Women and girls in resource poor countries experience much greater exposure to household air pollutants than men: results from Uganda and Ethiopia

23 Household Air Pollution (HAP) from burning biomass fuels is a major cause of mortality and 24 morbidity in low-income settings worldwide. Little is known about the differences in 25 objective personal HAP exposure by age and gender. 26 We measured personal exposure to HAP across six groups defined by age and gender (young 27 children, young males, young females, adult males, adult females, and elderly) in rural 28 households in two sub-Saharan African countries. 29 Data on 24-hour personal exposure to HAP were collected from 215 participants from 85 30 households in Uganda and Ethiopia. HAP exposure was assessed by measuring carbon 31 monoxide (CO) and/or fine particulate matter (PM2.5) concentrations using five types of 32


Abstract 23
Household Air Pollution (HAP) from burning biomass fuels is a major cause of mortality and 24 morbidity in low-income settings worldwide. Little is known about the differences in 25 objective personal HAP exposure by age and gender. 26 We measured personal exposure to HAP across six groups defined by age and gender (young 27 children, young males, young females, adult males, adult females, and elderly) in rural 28 households in two sub-Saharan African countries. There are substantial differences in exposure to HAP depending on age and gender in sub-42 Saharan Africa rural households reflecting differences in household cooking activity and time 43 spent indoors. Future work should consider these differences when implementing exposure 44 reduction interventions. There was a strong agreement between optical and gravimetric 45 devices measurements although optical devices tended to overestimate exposure. There is need to calibrate optical devices against a gravimetric standard prior to quantifying exposure. 47

Introduction 71
The World Health Organisation (WHO) estimates that over 4 million deaths annually are 72 attributable to exposure to Household Air Pollution (HAP) from biomass fuel smoke making 73 it a leading cause of global mortality (Lim et al. 2012;WHO 2016). Exposure to HAP is also 74 a leading cause of disability, being associated with a range of illnesses including acute and 75 chronic respiratory diseases, cardiovascular diseases, low-birth weight and cataracts (Gordon 76 et al. 2014). HAP is generated from the incomplete combustion of biomass fuels such as 77 wood, charcoal and crop residues and contains fine particulates often measured as Particulate 78 Matter less than 2.5µm in diameter (PM2.5) and gases such as Carbon Monoxide (CO). Currently there are a lack of high quality data on personal exposure to HAP in SSA with the 87 limited data available generally collected by non-comparable, study-specific methods. Most 88 epidemiologic studies have utilized indirect methods of exposure assessment such as 89 comparing household fuel use or housing type as proxies for personal exposure ( We conducted a multi stage sampling process where households in Kikati and Kumbursa 144 villages were invited for a meeting with the help of community leaders and research 145 assistants through market day announcements, worship day announcements and door to door 146 messages. A brief presentation explained the objectives and methodology of the study to all 147 those who voluntarily attended the meeting. An interpreter was used in Ethiopia. We also 148 conducted demonstrations using the various instruments with volunteers from the meetings. 149 From those households expressing an interest in participating, a random sample of those 150 using biomass fuel as the main fuel for cooking and/or heating was taken to identify those 151 who would participate in the study. The make-up of the household was confirmed by a 152 personal visit and where possible one person from each of the six age-gender groupings was 153 then invited to take part in the study. Age-gender groups were as follows: young children 154 The Dylos 1700 measured the number of particles at minute intervals for two particle size 183 ranges: >0.5 μm and >2.5 μm; with particles between 0.5-2.5 μm being calculated by 184 subtraction. Particle counts are expressed as a concentration per 0.01 cu ft of sampled air. 185 The Dylos particle count for particle sizes ≤2.5 μm was calculated subtracting the >2.5 mm 186 fraction from the total count number for particles >0.5 mm. Dylos particle count 187 concentrations were converted to PM2.5 mass concentrations using a previously published 188 conversion equation for combustion aerosol (Semple et al., 2015). 189 Two versions of Dylos devices were used for the study. One model was modified to have a 190 slower fan speed and could measure to approximately 6000 μg/m 3 while the standard version had a fan speed that provides maximum particle concentration data equivalent to about 1000 192 μg/m 3 . 193 The SidePak Personal Aerosol Monitor (AM510) measured airborne particle mass-194 concentration in mg/m 3 . TSI Sidepak AM510 was fitted was with a PM2.5 size selective 195 impactor which removed particles larger than 2.5 nm. The device draws air through a size- The PATS+ was deployed to measure real-time particulate matter (PM2.5) concentrations. 