Relation of repeated exposures to air emissions from swine industrial livestock operations to sleep duration and awakenings in nearby residential communities

Abstract

Objectives:

Since waste from swine industrial livestock operations (ILOs) produces air pollutants associated with negative health outcomes among nearby residents, we assessed the impact of odor emissions on sleep duration and awakenings.

Design:

A repeated-measures design

Setting:

16 residential communities in eastern North Carolina hosting swine ILOs

Participants:

80 participants residing in eastern North Carolina from 2003 to 2005

Intervention (if any):

Not applicable

Measurements:

Study participants completed twice-daily diaries in which they rated the strength of hog odors and indicated whether they were asleep or awake per hour for two weeks. Simultaneously, a monitoring trailer placed in a central location in each community measured the atmospheric concentration of hydrogen sulfide (H₂S). Subject-conditional fixed-effects regression models were used to estimate associations between two markers of swine ILO pollutant exposures (H₂S and swine odor) and two self-reported sleep outcomes (nightly sleep duration and awakening from sleep).

Results:

Among 80 participants, nightly (across a 12-hour period) swine odor was associated with lower-nightly sleep duration (mean difference = −14.3 minutes, 95% confidence interval − 25.0 to −3.3 minutes) compared to odor-free nights and detection of nightly hydrogen sulfide was associated with an increased risk of awakening (Hazard ratio = 1.23, 95% confidence interval 0.98 to 1.55) compared to nights with no detection of hydrogen sulfide.

Conclusions:

These results suggest that environmental odorants are important considerations for sleep health and highlight the importance of sleep as a potential mediator between environmental air pollution and health outcomes impacted by poor sleep.

INTRODUCTION

Swine Industrial Livestock Operations (ILOs) are a prevalent source of air pollutants in eastern North Carolina. Today, nearly all of the pork consumed and exported by the United States is produced by industrial livestock operations ¹ and North Carolina is a leading producer - with over 2,000 permitted operations and 9 million swine (National Agricultural Statistics Service, North Carolina Field Office 2016). The industry is highly concentrated in the southeastern part of the state ², where the two top-producing counties are also the two top-producing counties in the entire United States (National Agricultural Statistics Service 2014).

This context has produced an environment where residential communities hosting swine ILOs can face concentrated industrial air emissions not typically associated with rural areas. Rural areas consistently score poorly on population health indicators ², and rural communities are experiencing social transitions that create health challenges ³: a shift to corporate agriculture, job loss, outmigration of young and working people, and poorer access to nutritious food ⁴. Air pollution produced by swine ILOs contains complex mixtures of particulate matter, aerosols, and gasses that can vary by facility, time of day, weather, and season. A large proportion of these air pollutants are produced by lagoon-and-sprayfield systems, which are used for waste management at swine ILOs in North Carolina ⁵. In this system, wet swine wastes flow through the slatted floors of confinement buildings into open pits where they decompose anaerobically to produce mixtures of microbial metabolites including ammonia and hydrogen sulfide ⁶. These wastes are sprayed onto adjacent fields to encourage aerobic decomposition, but this process also produces waste aerosols that spread liquid pollutants into the air and groundwater.

In ethnographic research conducted in communities near swine ILOs, neighbors have reported that swine ILO air pollutants interfere with sleeping⁷ and time outdoors ⁸ but these associations have not been statistically quantified. ILO air pollutant exposures could be linked to sleep disruption could occur through several mechanisms: awakenings could be caused by olfactory or trigeminal nerve excitation by chemical components of the emissions mixture, exposures to the emissions mixture may produce disease symptoms like airway restriction that could increase the risk of sleep apnea or making falling asleep more difficult, or the psychological impacts of uncontrollable malodor could lead to stress and impair activities that enhance sleep. Exposure to ammonia odorants like those found in swine ILO pollutants have long been known to cause awakenings from sleep ^7,9; this property has been leveraged in the clinical context through the use of smelling salts ¹⁰. Many of the disease symptoms linked to swine ILO air emissions in past research are known to cause sleep impairment. Disrupted breathing can make falling asleep difficult ¹¹, cause awakenings from sleep ¹², interfere with outdoor activities ¹³, and produce psychological stress ¹⁴. Respiratory disease symptoms ¹⁵, the cultural and psychological meanings of malodor ¹⁴, and stress resulting from the inability to control odors could make falling asleep more difficult.

