Health impacts of a randomized biomass cookstove intervention in northern Ghana

Worldwide, approximately 3 billion people cook using solid fuels (e.g., wood, coal), producing household air pollution (HAP) that contributed to approximately 2.3–3.8 million premature deaths per year. Specific illnesses linked to HAP exposure include pneumonia, acute lower respiratory disease, stroke, ischemic heart disease, lung cancer and chronic obstructive pulmonary disease (COPD) [1‐5]. Cleaner cooking practices have the potential to reduce HAP exposure, improving people’s health and saving lives. While many so-called “improved” biomass stoves have shown promising results in laboratory tests, measuring the effectiveness of these stoves in real world settings is essential to assess whether or not these stoves are living up to their potential and should be more widely promoted.

In this paper, we examine indicators of the health impacts of the REACCTING (Research on Emissions, Air Quality, Climate, and Cooking Technologies in Northern Ghana) randomized cookstove intervention study [6]. This study provided two different types of biomass stoves, which were both intended to improve cooking efficiency and reduce emissions, for free to study participants. Given limited evidence on the effectiveness of these stoves for changing cooking behaviors and reducing HAP exposure in a field setting, the primary focus of the REACCTING study was on these intermediate impacts on the causal pathways towards health and environmental outcomes (see Fig. 1).

In prior papers, we have reported effects of the intervention on cooking behavior [7, 8], stove emissions [9], fuel use [10], personal exposure [11, 12], and ambient air quality [11, 13]. Briefly, we found that households did use both types of improved stoves, but also continued to use their traditional wood and charcoal stoves (a practice known as stove stacking) [7, 8]. Use of the lower-tech rocket stove (Gyapa) was higher than use of the higher-tech gasifier stove (Philips), and use of the latter declined substantially in the second year of the study due in large part to battery failures [7]. Particulate matter emissions were not significantly reduced relative to a traditional three stone fire for either the Philips or the Gyapa stove, while the Philips stove did reduce carbon monoxide (CO) emissions by roughly 50% [9]. We also observed some reductions in personal exposure to particulate matter for all intervention groups, though results vary by pollutant; organic carbon exposure was reduced for all intervention groups, while elemental carbon was only reduced in groups receiving the Philips stove, relative to the control group [11].

Building on these results, this paper estimates effects of the REACCTING intervention on a set of health indicators. As noted above, studies have linked exposure to HAP with a wide variety of health outcomes. In particular, respiratory illnesses and pulmonary health outcomes have been a key focus of prior work [4, 5, 14]. A systematic review of studies that looked at the association between biomass fuel exposure and chronic obstructive pulmonary disease (COPD, clinically confirmed or self-reported) found that the overall odds were 2.4 [95% CI: 1.47,3.93] for women exposed to biomass fuel compared to other fuel exposure. In the same systematic review, children exposed to biomass fuel had 3.53 [95% CI: 1.94,6.443] times the odds of acute respiratory infection compared to children exposed to other types of fuel [15]. Another systematic review and meta-analysis also concluded that there was a positive association between solid fuel use and COPD (OR = 2.80, 95% CI = 1.85,4) [16]. Systematic reviews have also linked use of biomass fuels for cooking to lung cancer [17, 18]. Exposure to fine particulate matter air pollution, a key component of HAP from biomass fuel burning, has also been shown to contribute to cardiovascular disease and acute cardiac events [19].

A growing number of studies examine whether interventions aimed at reducing exposure to indoor pollutants can reduce the risk of adverse outcomes in both adults and children. The majority of the interventions studied have not been able to show a significant health impact. One notable exception is the RESPIRE study, a randomized controlled trial conducted in highland Guatemala among households with either a pregnant woman or young infants [20]. The intervention (woodstove with a chimney) led to a slight reduction in the incidence of pneumonia in children compared to the control (open woodfires) (RR = 0.82, 95%CI = 0.70,0.98) [20]. Meanwhile, the Cooking and Pneumonia Study (CAPS) in Malawi showed no significant differences between the intervention (gasifier woodstove) and control groups for pneumonia episodes in children (IRR = 1.01, 95% CI = 0.91,1.13) [21]. Similarly, another study in Orissa, India, concluded that the cookstove intervention did not have a significant impact on health outcomes [22]. In both of the latter cases, low rates of improved stove use contributed to small reductions in personal exposure to pollutants, limiting the potential for health effects. Stove uptake was higher among households with pregnant women in the Ghana Randomized Air Pollution and Health Study (GRAPHS) who received either a liquefied petroleum gas (LPG) stove or an improved biomass stove. Air pollution (CO and particulate matter) were reduced in the LPG arm relative to the control, but these reductions did not lead to significant improvements in birthweight [23].

In the REACCTING study, we did not have access to data on clinical cases of respiratory infections, pneumonia, COPD, birthweight, or other illnesses. To assess whether observed changes in cooking behavior and exposure were able to “move the needle” on health outcomes, we therefore relied on three primary sets of data measuring health outcomes associated with HAP exposure: self-reported health symptoms, biomarkers of inflammation collected from dried blood spots, and child growth measures from anthropometrics. Here we discuss each of these measures and its connection to HAP-related health outcomes.

