Acute Effects on Blood Pressure Following Controlled Exposure to Cookstove Air Pollution in the STOVES Study.

Background Exposure to air pollution from solid fuel used in residential cookstoves is considered a leading environmental risk factor for disease globally, but evidence for this relationship is largely extrapolated from literature on smoking, secondhand smoke, and ambient fine particulate matter ( PM 2.5). Methods and Results We conducted a controlled human-exposure study (STOVES [the Subclinical Tests on Volunteers Exposed to Smoke] Study) to investigate acute responses in blood pressure following exposure to air pollution emissions from cookstove technologies. Forty-eight healthy adults received 2-hour exposures to 5 cookstove treatments (three stone fire, rocket elbow, fan rocket elbow, gasifier, and liquefied petroleum gas), spanning PM 2.5 concentrations from 10 to 500 μg/m3, and a filtered air control (0 μg/m3). Thirty minutes after exposure, systolic pressure was lower for the three stone fire treatment (500 μg/m3 PM 2.5) compared with the control (-2.3 mm Hg; 95% CI, -4.5 to -0.1) and suggestively lower for the gasifier (35 μg/m3 PM 2.5; -1.8 mm Hg; 95% CI , -4.0 to 0.4). No differences were observed at 3 hours after exposure; however, at 24 hours after exposure, mean systolic pressure was 2 to 3 mm Hg higher for all treatments compared with control except for the rocket elbow stove. No differences were observed in diastolic pressure for any time point or treatment. Conclusions Short-term exposure to air pollution from cookstoves can elicit an increase in systolic pressure within 24 hours. This response occurred across a range of stove types and PM 2.5 concentrations, raising concern that even low-level exposures to cookstove air pollution may pose adverse cardiovascular effects.

Clinical Perspective

What Is New?

We used a novel study design—a controlled human‐exposure study—to investigate acute responses in blood pressure following exposure to air pollution emissions from cookstove technologies.

Results demonstrated that short‐term exposures to cookstove‐generated air pollution can acutely perturb systolic blood pressure, with a small decrease immediately after exposure and a 2‐ to 3‐mm Hg increase 24 hours after exposure compared with filtered air control.

Responses were consistent across a range of stove treatment types, with fine particular matter (PM2.5) levels ranging from 10 to 500 μg/m3.

What Are the Clinical Implications?

Nearly 40% of the world's population that uses solid fuels for cooking, and replacement of traditional stove technologies with lower particulate matter–emitting technologies has been a major public health focus.

Our results suggest that household air pollution may be detrimental to cardiovascular health, even at low PM2.5 levels.

Given these findings, public health practitioners and researchers need to carefully consider the intended consequences of cookstove intervention programs and the timelines of exposure‐response observations.

Introduction

Nearly 40% of the world's population uses solid fuel for cooking.1 Exposure to the resulting household air pollution is a major contributor to global disease, particularly in the form of cardiovascular diseases.2 Although some studies have shown cardiovascular health benefits from improved stove designs that reduce emissions compared with traditional stoves,3, 4, 5 questions remain regarding the level of exposure reduction needed to reduce cardiovascular health burden.6, 7

The connections between household (ie, cookstove‐generated) air pollution and cardiovascular disease risk is extrapolated primarily from research on other pollution sources (ie, active cigarette smoking, secondhand smoke, and ambient air pollution).8, 9 Additional research is needed to explore emissions across a wide variety of cookstove practices, exposure levels, and health responses.

Blood pressure is an established marker of cardiovascular disease risk10, 11, 12 that can increase following acute and chronic exposure to ambient and household air pollution.4, 13, 14, 15, 16, 17, 18 Several field studies have investigated relationships between household air pollution exposures and blood pressure,4, 5, 17, 19, 20, 21, 22, 23, 24 with results ranging from null associations to upward of 10‐mm Hg increases in systolic pressure for traditional‐stove users compared with improved‐stove users. Few controlled wood‐smoke exposure studies exist,25, 26, 27, 28, 29, 30, 31, 32, 33, 34 and even fewer consider blood pressure; in 2 studies that did measure blood pressure, no effects immediately after exposures were found.35, 36 However, these studies did not include longer follow‐up times and observed other acute hemodynamic responses suggestive of vascular impairment and acute autonomic nervous system perturbations.

As part of the STOVES (Subclinical Tests on Volunteers Exposed to Smoke) Study, this work examines changes in blood pressure up to 24 hours following 2‐hour exposures to cookstove emissions from 5 stove technologies at characteristic fine particulate matter (PM2.5) mass concentrations between 10 and 500 μg/m3 and a filtered air control.