Poor sleep is associated with a host of poor health outcomes that have also been connected to air pollution, including obesity, hypertension, type 2 diabetes, cardiovascular disease, and premature mortality^16–19. This study seeks to expand the understanding of the transient exposure-response dynamics of ILO pollutant exposures by assessing their impact on sleep. Associations between two exposure markers (swine odors and atmospheric hydrogen sulfide concentration) and two outcome measures (nightly sleep duration and awakening from sleep) are estimated. We leverage the unique design of the CHEIHO (Community Health Effects of Industrial Hog Operations) study^14,20–23, which aimed to capture the acute impacts of exposures to hog operation emissions. The repeated measures of exposures and outcomes recorded as part of the CHEIHO study enable each participant to act as his or own control in assessments of potential sleep disruption, eliminating bias due to factors that remain constant over time (e.g. sex, race/ethnicity, co-exposures, pre-and existing conditions) and allowing adjustment for other measured time-varying factors.

PARTICIPANTS AND METHODS

Study population

We used data collected as part of the CHEIHO study, a community-engaged project combining environmental health education with mixed-methods qualitative and quantitative research in 16 communities in North Carolina. Potential CHEIHO communities were identified in collaboration with community organizations and had at least four residents interested in study participation.

One-hundred and one CHEIHO participants were recruited from 16 North Carolina residential communities hosting industrial swine operations from 2003 to 2005 ²³. To be eligible for CHEIHO, participants had to live within 1.5 miles of an active ILO containing swine, be at least 18 years old, and have access to a freezer to store saliva samples collected as part of the study. Potential participants were excluded if they smoked or could not read or write to complete the study materials.

As part of the original CHEIHO study design, participants kept a diary in which they rated odors they sensed for each hour of the day for two weeks. Each participant completed two diary data-collection sessions per day at times identified in conjunction with research staff. Each participants’ daily diaries were made 12 hours apart between 7:00 and 9:00 am and 7:00 to 9:00 pm (e.g. 8:00 am and 8:00 pm), providing 24-hour coverage if all diary entries were completed and also allowing data collection to occur at convenient times for each participant. Participants also used these diaries to document their sleeping hours by writing a Z in place of an odor rating for hours they were asleep after awakening. Based on the original n=101 CHEIHO participants’ odor diaries, twenty-one participants were excluded due to insufficient exposure or outcome data (11 participants) or atypical sleep patterns (10 participants reporting >12 hours/day of sleep, <4 hours/day of sleep, or >3 study nights with no reported sleep), including apparent night-shift work. Of the remaining 26,880 (80*24*14) person-hours recorded in diaries, 24,552 (91.3%) were used due to missing outcome data, covering 1,023 day/night person-periods. An initial training session in each community was used to obtain informed consent and train study participants in data collection procedures. This research was approved by the Institutional Review Board at the University of North Carolina at Chapel Hill.

Exposure assessment

Swine ILO pollutant exposures were characterized using participants’ hourly swine odor ratings, participants’ twice-daily outdoor swine odor ratings, and hydrogen sulfide concentrations recorded by monitors. Over a two-week study period, each participant completed a twice-daily diary form containing questions on the strength of swine odors. At the beginning of each diary session, participants rated the odor from swine operations during the preceding twelve hours using a 9-point scale (0–8); for each rated hour, participants also indicated if the rated odor was indoors, outdoors, or away from home. Participants did not report hourly odor ratings for periods when they were asleep. Each evening, participants spent ten minutes outside and rated the current strength of outdoor odor from swine operations on the same scale. The primary exposure variables of interest thus consisted of (1) hourly H₂S concentrations from a nearby monitor, (2) outdoor odor recorded by participants each evening over a 10-minute observation window, and (3) average odor ratings during the night (9pm-9am).

