1

Exposure reductions associated with introduction of solar lamps to kerosene lamp-using households in Busia County, Kenya.

A report prepared by:

Nicholas L. Lama Godfrey Muhwezib Fred Isabiryeb Kat Harrisonc,d Ilse Ruiz-Mercadoe Evans Amukoyef Tom Mokayaf Margaret Wambuaf Ian Baileyg Michael N. Batesh

Affiliations a. Civil and Environmental Engineering, University of Illinois, Urbana-Champaign, IL, U.S.A. b. CIRCODU, Kampala, Uganda c. SolarAid, London, England d. Acumen, London, England e. Instituto de Investigaciones en Ecosistemas y Sustentabilidad, UNAM, Morelia, Mexico f. Kenya Medical Research Institute (KEMRI), Nairobi, Kenya g. School of Optometry, University of California, Berkeley, CA, U.S.A. h. School of Public Health, University of California, Berkeley, CA, U.S.A.

Corresponding author Michael N. Bates, Ph.D., Division of Environmental Health Sciences, 783 University Hall, School of Public Health, University of California, Berkeley, CA 94720, U.S.A. Tel: +1 (510) 504-5424 Email: [email protected]

Date of publication: March 17, 2017

2

Table of Contents EXPOSURE REDUCTIONS ASSOCIATED WITH INTRODUCTION OF SOLAR LAMPS TO KEROSENE LAMP-USING HOUSEHOLDS IN BUSIA COUNTY, KENYA...... 1 EXECUTIVE SUMMARY ...... 3 INTRODUCTION ...... 3 METHODS ...... 3 RESULTS ...... 4 DISCUSSION ...... 5 INTRODUCTION AND BACKGROUND ...... 7 METHODS ...... 10 ETHICAL APPROVALS ...... 10 STUDY DESIGN ...... 10 RECRUITMENT OF PARTICIPANTS ...... 10 STUDY COMPONENTS ...... 11 Exposure and Health Questionnaires ...... 11 Exposure Measurements...... 12 Lighting and Cooking Device Usage Monitoring ...... 12 Micro-environmental Monitoring ...... 14 Personal Monitoring...... 15 Visual Acuity and Illuminance ...... 16 Statistical analysis of data ...... 17 RESULTS ...... 18 QUESTIONNAIRE DATA...... 18 LIGHTING DEVICE USAGE ...... 21 Kerosene Lamp Usage...... 21 Solar Lamp Usage ...... 23 Displacement of kerosene lamp usage by solar...... 26 MICRO-ENVIRONMENTAL MONITORING ...... 28 PERSONAL MONITORING ...... 34 DISCUSSION ...... 39 ACKNOWLEDGEMENTS ...... 45 REFERENCES ...... 46 APPENDICES ...... 49 APPENDIX 1: SOLAR LAMPS AND VOLTAGE LOGGERS ...... 49 APPENDIX 2: MICRO-ENVIRONMENTAL MEASUREMENTS IN THE KITCHENS AND PUPILS’ ROOMS ...... 50 APPENDIX 3: DEVELOPMENT OF THE LIGHT TYPE DETECTOR (LTD) ...... 51 APPENDIX 4: VISUAL ACUITY MEASUREMENT PROTOCOL ...... 54 APPENDIX 5: PERSONAL EXPOSURE PROFILE EXAMPLES ...... 55 APPENDIX 6: INTEGRATED EXPOSURE-RESPONSE RELATIONSHIPS ...... 56

3

Executive Summary

Introduction This is the report of a small study, carried out in Busia County, Kenya, and intended primarily to quantify changes in exposure to particulate matter (PM) and carbon monoxide (CO) associated with introducing solar lamps into household using kerosene lamps as their primary lighting source. Popularity and sales of solar lighting have been increasingly rapidly in developing countries, particularly in sub-Saharan Africa and South Asia. Whether there are health benefits to families using solar lamps rather than kerosene lamps has not been scientifically investigated. Evidence has been increasing, however, that uses of kerosene lighting and cooking devices in households are associated with serious health effects, particularly involving the lungs, but possibly also the eyes and to unborn babies. Evidence also suggests that these effects are caused by exposure to the pollutants emitted by these devices when they are operated.

The level of pollution to which an individual is exposed is often used as an indicator of health risk. Thus, any reduction in exposure resulting from the removal of kerosene sources would provide a first approximation of potential health benefits. The size of exposure changes is also a critical input in the design of any health study intended to directly measure health impacts. Such health-focused studies typically require large sample sizes. This input on exposure changes was what the present study was intended to obtain as well as to confirm the acceptability of the solar lamps as kerosene lamp replacements and to test some questions in a questionnaire and a few other procedures that might be used in such a larger study.

This research was commissioned by the London-based international non-government organization (NGO) SolarAid1 and funded by Google Ireland Limited.

Methods The study had a paired “before-and-after” design in which data were collected in the households of enrolled participants before and after the introduction of solar lamps (“baseline” and “follow-up”, respectively). Households were provided 3-4 weeks of “adoption” time with the solar lamps before follow-up exposure measurements.

As we were particularly interested in the impact for school pupils doing homework in the evening, the basis for selection of participating households was a single secondary school, located a few miles from the city of Busia. With the co-operation of the school, 20 pupils (and their households) were selected from among the senior students in the school (Forms 3 and 4) on the basis of several selection criteria. These included requirements that:

1. The household was not connected to the electric grid and was currently using kerosene lamps as its main source of lighting.

2. There was at least one non-smoking person, in addition to the selected pupil, who used a kerosene lamp for specific purposes (e.g., reading, studying or working).

1 SolarAid’s research and impact work is now housed and managed at the non-profit impact investor, Acumen.

4

3. Cooking was not conducted in the main house. This was to reduce the possibility that emissions from stoves would interfere with interpretation of the lamp emissions data.

After each head of household gave permission, we selected two lamp users for direct participation: the school pupil doing nightly homework and an adult kerosene lamp user. Household participation involved several activities that happened both at baseline, while they were still using kerosene lamps, and at follow-up, when they had had the solar lamps for a few weeks;

Questionnaires were administered separately to the head of household, the school pupil lamp user and the adult lamp user (in almost all cases a female household member involved in the cooking, often the pupil’s mother). These questionnaires obtained information on household circumstances, including means of cooking and lighting. They also inquired about symptoms experienced by the lamp users.

To measure household concentrations of both PM2.5 (particulate matter with an aerodynamic diameter less than 2.5 microns) and CO (carbon monoxide), instruments were affixed for 4 days in 3 rooms (main living room, school pupil’s room and kitchen). To measure personal exposure to PM2.5 and CO, lamp users wore for 48 hours a vest containing light-weight, unobtrusive PM2.5 and CO monitors. Personal monitoring provides a better measure of true exposure than the room concentrations because the monitoring devices move with the wearer and are thus affected by any pollutant source encountered.

Devices were fitted to all the kerosene lamps and the solar lamps to monitor their usage. The kerosene lamp monitors recorded temperature and the solar lamp monitors logged times the lamps were switched on or off. Kerosene lamp monitors were left in place for the duration of the study after first interaction with the participating family—up to 2 months. Solar lamps were monitored from the first day of deployment to the last day of the study – between 3 and 5 weeks.

