Introduction

Biomass fuels (wood, charcoal, crop residues, and dung) and coal are the primary source of domestic energy for one-half of the world's population (1). Combustion of biomass and coal, especially in open or poorly ventilated stoves, emits hundreds of health damaging pollutants (2) that cause a number of diseases (1, 3, 4). Indoor air pollution (IAP) from household use of solid fuels resulted in more than 1.6 million deaths and nearly 3% of the global burden of disease in 2000 (1, 5).

IAP exposure has multiple technological (e.g., fuel and stove), environmental (e.g., housing characteristics), and behavioral (e.g., energy use behavior and time-location-activity budgets) determinants. Although most indoor air pollutants are products of incomplete combustion, combus tion conditions such as temperature, moisture, and air flow are likely to affect the emissions of the various pollutants differently. The spatial dispersion of pollutants inside the house may also vary, depending on whether they are particles or gases and the physical characteristics of the house (e.g., whether the house consists of a single large room or multiple rooms separated by walls and doors). Finally, there may be day-to-day or longer term (e.g., seasonal) changes in emis sions of any pollutant or pollutant groups because of changes in energy use patterns or housing conditions.

A number of studies have examined temporal and spatial patterns of IAP, or the relationship between multiple pol lutants (6-13), but none has considered the combinations of spatial, geographical, day-to-day, and seasonal variability for multiple pollutants using comparable measurement methods. An understanding of the patterns of exposure to different mixtures of pollutants is required to design and evaluate interventions. In this paper, we use data collected in mid- and late-heating season (December and March) in multiple households in four poor provinces in China to assess the spatial and temporal distributions of three important indoor air pollutants (respirable particles, carbon monoxide, and sulfur dioxide). The results are being used to generate a baseline profile for the patterns and determinants of exposure to indoor air pollution from household energy use and to design more effective intervention technologies in the world's most populous nation. The implications of findings for interventions are discussed.

Study Location

More than 70% of China's households rely on solid fuels (biomass and coal) for their domestic energy (1, 14). Indoor air pollution caused an estimated 500 000 annual deaths in the developing countries of the Western Pacific region (approximately 85% of the region's population lives in China) in the year 2000, making IAP the fourth leading cause of regional mortality and fifth leading cause of regional disease burden (1, 5). The diversity of factors that affect multiple indoor air pollutants is particularly relevant in China. These factors include climate and geography, housing, locally available fuels and stoves, and socio-cultural factors such as income, food types, and food preparation. Although until the 1980s and 1990s biomass was the dominant source of household energy in China, economic development and deforestation-and policies to reduce and reverse it-have compelled many rural residents to switch to coal (14, 15). China has also implemented an ambitious program to disseminate improved (high-efficiency and low-emissions) stoves (15, 16). A recent evaluation of the program has illustrated that the design and performance of the new stoves was highly variable, with many of the stoves labeled as "improved" lacking flues or other characteristics necessary for reducing exposure (15). The study took place in Gansu, Guizhou, Shaanxi, and Inner Mongolia provinces of China. Heating season in all provinces lasts approximately from early November until late March. Winter temperatures are, however, lower in northern, colder provinces and the "intensity" of home heating is higher, particularly in Inner Mongolia; or windows may be open in Guizhou during the heating season but they are closed and sealed in Inner Mongolia. Detailed information on the study provinces, and on selection of study households, is provided in Supporting Information.

Housing. Table 1 provides basic information on housing in the study households. Most houses in Gansu province have a kitchen separated by a wall from the sleeping/living room, with separate entrances. In Guizhou most houses have 2-3 rooms-cooking/living, sleeping, and entrance/storage-connected with doors. Most houses have a separate cooking area, but cooking is done almost entirely in one of the main rooms (cooking/living room), especially during the heating season; the separate cooking area is used primarily for large family events (e.g., Spring Festival) and for making animal feed. Houses in Guizhou have an attic above the cooking/living and sleeping rooms, used for food drying and storage, and at times containing an additional bed. The lower rooms and the attic are separated by a porous ceiling (e.g., made of pieces of wood) which allows air flow between the two levels. Most houses in Shaanxi province have a cooking area connected to the main house by a door, a living room with a ground stove (fire-pit), and one bedroom which sometimes also has a ground stove. Older homes in Inner Mongolia are constructed inside a cavelike structure with a single room used for cooking, living, and sleeping. This room contains a "bed-stove" configuration (a cooking stove connected to a bed for heating). Some homes have an additional room used for sleeping or storage. Newer homes are made with modern construction material with similar structure. The newest homes in the study area have a wall with windows and door between the cooking and sleeping/living areas.