Methods

Methods are described briefly herein; more information is in the Supplemental Material. Data from this study are available from the corresponding author upon reasonable request.

Eligibility Criteria and Recruitment Methods

Forty‐eight healthy nonsmoking volunteers were recruited through articles in the local and university news, advertisements sent to various university email lists, and word of mouth. Eligibility criteria were based on age, weight, history of disease, drug use, and occupational or incidental pollution exposure, current cardiovascular health status, medication use, and ability to complete the study protocols (full criteria in Data S1). Individuals with occupations that may result in increased air pollution exposures were excluded. Individuals who were interested in participating in the study completed a screening questionnaire and attended an in‐person screening examination and physical to ensure they met study criteria (see Data S1). All study protocols were approved by the Colorado State University institutional review board; procedures followed were in accordance with institutional guidelines. All participants provided written informed consent.

Study Design

A sample size of 48 was chosen based on a number of factors including statistical power, budget, reasonability for recruiting in our target population, and maximizing our study design and facility capacities. Each participant underwent six 2‐hour exposure treatments over 13 to 16 weeks, with a minimum 2‐week period between treatments. We conducted the study in 3 rounds. Within each round, 2 groups of 8 participants alternated weeks until completion of all 6 treatments. Within each week, 4 participants started their study sessions on Mondays and 4 on Wednesdays. Treatment‐assignment sequences were determined following a Williams square design, a Latin square crossover that balances treatments and first‐order carryover effects.37 This design is robust for time‐invariant factors at the person level (ie, subject effects)—each person receives each treatment—and time‐variant factors that might differ across study sessions (eg, ambient conditions, caffeine or alcohol consumption) because the distribution of these variables is expected to be similar across all treatments when data are balanced.38, 39 Participants who missed a scheduled study session could make up the missed treatment at the end of their sequence. Participants were not told which treatment they were assigned to on each visit.

Study Session Protocol

The timeline of a study session is illustrated in Figure 1. Participants were instructed to abstain from medications, nutritional supplements, and vitamins starting 72 hours before each study day and from caffeine, alcohol, strenuous exercise, and smoke exposures (eg, campfires/wood stoves, secondhand smoke) starting 24 hours before and continuing through the 24‐hour follow‐up period. Participants were also asked to avoid high‐fat and high‐cholesterol foods on study days. Surveys were administered to determine compliance with these protocols (see Health and Additional Measurements). There were no restrictions on lifestyle, food, or activity between study sessions aside from maintaining compliance with the eligibility criteria.

Figure 1

Timeline of a study session. Participants arrived at the facility at the same time on each of their assigned study session dates (between 7:30 and 9 am) and completed the same protocols at each session according to the timeline shown. Participants completed sessions with a minimum of 10 days (typically 2–3 weeks) between sessions.

Participants arrived at the facility at the same time and followed the same protocols and schedule each study session. Participants were asked about current or recent illnesses at the start of each session, and an on‐site physician approved participation each day. Baseline health measurements were conducted on arrival (see Health and Additional Measurements). Participants then spent 2 hours in the exposure chamber receiving the treatment; the physician remained on‐site during exposures and confirmed that participants did not have any acute concerns on exiting the chamber; the physician was also available on call for 24 hours after the end of the exposure period. Additional rounds of health measurements were conducted starting immediately after exposure and 3 hours after exposure. Participants remained on‐site between measurements, and lunch was provided (low fat or low cholesterol; same each session). Participants returned for a final round of health measurements 24 hours after the end of the exposure treatment.

Treatments and Administration

The exposure chamber consisted of a main exposure room (2.7 m height×3.5 m width×2.8 m length) and an airlock/anteroom. Up to 4 participants could be in the chamber at the same time. Participants’ blood pressure, heart rate, and oxygen saturation levels were recorded by a registered nurse every 15 minutes during the exposure, for safety purposes.