Once baseline procedures had been completed in a household, it was provided at no charge with 3 Sun King Eco solar lamps and given instruction in their proper use.

Households participated in study procedures at the rate of 5 per week. Once baseline procedures had been completed in 20 households (4 weeks), the team returned for follow-up procedures, with households participating in the same order as for baseline procedures. Follow-up procedures were similar to the baseline procedures—questionnaires to the lamp users, room monitoring of PM2.5 and CO for 4 days and personal monitoring of lamp users for 48 hours.

At conclusion of follow-up procedures across all participating households, lamp use monitors were removed, but households were permitted to keep their solar lamps.

Data analysis focused on determining differences between baseline and follow-up for symptoms experienced, usage of lighting devices, and room concentrations of, and personal exposures of lamp users to, PM2.5 and CO.

Results All 20 households fully participated in the study with the exception that a solar lamp used by a pupil was misplaced, so that only 19 school pupils participated in the follow-up.

The median household size was 6 people, including 5 kerosene lamp users. All 20 households used cooking fires with biomass (19) or charcoal (1) as their primary cooking fuel. Of the 20 homes, 12 exclusively used kerosene for lighting; the other 8 reported use also of the light of the wood fire, cell

5 phone lights or rechargeable battery lights. Lamps were used for reading, studying, cooking and other work.

At baseline, high symptom prevalences were reported for respiratory and eye irritation symptoms. However, at follow-up, when inquiring about the same symptoms, all participants reported that they no longer experienced any of these symptoms.

Usage data showed that solar lamps almost completely (more than 90% in terms of the hours of usage) replaced kerosene lamp use at follow-up, for a daily average at follow-up of about 5 hours per solar lamp (15 hours per household for the 3 lamps provided). Solar lamp monitoring data showed peak usage in the evening (maximal at around 7:00 pm) and a smaller peak in the morning (maximal at around 6:00 am), similar to what was observed with the kerosene lamps.

Comparing baseline and follow-up PM2.5 concentrations in the three household rooms showed little change in the kitchens, since the concentrations there were dominated by cooking smoke, but there was a mean reduction of 79% in the pupils’ rooms and a 61% reduction in the main living rooms. Both of these reductions were highly statistically significant (p < 0.003). The average follow-up indoor PM2.5 levels in the main living rooms and pupils’ bedrooms were around likely ambient concentrations--which would be expected in the absence of indoor sources of smoke.

Both baseline and follow-up CO concentrations were low in terms of recognized health standards, and were unlikely to have been greatly influenced by the change to using solar lamps, as kerosene lamps have been shown to produce relatively little CO. Monitoring of CO was not carried out in the kitchens.

Personal monitoring of PM2.5 showed an average reduction in exposure of 73% between baseline and follow-up for school pupils (p = 0.003) and 50% reduction for adult lamp users (p = 0.002). Personal CO monitoring showed average reductions of 16% and 33% for school pupils and adults, respectively. P- values for both were > 0.1 and, again, both baseline and follow-up concentrations were below recommended health standards.

Discussion As far as we are aware, this is the first field-based assessment of personal exposures to PM2.5 from kerosene lamps, and the first to estimate the extent to which such exposures can be reduced by transitioning to solar lamps. Research into kerosene combustion exposures and their possible health impacts has been greatly overshadowed by research into exposures and health effects associated with cooking with solid fuels—biomass and coal—probably in part because of the clouds of visible smoke that solid fuels produce when burned in an indoor stove without improved combustion and/or a chimney.

However, epidemiological evidence has been increasing that PM2.5 produced by kerosene may be more toxic than the equivalent mass of PM2.5 produced by solid fuel combustion. The reason for that is presently unknown, but may be related to factors associated with the size or composition of the particles.

The longer-term objective of this study is to research the question of whether or not replacing kerosene lamps with solar lamps will bring health benefits to the lamp users. To do that it will be necessary to carry out a randomized trial, comparing health experience associated with kerosene lamp use with health experience associated with solar lamp use. To design such a study, particularly to decide on the appropriate sample size, information on the extent to which providing solar lamps would reduce exposure to PM2.5 is needed. This study has provided that information.

6

Another important objective of this study--to prove the acceptability of the solar lamps to families--has also been achieved, as is demonstrated by the usage data, showing an average of 5 hours per lamp per day, or 15 solar lamp hours per day per household (for the 3 lamps provided each household). Provision of 3 lamps ensured a high level of kerosene lamp displacement, although perhaps not 100% in all households, and ensured that the school pupils would have a solar lamp when doing their homework in the evenings.

A third study objective of this study was to test the usefulness of some basic symptom questions. We did not expect to be able to detect with statistical confidence any changes in health status associated with the change from kerosene to solar lamps. However, we found a complete remission of symptoms reported at baseline.

This reduction in ocular and respiratory symptoms is difficult to interpret. Usually symptoms of the type we inquired about have multiple causes and we would not expect replacement of the kerosene lamps by solar lamps to completely eliminate all such symptoms. Therefore, it is possible that the reported symptom reduction was, at least in part, a manifestation of the so-called “Hawthorne effect”, in which knowledge of the investigation and assumptions about what the investigators hoped to see influenced symptom reporting. In any case, there is no obvious way to distinguish this possibility from a real reduction in symptom etiology. Despite this uncertainty, we believe that the questions were properly understood and, with some modifications, would be useful in a further study. Such a study would likely include more objective measures of health status, less susceptible to a Hawthorne effect.

We believe that this study has been successful in almost all of the ways that we had hoped. Most importantly, this study has shown (i) that kerosene lamp use in Busia county, Kenya, is associated with substantial measureable exposure to PM2.5, both in adult and school pupil lamp users, (ii) these exposures are of such a magnitude that they have high potential to cause adverse health effects, and (iii) provision of at least 3 solar lamps per household provides a potentially very successful means of reducing these exposures and likely mitigating health impacts of household air pollution.

In conclusion, we believe that the results of this study provide prima facie evidence of likely health harm from kerosene lamp use and benefits of providing solar lamps to displace kerosene lamp use.

Demonstrating such health improvements and the sustainability of any such solar lamp intervention would require a much larger and more sustained study, for which the present study provides a basis for design and sample size calculation.

7

Introduction and Background

We report the results of a small study, carried out in Kenya, which sought to quantify changes in exposure to particulate matter (PM) and carbon monoxide (CO) associated with introducing solar lighting into non-electrified households relying on kerosene for light.

Sales of solar-powered lighting devices in the developing world, including sub-Saharan African countries, have risen exponentially in recent years, suggesting that it may be a promising near-term solution for providing clean lighting services to homes and business in non-electrified regions, or regions with frequent supply interruptions. Despite this, there is still little information on the impacts, particularly for health, of replacing kerosene lamps, candles and other inefficient and potentially harmful light sources, with solar lights or other electric light sources.

This study is a step towards obtaining information for decisions on whether introduction of solar lamps could provide meaningful human health benefits. It seeks to obtain the necessary inputs to justify the need for, and design of, a larger study to investigate potential health effects associated with kerosene lamp use. Building a stronger evidence base will quantify the benefits of investment in solar technology and provide a foundation for policy recommendations and advocacy to shape development objectives.