A baseline assessment of odor sensitivity was conducted for each participant in 2003 using a butanol dilution series ^7,23. Briefly, participants were presented with two vials (one contained a dilution of Butanol in distilled water and the other contained only distilled water) and asked to identify the vial with an odor. Starting at a baseline concentration of 10 ppm, the concentration of butanol was doubled with each successful trial until participants reached a point where they had identified five vials correctly (the lowest of these five trial was taken as the odor detection threshold for that participant).

Simultaneously, a mobile air monitoring trailer was used to make meteorological measurements (temperature, humidity, wind direction, and wind speed) and record atmospheric concentrations of hydrogen sulfide (H₂S). Atmospheric hydrogen sulfide (H₂S) is produced by the anaerobic decomposition of swine waste and has been used as a specific marker of ILO emissions plumes^21,24. The monitor was placed in a central location in each community, on average 0.2 miles from participants’ homes. H₂S was measured as a chemical marker specific to the complex mixtures of air pollutants produced by liquid swine waste management systems in rural areas. Mean 15-minute H₂S accumulations were originally measured by an MDA Scientific Single Point Monitor (Zellweger Analytics, Inc.) using a chemcasette with a detection limit of 1 part-per-billion volume (ppb) and converted to hourly averages to align with participants’ odor records. A HOBO microstation datalogger (Onset Computer Corporation) with temperature and humidity sensors was used to measure meteorological conditions.

Outcome ascertainment

During twice-daily data collection sessions, participants self-reported if they were asleep during each of the preceding 12 hours on a diary form. Each participants’ daily diary completions were scheduled 12 hours apart between the hours of 7:00 to 9:00 am and 7:00 to 9:00 pm (e.g. 8am and 8pm), which provided 24-hour coverage of sleep status if all diary entries were completed, but allowing data collection to occur at convenient times for each participant. Hourly sleep duration was classified as a binary variable on an hourly scale in each participants’ diary. Based on data availability, these hourly values were summed by night to produce a variable representing the number of hours of sleep each evening. Awakening from sleep was defined as a period when a participant reported at least one hour of wake time, following at least one hour of sleep time. In analysis, naps were defined as one or more hour of sleep from 9:00am to 9:00pm separated from nightly sleep by at least one hour. Due to the structure of the diary, sleep and wake periods were thus rounded to the nearest hour and sleep duration measures could only take discrete values.

Potential confounders

The combination of high relative humidity and hot temperature (humid heat) was treated as a potential confounder and used as a model covariate because it has the potential to disrupt sleep ²⁵ and odorant production ⁷. Based on experimental data, humid-heat was defined as a dry-bulb temperature above 80 °F and relative humidity above 60% ²⁶. Other potential confounders, such as season, time-of-day, sociodemographic variables and health status were addressed using the study design and were not used as covariates.

Statistical analysis

Sleep duration each night was modeled as a Quasipoisson distributed outcome following the form

$$\ln n_{ij}={\delta }_i+X_{ij}\beta$$ (1)

where $n_{ij}$ is the number of hours of sleep, i is the participant index, j is the night index, X is a vector of exposures and covariates, and ${\delta }_i$ represents a subject-conditional fixed effect ²⁷ for each participant that is conditioned out of the model likelihood by using a conditional likelihood function. In this fixed-effect model form, potential confounding factors are limited to those that vary with time and are associated with odor and nightly sleep duration. Therefore, in our fitted model the X vector consists of the potential confounder, humid-heat, and the exposure of primary interest. We estimated associations with hourly hydrogen sulfide level, nightly average swine odor, or evening outdoor odor. To facilitate interpretation of model parameters, average sleep duration for all study participants was substituted into the model using the formula $\widehat{n}\ast (\mathit{exp}(\beta )-1)$ to yield an estimate of the average exposure effect in the total population at the reference level of the confounder.