The REACCTING study protocol is detailed in Dickinson et al. [6], and the study area is described by Oduro et al. [51]. In brief, the Kassena-Nankana Districts, located in the Upper East Region in Ghana, have a population of about 156,000 with an area of 1657 km². The study was conducted through a partnership among researchers based in the United States and at the Navrongo Health Research Centre, an organization which routinely conducts a census of all households in the district (the Health and Demographic Surveillance Survey, HDSS). These data were used to carry out a cluster random sampling approach to select households for inclusion in the study (see Fig. 2). In this context, clusters consist of groups of up to 99 households defined for the purposes of HDSS data collection. The target population for the intervention included rural households that relied primarily on biomass for their cooking needs, and that included both adult women and young children (since these groups were widely thought to experience higher exposures to HAP). We thus used the HDSS data to apply a set of household-level eligibility criteria, which included: classified as “rural” according to the HDSS; reporting use of firewood, animal waste, or crop residue/sawdust as the household’s “main source of cooking fuel”; and having at least one child under five and one woman between the ages of 18 and 55 in the household. We then applied a set of cluster-level eligibility criteria: no more than 25% of households classified as urban according to the HDSS; accessible year-round (determined by field staff); and having at least 10 eligible households (after applying household-level criteria above). From the remaining set of clusters, we then randomly selected 25 from four geographic regions (North, South, East, West) in rough proportion to the distribution of population across these regions. Finally, we randomly selected eight target households and two replacements from each cluster. Target households were recruited for the study by the baseline interview team; if the household could not be located or declined to participate, a replacement was selected from the replacement list. The resulting sample consisted of 200 households.

The main traditional stove in the study area is a three stone fire and the majority of households also had a charcoal stove (“coal pot”). The REACCTING intervention involved two different types of improved biomass cookstoves, the Gyapa Woodstove and the Philips Smokeless Woodstove (HD4012-LS). Selection of these stoves for this intervention was based on formative research, described in more detail in the study protocol [6], which involved piloting stoves with households in the K-N Districts, assessment of stove testing results, and consultation with the international cookstove community. The Gyapa stove is a simple rocket stove that was made in Ghana specifically for this study. It represents a lower-tier, affordable stove with some potential for fuel efficiency improvements over a three stone fire. Meanwhile, at the time of the intervention’s launch, the Philips stove was widely regarded as the cleanest improved biomass stove model based on laboratory tests [52]. It had a forced draft design with a battery-powered fan (charged via a solar panel that was also provided to participants).

The study design involved randomizing the 200 participating households into one of four different cookstove groups, with randomization occurring within each cluster (i.e., stratified by cluster). Following the baseline survey, groups of households (eight households per cluster, as described above) were invited to community meetings where the two types of stoves were demonstrated and instructions on proper use were given. After the demonstration, randomization occurred by each participant (primary cook) drawing a slip of paper from a bucket with the intervention group assignment written on the paper. Within each cluster, two households were assigned to each of the four intervention arms. The first group received two Gyapa stoves, the second received two Philips stoves, the third received one of each type of stoves (one Gyapa and one Philips stove), and the fourth served as the control group for the duration of the study (and received their choice of two stoves at the study’s conclusion). Participants were not compensated beyond the stoves they received. The decision to provide two stoves to intervention households was based on our own prior work in this area and other studies that indicated a tendency for households to cook with multiple stoves and stove types [53], such that providing multiple stoves was thought to create a greater potential to displace traditional stoves.

Next, we examined a set of biomarkers that have been implicated in the inflammatory pathway linking exposure to PM with health outcomes. The most intriguing evidence for possible health effects involves adults’ CRP levels (lower in the Gyapa/Gyapa group relative to control) and SAA levels (lower in all three intervention groups relative to control). CRP and SAA have been linked to respiratory diseases including COPD [66], and CRP has also been associated with air pollution exposure and cardiovascular disease [19, 38]. A reduction in these inflammatory biomarkers and a trend in several other indicators provides suggestive evidence that intervention may have lowered the systemic pro-inflammatory status of these women. At the same time, in light of the number of tests that were run and the issue of multiple comparisons, we are cautious not to overinterpret our results. Any improvements that were achieved at the level of biomarkers did not manifest into improvements in child growth or self-reported health symptoms. Clinical outcomes (e.g., respiratory illness, mortality) were not measured in this study, and we would be underpowered to detect effects on these outcomes given our sample size.

Turning to child growth, although two stove groups significantly differed from the Control group in the endline HAZ model, this is likely an artifact of the sample. If stove group had a significant and realized impact, the parameter estimates for each stove group relative to the Control group should all be positive. Over time, individual HAZ tends to increase, especially mean HAZ for groups, sometimes irrespective of height deficits relative to a reference median increasing [67]. Since this is a low-income population, WHO Growth standards are somewhat biased as they were computed based on primarily Western populations, though no other appropriate references exist for this population at present.