Kerosene has been used for lighting since around the middle of the 19th Century. In developed countries it was mostly replaced long ago by electric lighting. Nonetheless, kerosene lighting remains the only option for many families in low and middle-income countries, particularly in rural areas of Asia and Africa. In some areas kerosene stoves are used for cooking. A benefit of kerosene is that it can be purchased in very small quantities if funds are limited. In some countries, household access to kerosene has increased as a result of long standing government subsidy programs that, once in place, are difficult to remove. Poisoning, burns and explosions from kerosene are widely recognized problems, but, until recently, kerosene had generally been regarded as a clean-burning fuel. This was especially so since kerosene lamps are used in darkness and do not obviously fill rooms with visible smoke, as do the biomass-burning stoves often used in the same households (Lam et al. 2012a). Despite that, the tendency of kerosene lamps to deposit soot on the ceilings of rooms in which it is burned, is well- established. In fact these soot particles, when heated, provide the incandescent light of the kerosene flame.

Recent epidemiologic studies have cast doubt on the assumption that kerosene is a clean burning fuel (Pokhrel et al. 2010; Bates et al. 2013; Epstein et al. 2013). Some studies suggest that kerosene is associated with health effects comparable with those of biomass burning for cooking, although a wider evidence base is needed to firmly establish this. If kerosene combustion products are confirmed as having health impacts comparable with, or even greater than, those of biomass burning, then this may be because it produces very fine particulate matter with compositional characteristics different to those from biomass combustion. There is also evidence that, when cooking with kerosene, mothers are more likely to remain in the kitchen than when cooking with biomass—and their children will stay with them (Bates et al. 2013; Saksena et al. 2003). This behaviour is likely to substantially increase exposure to kerosene combustion products.

Although there are still few epidemiologic studies of the health impacts of kerosene use for cooking and lighting, the existing evidence is strongly suggestive. A growing body of evidence has associated household kerosene use with adverse health outcomes – including tuberculosis (Pokhrel et al. 2010), low birth weight and neonatal death (Epstein et al. 2013), and pneumonia (Bates et al. 2013). This has

8 led the World Health Organization to recommend discontinuing household kerosene use, while calling for additional studies to strengthen the evidence base (WHO 2014).

Until recently, the options for poor families who wished to replace kerosene for lighting have been limited. Some families who could not afford kerosene have relied on the light of the cooking fire. Others have the option of candles, although a disadvantage of these is that they must be purchased as whole candles, unlike kerosene which can be purchased in extremely small quantities when funds are limited. Also, candles are similar to kerosene in some ways in that they burn and can produce sooty particles that may have health impacts (Zai et al. 2006; Fine et al. 1999; C Fan and J Zhang 2001), although that has been little investigated in a developing country context.

For several reasons, including local non-availability and unaffordable connection and maintenance costs, the electricity grid is beyond the reach of many or most poor families. This is particularly true in many parts of sub-Saharan Africa. Even when connected to the grid, frequent supply interruptions often require houses to continue reliance on kerosene for lighting (Lam et al. 2016). Because of battery costs, battery-powered lights are also often unaffordable for extensive periods.

More recently, an extensive range of pico-solar lamps have become available, often at prices (sometimes government subsidized) affordable by poor families. “Pico-solar” refers to small, portable solar units. Unlike household solar systems, pico-solar devices are often charged by a small independent photovoltaic cell, providing a smaller range of energy services at greatly reduced up-front cost of the device. These offer a clean alternative to kerosene lamps and candles, typically with much better light quality and no operating costs.

Although there exist many studies evaluating health benefits of “improved” household technologies in energy poor regions, most have focused on cooking activities. These studies have generally examined exposure to biomass combustion products, with little evaluation of kerosene or other cooking fuels. Even fewer studies have looked at lighting activities, let alone kerosene lighting.

Exposures to pollutants emitted from kerosene lights and their associated health risks are not well characterized. We know that the smoke emitted by lamps commonly found in African homes contains large quantities of fine particulate matter (PM2.5), probably the most important pollutant indicator of health risks. From measured emission rates of PM2.5 from kerosene lamps, it has been estimated that a single lamp used inside a typical room can easily exceed World Health Organization Indoor Guideline Concentrations (WHO 2006, 2010; C-W Fan and J Zhang 2001). Along with these particles, other potentially health-damaging chemicals, including sulphur dioxide (SO2), carbon monoxide (CO), and polycyclic aromatic hydrocarbons (PAHs), are also emitted (Lam et al. 2012a). PM2.5 emissions from kerosene lamps are also rich in black carbon, a highly potent climate warming pollutant associated with adverse health outcomes (Lam et al. 2012b). Nearly all of the existing evidence for lighting exposure, however, is based on laboratory experiments or indoor air quality modelling (Schare and Smith 1995; Apple et al. 2010). This information needs confirmation with measurements in homes of actual users in uncontrolled settings.

To the best of our knowledge, no previously published study has evaluated the impact on exposure or health of replacing household kerosene-based lighting with solar lamps. Apart from likely health (e.g., respiratory) benefits from eliminating exposure to kerosene lamp emissions, there may be visual and ocular benefits from the improved illumination. Although the evidence so far is limited, kerosene combustion emissions may irritate or otherwise damage the eyes (Lam et al. 2012a) and poor lighting

9 can make it difficult to read or perform tasks requiring the seeing of fine detail. Possible consequences are task avoidance, reduced efficiency and visual discomfort, with symptoms such as headaches and tired eyes.

Because of this lack of evidence for and understanding of the potential impacts of kerosene lamp use on household pollution and health, or of the possible health benefits of kerosene lamp replacement, the then Research and Impact division of SolarAid, an international NGO with a focus on social enterprise, distributing pico-solar lights in rural Africa, conceived of this study with the other authors, and received funding from Google to conduct the research2.

A key objective of the present study was to collect data on changes in exposure to PM2.5 and CO, but also preliminary data on changes in self-reported eye irritation and respiratory symptoms, associated with replacing kerosene lamps with solar lamps in rural households in Busia County, Kenya. We hypothesized that there would be substantial reductions in both exposures and symptoms associated with the change-over in lighting sources. However, it was not our intention to collect statistically robust symptom data, as the sample size was small.

Also sought was information for refinement of measurement methodologies, information on behavioral changes resulting from the introduction of solar lamps (e.g., adoption and usage habits), and some data on visual acuity changes associated with changing from kerosene to solar lighting.

The underlying purpose was to conduct a study that could act as proof of concept for the relevance of designing a larger study to investigate health impacts associated with the change from kerosene to pico- solar lamps and, at the same time, gather sufficient information for the design and sizing of that study. We envisage the larger study as a randomized intervention trial involving both exposure and health status measurements, including objective measures, such as respiratory function testing. Data from the present study, which we refer to as the exposure study, is important for the design, particularly determination of the necessary sample size, of the health benefits study.