To assess the potential acute impacts of air pollutants on awakenings sleep, we also fitted a model for estimating the association between awakening from sleep (or sleep stability) and H₂S concentration. Sleep stability was modeled as a binary outcome variable using a discrete-time hazard model following the logistic form

$$logit\left(P\left(y_{i,j,t}\right)\right)={\alpha }_{i,}+{\lambda }_t+X_{i,j}\beta$$ (2)

where $y_{ij}$ is awakening from sleep (taking a value of 1 if a person is awakened at hour j, conditional on having been asleep at hour j-1, and 0 otherwise), i is the participant index, j is the hour index, α_i are an participant-specific fixed-effects conditioned out of the likelihood, ${\lambda }_t$ are used to model the baseline hazard as a function of time-asleep, and X is the vector of humid-heat and hydrogen sulfide. Following a time-to-event data structure, only the first awakening of the night was counted as an outcome in the model. We report estimates of association between current hydrogen sulfide exposure and sleep stability, as well as associations with hydrogen sulfide exposure in the preceding hour (i.e., one-hour lagged exposure) due to the coarse sleep intervals available in the data structure. Because participants did not report odor ratings during periods they reported sleep, we did not model the associations between nightly odor and awakening from sleep.

Given the right-skewed distribution of hydrogen sulfide, with 92% of hourly measured below the limit of detection, in current analyses hydrogen sulfide level was modeled as a binary variable coded as 1 if above the limit of detection, and 0 otherwise. Confidence intervals are provided to estimate model precision, but do not represent test statistics as the CHEIHO data do not come from a random sample.

RESULTS

Among the 80 eligible adult participants, 65% were female, 85% were black, and 40% had an odor sensitivity threshold below 40 parts per million. Demographic characteristics of the 80 study participants are shown in Table 1. Among the cohort, 35% percent of participants had an odor sensitivity ≤40 ppm.

Table 1.

Distributions of demographic characteristics of study participants by grouped by eligibility and number of records

		Participants	Person-time
			Person-hours	Person-days
Variable		(n=80) n (%)	(n=24552) n (%)	(n=1023) n (%)
Age
	≥65	20 (25.)	6768 (27.6)	282 (27.6)
	24–64	60 (75.)	17784 (72.4)	741 (72.4)
Gender
	Women	54 (67.5)	16656 (67.8)	694 (67.8)
	Men	26 (32.5)	7896 (32.2)	329 (32.2)
Race
	Black	66 (82.5)	19296 (78.6)	804 (78.6)
	non-Black	14 (17.5)	5256 (21.4)	219 (21.4)
Odor Sensitivity
	≤40 ppm	28 (35.)	8256 (33.6)	344 (33.6)
	>40 ppm	48 (60.)	15360 (62.6)	640 (62.6)
	Missing	4 (5.)	936 (3.8)	39 (3.8)
Heat-Humidity1
	No		18606 (75.8)	867 (84.8)
	Yes		1978 (8.1)	66 (6.5)
	Missing		3969 (16.2)	90 (8.8)

^1.≥80°F and ≥60% Relative Humidity

The distributions of sleep and odorant exposures are described in Table 2. Hydrogen sulfide was above the detection threshold for 8.2% of all study hours and participants reported odors 14.5% of their time awake. Evening outdoor odorants were higher on average than morning outdoor odorants (1.50 vs. 1.34). Participants experienced at least one odor episode on 46.8% of days and 50.0% of nights. Atmospheric temperature ranged from 31 °F to 87 °F (mean 62 °F) and relative humidity ranged from 40.7% RH to 100% RH (mean 80.4% RH); 12.4% of nights were classified as hot-humid. Participants slept 7.3 hours per night on average (standard deviation 0.90 hours), and 8.3% of days contained at least one nap episode.

Table 2.