206 The device had lower particulate matter detection limit 10 μg/m 3 under most conditions and 207 an upper particulate matter detection limit of 30,000 to 50,000 μg/m 3 . The device logged PM 208 concentration, temperature, humidity, movement, and battery voltage. We used a logging 209 interval of one minute. The device was zeroed before and each sample measurement. 210 The MicroPEM device provides both time-resolved PM2.5 data via a micro-nephlometer and 211 gravimetric PM2.5 concentration for the whole sampling period through filter samples 212 collected on 3µm PTFE 25mm Teflon filters (Zefon International -Ocala, FL USA). The 213 device operated at a flow rate of 0.44 L/min. Pre-and post-sampling filter weights, and field 214 and lab blank weights, were determined in the temperature controlled Exposure Laboratory at 215 sampling on a ScalTec SBC 21 microbalance (Scaltec Instruments GmbH -Göttingen) 217 maintained at the exposure laboratory, to acclimatize to the temperature and humidity of the 218 lab for at least 24h. Filters were weighed in the same lab under a stable temperature (22 O C). 219 The MicroPEM instruments were deployed together with other devices to measure PM2.5 for 220 24h. After being retrieved, the real-time data were assessed with assistance from RTI and 221 filters were posted back to the exposure laboratory for gravimetric measurement. 222 We experienced saturation with the Sidepak and the Dylos. Whenever an out of range reading 223 was obtained, the highest concentration limit was used for the particular logged minute, i.e 224 6,000 µg /m 3 for Sidepak, 1,000 and 6,000 µg /m 3 for the lower concentration limit and the 225 higher concentration limit Dylos devices respectively. The 24-hour concentrations measured 226 by the optical devices were compared to the paired 24-hour measurements from the 227 gravimetric devices. 228 As part of an additional study to compare data and to assess field-use of different devices, 229 participants generally wore paired devices. In instances where a participant had two PM2.5 230 devices for 24h, the gravimetric data are reported where available, followed by data from the 231 Sidepak then Dylos and finally PATS+ in that order. This pragmatic hierarchy was employed 232 to reflect the likely quality of PM2.5 data from gold-standard (gravimetric) through to lower-233 cost instrumentation. The time-resolved data from the micro-nephlometer of the MicroPEM 234 device were not used due to download and interpretation problems. 235 236 Participants aged 18 years and above wore one or a pair of the particulate matter and carbon 237 monoxide monitors around the waist or chest with the help of a strap or bag all of the day 238 except during bathing and sleeping where devices were placed 1 to 2 metres away from the 239 participant. Participants aged between 15 and 17 years wore the devices in a similar manner to adults. Those aged between 2 and 15 did not wear the devices directly but instead kept 241 moving the devices to the microenvironment they were in, either by themselves or with help 242 of parents or older siblings. The devices were generally located within 1-2 metres of the 243 participant in each microenvironment. This group also received help from parents or older 244 siblings when completing the TADs. For participants under 2 years, their exposure was 245 assumed to be that recorded by a personal monitor worn by their mother or guardian. Two-way ANOVA was used to model HAP data and establish differences between gender-258 age groups in the two countries. Bonferroni and Tukey Post Hoc tests were used to establish 259 the trend of differences between the gender-age groups across Ethiopia and Uganda. Optical 260 and gravimetric paired devices were compared by using the average 24h PM2.5 measurement. 261 Agreement between the fine particulate matter measurement from the optical and gravimetric 262 devices was achieved by use of the Bland-Atman plots/analysis. The spearman's rank 263 correlation coefficient was used to quantify the relationship carbon monoxide and fine 264 particulate matter.  Crop residues 2 (5) 4 (9.0) 6 (7) 3.2 Agreement between gravimetric and optical devices 331 As part of the study to compare data from the various devices, participants generally wore 332 paired devices. Data to compare the PM2.5 exposures from the different device was obtained 333 from measurements over a 24-hour period using paired PM air monitoring devices. Table 2  334 shows the number of pairs in each of the countries. Bland-Altman plots were used to 335 compare the gravimetric measurements of PM 2.5 (from the filters used in the MicroPEM 336 devices) with the measurements from optical device (Figure 1) and agreement between 337 measurements from the optical devices (Supplementary data). 338 Table 2 showing pairing of the fine particulate matter measuring devices in Uganda and 339 Ethiopia. 340

10
SidePak -Lascar (1.89) in Uganda and Ethiopia respectively. There was a significant difference in 24h PM2.5 376 exposure between young males and young females (p<0.001) in Uganda; and similarly, 377 between young males and young females (p < 0.001) in Ethiopia. 378 There was also a significant difference in 24h PM2.5 exposure between adult males and adult 379 females (p<0.001) in Uganda; and similarly, in Ethiopia, between adult males and adult 380 females (p<0.001). 381 There was no significant difference in 24h PM2.5 exposure between children and adults 382 (p>0.001) in Ethiopia; whereas Uganda results showed a significant difference between 383 children and adults (p<0.001).