Energy Technology. Table 2 shows the summary of fuel and stove use in the study households. Multifuel use and multistove use are common features of household energy in China because of fuel availability and multiple uses of energy (e.g., cooking and heating) (see also ref 15). In particular, although coal is the nearly universal fuel for heating in Guizhou and Shaanxi, 18% and 52% of study households in the two provinces, respectively, used biomass as their main cooking fuel.

Methods

Measurement Locations and Times. For each province and for each pollutant, Table 3 shows the number of households and measurement days, measurement points, and time periods. All measurements took place by teams consisting of investigators from the Chinese National Center for Disease Control and Prevention (CDC) assisted by provincial and county CDC staff and health workers. On each day, measurements for all pollutants began in mid- or late-morning and continued until the next day, as close to a full 24-h period as possible.

Measurements took place in two different time periods: March through early April 2003 (late part of the heating season) and December 2003 through January 2004 (peak of the heating season); we refer to these two measurement periods as March and December throughout the paper. Therefore, the results of this study illustrate exposure during the heating season; exposure during summer months is expected to be substantially lower because less energy is used and because windows may be open more frequently. The households monitored in December were a subset of those from March. Repeated measurements were conducted in a small number of households to examine day-to-day variability of pollution.

Due to the cost of measurements, an optimal combination of pollutants and measurement locations was selected to best characterize the exposure conditions of each province, based on a pilot study in January 2003 which examined the relationship between different pollutants and different measurement points (17) and a survey on household energy use behaviors and time-activity budgets (18). On the basis of the pilot study, pollutants were measured at 2-3 points in the cooking, living, and sleeping areas which are the main exposure microenvironments. In a small number of homes and for selected pollutants, measurements were made at additional points for comparison.

Respirable Particles (RPM). Respirable particles were measured according to The National Institute for Oc cupational Safety and Health, NIOSH, protocol 0600, designed to capture particles with a median aerodynamic diameter of 4 m (PM₄) (19). Samples were collected using a 10-mm nylon cyclone equipped with a 37-mm diameter poly(vinyl chloride) (PVC) filter (pore size 5m supplied by SKC Inc., U.S.A.) at a flow rate of 2.5 l/min. Air was drawn through the cyclone preselectors using battery-operated constant flow pumps (model PCXR8 supplied by SKC Inc., U.S.A.). All pumps were calibrated prior to and after each sampling day using a field minimeter, itself calibrated by a soap bubble meter in laboratory. Pumps were also calibrated in the laboratory after each field exercise using the same minimeter. To maintain battery power throughout the sampling period, pumps were programmed to cover the 24-h interval through intermittent sampling (1 min out of every 4-6 min). One field blank was taken on each sampling day.

Gravimetric analyses were conducted at the laboratory of the National Institute for Environmental Health and Related Products Safety, China CDC using an analytic microbalance (1/100,000, Sartorius 2004 MP, Germany) calibrated against standards provided by the Bureau of National Technological Control. All filters (field blanks and samples) were conditioned for 24 h before weighing. Respirable dust concentrations were calculated by dividing the blank-corrected increase in filter mass by the total air volume sampled.

Carbon Monoxide and Sulfur Dioxide. Carbon monoxide (CO) and sulfur dioxide (SO₂) were measured using long-term diffusion tubes (manufactured by GASTEC, U.S.A.), with detection ranges of 10-200 or 50-1000 ppm for CO and 2-100 ppm for SO₂.