Treatments consisted of a high‐efficiency particulate air–filtered control and pollution generated from 5 different cookstoves, chosen to represent commonly used technologies and span the International Standard Organization's (ISO's) cookstove performance tiers.40 A target PM2.5 exposure concentration was chosen for each stove. Setting target concentrations with a narrow tolerance for each stove allowed for increased statistical power to resolve between‐stove differences while abiding by protocols for participant safety and informed consent. Target concentrations were aligned with the ISO performance tiers and values realistically expected for the stove when used in the real world20, 41, 42, 43 while considering the feasibility of achieving the level with each stove within our facility and maintaining distinct distributions of exposures for each treatment. Cookstoves were a liquefied petroleum gas (LPG) stove (10 μg/m3), a gasifier (35 μg/m3), a forced‐draft (fan‐powered) rocket elbow (“fan rocket,” 100 μg/m3), a natural‐draft rocket elbow (“rocket elbow,” 250 μg/m3), and a three stone fire (500 μg/m3). Pollution was generated within a total‐capture fume hood, diluted with high‐efficiency particulate air–filtered laboratory air, and then drawn into the exposure chamber. Carbon monoxide, PM2.5, oxygen, temperature, and humidity in the chamber were monitored in real time; a dynamic control system (LabVIEW, v15.0 32‐bit; National Instruments) automated the real‐time PM2.5 averaging and dilution process. Real‐time PM2.5 was measured using a DustTrak DRX (model 8533; TSI Inc).

Additional treatment emissions characterization was conducted at the end of the study. The facility was operated for 2 hours under the same conditions as during human exposures but without participants present, on at least 2 occasions per treatment. Air was sampled from the facility for measurement of PM2.5 mass, particle‐number size distributions (10 to 500 nm), organic and elemental carbon (EC) concentrations, nitrogen oxide, nitrogen dioxide, volatile organic compounds, and carbonyls (see Data S1).

Health and Additional Measurements

Brachial blood pressure was measured 4 times per session: before exposure, immediately after exposure, 3 hours after exposure, and 24 hours after exposure. Measurements were performed on the left upper arm with participants in a supine position after a minimum 10‐minute rest period using an automated oscillatory monitor (SphygmoCor XCEL; AtCor Medical Pty Ltd). Three readings were taken 1 minute apart; the average of the last 2 measurements was used in the analysis.44

Questionnaires were administered to assess compliance with protocols and other factors across study sessions, such as the participant's mode of commute to our facility and incidental smoke exposures. Hourly ambient data for the 24 hours before and throughout each study session were downloaded from the US Environmental Protection Agency's Air Quality Data air pollution index and a local weather station.45, 46

Statistical Analysis

Data processing and statistical analyses were performed in R (v3.3.1; R Foundation for Statistical Computing).

Summary statistics (mean±SD, range) were calculated for anthropometric values for the total population and by sex. Each participant's mean PM2.5 and CO exposures were determined by averaging the 1‐second data over the 2‐hour exposure window; the population standard deviation and range were determined from the 2‐hour averages. Additional emissions‐characterization measurements (collected after the study ended) were averaged across each treatment.

Linear mixed‐effect models were employed (using the lme4 package47) to estimate the difference between blood pressure at each post‐exposure time point for each stove treatment compared with the control. Separate models were run for each time point (immediate, 3‐hour, 24‐hour) and for each blood pressure metric (systolic, diastolic). Model assumptions were evaluated by examining the normality of the model's residuals, linearity of the fitted models, and equality of the error variance. We also identified potential outliers in the data and examined the impact of the outliers on model fit.

The primary models contained a fixed effect of categorical treatment, a random person intercept to account for nonindependence across repeated measures within the crossover design, a random effect for date to account for within‐day correlation for individuals who received treatments on the same day, and the pre‐exposure blood pressure value to account for differences in individuals’ starting blood pressure across treatments or study sessions (which captures information similar to a pre‐/post‐exposure change model but is more efficient and easier to interpret).48, 49 By using stove treatment type in the model, we capture the combined effect of all the emissions from the stove (eg, particle and gases) on blood pressure compared with control. The study design eliminates the need to control for individual‐level confounders (eg, age, sex), as each person participates in each treatment, and external confounders that might vary across study days (eg, ambient conditions, caffeine or alcohol consumption), as each person participates in each treatment and treatments are balanced across time.38, 39 Descriptive statistics and bivariate analyses were conducted to confirm that associations between these covariates and the treatment groups did not occur by chance or because of imbalances caused by missing data.

Additional models were evaluated as alternatives to the main model. We developed a mixed‐effect model that considered more structured study design parameters relevant to our Williams square, including assigned sequence group and day of the week (Monday versus Wednesday). Only data that were collected within the intended sequence (ie, not including makeup sessions) were used in this model. We also ran the same model as the primary model but (1) excluding data collected outside of the intended sequence and (2) excluding data from study sessions in which the exposure mean was outside of a narrowed range around the target value.

