Impact of Improved Biomass and Liquid Petroleum Gas Stoves on Birth Outcomes in Rural Nepal: Results of 2 Randomized Trials

Population

Two sequential trials were conducted in the low-lying plains of southern Nepal (Sarlahi District) contiguous with Bihar, northern India. This area is rural and has almost universal use of traditional mud brick biomass stoves (wood, dung, crop waste). The population consists primarily of subsistence farmers with low literacy rates. About half of women deliver at home and the prevalence of LBW, preterm, and SGA births is high.^13,14 In trial 1, the study area consisted of eligible households in 4 village development committees (VDCs) in Sarlahi. Eligible households in 2 of these 4 VDCs were enrolled in trial 2. Trial 1 was conducted between March 2010 and August 2012, and trial 2 between March 2013 and March 2014.

Study Design

The study methods have been described in detail elsewhere.¹⁵ The aim of trial 1 was to examine whether lowering indoor air pollution by using an improved biomass stove could reduce acute lower respiratory illness in children and adverse birth outcomes among pregnant women. At the end of the first trial, indoor particulate matter (PM) was reduced by the improved stove but was still very high. Therefore, a second trial was designed to examine whether cleaner fuel, such as liquid petroleum gas (LPG), could further decrease the exposure and increase the likelihood of improved health outcomes.

For trial 2, because households from trial 1 and newly eligible households were enrolled and randomized, some pregnant women who were previously enrolled in trial 1 may have changed the type of stove they used, depending on the randomization in trial 2. A total of 126 women were still pregnant at the end of trial 1 and were eligible to enroll in trial 2.

Trial 1 was a randomized step-wedge community-based trial. Households were eligible if they did not have an LPG, electric, or improved vented biomass stove; had house materials that did not constitute a fire hazard with chimney installation; had at least 1 married woman 15–30 years of age or at least 1 child <36 months of age in the household at the start of the trial; and consented to participate. Women who were pregnant at the start of the trial were included and were assumed to have been exposed to traditional biomass stoves prior to enrollment in the trial. Over a 2-year period, these women were monitored for pregnancy with regular visits every 5 weeks. If women reported missing a period since the previous visit, project staff offered a pregnancy test; newly identified pregnant women were then followed until an outcome (miscarriage, live birth, or stillbirth). This process meant that for incident pregnancies, the recall time for date of last menstrual period was not very long, improving the quality of gestational age compared with asking about date of last menstrual period at delivery. The data collection for reproductive outcomes began with a 6-month trial of uniform surveillance in all enrolled households. This period was followed by a 12-month step-in period in which each household eventually had their open burning stove replaced with an improved version that had a chimney venting to the outside. The timing of the replacement from traditional biomass to improved biomass stove was randomized. Another 6-month period of surveillance for reproductive outcomes followed the 12 months of stove replacement. This design provided a similar time period and seasons before and after improved stove installation in the enrolled households (Figure 1).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Nepal Cookstove Intervention Trial, Phase 1 CONSORT Diagram

In trial 2, households were eligible if they were in 2 of the 4 VDCs from trial 1, had at least 1 married woman 15–30 years of age or at least 1 child <24 months of age, and consented to participate. Study area households not enrolled in trial 1 were eligible for trial 2 if they met the same eligibility criteria as trial 1 (Figure 2). Households were individually randomized to use the improved biomass stove from trial 1 or to receive an LPG stove with a free year’s supply of gas. These households were followed for 1 year from the time of installation and birth outcomes within this time period were compared. Although this design was a traditional randomized trial, the length of time a pregnancy was exposed to the intervention varied.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Nepal Cookstove Intervention Trial, Phase 2 CONSORT Diagram

Randomization

In trial 1, 51 clusters of 20–30 households comprising about 25–40 children <3 years of age were identified. This number of households was the approximate number in which a single installation team could install improved stoves in 1 month. The clusters were then grouped such that all enrolled households could have stoves installed over 12 months, and those groups were randomized to the timing of receipt of the intervention. The numbers 1 through 12 were written on slips of paper, and senior project staff selected a number from a hat. The numbers represented the month of the year in which the intervention would be provided for each of the 12 groups of clusters. In trial 2, households were randomized stratified on VDC and ward (a smaller administrative unit, with 9 in each VDC) using the sample procedure in STATA.