There was significant difference in CO exposure between similar gender-age groups namely 385 young children, young females and adult females with p values of 0.008, <0.001 and 0.006 386 respectively. Similar exposure profiles and age-gender differences were evident from the 387 carbon monoxide results: measurements were highest among adult females and lowest among 388 young males. The difference in GM CO concentrations between adult females and adult 389 males was approximately 5-fold in Uganda and 11 times in Ethiopia. * Shows significant difference in PM2.5 exposure (p < 0.01) between young males and young females in both countries. 392 ** Shows there is significant difference in PM2.5 exposure (p< 0.001) between adult males and adult females in Uganda and Ethiopia. 393 # show significant difference in CO exposure (p< 0.001) between young males and females; and adult males and adult females in both countries 394 ^ shows significant difference in CO exposure (p< 0.001) between similar gender age groups across both countries. 395 The two-way ANOVA showed no significant differences in 24h CO exposure between 396 Young children and young females, young children and adult females, young males and adult 397 males, young males and elderly, and young females and adult females across the two 398 countries. Bonferroni and Tukey Post Hoc tests showed similar trend of differences between 399 the gender-age groups across Ethiopia and Uganda. The highest intensity exposure where PM2.5 concentration in homes was >250µg/m 3 among 438 adult females in both countries were experienced from 1 to 1:59pm (61.5%) followed by 11 439 to 11:59 am (51.3%) and then 12noon to 12:59 pm (48.7%). 45% of the households 440 experienced PM2.5 concentrations above 250 g/m 3 between 7 and 8 am. 441 The highest hourly median PM concentration 386 g/m 3 was recorded from 1pm to 2pm 442 followed by 308 g/m 3 recorded between 11am to 12noon. The lowest hourly median of 443 PM2.5 concentration 11g/m 3 was recorded between 4 am and 6am. 444 An assessment on the effect of fuel type used for cooking in the household (Figure 4)  This is the first study to report data on personal exposure to HAP for rural households in 468 Uganda and Ethiopia and the first to consider differences in exposure by age and gender for 469 PM2.5 measurement in SSA. We found that women and girls had much higher exposures than 470 men. 471 Our study is similar to the work Saksena et al (1992) and Ezzati et al (2000).   Our study demonstrates that adult females had considerably greater exposure compared to 499 adult males of the same age group category (4.5 times greater in females for PM2.5 and 5.4 500 times greater for CO in Uganda; 4.4 (PM2.5) and 5.1 (CO) times in Ethiopia). These trends are 501 similar to a study conducted by Ezzati et al. (2000) that reported young and adult females to 502 have the highest exposure among the age groups and considerably greater exposures to HAP 503 than males of the same age group (2.5 and 4.8 times, respectively) in Kenya. 504 Measuring PM2.5 and CO for 24h enabled us to measure the concentration experienced by the 505 various participants throughout the full day and makes our data easier to compare with future 506 studies that also measure for 24h periods. From our data, we found that PM2.5 concentrations 507 can be higher even after 8.30pm depending on the household/participant activity and the time 508 a participant slept. For example, some coffee ceremonies in Ethiopia which involved roasting 509 of coffee and burning of incense, took place after 7pm in the house. Therefore, it is desirable 510 to carry out at least 24 h personal exposure measurements. 511 Fixed site monitoring is also likely to lead to exposure misclassification as people move 512 about from one micro-environment to another. This spatial variability makes it difficult to 513 rely on time-activity diaries especially since people sometimes forget the exact times they 514 were in specific microenvironments.