2 SolarAid’s research and impact work is now housed and managed at the non-profit impact investor, Acumen.

10

Methods

Ethical approvals No human subjects work was conducted before approvals for the study procedures were obtained from the Institutional Review Board of the University of California, Berkeley and from the Scientific and Ethics Review Unit of the Kenya Medical Research Institute (KEMRI). After the Head of Household gave permission for the family to participate, prior written informed consent was obtained from all adult participants. After the consent of their parents was obtained, written assent to participate was obtained from all participants less than 18 years of age. Study design The study had a paired “before-and-after” design, where exposures in the households of all enrolled participants were measured before and after the introduction of solar lamps (“baseline” and “follow-up”, respectively). Households were provided 3-4 weeks of “adoption” time with the solar lamps before follow-up exposure measurements. The usage of both baseline and intervention technologies (kerosene and solar lamps, respectively) was monitored in each exposure study household from its time of entry until the end of the study. Recruitment of participants The basis for participant recruitment was St. Peter’s Budokomi Mixed Secondary School, located a few miles outside the city of Busia, in Busia County, Kenya. As well as the agreement to participate of the school’s principal and teaching staff, approval to involve the school in the study was obtained from the County Director of Education, Busia County. The focus around a school was both because of the partnership with SolarAid and their distribution method, through schools, and because we considered it important to obtain some assessment of the benefits to school pupils doing their homework in the evening, although we were also interested in the benefits to adults in the same households.

Twenty pupils were selected from among the senior students in the school (Forms 3 and 4) on the basis of the following selection criteria: 1. The head of household was 18 years of age or older. 2. The household was not connected to the electric grid and was currently using kerosene lamps as its main source of lighting.3 3. There was at least one non-smoking person, in addition to the selected pupil, who used a kerosene lamp for specific purposes (e.g., reading, studying or working). 4. The family must be willing to replace use of their kerosene lamps with solar lamps provided free of charge.4 5. Cooking was not conducted in the main house. This was to reduce the possibility that emissions from stoves would interfere with interpretation of the lamp emissions data5.

3 From SolarAid research in Kenya, 60% of rural households interviewed through public surveys (n=2,485) used kerosene as their main source of lighting. The rural electrification rate for Kenya is 7% (IEA World Energy Outlook 2015). 4 Kerosene lighting devices were not removed from households. 5 70% of rural houses in Kenya cook in buildings separate from the main living area or outside (USAID. 2015. Demographic and Health Surveys.

11

Selection of potentially participating pupils and households was done purposively in conjunction with the Head Teacher and head science teacher, who were familiar with parents and their household arrangements. Households were approached and a screening questionnaire was used to interview the Head of Household, to confirm participation eligibility. There were no exclusion criteria based on gender, race or ethnicity. However, the youngest age at which participation was permissible for the purposes of obtaining personal exposure measurements was 13 years.

In return for participation, households were allowed to keep the provided solar lamps (paid for by the research grant) at no charge. Other than that, participating families received no other form of compensation for their participation.

During the baseline period, households were enrolled at a rate of 5 per week, with two or three households sampled simultaneously, to obtain 48-hour personal exposure measurements. One week after completion of its baseline procedures, each household was provided with three Sun King Eco solar lamps (Greenlight Planet Inc., U.S.A., Appendix 1) and given instruction in their proper use. Three lamps were provided, as preliminary investigations had shown that often several kerosene lamps were simultaneously used in a household. We wanted to have reason to be confident that kerosene use for lighting could be fully displaced in the household and that there would be at least one solar lamp available for the school pupil doing his/her nightly homework. When all 20 households had completed baseline procedures, follow-up procedures began, at a rate of 5 households per week--in the same order that they completed baseline procedures. This ensured that each household had a period of 3-4 weeks to become accustomed to using the solar lamps. This “stabilization” period was to ensure that households were accustomed to using the new lamps when follow-up testing began. Experience from previous studies of cookstoves suggested that a 3-4 week period should be sufficient for finding a new equilibrium of usage practices. We were able to examine this objectively by using usage sensors in the solar lamps.

Study Components As planned, the exposure study comprised four components:

1. Exposure and health questionnaires. 2. Personal exposure and indoor air quality measurements. 3. Monitoring of kerosene and solar lamp use. 4. Visual acuity and illuminance measurements.

Each component is described in detail below.

Exposure and Health Questionnaires Study questionnaires were developed in English, then translated into Swahili and back-translated into English. The two English versions were compared to ensure their comparability and some adjustments made accordingly to the Swahili version. Further adjustments for comprehensibility and cultural appropriateness were made after local consultation. The study questionnaires were administered in the field in paper-based form. They were administered by a Busia-based female interviewer, fluent in English and Swahili, as well as the two local languages, Luo and Luhya.

The study used 3 short questionnaires. The first questionnaire was directed to the head of household, to obtain information on household circumstances, including who were the kerosene lamp users; the other two questionnaires were the baseline and follow-up questionnaires for actual lamps users. Some of the questions were about use of lighting devices in the household—lamp type, frequency of use, and tasks

12 for which the lamp was used. Other questions were about symptoms, particularly respiratory and vision-related symptoms, and when they were typically experienced in relation to use of lighting devices--such as when reading. Questionnaires took about 10 minutes each to administer. Apart from their use in collecting basic participant data, another purpose of using the questionnaires in this study was to examine the usefulness of some of the questions. With the small study sample size, we did not expect to see statistically significant differences between baseline and follow-up questionnaire results.

Exposure Measurements Household exposure estimates were generated using three approaches: usage of lighting devices, measurement of indoor pollutant concentrations in selected rooms, and personal exposure measurements.

Measuring device usage is a minimally invasive, but objective, measure of whether people are operating a light source—kerosene or solar. In addition to being an indicator of exposure, objective usage measures can complement pollutant-based exposure methods in several ways: they are a direct measure of whether a light source is being used, and they can be deployed unattended for weeks. Usage monitoring was coupled with measurements of actual pollutant concentrations--PM2.5 and carbon monoxide (CO)--within the kitchens, the main living rooms and the school pupils’ bedrooms (“environmental measurements”) and on two participants in each house using lightweight, wearable monitors (“personal exposure measurements”). All three monitoring methods were deployed within all enrolled households. The one exception was that CO was not measured in kitchens because of insufficient available instruments. For a similar reason, the kitchens of the first five households were not measured for PM2.5 at baseline.

While each exposure assessment approach has unique advantages and disadvantages, personal monitoring is generally considered the most accurate measure of what a person is actually exposed to. This is because measurement devices move with the participant across all micro-environments, and thus measure exposure to relevant pollutants from all sources, which may or may not be observed or monitored by the study investigators.

Lighting and Cooking Device Usage Monitoring Sensor-based measurements of lamp and cookstove usage were collected continuously in all houses starting from when baseline measurements were taken (individual dates, according to house) and ending upon completion of the study--at the same time for all participating households.