Results

Participants

The 48 participants (26 male, 22 female) ranged from 21 to 36 years old (mean±SD: 27.5±3.6 years), were within the normal or low‐overweight body mass index (kg/m2) categories (mean±SD: 23.4±2.2), and on average had nonhypertensive baseline blood pressure (mean±SD for systolic/diastolic: 116±9/69±6 mm Hg; Table 1). Values were comparable between men and women. Participants predominantly identified as non‐Hispanic white (42/48; 88%). Reported use of alcohol, caffeine, and medication were low throughout the study; bivariate analyses indicated no meaningful associations between these or other potentially confounding covariates (eg, ambient PM2.5 and CO) and the various treatments (see Data S2 and Tables S2–S11).

Table 1

Description of Study Participants

Variable	All (n=48)	Female (n=22)	Male (n=26)
BMI, kg/m2	23.4 [2.2], 19.4, 28.7	23.5 [2.6], 19.7, 28.7	23.3 [2.0], 19.4, 26.0
Age, y	27.5 [3.6], 20.5, 36.1	27.5 [3.4], 22.8, 34.0	27.4 [3.9], 20.5, 36.1
Baseline SBP, mm Hg	116 [9], 99, 135	113 [9], 100, 135	118 [8], 99, 135
Baseline DBP, mm Hg	69 [6], 59, 86	69 [7], 60, 86	69 [5], 59, 80
Participants with data for all 6 treatmentsa	79	82	77
Participants with data for at least 5 treatmentsa	94	100	88

Data are shown as mean [SD],* minimum, maximum, or percentage. BMI indicates body mass index; DBP, diastolic blood pressure; SBP, systolic blood pressure.

Mean calculated as the population mean of each individuals’ average baseline health measurement across their completed study sessions.

Participant was counted if he or she had data for baseline measurement and at least 1 post‐exposure measurement.

The total missing data rate was 6% (see Data S2). Of the 48 participants, 22 (46%) completed the study in the intended, assigned order with no missed treatments. Age, sex, and body mass index were comparable between participants who missed sessions and those who did not (see Data S2). Using makeup dates at the end of a study round to complete missed treatments, 79% of participants (38/48) contributed data relevant to all 6 treatments, and 94% (45/48) had data for at least 5 treatments.

Exposure Conditions

The means and ranges of individual 2‐hour exposure averages within each treatment are provided in Table 2. The means of participants’ averaged PM2.5 mass exposure concentrations were within 10% of the target for the fan rocket, rocket elbow, and three stone fire treatments (+5, +4, and −37 μg/m3, respectively), 20% for the LPG treatment (+2 μg/m3), and 30% for the gasifier treatment (+16 μg/m3). The mean of participants’ averaged CO exposures per treatment type generally increased with increasing PM2.5, ranging from 2 ppm for the control up to 9 ppm for the three stone fire.

Table 2

Distributions of the Individual Mean 2‐Hour Pollutant Exposures Measured During Treatments

Treatmenta	Fuel	Participants Completing Treatment (n)	PM2.5 (μg/m3)		CO (ppm)
			Mean [SD]b	Min, Max Individual Exposureb	Mean [SD]b	Min, Max Individual Exposureb
Control	None	47	1 [2]	−1c, 9	2 [2]	1, 10
LPG	Propane	44	8 [3]	3, 13	3 [1]	1, 6
Gasifier	Wood chips	44	46 [9]	30, 76	5 [3]	1, 14
Fan rocket	Wood sticks	44	95 [9]	77, 111	8 [2]	5, 12
Rocket elbow	Wood sticks	45	254 [9]	236, 276	6 [2]	3, 11
Three stone fire	Wood sticks	47	463 [41]	367, 531	9 [4]	4, 20

LPG indicates liquefied petroleum gas; max, maximum; min, minimum; PM2.5, fine particular matter.

Target PM2.5 levels for each treatment were high‐efficiency particulate air‐filtered air (0 μg/m3), LPG (10 μg/m3), gasifier (35 μg/m3), fan rocket (100 μg/m3), rocket elbow (250 μg/m3), and three stone fire (500 μg/m3). CO did not have a target level and was not controlled but rather varied naturally.

Measured pollutant mean is of the participants’ 2‐h average values, calculated by determining the 2‐h average of the 1‐s exposure data for each participant and then averaging across all participants for each treatment. Min and max individual values are the lowest and highest 2‐h average value measured for a single participant.

Negative values are a result of a DustTrak calibration artifact.

