This study was conducted within the framework of a c-RCT involving 51 communities that used wood for cooking and heating in Peru, hereafter referred to as the parent study (Hartinger et al., 2011). The parent study aimed to evaluate reduction of childhood illness through reducing HAP and improving drinking water and kitchen hygiene conditions. The households in the intervention arm received an intervention package that included a new stove, a kitchen sink, and a solar disinfection home-based water treatment. The new stoves were built from red burnt bricks, plastered with a mixture of mud, straw and donkey manure; the stoves consisted of three pot-holes for cooking, a closed combustion chamber, a metal chimney with a regulatory valve, a hood, and metal rods for support (Hartinger et al., 2012). In the control arm, households used their existing stoves, most of which were traditional open-fire stoves. To reduce potential dropout and non-blinding bias in the control arm, households also received a psychomotor stimulation package focusing on early child development, a package that was unrelated to the environmental factors targeted in the study. The new stoves were installed in intervention households between September 2008 and January 2009. No exposure assessment was conducted before the installation of the intervention package in the parent study.

Starting February 2009, 503 households (250 and 253 in the intervention and control arms, respectively) entered the follow-up evaluation phase of the parent study (Hartinger et al., 2011). From June to August 2009, we conducted this cross-sectional study evaluating exposure to HAP (Commodore et al., 2013). Female members of the households (one per household) were eligible for this study if they met the following criteria: 1) were the mother or primary caregiver of the children enrolled in the parent study, 2) used an indoor wood stove, and, 3) agreed to participate in this study and agreed to comply with the project instructions during the 48-h sampling period. Eligible and enrolled participants provided a first-morning urine sample, a 48-h personal CO measurement, and filled out a questionnaire on demographics, smoking status, daily activities, household and community characteristics, and health symptoms, including headache, respiratory and eye-related symptoms. Although field workers visited all households in the parent study, subjects’ availability and willingness for participation, and time and budget constraints limited the total sample size of this study.

After the HAP exposure measurement, we classified post-hoc the intervention group into two sub-groups—“no-repair” sub-group with stoves in good running conditions at the time of the assessment, and, “need-repair” sub-group with stoves that needed repairs, e.g., re-plastering, filling small cracks, and chimney valve replacement. The control group was stratified into three sub-groups based on the type of wood-burning stoves: 1) traditional three-stone open-fire stoves and non-vented stoves (“traditional”), 2) chimney-stoves built by a non-governmental organization (“built-by-NGO”), and, 3) chimney-stoves built by the household members (“self-improved”). A flow diagram for this cross-sectional study is given in Supplemental Material, Fig. S1.

Time-integrated 48-h personal CO measurements were taken from the participants as described elsewhere (Commodore et al., 2013). In brief, the CO measurements were collected using passive CO diffusion tube, i.e., Dräger Diffusion Tube for Carbon Monoxide (Dräger Safety Inc., USA). The sampler uses principles of diffusion and colorimetry where CO passively diffuses into the tube and causes the reduction of sodium palladosulfite to palladium metal, which results in a grayish stain corresponding to a cumulative dose of CO. During the 48-h sampling period, the participants wore the passive CO diffusion samplers in the breathing zone. Field workers recorded the time of tube breakage and capping, which marked the beginning and ending of the CO sampling period, respectively. Upon return to the field station, tubes were read independently by two of the authors (AAC and SMH) and an arithmetic mean was taken. Additional information on the personal CO measurement is given in Supplemental Materials.

At the end of the 48-h personal CO sampling period, the participants collected a morning urine void between 5:00 am and 7:00 am in a pre-labeled sterile polyethylene container and placed the container in an insulated bag with ice packs. Field workers retrieved the urine samples from the participants, recorded the date and volume of the void, and delivered them to the study base, whereupon the samples were transferred into polypropylene tubes and stored at −20 °C until the end of the field work. Samples were then shipped frozen on dry ice to the University of Georgia and the Centers for Disease Control and Prevention (CDC) and stored at −70 °C until analysis. The study protocol was approved by the Human Research Protection Office at the CDC, the Human Subjects Division at University of Georgia, the Ethical Review Board of the Instituto de Investigacion Nutricional and the Universidad Peruana Cayetano Heredia in Lima. Written informed consent was obtained from all participants prior to enrollment in the study.