Usage of all kerosene lamps and cookstoves in each household was measured using small, button-sized, temperature loggers, called “ibuttons” (DS1922L, Maxim, USA), which record surface temperature and a corresponding time stamp. These loggers have been deployed in numerous studies to assess the adoption of cookstoves in developing countries (Ruiz-Mercado et al. 2012; Ruiz-Mercado et al. 2013; Pillarisetti et al. 2014; Lozier et al. 2016). Using duct tape, the ibuttons were attached to the metal necks, below the flames, of simple kerosene wick lamps and to the support arms of hurricane lamps (Figure 1). Placement on stoves varied according to stove design, but was generally on the side of metal stoves or in a metallic casing beside open fire stoves. All ibuttons were deployed with a sampling rate of ten minutes, meaning that temperature was recorded once every ten minutes. This was a practical sampling period, based on a need to ensure that the ibutton memory capacity of about 8 weeks was sufficient to store all the collected data before download. iButton sensor data were downloaded as text files using software provided by the manufacturer (Maxim 1-Wire). This occurred 4 times: (i) at the end of Baseline, (ii) at the beginning of the follow-up monitoring week, (iii) at the end of the follow-up

13 monitoring week, and (iv) at the end of the study. What resulted was a continuous data log of kerosene and stove usage across all phases of the study.

Usage of solar lamps was monitored using custom-built “Lamploggers” (Bonsai Systems GmbH, Zurich, Switzerland). Lamploggers capture usage data by monitoring changes in the voltage of the lamp light- emitting diode (LED) and recording the times when the lamp is turned on and turned off. The entire logger was mounted inside the lamp casing and not visible to the lamp user (Appendix 1). Usage monitors were installed by opening the lamp casing and soldering three wires from the monitor to the board of the lamp (voltage, ground, side of the LED). The logger draws a tiny amount of power directly from the lamp battery, with minimal effect on lighting duration, so that the logger functions as long as the lamp battery is charged. If the battery falls below a critical threshold, the logger enters a “sleep mode”, in order to avoid draining the battery and risking damage to the lamp. The record from each Lamplogger contains a unique logger identifier and records event date and time (to the minute) for events lasting longer than 2 seconds. The data record also includes the average voltage reading from the LED over the event period, and the average voltage of the battery, although these items were not used in the data analysis.

Data from the solar lamp loggers are stored directly on the loggers and downloaded via Bluetooth using a custom iPhone application provided by Bonsai Systems. Field staff visited households and downloaded data directly from the Lamplogger to the phone. When Wi-Fi connectivity is available and the Lamplogger iPhone application is active, all data on the phone are automatically uploaded to a password-protected server maintained by Bonsai Systems where they can be accessed and logger data downloaded as a text file. The serial numbers of monitors are assigned to the researchers who purchase them, and the lamp data for a specific logger can only be viewed through a user (researcher) account that has been granted access permission.

Additionally to the use of ibuttons and solar lamp loggers, in the course of this study we investigated the possibility of developing a small device that could be placed in a room and simultaneously monitor and log the use of solar lamps (or other electric lighting devices) and kerosene lamps in the room. The device was provisionally named the Light Type Detector or LTD. Two prototype LTDs were developed and subjected to some testing in the field. Although these devices were not used in the actual fieldwork, they showed promise and results of the prototype testing are reported in Appendix 3. Additional funding would be necessary to further progress the development of these devices.

14

Figure 1. Hurricane (left) and simple wick (right) kerosene lamps with ibutton temperature sensors (indicated by yellow circles). Sensors detect changes in surface temperature of the lamp, specifically the heat generated by the flames. These simple temperature signals can then be analyzed to estimate various metrics of lamp usage.

Micro-environmental Monitoring Micro-environmental monitors were deployed in each household for four days (Monday to Friday) during both the baseline and follow-up periods. Three rooms in each household were monitored: the main living area, the participating school pupil’s room, and the kitchen. Because of a temporary equipment shortage, kitchen measurements were not included in the first five households at baseline.

For environmental monitoring, we measured fine particulate matter (PM2.5) and carbon monoxide (CO) with small, lightweight, unobtrusive air pollution monitors that have frequently been used in developing country households for similar monitoring:

Particulate Matter (PM2.5) Micro-environmental PM2.5 concentrations were measured every two minutes with UCB Particle and Temperature Sensors (UCB-PATS, Berkeley Air Monitoring Group, Berkeley, CA, Figure 2), which use the light scattering characteristics of particles to infer a mass concentration in the air. The mass concentration of particles is typically reported in units of milligrams (or micrograms) of particles of aerodynamic diameter of 2.5 micrometers (2.5 x 10-6 meters) or less per cubic meter 3 of air (e.g., mg PM2.5/m ). This optical-based device is a smoke detector modified to log sensor readings, rather than trigger a smoke alarm. The appearance of the UCB is identical to a smoke detector, with the exception of a serial port used to download and program the internal logger.

15

While all UCBs were calibrated prior to deployment, particle mass concentrations inferred from light scattering are sensitive to differences in particle composition (e.g., size, shape, color). Thus, a subset (75%) of the UCB-PATS placed in houses were co-located for 48-hrs with a filter-based, integrated (gravimetric) measurement system using 37mm Teflon filters (Pall Corp.) and BGI Triplex Cyclones to selectively capture only PM2.5. Pre-sampling and post-sampling filter weighing was performed at the University of California, Berkeley on a sensitive balance (Mettler Toledo XP2U, repeatability of 0.15 µg). The difference in weights provides a direct measure of particle mass. Comparisons between co-located UCB-PATS data and filter-based particle masses were then used to derive adjustment factors separately for the main living areas and school pupils’ rooms. As many studies have deployed UCB-PATS in wood-burning kitchens, the correction factors for this microenvironment were obtained from previous studies and experiments (Armendáriz-Arnez et al. 2010).

Figure 2. UCB Particle Monitor (left) used to measure real-time PM concentrations and a BGI Triplex cyclone (right) used for gravimetric measurement of PM2.5 (pump not shown).

Carbon Monoxide (CO): Micro-environmental CO concentrations were measured in real time using Lascar CO Loggers (EasyLog EL-USB-CO300, Lascar Corp.). Carbon monoxide is typically measured in the concentration units of parts per million by volume in air, abbreviated in this report as ppm. The logger response was checked prior to field deployment using laboratory span gas and logger- specific adjustment factors were calculated. Loggers recorded CO concentrations once every two minutes.

Personal Monitoring Personal monitoring was conducted for 48 hours during the micro-environmental sampling period, in both the baseline and follow-up periods. In each family we carried out personal monitoring on the school pupil and one other person in the household who, at baseline, used a kerosene lamp for a specific task. With one exception, this was the mother or another female family member who performed cooking tasks in the house.

As with the micro-environmental monitoring, both PM2.5 and CO were measured, although with a different device for the PM2.5:

16

Particulate Matter (PM2.5) Personal monitoring of real-time PM2.5 was conducted with Micro-Personal Exposure Monitors, or MicroPEMs (RTI, USA). Like the UCB, the MicroPEM provides a real-time, optical-based measure of particle concentrations, but also obtains an internal filter-based gravimetric measure for deriving sample-specific correction factors (Chartier et al. 2016). The MicroPEM is small enough to fit inside a shirt pocket (Figure 3) and can run for 48-hrs continuously on four AA batteries. The MicroPEM has an accelerometer to measure movement, which can be used to calculate compliance with use instructions.