Distributions [n (%)] of odor ratings, hydrogen sulfide concentration (H₂S) and sleep, on hourly, daily, and nightly scales

Variable (Unit)	Hourly	Daily a	Nightly b
	n (%)	n (%)	n (%)
Odor (0–8)	Average (1h)	Outdoor (10m)	Outdoor (10m)
0	14566 (59.3)	474 (46.3)	465 (45.5)
1–2	1254 (5.1)	273 (26.7)	278 (27.2)
3–4	681 (2.8)	142 (13.9)	123 (12.)
5–8	527 (2.1)	71 (6.9)	107 (10.5)
Missing c	7524 (30.6)	63 (6.2)	50 (4.9)
Mean (sd)	0.39 (0.63)	1.34 (1.82)	1.50 (2.00)
H 2 S(ppb)	Average (1h)	Average (12h)	Average (12h)
0	20804 (84.7)	826 (80.7)	663 (64.8)
0–2	1143 (4.7)	189 (18.5)	319 (31.2)
>2	726 (3.0)	8 (0.8)	41 (4.0)
Missing	1879 (7.7)	0 (0.0)	0 (0.0)
	0.25 (1.85)	0.10 (1.06)	0.42 (2.46)
Sleep duration (h)	Length (1h) d		Total (12h)
1–3	239 (1.0)		7 (.7)
4–6	1488 (6.1)		251 (24.5)
7–9	5554 (22.6)		728 (71.2)
10–12	365 (1.5)		37 (3.6)
Awake	16906 (68.9)		0 (0.0)
Mean(sd)			7.27 (1.32)

^a9am to 9pm

^b9pm to 9 am

^pParticipants did not record odors during sleep

^dProportion of all hours by sleep episode length

h=hours, m=minutes

Nightly H₂S concentration, nightly swine odor, and evening outdoor ratings were each associated with lower nightly sleep duration (Table 3). Using the formula above, models estimated that the presence of nightly swine odor decreased nightly sleep duration in the study population by an average of 14.3 minutes (3.3 to 25.0, 95% interval). Through a similar calculation, the presence of nightly hydrogen sulfide decreased nightly sleep duration by 5.0 minutes on average (−5.8 to 15.6, 95% interval). These estimated associations did not differ significantly by participants’ sensitivity to odors or after excluding periods during which participants reported cold/flu symptoms, although the magnitude of the association between evening outdoor odor and sleep duration was lower after excluding participants experiencing symptoms consistent with cold/flu. Though not statistically significant, the magnitude of the association between H₂S concentration and sleep was higher among participants with lower sensitivity to odors. Results among participants by exclusion of cold/flu symptoms and by odor sensitivity stratification are shown in Supplemental Table 1.

Table 3.

Estimated change in daily sleep duration by hydrogen sulfide, evening, and nightly odors in the Community Health Effects of Industrial Hog Operations (CHEIHO) study, North Carolina

All Participants	β a (95% CI)	n	RD b (95% CI)
H2S	−0.012 (−0.036, 0.013)	865	−5.0 (−15.6, −5.8)
Odor (swine ILO)
Evening (outdoor)	−0.005 (−0.029, 0.020)	883	−2.0 (−12.4, 8.7)
Nightly	−0.033 (−0.059, −0.008)	933	−14.3 (−25.0, −3.3)
Low Odor Sensitivity	β (95% CI)	n	RD (95% CI)
H2S	−0.023 (−0.053, 0.008)	541	−9.8 (−22.6, 3.4)
Odor (swine ILO)
Evening (outdoor)	−0.006 (−0.036, 0.024)	557	−2.6 (−15.2, 10.4)
Nightly	−0.043 (−0.074, −0.012)	596	−18.3 (−30.1, −5.2)
High Odor Sensitivity	β (95% CI)	n	RD (95% CI)
H2S	0.018 (−0.028, 0.064)	285	8.1 (−12.0, 29.1)
Odor (swine ILO)
Evening (outdoor)	−0.007 (−0.053, 0.040)	287	−2.3 (−22.7, 17.9)
Nightly	−0.024 (−0.075, 0.026)	298	−10.5 (−31.5, 11.6)

Model design adjusts for all time-invariant factors and heat-humidity.

^aQuasipoisson rate parameter.

^bRD = estimated absolute change in sleep duration per night in minutes.