Additional pollutant measurements conducted after the end of the study were used to characterize particle properties and quantify the coemitted gases in the cookstove smoke compared with the control filtered air. Nitrogen oxide concentrations were elevated for all stove treatments compared with the control, with the largest differences for the fan rocket and rocket elbow stoves (24 ppb each, compared with 1 ppb for the control, 4 ppb for LPG and three stone fire, and 2 ppb for gasifier). Nitrogen dioxide levels were similar for all treatments, including the control (range: 8–12 ppb). Gaseous carbonyls were measured in all treatments including the control, with the highest levels for LPG, rocket elbow, and three stone fire (197, 194, and 293 μg/m3, respectively, versus 107, 128, and 131 μg/m3 for control, gasifier, and fan rocket). EC concentration was notably higher for the rocket elbow stove (94 μg/m3; EC:PM2.5 ratio: 0.7) compared with the other treatments (0, 3, 29, 38, and 30 for control, LPG, gasifier, fan rocket, and three stone fire; EC:PM2.5 ratios of 0.1–0.5). Although total particle number generally increased in a PM2.5 mass‐dependent manner, ultrafine particle number fraction was considerably higher for the LPG treatment than all others (<95% of particles were <100 nm versus 60–70% for all others). Within the smallest measured size, 10 to 30 nm, the absolute particle number for the LPG treatment was ≈4 times higher than for the gasifier and three stone fire (which were similar), 65% higher than for the rocket elbow, and 25% higher than for the fan rocket. Additional detail is provided in Data S2, Table S1, and Figure S1.

Differences in Blood Pressure for Stove Treatments Compared With Control

Blood pressure measurements occurred on average 30 minutes (SD: 4.2 minutes) after exposure for the immediate time point, 3 hours and 26 minutes (SD: 4.8 minutes) for the 3‐hour time point, and 24 hours and 13 minutes (SD: 30 minutes) for the 24‐hour time point (see Data S2). Mean blood pressure across all participants, treatments, and time points was nonhypertensive (average: 115.7/68.9 mm Hg), although some individual measurements were within a hypertensive range (measurements with systolic pressure ≥130 mm Hg: 9%; measurements with diastolic pressure ≥80 mm Hg: 8%). Average pre‐exposure blood pressure varied by treatment type (highest: three stone fire, 117.0/70.2 mm Hg; lowest: fan rocket, 115.0/68.2 mm Hg).

Effect estimates and 95% CIs for the difference in blood pressure after exposure for each stove treatment compared to the control from the main model are presented in Table 3 and Figure 2. Alternative model results were consistent with the main model (see Data S2, Table S12, and Figures S2–S5).

Table 3

Mean Difference in Blood Pressure for Stove Treatments Compared With Control at Each Measurement Time

Treatment	BaselineaValue mmHg [Mean (SD)]	Effect Estimate (95% CI) [mm Hg Difference Compared With Control Treatment]b
		Immediately After Exposure	3 h After Exposure	24 h After Exposure
Systolic pressure
LPG	116.5 (10.7)	−0.2 (−2.5 to 2.0)	1.1 (−1.1 to 3.3)	3.1 (1.0–5.3)
Gasifier	115.7 (10.8)	−1.8 (−4.0 to 0.4)	1.0 (−1.2 to 3.2)	2.3 (0.1–4.5)
Fan rocket	115.0 (9.2)	−0.4 (−2.7 to 1.8)	−1.8 (−4.0 to 0.5)	2.5 (0.4–4.7)
Rocket elbow	115.6 (9.7)	−0.58 (−2.8 to 1.6)	−0.5 (−2.7 to 1.7)	−0.1 (−2.2 to 2.1)
Three stone fire	117.0 (11.3)	−2.3 (−4.5 to −0.1)	−2.1 (−4.3 to 0.2)	2.4 (0.3–4.5)
Diastolic pressure
LPG	69.2 (6.7)	−0.7 (−2.2 to 0.8)	−0.0 (−1.7 to 1.7)	0.3 (−1.6 to 2.2)
Gasifier	69.1 (6.9)	−0.8 (−2.2 to 0.7)	0.25 (−1.5 to 2.0)	−0.4 (−2.3 to 1.5)
Fan rocket	68.2 (7.3)	−0.1 (−1.6 to 1.4)	−0.4 (−2.2 to 1.3)	−0.1 (−1.9 to 1.8)
Rocket elbow	69.1 (7.3)	0.4 (−1.1 to 1.8)	0.2 (−1.5 to 1.9)	−1.7 (−3.6 to 0.2)
Three stone fire	70.2 (7.6)	−0.9 (−2.3 to 0.60)	−0.8 (−2.5 to 0.9)	0.8 (−1.0 to 2.7)

LPG indicates liquefied petroleum gas.