It is informative to interpret these results in light of the study’s prior findings for measures along the causal pathway linking improved stove introduction to health outcomes. First, using both surveys and electronic stove use monitors, we found high levels of “stove stacking” (continued use of traditional stoves) across study groups throughout the study [7, 8]. Use of the lower-tech Gyapa stove was substantially higher than use of the more advanced Philips stove, and high rates of battery failure in the Philips stove contributed to a declining trend in use of this stove over time. Next, the health impacts of “improved” stoves are contingent on the extent to which these stoves reduce emissions relative to traditional stoves when operated by end-users in the field. The Philips stove in particular had shown promising emissions results in lab testing [52], which was a major factor in our decision to include this stove in our study (and in others’ decisions to use this stove in studies initiated around the same time [21, 68]). However, results from uncontrolled cooking tests (encompassing variability of a real world setting) showed mixed results across performance metrics [9]. As tested in the field, particulate matter emissions were not significantly reduced relative to a traditional three stone fire for either the Philips or the Gyapa stove, while the Philips stove did reduce carbon monoxide (CO) emissions by roughly 50% [9]. Finally, a growing body of evidence suggests that improved health outcomes are likely only realized with significant and substantial reductions in exposure to air pollution given a steep exposure-response curve [2, 69]. Personal exposure measures from the REACCTING study did show some reductions (49–63%) in organic carbon PM_2.5 exposure for all intervention groups relative to the control, while elemental carbon PM_2.5 exposure was reduced in groups receiving the Philips stove (44–59%) [11]. However, factors specifically linked to cooking / biomass burning (polycyclic aromatic hydrocarbons (PAHs) and methoxyphenol) did not differ significantly across intervention arms [11]. Additionally, source apportionment results from speciated personal and ambient PM_2.5 point to large contributions of PM_2.5 from sources other than stoves, reducing the potential for exposure reductions from a purely cookstove-targeted intervention [11]. Putting these pieces together, the story that emerges is that these stoves resulted in small reductions in exposure to HAP at best, leading to limited changes in the health measures examined here.

Our results are largely in line with the accumulating evidence from a broader collection of “improved” biomass stove studies [21, 23, 24, 70, 71] suggesting that low use levels, modest emissions and exposure reductions from “improved” stoves, and contributions of other community-level sources limit the potential for these interventions to substantially reduce the health burden related to HAP. There are exceptions to this general trend, including the RESPIRE study mentioned in the introduction [20]. A key difference between our study and RESPIRE is that the latter used a chimney stove in Guatemala, which was not feasible given kitchen design and cooking patterns in Ghana (e.g., both indoor and outdoor cooking). Closer to our study context, the Ghana Randomized Air Pollution and Health Study (GRAPHS) examined impacts on CO exposure and systolic blood pressure among pregnant women following an intervention that distributed improved biomass stoves (arm 1) and liquefied petroleum gas (LPG) (arm 2) stoves to participants [72]. Reductions in CO exposure were observed in the LPG arm only, and blood pressure was lower but not to a statistically significant degree across both study arms. Birthweight was not significantly different across study arms [3]. These results highlight that study context, cooking patterns, and the specifics of the cookstove intervention are important determinants of program impacts, and also that multiple potential health pathways (e.g., impacts on respiratory and cardiovascular health) should be considered.

The REACCTING study has several key limitations. First, the study was underpowered to assess impacts on clinical health endpoints and did not include these measures, limiting our ability to definitively determine whether or not these stoves improved participants’ health. As discussed above, the self-reported health measures we included did not generate reliable data on respondents’ health outcomes, and both models of “improved” cookstoves we employed ultimately showed limited real-world improvements over traditional stove models. We are also careful to note that these study’s findings may not generalize to all cookstove interventions or settings. The study population was randomly selected from the rural, biomass-reliant households of Ghana’s Kassena-Nankana Districts, such that results may generalize to this specific population and other similar populations in this region. However, a wide variety of factors, from the types of fuels and cookstoves used to cooking tasks performed, will influence the outcomes of cooking interventions in different settings.

Recent enthusiasm for improved biomass cookstoves has been driven by their potential to provide win-win or triple-win scenarios, offering benefits for local users (health, time savings, development), local environments (reduced deforestation, local ambient air quality), and global climate [73, 74]. A growing number of studies, including this one, suggest that the human health impacts of these stoves should not be assumed. In light of this evidence, the health community and organizations like the Clean Cooking Alliance have shifted their focus to promotion of clean fuels like liquefied petroleum gas and electricity [75, 76], as well as advocating for community-level interventions to address both household-level and ambient air pollution sources [71].

Despite its limitations, we believe the REACCTING project provides a useful model for interdisciplinary, policy-relevant research in the household air pollution and household energy sectors. Households’ reliance on solid fuels to meet their basic energy needs continues to impose a significant health burden on populations around the world, along with environmental and social impacts. The evidence base to which this study contributes shows that simple solutions to this problem are elusive. Understanding and learning from both successes and failures requires expertise in a wide range of disciplines implicated in the causal pathway linking new technologies with desired impacts. Making progress towards transitions to cleaner household energy systems will require teams of social scientists, exposure scientists, environmental scientists, and health scientists working together, alongside affected community partners.