Hourly H₂S concentration was associated with a greater risk of awakening from sleep (Table 3). The presence of hydrogen sulfide above increased the risk of awakening by 24% during that hour (HR=1.24, 95% interval: 0.99 to 1.55) and by 23% (HR=1.23; 95% interval: 0.98 to 1.55) during the following hour (i.e., a 1-hour lagged analysis, Table 4). H₂S concentration was associated with an increased risk of awakening among all participants (HR 1.23, 95% interval 0.98, 1.55). This effect was higher among participants with higher sensitivity to odor (threshold <40 ppm; HR=1.62, 95% interval 1.10 to 2.40) compared to participants with a lower sensitivity to odor (threshold >=40 ppm; HR=1.08, 95% interval 0.82 to 1.43).

Table 4.

Estimated association between awakening from sleep ^a and pollutant exposure indicators by odor sensitivity and exposure lag time

All Participants	HRb	(95 % CI)	nc
H2S (current measure)	1.23 (0.98, 1.55)		6643
H2S (lagged 1-hour)	1.24 (0.96, 1.53)		6645
Low Odor Sensitivity
H2S (current measure)	1.08 (0.82, 1.43)		4119
H2S (lagged 1-hour)	1.08 (0.81, 1.44)		4118
High Odor Sensitivity
H2S (current measure)	1.62 (1.10, 2.40)		2270
H2S (lagged 1-hour)	1.57 (1.04, 2.38)		2273

^aNight-time interruption in sleep (≥1 hr) preceded by one or more hours of sleep; model design adjusts for all time-invariant factors, heat-humidity, and time-asleep.

^bHazard ratio.

^cNumber of sleep periods during which awakening could have occurred.

DISCUSSION

In this study we estimated the effect of exposures to swine ILO pollutants on sleep using two measures of odor perception and ambient outdoor H₂S concentration. Night-time ILO pollutant exposures (hydrogen sulfide and odor from swine operations) were associated with adverse sleep. Episodes of nightly hydrogen sulfide exposure and odor, two markers of swine ILO pollutant exposures, decreased participants sleep by 14.3 minutes on average and increased the risk of awakening from sleep by 23%. In the context of chronic daily exposures, these impacts could lead to substantial sleep losses over time.

Observed associations between ILO pollutants and various sleep dimensions could be attributable to several causes. Hydrogen sulfide, amines, and other emissions components could have direct chemical effects - prompting awakening ⁹ or disrupting homeostatic (sleep-wake) or circadian regulation ²⁸. Studies of communities exposed to ILO pollutants have documented sensory effects consistent with olfactory and trigeminal nerve irritation ²⁹ including nausea ²¹, burning nose and eyes ^21,30, and headaches ³⁰ that could cause awakenings from sleep. Inability to control exposures during the day and night could also cause annoyance, stress, negative thoughts, and anger that could make sleeping difficult. Respiratory effects of exposures could have secondary effects on sleep by lowering lung function or exacerbating existing conditions like asthma, obstructive pulmonary disease, or sleep apnea ¹². Respiratory symptoms consistent with sleep impairment have been linked to ILO pollutant exposures in Western Europe and the United States, including excessive coughing ³⁰, asthma ³¹, wheezing ²¹, difficulty breathing ²¹, runny nose ³⁰, sore throat ^21,30, and chest tightness ²¹.

Generally, we saw stronger associations between sleep impacts and odor measures as compared to H₂S concentration. This could be explained by the presence and potential effect of ammonia on sleep, which was not directly measured in this study but is a component of ILO emissions and could also cause the experience of malodor. In this sense, the sum of sleep deficits attributable to odor could be thought of as the sum of the deficits of the emissions mixture components. Prior work with CHEIHO study data suggesting that H₂S and odor were strongly correlated during certain periods and discussions of estimated respiratory impacts impacts are consistent with this hypothesis⁷. In awakenings models, the magnitude of the association among those sensitive to odor appeared much larger than those not sensitive to odor, suggesting an important role of odor perception as a link between pollutant exposures and impacts on sleep.