Sample Size

The sample size for both trials was based on the number of households needed to detect a difference in acute lower respiratory infection rates. The number of live births estimated to occur in trial 1 households was 2350 and 600 in trial 2. Given these live births, the detectable difference in birth weight was 57 g in trial 1 and 83 g in trial 2 with power of 80% and a type I error of 5%. Assuming 29% of infants would be born LBW (<2500 g) before the intervention (based on prior studies in this area), we had >80% power to detect a 20% relative reduction in LBW in trial 1, and 80% power to detect a 34% relative reduction in LBW in trial 2. Correlation of the outcomes within clusters was considered in the sample size calculations. Positive correlation increased the sample size required. The prevalence of adverse birth outcomes observed before the intervention in trial 1 was 39% LBW, 48% SGA, and 22% preterm. Trial 2 may have been underpowered because 9% fewer pregnancies were enrolled than the anticipated number of 600.

Interventions

In trial 1, the intervention was a 2-burner biomass stove with a chimney to vent smoke to the outside. This stove was manufactured by Envirofit International (Colorado Springs, CO, USA), model G-3300 with G-3355 2 pot attachment.¹⁵ The stoves were installed by a specially trained study team, and household members were trained in their use and maintenance. Stove monitors visited households weekly to encourage and record stove use.

In trial 2, households were randomly assigned to either continue using the same Envirofit stoves as in trial 1 or were provided an LPG 2-burner stove with gas tank. Study workers provided each household with 1 tank per month, which was estimated to provide gas for an average household of 5 for the 12-month duration of the trial. At the end of the study, households were also provided an additional month of fuel. Households assigned to continued use of the Envirofit stove were provided LPG stoves and a 1-month supply of gas at the end of the trial.

Exposure Measures

In trial 1, monitors to collect daily particulate matter of 2.5 μm or less (PM_2.5), CO, humidity and temperature were deployed in each household at least once prior to and once post installation of the vented stove. On average, indoor air pollution was measured 6 months prior to and 6 months after stove installation. Measurements were made every 10 seconds over an average of a 21.7-hour period (from approximately 3:00 pm through 12:00 pm the next day). Lunch is not a usual meal in Nepal, so this monitoring period likely covered nearly all cooking events in a day. The package of instruments was placed approximately 1.5 m above the floor and about 1 m in front of the stove to best represent exposure of the person cooking. Details of the indoor air pollution monitoring and impact have been described elsewhere.¹⁶ We measured CO and PM_2.5. We included CO measurements because we hypothesized that CO might be directly associated with adverse birth outcomes through the mechanism of oxygen displacement. In trial 2, the measurement taken post installation of the Envirofit stove in trial 1 was used as the PM_2.5 exposure measure for trial 2 households randomized to retain the Envirofit stove. Exposure measures were obtained in LPG households post installation (between 1 and 12 months later, average 6 months) in trial 2 as a comparison. Temperature and humidity were measured with a HOBO U10 Temperature and Humidity (TH) Data Logger (Onset Computer Corporation, Pocasset, MA). CO was measured with the LASCAR CO data logger (EL-USB-CO300, Erie, PA). PM_2.5 was measured using a DataRAM pDR 1000AN (Thermo Fisher Scientific Inc., Franklin, MA) as a nephelometer. These measures were then corrected for humidity based on calibration with gravimetric measures.¹⁷

Outcome Measures

Pregnancies were followed to their outcomes. The family was asked to notify a local female study worker if the enrolled woman went into labor and a special team trained in infant anthropometry was dispatched to the household as soon after birth as possible to measure live born infants. Birth weights were measured with a Tanita BD-585 Pediatric Scale (Tanita, Tokyo, Japan) with an accuracy of 10 g. The scale was regularly calibrated prior to use. We used only weights obtained within 72 hours of birth. LBW was defined as <2500 g. Gestational age was estimated as weeks completed from date of last menstrual period (collected every 5th week by local female study workers visiting the homes of enrolled women) to birth. Gestational ages that fell outside a feasible range of 24 to <43 weeks were treated as missing. Preterm births were those occurring before 37 weeks completed gestation. SGA births were defined as those whose sex- and gestational-age-specific birth weights fell below the 10th percentile of the intergrowth population distribution using the upper bound of the weekly published data.¹⁸ Primary outcomes compared between stove interventions were mean birth weight, mean gestational age, LBW, preterm birth, and SGA.