We analyzed the urine samples for 10 OH-PAHs, i.e. 1-, 2-naphthol, 2-, 3-, 9-hydroxyfluorene, 1-, 2-, 3-, 4-hydroxyphenanthrene and 1-hydroxypyrene. The detailed laboratory method was described elsewhere (Li et al., 2014). Briefly, urine samples (1 mL) were spiked with a ¹³C-labeled internal standard solution mixed with buffer and enzyme, and hydrolyzed overnight at 37 °C. The OH-PAHs were then extracted by a solvent mixture through semi-automated liquid-liquid extraction. The extracts were evaporated, derivatized, and analyzed by isotope dilution gas chromatography triple quadrupole tandem mass spectrometry. All analyses were subjected to a series of quality control and quality assurance checks as described previously (Li et al., 2014). The limits of detection (LODs) were 0.001–0.019 μg/L, depending on the analyte, and the detection frequency was 100% for all 10 OH-PAHs in this study. The coefficient of variation from 36 quality control samples analyzed with the study sample ranged 1.8–7.3% depending on the analyte. Urinary creatinine was measured according to Roche’s Creatinine Plus Product Application # 03,631,761,003.

All urinary OH-PAH concentrations were blank-subtracted. 1-Naph-thol results in two participants in the control group were non-reportable because of chromatographic interference on the ¹³C-labeled internal standard peak. Creatinine adjustment was made to correct for urine dilution that is known to vary with the hydration status of the individual (Barr et al., 2005). Two urine samples did not have creatinine results and therefore, did not have creatinine-adjusted OH-PAH results.

Statistical analyses were performed using SAS 9.3 software. The Shapiro-Wilk test, histograms and Q-Q plots indicated that OH-PAH concentrations were log-normally distributed. Therefore, creatinine-adjusted OH-PAH concentrations were log-transformed before all statistical analyses. Based on influence statistics and corresponding plots of OH-PAH concentrations, we determined that one participant in the control group was an extreme outlier and excluded the participant in subsequent analyses. Least square geometric means of PAH biomarker by stove type were calculated, controlling for second-hand smoking status, distance between home and road, community traffic, and recent consumption of food cooked directly on open fire (hereafter referred to as “grilled food”). We used t-tests to compare geometric means (GMs) of OH-PAH concentrations by stove type. Both un-adjusted and least square GMs gave similar results. Hence, only un-adjusted GMs are presented here. We calculated Pearson correlation coefficients to assess correlation between urinary OH-PAHs and personal CO levels (log-transformed). Chi-square test and Fisher’s exact test were used to test for differences on the prevalence of health symptoms in the control and intervention groups, and both tests gave the same results regarding statistical significance. Results are considered statistically significant at p < 0.05.

Logistic regression was used to model the association between self-reported health symptoms and OH-PAHs as HAP exposure biomarkers. Health symptoms studied included several eye-related symptoms, several respiratory symptoms, and headache. A stepwise selection procedure was used in which the entry and removal p-values were 0.10. Final model controlled for subject’s age, stove type, CO, wood type, second-hand smoke, distance between home and road, community traffic, pesticide use, fertilizer use, and kitchen type. Urinary OH-PAHs were forced into the model. Sums of fluorene metabolites (summed concentration of 2-, 3- and 9-hydroxyfluorene) and phenanthrene metabolites (summed concentration of 1-, 2-, 3- and 4-hydroxyphenanthrene) were used in place of individual components to reflect exposure to fluorene and phenanthrene, respectively.