Despite the relatively small sample size compared to a typical cohort study (n=80), this repeated-measures study design provides good control of time-invariant potential confounding factors and some control of time varying-factors. As observations are only directly compared within each person’s records, estimated associations cannot be explained by factors that remained constant for each participant over the study period. For instance, neither participants’ pre-existing medical conditions nor seasonal effects could affect the association between ILO pollutant exposures and lower sleep duration because they remained fixed over the short two-week study period for each participant. Other factors that influence sleep duration and awakenings from sleep, like noise, care for young children, and illnesses could confound the study results if they were also associated with odor, odor perception, and H₂S concentrations on a short time scale.

We adjusted our models for heat-humidity, a time-varying factor that we expected might have (1) a strong relation with both sleep and received exposures by influencing pollutant production or environmental transport, and (2) hourly and daily variability within the 2-week study period. Although heat and humidity can have independent effects of sleep, we had insufficient data to model these factors independently and models using these factors independently failed to converge; briefly, we would have required observations across multiple heat-humidity states for each participant across the two-week period for these models to converge (high-heat/high-humidity, high-heat/low-humidity, low-heat/high-humidity, and low-heat/low-humidity). A longer study period could address this limitation in future research.

Despite strong internal validity, it is difficult to estimate how closely these results might generalize to other populations (for instance urban populations) because rural communities hosting CAFOs face unique co-exposures. As potential participants were excluded from the study if they smoked or were unable to participate due to functional limitations, it is possible that the true effect is stronger than estimated here due to participants being healthier than the population exposed to swine ILO air emissions. Similarly, participants were excluded if they had atypical sleep schedules (e.g., shift work), which could have made the analytic sample appear healthier than the overall population exposed to CAFO pollutants. In sensitivity assessments, replicating the analysis without any participant exclusions yielded similar but less precise results (based on estimated effect direction and magnitude).

The study is limited by the quality of self-reported exposure and outcome data. The study used two exposure assessments: participants’ perceptions of odor from swine operations, and hydrogen sulfide concentrations measured in participants’ neighborhoods. In the present study context, odor perception can be understood as a biomarker of exposure and is limited by potential hysteresis, between-person variability, and non-linearity. Although environmental monitors were placed close to participants’ homes (mean 0.2 mi), they only measured one marker of a complex mixture of pollutants and did not quantify individually-received exposures and could differ from the air quality in participants’ sleeping environments indoors. The outcome data used for this study were collected from self-report in a parent study that was not originally designed to assess sleep disturbances, and have more missing values compared to other study data.

Additionally, our estimates of sleep duration are based on self-report and may overestimate true sleep duration. For instance, in the Coronary Artery Risk Development in Young Adults study, self-reported sleep duration overestimated 3-day actigraphy-measured sleep duration by 48 minutes based on the question: “How many hours of sleep do you usually get at night (or when you usually sleep)?” and these measures were only moderate correlated (0.45; 0.29 for blacks and 0.56 whites) ³². Similarly, Based on questions estimating the difference between bed and wake times on weekdays in the Multi-Ethnic Study of Atherosclerosis, Jackson et al. found that self-reported sleep duration overestimated sleep duration by, on average, 64 minutes compared to 5-day actigraphy-measured total sleep time, 49 minutes for 5-day actigraphy-based time in bed, and 64 minutes for sleep time based on in-home polysomnography (the alloyed gold standard) ³³ and all racial/ethnic groups overestimated sleep duration when comparing self-report of bed and wake time to both wrist actigraphy and PSG. In this study, self-reported sleep duration also only moderately correlated (ρ = 0.38 overall, with correlations significantly lower only among blacks compared with whites (ρ = 0.45 for whites, ρ = 0.28 blacks, ρ = 0.38 Hispanics/Latinos, and ρ = 0.35 Chinese) with 5-day actigraphy-based assessments ³³.

