (e.g. wheeze, dyspnea) were less prevalent among people using natural gas compared to those using other types of gas cooking stoves, including liquefied petroleum gases and bottled gases, which might produce more pollutants from combustion due to insufficient burning [28, 64]. In addition, other pollutants emitted from gas cooking stoves might interact with NO₂ and alter its effect on sleep health [61].

The mechanisms underlying the association between NO₂ and short sleep duration are unclear. However, NO₂ is a recognized irritant that can affect the mucosa of the eyes and upper and lower airways, including nasal passages and the respiratory tract [65]. Nasal and airway inflammation, often are worse at night, may disrupt and curtail sleep. Short term exposure to NO₂ also can lead to increased bronchial reactivity, which leads to sleep disruption [66–68]. Neuroinflammatory effects also may directly influence central sleep–wake mechanisms [69].

It is estimated that over 75% of the population globally lives in areas with NO₂ concentrations exceeding the recently updated air quality guidelines (i.e. annual average exposure of 10 μg/m³) proposed by the World Health Organization [70], Moreover, 35% of households in the USA have gas cooking stoves, and in some states this proportion exceeds 60% [71]. Indoor NO₂ levels of households with gas appliances typically exceed the outdoor NO₂ levels, which could lead to acute NO₂ spikes [28]. For example, a simulation study found that using gas appliances could increase the median highest simulated one-hour indoor NO₂ concentration by 100 ppb [72]. Our study findings highlight that NO₂ levels vary with ventilation, suggesting the need for ensuring optimal stove venting.

Study strengths include the study of a diverse pediatric population living in neighborhoods with primarily low SES and measurement of multiple relevant sleep outcomes. Unlike previous studies, we directly measured indoor NO₂ levels continuously for one week, in addition to considering gas cooking use, and also considered effects of outdoor NO₂ level and indoor PM_2.5. The measured indoor values offer more precise estimates for participants’ NO₂ level exposure than the presence of a gas cooking stove, which does not capture information on intensity and frequency of use, stove ventilation, and household ventilation. We also assessed different sets of covariates and multi-level risk factors, including household and neighborhood SES, health conditions, physical activity, various sources of indoor NO₂, and meteorological factors. Sleep assessments were made in the children’s typical home environment concurrently with indoor air quality monitoring.

We acknowledge several study limitations. Residual confounding from other sources of poor air quality including other irritants, pollutants and allergens, and housing conditions, is possible. We defined gas cooking exposure based on the type of cooking fuel, rather than incorporating other metrics, such as cooking methods and frequency, that may capture additional aspects of gas cooking potentially relevant to sleep health. We assessed the indoor NO₂ level for seven days. The critical time window and duration of exposure as relevant to different sleep measurements are unknown. The focus on one geographic region limits the generalizability of the study. We did not measure children’s individual-level NO₂ exposure, which may vary depending on their time spent outdoors and in other indoor settings such as schools. While we used standardized sleep assessments, single night in-home SDB assessment and multi-night actigraphy sleep–wake studies may have led to some degree of measurement misclassification. However, prior studies have reported that the AHI from wrist-worn devices was highly correlated with AHI measured by in-laboratory polysomnography among children aged 8–15 years [33].

Conclusion

An elevated level of indoor NO₂ was common in an urban, predominantly low-income sample, and was associated with an increased odds of short sleep duration in children, even after accounting for multiple potential confounders. While indoor environmental exposures have long been implicated in asthma and respiratory tract infections, this novel finding suggests that elevated level of indoor NO₂ also may adversely affect children sleep health. Interventions targeted at improving indoor air quality, including better stove venting, might provide innovative approaches for improving the sleep health in children and reducing pediatric sleep disparities. Moreover, given that the majority of our time is spent indoors, and sleep occurs indoors, future research is needed to specifically examine indoor air quality and to more clearly define the mechanistic pathways underlying association for sleep disturbances.

Funding

This work was supported by the following National Institute Health grants, R01HL137192 (W.P.), L40HL165622 (S.G.N.), R35HL135818 (S.R.), P30ES000002 (D.R.G. and G.A.), and K24AI106822 (W.P.).

Disclosure statement

Financial disclosure: None declared.

Non-financial disclosure: None declared.

Ethics approval

The study was approved by the Boston Children’s Hospital Institutional Review Board (No. IRB-P00025433). Written informed consent was obtained from participants’ guardians, and assent was obtained from participants before enrollment.

Data availability

The data underlying this article cannot be shared publicly for the privacy of individuals that participated in the study.