To maintain privacy by reducing visibility of the monitors and possible embarrassment to participants, personal measurement devices for both PM2.5 and CO were integrated into fabric vests that the participants wore (Figure 3). A number of such vests of different sizes were produced by a local tailor. School pupils were not required to wear their vests when at school, but they were asked to put it on again as soon as they returned home after school. Participants were asked to place the vest beside their bed during sleeping hours.

Figure 3. Personal exposure monitoring vest, showing Lascar (top right) and MicroPEM with sampling tube (bottom right) sampling devices, worn by adults and school pupils.

Visual Acuity and Illuminance We had proposed to include some testing and assessment of vision-related aspects in the exposure study, partly to assess the acceptability and feasibility of the testing, but also to obtain some information that would help with determination of sample size requirements for the main study. However, in the baseline measurement period, it quickly became apparent that there was an unanticipated hazard: because of the way that the close-up visual acuity chart and kerosene lamp needed to be relatively positioned, it would have been quite easy for a participant carrying out the procedures to have accidentally knocked over the lamp, causing kerosene spillage and a fire, with potentially very serious consequences. As a result of this observation, we decided to abandon this component of the exposure study, but continue with the other procedures. The proposed illuminance and visual acuity study procedures are contained in Appendix 4.

17

We believe that for any further study to investigate health effects it would still be desirable to have a component testing visual acuity and measuring illuminance under kerosene and solar lamp light conditions, but these procedures need to be further refined and pilot-tested at a different time.

Statistical analysis of data Questionnaire data were analyzed using simple descriptive statistics, comparing baseline and follow-up results. Lamplogger, micro environmental and personal monitoring data were cleaned, assembled, and analyzed using R software. Downloaded ibutton data were analyzed using the signal processing software SoftSUMit (UNAM, Mexico), which separates event peaks from diurnal ambient temperature variation, then calculates key summary statistics, such as average usage duration. For comparisons of PM2.5 and CO concentrations, non-parametric statistical tests were used because of the skewed nature of the data.

18

Results Twenty eligible families with pupils attending St. Peter’s Budokomi Secondary School were identified and permission was given by the heads of household for their families to participate. No eligible households refused to participate. Table 1 provides basic demographic information on the individual participants and their use of kerosene lamps at baseline. There are 20 adults and 20 school pupil participants, one of each from each participating household. All 20 households fully participated in the study, with the single exception of one household in which the solar lamp used by one school pupil was misplaced. For that reason, there are follow-up data from only 19 pupils.

Questionnaire data. Table 1 shows basic descriptive data for participating households at baseline, as obtained by interview of Heads of Households. All houses had a separate private pit latrine and a separate cooking building from the main house where cooking was done with an open fire. None of the houses were connected to the electric grid and all used kerosene as their main lighting source, although supplementary sources were sometimes used (see Table 1). No household used solar lamps or candles.

Table 1. Household characteristics at baseline. Characteristic of household (N = 20) Numbers People sleeping there at least 4 nights per week Mean: 7.4; SD: 3.1; median: 6; range: 2, 15 Number of kerosene lamp users Mean: 5.1; SD: 1.8; median: 5; range: 2, 10 Adults Mean: 3.6; SD: 1.5; median: 4; range: 0, 6 Children Mean: 1.5; SD: 1.3; median: 1; range: 0, 4 Number of kerosene lamps available in Mean: 2.6; median: 2, range: 1, 5 householda Floor in living room Cement 4 (20%) Earth 16 (80%) Roof material Corrugated iron 18 (90%) Grass, reeds, palm leaves, branches or 2 (10%) mud Main cooking fuel Biomass 19 (95%) charcoal 1 (5%) Supplementary lighting sources (to kerosene)b Wood fire light 4 (20%) Rechargeable battery light 2 (10%) Cell phone light 4 (20%) No other lighting source used 12 (60%) a Based on numbers of kerosene lamps in households to which ibuttons were attached. b Sums to >100% as two families used both cell phones and wood fires as alternative sources of light.

Table 2 shows demographic data and statistics for kerosene and lamp use data for all study participants, both parents and school pupils. All participants used kerosene lamps at baseline, but at follow-up only one reported doing so. For reasons not specified, she advised that the head of household would not permit her to use a solar lamp in the kitchen. It is not clear why in the adult lamp user group there is a

19 lower reported prevalence of reading and studying with solar lamps than with kerosene lamps at baseline.

Table 2. Demographic and lamp use data for participants in 20 households. School pupil lamp users Adult lamp users Sex: Male 14 (70%) 1 (5%) Female 6 (30%) 19 (95%) Age (yrs): Mean, median 18.3, 18.0 44.8, 46.5 Range 16 - 21 22 - 69 Lamp use: Baseline Follow-up Baseline Follow-up (kerosene) (solar) (kerosene) (solar) Reading 20 (100%) 19 (100%) 12 (60%) 3 (15%) Studying 18 (90%) 19 (100%) 2 (10%) 0 Cooking 16 (80%) 11 (58%) 18 (90%) 19 (95%) Other work 20 (100%) 18 (95%) 2 (10%) 20 (100%)

Participants were asked a series of questions, both at baseline and follow-up, about eye symptoms. As shown in Table 3, high prevalences of eye symptoms were reported at baseline (during kerosene lamp use), but all symptoms were reported to have abated entirely at the time of follow-up—three or more weeks after solar lamps were provided to the family.

Table 3. Prevalence of self-reported eye symptoms associated with lamp use, at baseline and follow-up. Symptom. Number reporting symptom (%)a School pupil lamp users Adult lamp users Baselineb Follow-upc Baselineb Follow-upc (N=20) (N=19) (N=20) (N=20) Tired eyes when reading or studying. 20 (100) 0 20 (100) 0 Eyes itchy or sore when reading or 18 (90) 0 20 (100) 0 studying. Over the last two weeks: -Dryness in the eyes 14 (70) 0 12 (60) 0 -Grittiness (having sand) in the eyes 18 (90) 0 20 (100) 0 -A burning feeling 18 (90) 0 17 (85) 0 -Redness of eyes 16 (80) 0 14 (70) 0 -Crusting with yellow discharge 13 (65) 0 10 (50) 0 -Sticking together of eyelids when 13 (65) 0 7 (35) 0 waking in morning a. For the purposes of this analysis, responses of “every day”, “most days” and “some days” were classified as “yes”; and “rarely” and “never” were classified as “no”. b. Baseline—when using kerosene lamps before solar lamps provided. c. Follow-up—3 weeks or more after family provided with 3 solar lamps.

A number of questions on respiratory symptoms were also asked of all participants and results are shown in Table 4.