Control value at baseline: systolic: 115.2 (9.6) mm Hg; diastolic: 68.6 (6.6).

All estimates are adjusted for baseline (pre‐exposure) blood pressure.

Figure 2

Effect estimates and confidence intervals for difference in blood pressure for stove treatment compared with control, by stove type and post‐exposure time point. Top: Systolic pressure. Bottom: Diastolic pressure. LPG indicates liquefied petroleum gas.

At the immediate post‐exposure measurement, systolic pressure was significantly lower compared with the filtered air control for the three stone fire treatment (500 μg/m3 PM2.5: −2.3 mm Hg; 95% CI, −4.5 to −0.1) and suggestively lower for the gasifier (35 μg/m3 PM2.5: −1.8; 95% CI, −4.0 to 0.4). Other treatments were not meaningfully different from the control at the immediate post‐exposure time point.

No significant differences were observed for systolic pressure between the control and treatments at 3 hours post‐exposure. However, effect estimates were ≈2 mm Hg lower than the control for the fan rocket (100 μg/m3 PM2.5: −1.76 mm Hg; 95% CI, −4.02 to 0.50) and three stone fire (500 μg/m3 PM2.5: −2.05; 95% CI, −4. to 0.15) treatments. Effect estimates were ≈1 mm Hg higher than the control for the LPG (10 μg/m3 PM2.5: 1.10 mm Hg; 95% CI, −1.1 to 3.33) and gasifier (35 μg/m3 PM2.5: 0.99 mm Hg; 95% CI, −1.2 to 3.23) treatments.

At 24 hours post‐exposure, systolic pressure was significantly higher than the control by 2 to 3 mm Hg for all treatments except the rocket elbow. These large significant effects followed a consistent pattern across stoves, with effect estimates ranging from 2.3 to 3.1 mm Hg and 95% CIs ranging from 0.1 to 5.3 mm Hg for the LPG, gasifier, fan rocket, and three stone fire treatments (LPG: 3.11 mm Hg [95% CI, 0.65–5.27]; gasifier: 2.3 mm Hg [95% CI, 0.11–4.48]; fan rocket: 2.54 mm Hg [95% CI, 0.39–4.70]; three stone fire: 2.41 mm Hg [95% CI, 0.28–4.53]).

Differences were consistent with the null for diastolic pressure at every time point for all stove treatments compared with the control except for the rocket elbow treatment at 24 hours after exposure, which was suggestive of lower diastolic pressure compared with the control.

Discussion

Exposure to household air pollution is a leading contributor to disease worldwide, yet there are many gaps in our understanding of how different stoves and exposure levels contribute to health effects. We observed evidence that short‐term exposures to cookstove emissions resulted in a 2‐ to 3‐mm Hg increase in systolic pressure compared with filtered air control at 24 hours after exposure. Conversely, 30 minutes after exposure we observed small nonsignificant decreases in systolic pressure compared with control that generally returned to no difference 3 hours after exposure. These differences were seen across stove types at PM2.5 levels from 10 to 500 μg/m3 and did not appear to follow an exposure‐response pattern that corresponded with increasing PM2.5 or CO concentrations. Results for diastolic pressure were generally consistent with the null hypothesis for all times and stove treatments.

Particulate matter air pollution is hypothesized to elicit vascular dysfunction through a variety of mechanistic pathways including activation of the autonomic nervous system and increased parasympathetic responses, proinflammatory responses leading to oxidative stress and inflammation, and direct interaction of particles with molecules in blood circulation that regulate endothelial function and cell signaling.13 Recent evidence supports an adrenal stress response (eg, increased glucocorticoids due to hypothalamic–pituitary–adrenal axis activation) may also be involved in particulate matter–induced blood pressure elevations.50 Our observed responses suggest that short‐term exposure to air pollution from most cookstoves (regardless of PM2.5 levels) produced a delayed increase in systolic pressure that was observable within 24 hours. As reviewed elsewhere,18 a delayed increase is suggestive of biological pathways with slower onset but more persistent actions. This could include hypothalamic–pituitary–adrenal axis activation and/or vasomotor dysfunction induced by slower proinflammatory (eg, cytokine‐mediated) mechanisms. Further work to investigate circulating inflammatory or hypothalamic–pituitary–adrenal axis markers would help confirm this hypothesis. The lack of acute increase in blood pressure does not support autonomic nervous system activation.14 There are multiple mechanisms through which air pollution exposures could result in biological changes that affect systolic pressure more than diastolic pressure; for example, if air pollution exposure causes increased arterial stiffness (as suggested by some work51), this would favor greater changes in systolic pressure. More work is needed to elucidate these pathways.