Among all urinary OH-PAHs, 2-naphthol, a proposed biomarker for inhalational exposure (Kang et al., 2002; Li et al., 2016), had the largest reduction in the intervention group compared to the control group (p < 0.001). Moreover, the extent of reduction increased by 42% when comparing the two sub-groups with presumably the most and least HAP exposure, i.e. traditional stove users within the control group vs. no-repair stove users within the intervention group. The findings are encouraging, especially considering the relatively small sample size, the cross-sectional study design, and considerably large within- and between-person variabilities reported for urinary metabolites (Li et al., 2010b; Siwinska et al., 1998). Urinary naphthols (1- and 2-naphthols) had been reported to give better selectivity and sensitivity for route-specific inhalation PAH exposure than 1-hydroxypyrene (Jansen et al., 1996; Yang et al., 1999). While 1-naphthol can result from naphthalene exposure, it is also a major metabolite of the wide-spectrum carbamate insecticide (Maroni et al., 2000). In contrast, 2-naphthol only results from naphthalene exposure. Additionally, 2-naphthol has shown the expected rise-fall excretion pattern and demonstrated the largest increase compared to other OH-PAHs after a controlled low inhalation exposure (Li et al., 2016). Therefore, 2-naphthol has been suggested as a more suitable biomarker for inhalational PAH exposure than 1-hydroxypyrene and other OH-PAHs (Li et al., 2016; Preuss et al., 2003; Yang et al., 1999). Hence, it is not surprising that 2-naphthol showed the largest difference between control and intervention in this study on HAP exposure. In another stove intervention project in Peru, 2-naphthol also reached a larger reduction than other PAH metabolites three weeks after installing new chimney-equipped stoves (Li et al., 2011).

For the remaining OH-PAHs, there was no significant difference between control and intervention groups, which is consistent with the personal CO results from these participants reported previously (Commodore et al., 2013). Among the 313 participants with both urine and 48-h personal CO measurements, urinary OH-PAHs were significantly associated with the CO. This is consistent with the fact that both PAHs and CO are components in biomass smoke from woodstoves. The weak correlations between urine OH-PAHs and CO can be explained by several factors. The urinary PAH metabolites reflect mostly exposure occurred within a day, due to their short excretion half-lives (4.0–23.5 h) after inhalation exposure (Brzeznicki et al., 1997; Lafontaine et al., 2000; Li et al., 2016; St Helen et al., 2012). In contrast, the CO data in the study were collected over the 48-h period before the urine sampling. In addition, while CO is in the gaseous form entirely, airborne PAHs exist in both gaseous and particle phases. Finally, urinary PAH metabolites reflect PAH exposure from all routes including inhalation, diet and dermal absorption.

The differences in urinary OH-PAHs between the control and intervention in this study are less than those in other stove studies that used PAH metabolites as HAP exposure biomarkers. In an un-related stove intervention study in Peru, the same suite of urinary OH-PAHs were significantly reduced by 19–52% in 57 women three weeks after the installation of new stove to replace the open-fire stoves (Li et al., 2011). In a randomized control trial in Mexico, these 10 biomarkers were reduced by 20–48% post-intervention in 47 women, compared to pre-intervention (Riojas-Rodriguez et al., 2011). A three-stage stove intervention program (removing indoor soot, paving dirt floors, and installing new stoves with chimney) in Mexico found a 29% reduction of 1-hydroxypyrene in 20 residents one month after intervention (Torres-Dosal et al., 2008).

Several factors could lead to the smaller observed reduction in our studies compared to other studies. First and foremost, we did not conduct exposure assessment before the installation of the new stove, and therefore, did not have pre- and post-measurements (self-control) that were available in the above-mentioned studies. Hence, it is not surprising that we observed less differences between intervention and control groups in this cross-sectional study. In addition, our investigation was conducted 6–8 months after the new stoves were installed, compared to few weeks in other studies. The longer lag between intervention and follow-up would likely diminish the observed difference between new and old stoves due to the wear and tear of the new stoves; however, the longer lag can provide valuable information on the long-term effectiveness of the intervention. As shown in this study, durability of the new stoves may be of concern given that over a third of the new stoves were in need of repair at the time of the study.