Titcome and Simmick (2011) reported high personal exposure for the primary cooks using 516 biomass fuel in a study carried out in Tanzania (Table 4). It is difficult to compare our results 517 with these given that their study collected data for just 8 hours and typically during high-518 exposure cooking periods. Again, this highlights the need for exposure measurement that 519 spans a full 24-hour period to enable direct comparisons and to enable consideration of 520 It is worth noting that average 24h PM2.5 exposures were higher for each gender based age 547 group in Ethiopia compared to Uganda. This could be attributed to difference in fuels used, 548 poorer ventilation levels and a comparative lack of windows in the kitchen areas in Ethiopia 549 (88% of Ethiopian homes had no windows in the kitchen area compared to 33% in Uganda). 550 Higher PM2.5 and CO levels in Ethiopia across age groups could also be attributed to the 551 "coffee ceremony" process where coffee beans are roasted over traditional stoves. This was 552 often combined with burning incense in a small burner made from clay. The ceremony was 553 generally carried out in most Ethiopian households in the morning and evenings. 554 Our study showed a 5-fold and 11-fold difference for 24h CO exposure between adult males 555 and females in Uganda and Ethiopia respectively yet PM2.5 the difference is approximately 5-556 fold in both countries. One reason could be attributed to the use of dung cake which was 557 predominant source of cooking fuel in Ethiopia. Combustion of dung cake has been found to 558 have high emission of CO ranging 14-29 g kg -1 compared to 11-12 kg -1 of wood 559 (Venkataraman and Rao, 2001) 560 Chakraborty et al also reported higher CO content produced from dung than wood in their 561 work to assess effect of exposure to biomass smoke on various health status (Chakraborty et 562 al. 2014). Another reason is, burning animal dung efficiently is more difficult than burning wood efficiently due to conversion efficiency. The frequent evening rituals/ceremony of 564 making coffee could have also been a major factor. The coffee ritual involves roasting beans 565 for some time which occurs more often (4 to 5 times a week). 566 A general comparison of fine particulate matter exposure between children (0 to 17 year) and 567 adults (18 to 49 years) showed a significant difference in Uganda but not in Ethiopia. This 568 further justifies the needs of gender-age quantification to avoid misclassification. 569 We were also able to conduct paired measurements in the field. Our results indicated that 570 optical devices (Sidepak, Dylos and PATS+) generally overestimated exposure compared to 571 the paired gravimetric device (MicroPEM). Although the optical devices tended to 572 overestimate exposure by 21 µg/m 3 on average and the data points were within the 95% 573 limits of agreement, the overestimate was as high as ~40 µg/m 3 which is important if a study 574 only used optical devices to quantify exposures and represents one of the limitations of using 575 optical instead of gravimetric devices. It is therefore important to always calibrate the optical 576 measuring devices against a gravimetric devices or methods. It is worth noting that there was 577 a strong agreement between optical and gravimetric fine particulate matter measurements. 578 There was a weak correlation between CO and PM2.5. We emphasize that this was observed 579 for personal exposure among the different gender-age groups. This should not be generalized 580 with correlation during cooking episodes. 581 The strengths of this study include the large dataset of over 5,160 hours of measurement 582 across 215 subjects in two countries. This is one of the largest datasets of personal exposure 583 to HAP in SSA. Previous studies that have measured personal exposure to HAP in SSA 584 Our age cut-offs were arbitrary and were intended to reflect differences in activities and 605 behaviours but we acknowledge that our definition of young child (0-5 years) masks large 606 differences between infants (<2y) who spend almost all their time close to their mother, and 607 children aged 4-5 who may have much greater freedom to move in and out of the home 608 setting. Similarly, our decision to include both male and female participants in a single 609 elderly age group may have hidden important differences here with elderly women perhaps 610 more likely to provide some assistance and experience during cooking activity. Elderly men are more likely to be involved in work around the homestead like sweeping and weeding 612 around the compound. 613 The study was restricted to rural households. Research has shown that many urban 614 households in developing countries also use biomass fuels for cooking with the proportions 615 of use generally varying according to socioeconomic status. Zhou  There are substantial differences in personal exposure to HAP from biomass fuel smoke 643 depending on the age and gender of an individual in sub-Saharan Africa rural households. 644 Exposure to HAP in a single household can vary from day to day because of fuel 645 characteristics such as moisture content or density, air flow, cooking method, fuel stacking is 646 being practiced. Individual exposure can also vary because of the different activities carried 647 out on different days including church days, market days, going to school, etc. 648 It is vital for future work to consider these differences in exposure to HAP across the life-649 course and to characterise age and gender differences when implementing exposure 650 reductions interventions. Health education interventions should target females and explain the 651 benefits to their health of reducing HAP concentrations. 652 There was an agreement between measurements from optical and gravimetric devices though 653 optical devices overestimated exposure. There is need to calibrate an optical device against a 654 gravimetric device prior to quantifying exposure.