The post‐exposure times for health measurements were chosen because of a combination of logistical considerations within our study protocols and because they represent potentially key response times within the mechanistic pathway for cardiovascular effects from air pollution.14 However, it is possible that the timing of our measurements (starting 30 minutes after the exposure ended) missed an immediate blood pressure increase during particle inhalation. Prior controlled inhalation studies with diesel particles and concentrated PM2.5 have shown that blood pressure can increase by ≥2 mm Hg immediately (within minutes) during short‐duration exposures in the 100‐ to 200‐μg/m3 PM2.5 range but does not stay elevated after particle inhalation ceases (subsiding within a few minutes to hours).16, 18, 52, 53, 54, 55 Only 1 of the identified studies maintained follow‐up through 24 hours; no effect was observed at this time.52 If this acute autonomic blood pressure elevation occurred in our study, we may not have observed it given our design. Perturbations that may have occurred during the exposure window, which are likely to be through an immediate autonomic nervous system activation pathway, might have subsided by the time post‐exposure measurements were conducted. Future studies that assess the blood pressure responses concomitant with exposure can help clarify this issue. Alternatively, if small changes in systolic pressure occurred at the immediate or 3‐hour time point, our study may not have been sufficiently powered to detect them. Our study may have been underpowered to detect effects in diastolic pressure at any time point because diastolic pressure is measured with less accuracy than systolic pressure, is more variable, and has a smaller absolute value and range.

Previous work suggests that the adverse cardiovascular effects of diesel exhaust exposure are entirely attributed to the particulate phase.56 It is possible that the complex mix of gaseous and particle pollutants in cookstove combustion results in competing vasoconstricting and vasodilatory effects that manifest differently across different stove types. For example, coemitted NO may have elicited a vasodilation response that obfuscated an immediate PM2.5‐induced blood pressure elevation. However, we do not have sufficient data on the multipollutant exposures to further support this hypothesis. Further work to assess blood pressure changes in studies that remove gaseous coexposures, leaving only cookstove particles, could help clarify these speculations.

Moreover, it is possible that we did not observe immediate effects on blood pressure (as is seen in ambient studies) because compositional differences in LPG and wood combustion emissions compared with ambient pollution result in different responses. The pollutant characterization tests conducted after the study ended demonstrated differences in particle composition (eg, EC and ultrafine levels) across stove types, suggesting that acute responses to air pollution of similar PM2.5 levels but from different sources may not be comparable. Few controlled wood‐smoke exposure studies exist and none use advanced cookstoves or nonbiomass LPG fuel; investigation of blood pressure in these studies is further limited. However, results generally align with our findings. Unosson et al35 found no changes in systolic or diastolic pressure during the 1 hour after a 3‐hour exposure to birch wood smoke (300 μg/m3 PM2.5) generated by a Nordic chimney stove compared with filtered air exposure. Evans et al36 reported no immediate effects on systolic pressure following 20‐minute exposure sessions to environmental tobacco smoke, cooking oil fumes, and cedar wood smoke (peak concentration target 350 μg/m3, generated by open burning) compared with a water vapor control. Neither study included a delayed follow‐up measure (eg, 24 hours). Hunter et al57 found no changes in blood pressure among firefighters during a 1‐hour controlled exposure to birch wood smoke generated by a Nordic chimney stove (1000 μg/m3) or at follow‐up 6 and 24 hours after exposure ended; however, this population may be less susceptible to acute impacts of smoke than a general population.

We did not observe an acute exposure‐response relationship between PM2.5 mass and blood pressure. A possible explanation for these findings is that other smoke constituents (besides PM2.5 mass) may be responsible for eliciting some or all of the observed blood pressure responses. No single pollutant provides an obvious explanation for the similar systolic pressure responses across most of the stove treatments or the null response for the rocket elbow treatment compared with the control at 24 hours post‐exposure. Alternatively, an exposure‐response curve for cookstove smoke and blood pressure may not exist on the timescale studied (2‐hour exposures, 24‐hour follow‐up). Previous work suggesting supralinear exposure‐response curves for air pollution are for different cardiovascular end points (eg, ischemic heart disease, cardiovascular mortality) and long‐term exposures;9, 58 although some limited cross‐sectional analyses of in‐field cookstove exposures and blood pressure suggest that a nonlinear relationship exists for individuals with chronic exposures.20, 51 Studies of acute cigarette exposures suggest that changes in subclinical cardiovascular function may occur at similar levels for active and passive smoking.59, 60 Our results suggest an acute threshold effect may occur for cookstove air pollution, with similar responses in blood pressure following exposure regardless of PM2.5 concentration levels or source (eg, LPG versus wood). It is unclear how the results of our study might translate under long‐term‐exposure scenarios.