Results and Discussion

Figures 1-3 show the distributions and descriptive statistics for the measured pollutants by measurement point, measurement period, and province. [For Figures 1-3, note that if the concentrations of CO and SO₂ were outside the detection range of the tubes (10-200 or 50-1000 ppm for CO and 2-100 ppm for SO₂), the following assumptions were made: (i) Those measurements that were only slightly higher than the measurement range were set to the maximum value. The number of measurements was (a) Gansu (March), 20 measurements for CO (10 in the cooking room and 10 in the living/bedroom); Gansu (December), 6 measurements for CO (3 in the cooking room and 3 in the living/bedroom); (b) Guizhou (March), 0 measurements for CO and 0 measure ments for SO₂; Guizhou (December), 0 measurements for SO₂; (c) Shaanxi (March), 5 measurements for CO (0 in the cooking room, 3 in the living room and 2 in the bedroom); Shaanxi (December), 30 measurements for CO (7 in the cooking room, 14 in the living room and 9 in the bedroom); (d) Inner Mongolia (December), 27 measurements for CO (13 in the cooking/living/bedroom point 1 and 14 in the cooking/living/bedroom point 2). (ii) Those measurements that were substantially higher than the measurement range were set to 150% of the maximum value. The number of measurements was (a) Gansu (March), 16 measurements for CO (7 in the cooking room and 9 in the living/bedroom); Gansu (December), 4 measurements for CO (1 in the cooking room and 3 in the living/bedroom); (b) Guizhou (March), 0 measurements for CO and 0 measurements for SO₂; (c) Shaanxi (December), 7 measurements for CO (1 in the cooking room, 2 in the living room and 4 in the bedroom); (d) Inner Mongolia (December), 19 measurements for CO (11 in the cooking/living/bedroom point 1 and 8 in the cooking/living/bedroom point 2). (iii) Those measurements that were lower than the measurement range were set to the minimum value. The number of measurements was (a) Gansu (March), 13 measurements for SO₂ (9 in the cooking room and 4 in the living/bedroom); (b) Shaanxi (March), 3 measurements for CO (1 in the cooking room, 1 in the living room and 1 in the bedroom) and 86 measurements for SO₂ (34 in the cooking room, 1 in the living room and 51 in the bedroom); Shaanxi (December), 21 measurements for SO₂ (11 in the cooking room, 4 in the living room and 6 in the bedroom). The overall results and conclusions of the analysis were not sensitive to this assumption.]

	Figure 1 Average 24-h concentration of RPM at different measurement points and measurement periods (March and December) in Gansu, Guizhou, Shaanxi, and Inner Mongolia provinces. n: number of observations. : mean. Numbers in brackets give the 95% CI for the mean.
	Figure 2 Average 24-h concentration of CO at different measurement points and measurement periods (March and December) in Gansu, Guizhou, Shaanxi, and Inner Mongolia provinces. n: number of observations. : mean. Numbers in brackets give the 95% CI for the mean.
	Figure 3 Average 24-h concentration of SO2 at different measurement points and measurement periods (March and December) in Gansu, Guizhou, Shaanxi, and Inner Mongolia provinces. n: number of observations. : mean. Numbers in brackets give the 95% CI for the mean.

Pollutant Concentrations in Different Provinces. RPM concentrations in all provinces were in excess of health-based standards and guidelines for particulate matter in ambient (outdoor) environment (e.g., the U.S. EPA requires the 24-h mean concentration of PM_2.5 to be below 65 g/m³ and annual mean concentration below 15 g/m³). The two provinces where biomass is the primary fuel (Inner Mongolia and Gansu) had the highest RPM concentrations (>700 g/m³ at both points in the cooking/living/bedroom in Inner Mongolia; 351-661 g/m³ in different rooms and months in Gansu); the higher concentration in Inner Mongolia reflect its colder temperature and longer heating hours, with a stove and housing arrangement that combines heating and cook ing. Lower concentrations were observed in the primarily coal-burning provinces of Guizhou and Shaanxi (202-352 g/m³ in different rooms and months in Guizhou; 187-361 g/m³ in different rooms and months in Shaanxi). RPM concentrations in Guizhou were significantly lower than those measured in the pilot study (17) in four homes in a single village. This difference is partly because the pilot measure ments were intended for selecting appropriate sampling points and pollutants for the larger study, rather than being from a representative sample households.

Except for a small number of observations in Gansu, Inner Mongolia, and Shaanxi, 24-h mean CO concentrations were consistently below available health-based standards and guidelines (e.g., WHO guideline values of 10 ppm and American Conference of Governmental Industrial Hygienists, ACGIH, guideline value of 25 ppm for 8-h exposures), and in some cases CO concentrations were close to the detection limits of the diffusion tubes. Since these were 24-h concentrations, concentrations may have been higher during cooking or when doors and windows were closed at night, not observable in our data. Guizhou had the lowest CO concentrations (<2 ppm), partly due to the specific type of coal and stove, and partly because the attic, where the chimney ends, provides ventilation for removing CO from the house (the chimney and the porous attic provide better ventilation for CO which is gaseous compared to particles). Mean 24-h concentrations were 3.7-13.8 ppm at difference points and months in the attic.

Due to cost, we measured SO₂ only in the two coal-burning provinces of Guizhou and Shaanxi, where detectable concentrations were expected based on the result of the pilot study (17) (a small number of measurements in Gansu showed significantly lower concentrations of SO₂; in Inner Mongolia SO₂ concentrations in a small number of measurements were comparable to those in Shaanxi and Guizhou because coal was used for heating at night). SO₂ concentra tions were higher than the WHO guideline value of 0.04 ppm at all locations in both provinces. SO₂ concentrations were substantially higher in Shaanxi than the corresponding points in Guizhou (e.g., the area near the coal-burning ground stove in the heated living room in Shaanxi had a mean 24-h concentration of 0.97-1.44 ppm compared with 0.16-0.20 ppm in the cooking/living room in Guizhou). Higher SO₂ concentrations in Shaanxi are likely due to both the type of coal and the use of a chimney in Guizhou (mean 24-h SO₂ concentrations in the attic in Guizhou were higher than those in Shaanxi, 1.08-3.40 ppm at difference points and months).