To the best of our knowledge, this is the first study that investigated the relationship between urinary PAH metabolites (as HAP exposure biomarkers), potential risk factors and self-reported respiratory and eye symptoms in a non-smoking population. We found reduced headache and eye-related symptoms in the intervention group compared to the control group. Stove type (control vs. intervention) and/or 2-naphthol were positively associated with eye symptoms and headache. This is consistent with previous reports that HAP exposure is significantly associated with eye irritations (West et al., 2013). Our finding also demonstrates that the ocular symptoms may be reduced or eliminated through the use of improved stoves.

We did not find statistically significant associations between urinary exposure biomarkers and/or stove type with respiratory symptoms or illnesses. Potential explanations include small number of participants reporting respiratory symptoms or illnesses (Table 3) and possibly longer latency period for respiratory symptoms than eye irritations. As expected, several risk factors were significantly associated with health symptoms. For example, participant’s age was positively associated with red eyes, humid eyes, achy eyes while cooking, and headache. Fertilizer and/or pesticide use were positively associated with a number of respiratory and eye symptoms, potentially due to irritations from fertilizer/pesticide application and usage.

Surprisingly, 1-hydroxypyrene, the most used biomarker for PAH exposure, was negatively associated with cough while cooking. It has been reported that in general populations with no occupational exposure, pyrene is mainly taken up through diet (Jansen et al., 1996; Li et al., 2010a), and urinary 1-hydroxypyrene can act as a route specific biomarker for oral or dermal exposure (Jansen et al., 1996). In this study on inhalational HAP exposure, we did not observe a difference on 1-hydroxypyrene levels between the intervention and control groups, and found opposite effect on association with certain respiratory systems, which suggested that non-inhalational route, e.g., touching surfaces with soot deposit and eating burnt food or food cooked in pots with soot deposit, may be important additional factors affecting 1-hydroxypyrene levels.

Our findings on HAP exposure and sensory irritative symptoms are generally consistent with other stove intervention studies that investigated the impact of improved stoves on women’s respiratory and/or eye health, although those studies used other surrogates to characterize HAP exposure, such as stove type or PM_2.5/CO. In a study titled “RESPIRE” in Guatemala, replacing traditional three-stone stoves with chimney-equipped stoves significantly reduced sore eyes and headache 6–18 months after the intervention (Diaz et al., 2007). The study also found reductions of the risk of wheeze and total number of respiratory symptoms, but did not observe significant effects on lung function within a 1.5-year follow-up period (Smith-Sivertsen et al., 2009). In a follow-up study within the RESPIRE cohort, exhaled CO and/or personal CO was significantly associated with several respiratory symptoms, such as wheeze and chronic phlegm (Pope et al., 2015). In a randomized control trial in Mexico, women who used new chimney-equipped stoves most of time had significantly lower risk of respiratory symptoms, eye discomfort, and headache; the use of new stoves was significantly associated with improved lung function comparable to smoking cessation (Romieu et al., 2009). In a cross-sectional survey among 79 Honduran women, users of improved stoves had lower levels of PM_2.5 and CO and lower prevalence of self-reported respiratory symptoms compared to open-fire stove users, but no association was found between stove type or air measurements with lung function or C-reactive protein (Clark et al., 2009). Our study findings demonstrate the utility of urinary PAH metabolites as effective exposure biomarkers for evaluation and monitoring purposes in stove intervention projects.

The urinary PAH levels in this study far exceeded the general population levels in national surveys in the U.S. (CDC, 2015), Canada (Health Canada, 2015), and Germany (Becker et al., 2003), c.f. Supplemental Material Table S4. As shown in Fig. 2, median concentrations of the 10 OH-PAHs were equivalent or several times higher than the 95th percentile in the U.S. adults reported in NHANES 2011–2012 (CDC, 2015).