Participants were young, predominantly white, healthy individuals with limited air pollution exposures in their daily lives; therefore, the generalizability of results to cookstove users globally may be limited. This population was feasible to study in this context in terms of participant safety and allowed us to minimize confounding or interactions by age, comorbid disease status, or other pollution exposures. Our study has strong internal validity accomplished by the controlled exposure design, and this strengthens the study's ability to balance data gaps of potentially more generalizable but less internally valid observational studies.

Our study expands on previous air pollution controlled exposure studies by incorporating more exposure levels, allowing for confidence in statistically suggestive trends; including more participants for greater power; and generating treatment exposures from multiple cookstove types. The Williams square crossover design and restrictive study protocols allowed for within‐person comparisons and eliminated many potential confounders, resulting in efficient analyses comparing more stove types and exposure levels than observational designs.

We demonstrated that short‐term exposures to cookstove‐generated air pollution can acutely perturb systolic pressure, with a small decrease immediately after exposure and a 2‐ to 3‐mm Hg increase 24 hours after exposure compared with filtered air control. Responses were consistent across a range of stove treatment types, with PM2.5 levels ranging from 10 to 500 μg/m3, which suggests that household air pollution may be detrimental to cardiovascular health, even at low PM2.5 levels. Given these findings, public health practitioners and researchers need to carefully consider the intended consequences of cookstove intervention programs and the timelines of exposure‐response observations. Further work is needed to better characterize the multipollutant exposures from household air pollution and aid in understanding the relationship to blood pressure across a range of smoke exposure compositions. Researchers must also carefully consider how acute exposure‐response relationships seen in controlled exposure studies translate to real‐world, chronic exposures because different exposure and response timelines and populations may affect results.

Sources of Funding

This research was funded by the US National Institute of Environmental Health Sciences grant ES023688.

Disclosures

None.

Data S1. Supplemental methods.

Data S2. Supplemental results.

Table S1. Characterization of Exposure Treatments: Average Emissions

Table S2. Alcohol, Caffeine, and Medication Intake by Treatments: 24 Hours Before Session Start

Table S3. Alcohol Caffeine, and Medication Intake by Treatments: During the Study Session

Table S4. Mode of Commute to Facility by Treatments: Before Session Start

Table S5. Mode of Commute to Facility by Treatments: Before the 24‐Hour Health Measurements

Table S6. Sleep Quantity by Treatment: Night Before Start of Study Session

Table S7. Sleep Quality by Treatment: Night Before the 24‐Hour Health Measurements

Table S8. Ambient Fine Particular Matter (PM2.5) Levels* by Treatment: 24 Hours Before Session Start

Table S9. Change in Systolic Pressure per Interquartile Range Change in Ambient Fine Particular Matter (PM2.5)

Table S10. Ambient CO Levels* by Treatment: 24 Hours Before Session Start

Table S11. Mean Temperature* (°C) by Treatment: 24 Hours Before Study Session

Table S12. Comparison of Model Results for 3‐Model Options: Effect Estimates and 95% CIs for All Model Parameters

Figure S1. Pollutant characterization in exposure facility, by treatment.

Figure S2. Effect estimates and 95% CIs for mean difference in systolic pressure (mm Hg) for stove treatments compared with control for main model compared with the model including ambient fine particular matter (PM2.5) variable.

Figure S3. Effect estimates and 95% CIs for mean difference in systolic pressure (mm Hg) for stove treatments compared with control for the main model vs the fully adjusted model. Fully adjusted model contains additional variables of alcohol consumption, caffeine consumption, medication use, sleep quantity, ambient fine particular matter (PM2.5), and ambient temperature.

Figure S4. Effect estimates and 95% CIs for mean difference in systolic pressure (mm Hg) for stove treatments compared with control for the 3 model types.

Figure S5. Effect estimates and 95% CIs for mean difference in systolic pressure (mm Hg) for stove treatments compared with control: comparison of main model to model with exposure outliers removed.

Click here for additional data file.