Pollutant Concentrations at Different Measurement Points. The concentrations of pollutants in different rooms were generally determined by whether a stove was used in that room for cooking versus heating and by housing characteristics that affect dispersion. The role of these factors also varied across pollutants. In Gansu, the cooking room had a higher RPM concentration than the bedroom (518 vs 351 g/m³ in March and 661 vs 457 g/m³ in December; the differences between the two points were not significant at p = 0.05). CO concentrations, however, did not show the same consistent ordering, and the relative levels in the two rooms changed in March and December. If outlier measure ments (>99th percentile) are dropped, the mean CO concentrations in March were 5.5 and 4.8 ppm in the cooking room and bedroom, respectively; in December they were 8.4 and 9.3 ppm. The reason for this differential pattern of the two pollutants may be that shorter periods of more intense combustion, intermittent with longer periods of no combus tion, in the cooking room result in similarly high emissions of RPM and CO. In the bedroom, the heated bed contains hot ashes for many hours with closed windows, and RPM emissions are lowered more than CO. As a result RPM concentrations are consistently higher in the cooking room than the bedroom, but those of CO are comparable between the two rooms.

In Guizhou, the concentrations of all pollutants in the bedroom-where there was no stove and which was con nected by a door to the cooking/living room-were consistently similar to, or only slightly lower than, the cooking/living room where the stove was located. This result illustrates that direct dispersion inside the house, and probably more importantly the transport of pollutants through the chimney and subsequent dispersion via the attic, make the bedroom an important exposure microenvironment. The small number of measurements in the attic shows that it consistently had the highest concentrations of all three pollutants (RPM, 453-649 g/m³; CO, 3.7-13.8 ppm; SO₂, 1.08-3.40 ppm in different months). Therefore, although the attic, where the chimneys ended, is not directly an important exposure microenviron ment because of the time-location-activity patterns of household members (i.e., despite its high concentrations, little time is spent there), its porous floor creates an important role in dispersion of pollutants into the rooms on the main floor.

In Inner Mongolia, cooking and heating take place in the same room, making it the main exposure microenvironment. In Shaanxi, the cooking room, the heated living room, and the bedroom all had relatively similar RPM concentrations. But the concentrations of CO and SO₂ were highest in the heated living room. The reason for the differential patterns is that biomass was the primary cooking fuel, and was used for a shorter time, resulting in high RPM concentration but limited contribution to CO and SO₂. In the heated living room on the other hand, coal with a high sulfur concentration was used for longer durations, therefore resulting in higher emissions of CO and SO₂. Pollution in the bedroom was determined by a combination of direct emission and disper sion from other locations (a stove was used in 9 out of 24 household-days of measurement in the bedroom in December; 89 out of 98 household-days of measurement in the bedroom in March) and was therefore highly dependent on household-specific characteristics. The high concentrations in the heated living room, and the nonnegligible levels in the bedroom, illustrate the important role of heating as a source of exposure in winter.

Pollutant Concentrations in Mid- and Late-Heating Season. The data in this study did not provide evidence on differences between pollution levels in mid- and late-heating seasons (December vs March) that are generalizable across provinces. In Gansu and Shaanxi, the concentrations of all measured pollutants were higher in December than in March, at all measurement points (Figures 1-3) (many of the differences were not statistically significant at p = 0.05). In Guizhou, RPM concentrations were nearly equal in December and March (352 vs 301 g/m³) in the cooking/living room; bedroom concentrations in December were substantially, and significantly, lower than March (202 vs 315 g/m³) (without outliers (>99th percentile) the concentrations in the two seasons would be 202 vs 256 g/m³). Concentrations of CO and SO₂ were however higher in December than in March in both the cooking/living and bedroom. The more consistent pattern of mid- and late-heating season in Gansu and Shaanxi may be because in both provinces energy is used for home heating in winter. Although heating is also important in Guizhou, winter in Guizhou is characterized with high humidity, creating a need for keeping stored food dry (17, 18). Therefore, warmer temperatures in March may have systematically resulted in lower energy use (and pollution) in the living rooms in Shaanxi and Gansu but not those in Guizhou (Table 4; the differences are not statistically significant). Because of the logistics of the project initiation, data collection in Inner Mongolia began only in December 2003, and hence no similar measurements from March were available for comparison.