Compared to other populations’ exposure to woodsmoke (Table 4), the 1-hydroxypyrene level in our study (median: 2.63 μg/g creatinine) was consistent with those in HAP-exposed populations in Afghanistan (Hemat et al., 2012), Burundi (Viau et al., 2000), Poland (Siwinska et al., 1999), and Mexico (Pruneda-Alvarez et al., 2012), but was several times higher than several Mexican studies on non-smoking women exposed to HAP (Pruneda-Alvarez et al., 2016; Ruiz-Vera et al., 2014, 2015). Compared to occupational exposure to biomass smoke, the 1-hydroxypyrene level in our study was similar to that of workers exposed to rubber woodsmoke in Thailand (Choosong et al., 2014), and was 10 times higher than among charcoal workers in Brazil (Kato et al., 2004).

Other OH-PAHs were rarely reported in published woodsmoke exposure studies (Supplemental Materials, Table S4). The levels of the 9 other OH-PAHs in this study were similar to another Peruvian stove intervention study (Li et al., 2011). Summed phenanthrene metabolite concentration in this study was consistent with that in a study in Afghanistan (Hemat et al., 2012). However, 2-naphthol in this study (GM: 14.9 μg/g creatinine) was higher than a study on charcoal worker (GM: 9.14 μg/g creatinine) with occupational exposure to woodsmoke (Kato et al., 2004).

These comparisons suggested that people exposed to HAP from indoor woodstoves are among the populations with the highest non-occupational exposure to PAHs (Martinez-Salinas et al., 2010; Pruneda-Alvarez et al., 2016), a group of carcinogenic and mutagenic pollutants formed from incomplete combustions. Furthermore, the median 1-hydroxypyrene in this study was higher than 1.93 μg/g creatinine (1.0 μmol/mol creatinine), a proposed no-observed-genotoxic-effect-level for occupational exposure to PAHs (Jongeneelen, 2014), which indicates an increased risk of cancer among the study subjects.

Several factors may explain the high HAP exposure in this study. First, the chimney-equipped stoves generally do not completely release woodsmoke outside of the house/kitchen, do not improve the combustion efficiency and thus cannot reduce the formation of harmful air pollutants such as PAHs. Second, 36% of the new stoves in this study required some repair at the time of the study (6–8 months after installation), which could affect the effectiveness of exposure reduction. The new stoves were built from materials such as mud, straw and donkey manure. These materials, although readily available, may not be durable for long-term usage. Third, some participants in the intervention group may have used both the new and old stoves at the same time, a practice called “stove-stacking” (Thomas et al., 2013). Although we did not assess the stove usage and adherence in this study, a subsequent study conducted in the same area—including the households in this study—revealed that 32.6% used a traditional stove as the secondary stove (unpublished data). Stove-stacking and adherence issues have also been noted in other stove projects (Romieu et al., 2009; Thomas et al., 2013). Many factors, such as cultural and personal cooking habits, needs and preferences, user training and support, could adversely affect adopting of the new stoves (Lewis and Pattanayak, 2012; Rehfuess et al., 2014) and reduce the benefit of stove-intervention programs. Fourth, although the stoves were replaced, the surfaces in the house may still be contaminated with soot known to contain PAHs, which would lead to continued dermal PAH exposure. Lastly, most study households had permeable roofs and open windows. Therefore, biomass smoke from the chimney and from neighbors could enter or re-enter the kitchens in the intervention households.

There are several limitations in the current study. First, exposure assessment was not conducted before the new stove installation in the intervention households. Pre- and post-intervention assessments would allow for stronger evidence than the cross-sectional data on the effectiveness of the intervention. Second, the current study did not have data to describe stove-stacking and could not assess factors and barriers affecting the adoption of the new stoves. Third, potential recall bias could occur in participants’ self-reporting of health symptoms. Lastly, the study had a small number of participants and an even smaller number of women reporting symptoms. Larger studies with more participants would increase the power of the analyses.