Statistical Analysis

For trial 1, descriptive demographic and socioeconomic characteristics are provided at baseline, prior to any stove installation. Given the step-wedge design, these characteristics were assumed not to change from pre to post installation. For trial 2, these same characteristics were compared between households that were randomized to the LPG intervention and those randomized to the improved biomass stoves to assess comparability between the 2 groups.

To characterize the impact of the vented biomass and LPG stoves, the mean daily PM_2.5 in micrograms per cubic meter and CO in parts per million (ppm) were compared with traditional stove measures in both trials. In addition, we examined the number of hours the kitchen measures of PM_2.5 were above 100 μg/m³ and above 1000 μg/m³ and CO measures were above 6 ppm and above 9 ppm and compared these measures of exposure between intervention groups. We chose these cut points for PM_2.5 and for CO because we wanted to describe average daily exposure and an ambient level in the absence of a cooking event. The median daily exposure to PM_2.5 was 1070 μg/m³ and the mean was 1380 μg/m³. Therefore, 1000 μg/m³ was chosen as a cut point.¹⁶ For CO, the median was 8.1 ppm and the mean was 11.0 ppm. We also used a statistical approach to determining when a cooking event was occurring (spikes in exposure could be seen). The lower cut point thus examined average ambient kitchen levels at times when cooking was not occurring. This cut point corresponded to PM_2.5 and CO at about 100 μg/m³ and 6 ppm, respectively.

In trial 1, women could have experienced various lengths of exposure to the intervention stove while pregnant, ranging from no exposure to exposure for the entire pregnancy, given data collection occurred over 2 years. This exposure calculation applied to both existing (prevalent) and new (incident) pregnancies. Exposure for incident pregnancies was based on when in pregnancy the improved stove was installed. Given that the length of time exposed to the improved stove could affect the impact of the intervention, we stratified the effect of the intervention by time of exposure in utero (no exposure in pregnancy, >0 to <1/3, 1/3 to <2/3, 2/3 to <1, fully exposed). This was calculated as the time in pregnancy spent using each type of stove rather than trimesters exposed. Similarly, in trial 2, not all pregnancies were exposed to each randomized intervention for the entire length of pregnancy. However, given that trial 2 was a parallel individually randomized trial, we analyzed the birth outcomes comparing the 2 randomized groups without regard to length of exposure in utero because these were equal in both groups. In trial 2, 55% were enrolled in the first trimester, 29% in the second trimester and 16% in the third trimester.

Intervention effects in trial 1 were estimated using linear regression to compare the mean differences and 95% confidence intervals (CIs) in birth weight and gestational age between levels of exposure to the improved stove using no exposure as the reference. Secular trends in birth outcomes were adjusted for, using a cubic spline with knots every 100 days. We used this approach because there is strong seasonality in birth weights and to some extent in gestational age in this population.¹³ For LBW, preterm, and SGA, relative risks and 95% CIs were calculated with no exposure as the reference, using Poisson regression with robust variance estimation. In trial 2, the same birth outcomes were compared between the 2 treatment groups (Envirofit and LPG stoves) using linear regression for continuous outcomes and Poisson regression for binary outcomes.

In trial 1, the study design randomized the timing of intervention rather than geographic clusters. Groups that received the intervention each month during stove installation were deliberately scattered geographically across the study area to reduce within cluster correlation. Therefore, we did not adjust for cluster randomization. In trial 2, the unit of randomization was the household. During this 1-year period, 5 sets of twins were born. Hence we do not believe adjusting for multiple births in the same household would impact the outcomes.

Both trials were approved by the institutional review boards of the Johns Hopkins Bloomberg School of Public Health and the Institute of Medicine, Tribhuvan University, Kathmandu, Nepal. Verbal informed consent was obtained from all participating households and individuals and documented on data collection forms. The trials are registered at ClinicalTrials.gov (NCT 00786877).