20

Table 4. Prevalence of self-reported respiratory symptoms associated with lamp use, at baseline (during the period of kerosene lamp use) and follow-up (the period since solar lamps supplied to the family). Symptom. Number reporting symptom (%) School pupil lamp users Adult lamp users Baselinea Follow-upb Baselinea Follow-upb (N=20) (N=19) (N=20) (N=20) Wheeze or whistling when don’t have a cold 3 (15) 0 5 (25) 0 Woken with chest tightness 7 (35) 0 7 (35) 0 Shortness of breath at rest 7 (35) 0 7 (35) 0 Woken by shortness of breath 3 (15) 0 5 (25) 0 Woken by coughing 12 (60) 0 10 (50) 0 Short of breath walking on level ground 2 (10) 0 7 (35) 0

Ever seen a doctor about breathing difficulties 1 (5) - 3 (15) - Ever asthma diagnosis 0 - 0 - Ever COPD (chronic obstructive pulmonary 0 - 0 - disease) diagnosis

Believes the lamp affects his/her breathing: 19 (95) 0 18 (90) 0 - Increased cough 9 (45) 0 12 (60) 0 Sneezing 2 (10) 0 1 (5) 0 Tightness in chest 2 (10) 0 4 (20) 0 Difficulty 16 (80) 0 16 (80) 0 breathing a. Baseline—when using kerosene lamps before solar lamps provided. b. Follow-up—3 weeks or more after family was provided with 3 solar lamps.

The comparison between baseline and follow-up in Table 4 is not exact. The questions at baseline apply to the period until the interview of kerosene lamp use (some questions apply just to the previous 12 months), but the follow-up questions apply only to the approximately 3- week period in which the families had solar lamps. This increases the likelihood that symptom reporting will be higher for the baseline period, irrespective of lamp type. This may in large part account for the apparently very substantial reduction (to zero) in symptom reporting at follow-up. For this reason, a direct statistical comparison between baseline and follow-up results would not be appropriate or meaningful. If we had applied such statistical tests, however, then it is clear that the difference between baseline and follow-up would be associated with a low p-value for every symptom.

There is suggestive evidence from the reported respiratory symptoms at baseline, particularly wheeze in the absence of a cold, that some of the school pupils and adults may have had asthma, although we asked about this and no participant reported ever having been diagnosed with asthma by a doctor. However, few of the participants had ever seen a doctor about breathing difficulties.

21

Lighting Device Usage

We present below data obtained from ibuttons and lamploggers for kerosene and solar lamps, respectively.

Kerosene Lamp Usage Kerosene lamp usage assessments, while not direct measures of exposure to health-damaging pollutants, are informative for exposure assessments and health intervention planning. First, data can be used as a long-term indicator of likely exposure changes and trends. In this study the ibutton loggers provided a measure of kerosene lamp usage across the entire study duration for each house, covering baseline, transitional and follow-up phases. They also provide an objective measure of service needs, in terms of hours of lighting, and characteristics of lighting events, such as duration and time of day. These measures can be informative for intervention design.

Figure 4 illustrates the raw kerosene lamp data, from 5 lamps in a single home, read by ibutton sensors during the baseline period. Two patterns can be seen in each panel: the natural temperature fluctuations of the ambient air (diurnal trends) and lighting events. The sharp and sudden peaks are indicative of the lamp being in operation (the lamp is lit and gets hot). These sharp peaks rise above a smoother temperature signal that reflects the diurnal temperature pattern of the ambient air, reaching a maximum around mid-afternoon. Looking only at light events and before any data processing, it can be seen, for example, that some lamps are used only in evenings, while others have morning and evening usage cycles. Although not presented here, weeks of these traces can be processed to obtain robust estimates of the hours of service each lamp provides to a house.

22

Figure 4. Illustration of real-time kerosene lamp usage data from ibutton sensors on 5 kerosene lamps in a single home before distribution of solar lamps. Steep temperature spikes indicate periods that the lamp was in use. The time periods covered by the yellow highlighting correspond to the evening hours--between 6:00pm and 12:00am--each day.

Figure 5 illustrates usage of all kerosene lamps in a single home over three phases of the study: baseline, a transition period starting when three solar lamps are distributed (beginning after the dashed red line), and follow-up. A simple qualitative assessment clearly shows a sudden cessation of use in the first three kerosene lamps after distribution, but continued use of the fourth (bottom panel). This suggests, in this house at least, a 1:1 replacement ratio of the kerosene lamps by solar lamps. It also suggests that kerosene lamps may continue to be used until all lighting-related services originally met by kerosene are met.

23

Figure 5. One month of real-time kerosene lamp usage data from a single house with four kerosene lamps. Sharp peaks (approximately above 35oC) correspond to lamp usage events. The red vertical line indicates the date that three solar lamps were introduced. The relatively soft peaks after solar lamp deployment, most obvious in the first three panels (SW1-SW3), correspond to diurnal (ambient) temperature patterns (no kerosene lamp usage). The continuing sharp peaks in the signal of the fourth lamp indicate continued use of that kerosene lamp.

Solar Lamp Usage Solar lamp usage data were collected from 60 solar lamp voltage-loggers covering all 20 households enrolled in the study6. Data collection began the first day solar lamps were given to the household, thus capturing transition trends as the households accustomed themselves to the new lamps. Figure 6 shows the average number of solar lamp events occurring in each

6 One solar lamp with its logger was lost before follow-up monitoring was conducted. However, some data from the period between baseline and follow-up were collected.

24 house (all three lamps) by hour of day, and the house average (blue dots). Here, a lamp event is defined as any instance that the lamp was turned on. It is of course possible that a single lamp has, on average, multiple events in a single hour, since a lamp can be turned on and off any number of times. As with kerosene lamps, there was a clear diurnal rhythm of usage, generally with the largest number of events in the evening (around 19:00 hrs) and another peak with fewer events in the morning (around 6:00 hrs).

Figure 6. The average number of solar lamp events for each hour of the day across households (grey lines). The averaging window for each house covers the day the solar lamps were provided to a household (varies) to the final day of the study when all loggers were collected (same across all houses, with the exception of one lamp that was lost between follow-up and study end). The blue dots are the averages across all houses during each hour of a day.

25

All the homes used the solar lamps. Figure 7 shows the average number of hours per day across all monitored days that each of the 60 solar lamps was used, standardized to start on the day the solar lamp was distributed. It shows a very modest rise in hours of average usage after the lamps are distributed, but quickly stabilizes at about 5 hours per lamp per day, equivalent to roughly 15 hours of solar lamp-hours per day, per house. It’s unclear whether this represents a full discharge of the lamp battery each day. Estimated service hours for the Sun King Eco vary from 4 to 30 hours per charge, depending on the lamp intensity selected (3 levels are available).

Figure 7. Number of daily hours of use of each solar lamp in the study, standardized to start from the day each lamp was distributed, and for the 22 days following. Blue dots represent daily averages.

A transition in the pattern of solar lamp usage was observed over the first 5-6 days following solar lamp deployment (Figure 8). The first few days see a higher proportion of very short events (less than 1 minute), which might be expected as household members accustom themselves to the new lamps and how they work. After the first week of ownership, the number of events less than 1 minute in duration stabilizes to what might be considered a steady-state level.

Solar lamp usage data showed that an average of 30% of lighting events lasted less than a minute. These short events did not comprise a major fraction of total usage (minutes of use), but may highlight a lighting task previously less available to the home—the ability to switch a lamp on for a short period, perhaps to locate an item or to walk between buildings in the household compound. The sampling rate for the ibuttons on kerosene lamps was quite coarse (every 10 minutes), so from our data we cannot directly infer that kerosene lamps were not used for short events. However, from field observation and discussion with households, it

26 seems unlikely that users would so frequently go through the effort of lighting a kerosene lamp for such a short period of use.