Measurement error in self-reported sleep status and the difference between odorant chemical concentration and participants’ experiences of odor could lead to bias if the accuracies of these self-reports are associated with each other, with the exposure, or with other confounders. For instance, long exposures to hydrogen sulfide and strong odorants could lead to olfactory fatigue and result in lower odor ratings during time periods in which the concentration of odorants were actually high. In the case that a true association between odorant chemical concentration and sleep duration was present, any exposure impacts observed during the period would be incorrectly assigned to the “unexposed” group, reducing the apparent association between odor and these impacts. Conversely, the apparent association resulting from a true effect could be magnified if (1) there were confounding factors linked to both under-estimation of odor and over-estimation of sleep duration, and (2) these factors systematically varied within participants over the two-week time period. We suspect that most within-person measurement errors in this study are nonsystematic with respect to the exposure and thus would expect the model results on balance to under-estimate any underlying associations.

We used independent measurement of H2S to address the potential for measurement errors in using odor reports as a surrogate of ambient odorant chemical concentration but did not have a comparable measure available for sleep duration. A study design using more objective, and comprehensive exposure assessment and an automated method of recording sleep data with improved temporal specificity (e.g. personal accelerometers) could help overcome the limitations resulting from self-reported data.

The repeated-measures models used for the study address some of these limitations and also reduces the potential for certain confounding biases that can be difficult to control in observational studies. These models estimated associations between within-person variations in exposure and outcome and are comparable to models adjusted by participant. This design removed the influence of factors that remained constant for each participant throughout the two-week study period, including demographic characteristics, season, odorant sensitivity, baseline sleep quality, and home permeability to odorants due to differences in construction and ventilation, which could differ by housing type and be worse for racial and ethnic minorities ³⁴. Potential confounding by time-of-day was addressed using hazard functions and comparison of identical time periods. Of the remaining potential time-varying potential confounding factors, heat-humidity was the most concerning and could be adjusted for in each model. Other potential pathways through which industrial livestock operations could impact sleep, like noise, were not assessed in this study.

The associations observed between swine ILO emissions exposure markers and sleep duration in this study suggest that ILO emissions have negative impacts on sleep among those living nearby. Sleep influences an array of disease risk factors and diseases, but is also an important part of health in its own right. Sleep is important for DNA repair ³⁵, cellular metabolism, tissue maintenance, immunological response, mood regulation, and memory consolidation ³⁶. Based on the importance of sleep to health, the National Sleep Foundation recommends 7 to 9 hours of sleep per night for adults 18–65 and 7 to 8 hours per night for adults over 65 ³⁷. Getting less than this recommendation (<7 hours) has been linked to increased risks of diabetes and obesity ¹⁶, cardiovascular disease ³⁸, accidents ³⁹, poor quality of life ⁴⁰, cancer ⁴¹, and premature death ⁴². The average sleep duration observed among participants in this study (7.3 hours per night) was not far from the 7-hour recommended minimum but could be below this minimum value given that self-reported sleep is typically under-estimated.

The social context of pollutant production in swine industrial livestock operations has proven resistant to public health intervention. Swine industrial livestock operations offer economic benefits to a select few that can use their political influence to secure their legal right to pollute ⁴³, leading to regulatory capture favoring larger operations ⁴⁴. Emissions abatement through technology or policy improvements could offer relief ⁴⁵, but have proven difficult to implement through traditional regulatory channels ⁵. In conjunction with existing data on the community health impacts of industrial animal operations, the results from this study could help guide public policy recommendations on health-based emissions controls for livestock waste treatment systems ⁴⁶. Odor abatement, in addition to nutrient management, should be an important consideration in designing systems to manage large-scale animal wastes.

CONCLUSION

Ultimately, the night-time ILO pollutant exposures investigated were associated with adverse sleep, which suggests that emissions reductions and odor abatement are important public health goals. Although we focused on odor and hydrogen sulfide from industrial livestock operations, environmental odorants from industrial sources in other areas could also be important for sleep hygiene and the secondary health effects associated with sleep. From a public health perspective, greater community control over local industrial development (for instance, increased representation in zoning and regulatory decisions) could help reduce the high concentrations of swine ILOs and other facilities that create environmental pollutants and positively impact population health, especially in vulnerable populations. Scientific evidence about the health impacts of ILOs could help facilitate meaningful public engagement around the health effects of environmental quality and promote public health by demonstrating the need for more sustainable methods of production.