Pollutant Concentrations across Multiple Measurement Days. We examined day-to-day variation in pollution using data from those households with multiple measurements in each province and in each monitoring period (Table 3). For each province and each measurement point, Table 5 shows minimum and maximum concentrations in the multiple measurement days, calculated separately for each household with multiple measurements and then averaged over all such households. Table 5 also shows the coefficient of variation (defined as standard deviation divided by mean) for multiple measurements in each household, calculated separately for each household and then averaged over all such households. Table 5 shows that, on average, pollutant concentrations varied by a factor of 2-10 across measurement days, for different pollutants, measurement points, and provinces. Standard deviations of multiple measurements in the same household varied between 10% and 100% of their mean. With the exception of SO₂ in Shaanxi, the variation was consistently smaller in December than in March, possibly because longer hours of stove use in December reduced heterogeneity across different measurement days.

Because of measurement difficulty and cost, most epidemiological studies rely on indirect indicators of exposure (e.g., fuel, stove, energy use behaviors, and time-activity budgets) or a small number of actual measurements of pollution and/or exposure. It is therefore important to compare the variability of pollution across households (i.e., interhousehold variation) with variability across different measurement days (interday variation) in the same household. For households with multiple measurements, Table 6 shows the ratio of proportion of variance explained by interhousehold variation to interday variability. In all cases, except CO in the cooking/living room in Guizhou in March, there was more variation in pollution across households than within households. This result indicates that the duration of stove use, quantity of fuel, ventilation, and stove use behavior that determine pollutant concentrations are likely to vary more across households than from day to day in individual households.

Implications for Interventions. In China, where rapid economic growth and infrastructure expansion have contributed to near-universal access to electricity for lighting (20), more than 70% of households continue to use coal or biomass as their primarily fuel for cooking and heating (1). Because rapid transition to clean fuels is not a feasible short-term option, there is a need for interventions that lower emissions by modifying current fuel-stove-housing-behavior combinations. Understanding exposure routes will inform the design of more effective interventions. Some of the main exposure paths in these four Chinese provinces are summarized below:

In households with separate cooking and living/sleeping areas, and distinct cooking and heating stoves (most households in Gansu and Shaanxi), the cooking stove is a source of exposure throughout the year for women, and their young children, who may spend time near their mothers. The duration of exposure from cooking is a few hours per day, and stove improvement alone can reduce exposure. Even in these households, smaller cooking tasks (e.g., making tea) take place in the living area, for example on a fire pan or ground stove, which results in IAP exposure for all household members. In these households heating is a source of exposure, possibly to a larger extent than cooking, for all household members in winter. Therefore, reducing exposure requires improvements in the heating stove above and beyond that used for cooking.

In households where cooking and living areas are the same, and there is no physical distinction between cooking and heating stoves (most households in Inner Mongolia), cooking is a source of exposure throughout the year for all household members. In winter, heating requires stove use, and hence causes exposure, for a period as long as or longer than cooking. In principle, stove improvement can reduce indoor concentrations and exposure. Stove improvements may also have to be accompanied with housing changes that separate the main stove body from the living area (i.e., create a specialized kitchen). With this alternative housing design, the stove's heating function is performed only by heat transfer to the bed.

In most households in Guizhou, a separate cooking area and stove exist, but it is used only occasionally. Most cooking takes place in the living area, combined with heating in winter. These households in practice have the same exposure patterns as those with combined cooking/living areas and cooking/heating stoves. The stove in Guizhou can be improved without housing changes. Because the porous ceiling between the ground floor and the attic allows pollutant dispersion back into the main floor, stove improvements must be accompanied with increased chimney length to limit the dispersion of pollutants inside the house and reduce exposure.

Acknowledgment

We thank the households who participated in the study for their help and hospitality. The staff of the Health Department and the Centers for Disease Control and Prevention in Gansu, Guizhou, Inner Mongolia, and Shaanxi provinces provided tremendous assistance in the field study design and data collection. Funding and technical assistance for this work was provided by the World Bank through financial support from the Energy Sector Management Assistance Program (ESMAP), Department for International Development, U.K. (DFID) and Swedish International Development Agency (SIDA). M. Ezzati was supported by the World Bank and the National Institutes of Health (Grant PO1-AG17625). The findings, interpretations, and conclusions expressed here are those of the authors and do not necessarily reflect the views of the Board of Executive Directors of the World Bank or the governments they represent.