Figure 8. Proportion of solar lamp usage events lasting less than 1 minute across all households (grey lines), standardized to start from the first day of lamp deployment. Blue dots correspond to the average across all houses each day from launch, and the black line a loess fit with 95% confidence bands.

Displacement of kerosene lamp usage by solar.

A key question that requires combined consideration of both kerosene and solar lamp usage data is: to what extent did solar lamps displace kerosene lamp usage in the study households? For this we considered 3 time periods: the baseline period when exclusively kerosene was used for lighting, the follow-up period about 4 weeks after solar lamp distribution, when households had contact with the field team again and personal and micro-environmental monitoring was carried out, and the week before the follow-up monitoring took place. The reason for the latter was because of the at-least-theoretical possibility that households could change their lamp- using behaviour if they were in contact with the field team during the follow-up monitoring period.

27

Figure 9 graphically represents the lamp usage data obtained from sensors placed on the kerosene and solar lamps of the households in the study. The vertical axis represents average daily household hours of light (including any non-use days) across the three study periods used for this analysis. It shows a number of things: (1) A high degree of displacement of the kerosene lamps between baseline and follow-up, with very little continuing kerosene lamp use; (2) virtually no difference in lamp use characteristics between the follow-up period and the week before follow-up; and both groups of houses having a similar level of lighting use at follow-up. As a frame of reference, the time window encompassing baseline to the end of follow-up for any home was five weeks.

Figure 9. Boxplots of the average household hours of lamp operation per day during the baseline phase, one week prior to follow-up measurements, and the follow-up phase, based on lamp usage sensors on kerosene and solar lamps. Data are paneled by the number of kerosene lamps in the home at baseline: either two lamps (N = 12 households) or three to five lamps (N= 8). “Kero” and “Solar” represent the exclusive lighting contribution of each lamp type, while “Total” represents the sum of lighting contributions by both solar and kerosene lamps in the homes. All households received three solar lamps regardless of the number of kerosene lamps at baseline.

At baseline, total household (kerosene) lamp use across the full cohort averaged 10.8 hours per day (SD = 6.0). Total lamp usage was greater among houses with 3-5 lamps than in those with two lamps by an average of 7.4 hours per day (p < 0.02), but the usage per lamp was the same in both house groups (3.9 hours per lamp per day, SD = 1.3). Simply put, lamp use averages approximately 4 hours per lamp, irrespective of the total number of lamps in a household. This does not imply that lamp use is balanced across all the lamps in a home.

Table 5 summarizes the underlying data on which Figure 9 is based. Differences between kerosene usage at baseline and both follow-up periods indicate a kerosene lamp use displacement of over 90% in terms of the average hours of usage. In the houses with two kerosene lamps, total lamp usage in the follow-up period roughly doubled (7.8 to

28 approximately 15 hours/day) with over 95% of those service hours met by solar lamps. Houses with 3 to 5 lamps exhibited a similar level of kerosene displacement, but with little change from baseline to follow-up in the total household hours of lighting used per day.

Table 5. Hours of lamp use per day for kerosene and solar lamps and related metrics by study period and number of kerosene lamps in the household at baseline. Values in parentheses correspond to one standard deviation. FollowUp - Kerosene lamps Baseline 1week FollowUp 2 Lamps at Baseline Kerosene Hours Only hrs/day 7.8 (2.6) 0.7 (0.8) 0.4 (0.6) (N=12) Solar Hours Only hrs/day - 15.2 (5.1) 14.1 (6.2) Total Hours hrs/day 7.8 (2.6) 15.8 (5.2) † 14.5 (6.4) † Change in Total1 hrs/day - 8.0 (4.6) † 6.7 (5.5) † Kerosene Percent of % of Total2 Total 100% (0%) 4% (6%)† 3% (5%)† 3-5 Lamps at Baseline Kerosene Hours Only hrs/day 15.2 (7.1) 1.6 (2.3) 0.5 (0.7) (N=8) Solar Hours Only hrs/day - 15.3 (6.2) 14.3 (5.1) Total Hours hrs/day 15.2 (7.1) 16.9 (7.9) † 14.8 (5.6) † Change in Total1 hrs/day - 1.6 (6.2) † -0.5 (5.0) † Kerosene Percent of % of Total2 Total 100% (0%) 7% (8%)† 3% (3%)† All (2-5 Lamps) Kerosene Hours Only hrs/day 10.8 (6.0) 1.0 (1.6) 0.4 (0.7) (N=20) Solar Hours Only hrs/day - 15.2 (5.4) 14.2 (5.7) Total Hours hrs/day 10.8 (6.0) 16.2 (6.2) † 14.6 (5.9) † Change in Total1 hrs/day - 5.5 (6.0) † 3.8 (6.3) † Kerosene Percent of % of Total2 Total 100% (0%) 5% (7%)† 3% (4%)† † Significantly different from baseline study phase (p-value < 0.05, Student’s Two-Tailed Paired T-test) 1 Change in total light hours from baseline 2 The percent of total lamp use attributable to kerosene after baseline was not significantly different by lamp group or study periods (p-value > 0.05) Micro-environmental Monitoring Four-day baseline measurements of PM2.5 indicated a strong influence of particulate sources in all three microenvironments—the kitchen, the main living area, and the school pupil’s bedroom. All kitchens were in separate buildings from the main living areas and bedrooms. Baseline PM2.5 events in both the main living spaces and school pupil’s bedrooms were consistent with the use-monitor (ibutton) patterns for kerosene lighting, both in terms of their peak shapes and the times of day that the peaks occurred.

Figure 9 shows examples of typical PM2.5 profiles from the main living area where kerosene lamps are the dominant PM source, and the kitchen where wood fires are the dominant source.

29

The generation of light from combustion is predicated on the production of particles7 (Lam et al. 2012b); thus kerosene lamps have a relatively stable rate of particulate emissions resulting in more sustained (longer) peaks. By contrast, a wood fire transitions through several combustion phases, with the majority of service coming from heat, not light. Thus, the flaming phase occurs at ignition, but quickly transitions to a char phase where heat, but less PM, is generated. PM emissions from fires are characteristically less stable because of how the fuel is broken down during the combustion process. This is reflected in the PM2.5 concentrations as more variable, with less sustained peaks. As substantially more fuel is burned in a wood fire, PM2.5 levels are typically several times higher than from a kerosene lamp, given the same room conditions8.

Figure 9. Examples of real-time PM2.5 concentration profiles from the main living space in which the dominant PM source is the kerosene lamps, and the kitchen in which the dominant source is the wood stove. The red bar corresponds to the period during which personal samples were also being taken. Note the order of magnitude difference in vertical scales between the main living room and the kitchen.

The pre-intervention pattern of PM2.5 in the main living area is shown in Figure 10, for each household, averaged across 5 days of monitoring. Blue dots show the average concentration

7 A concept first documented over 150 years ago by English scientist Michael Faraday: “all bright flames contain these solid particles; all things that burn and produce solid particles, either during the time they are burning, as in the candle, or immediately after being burnt, as in the case of the gunpowder and iron filings—all these things give us this glorious and beautiful light” (The Chemical History of a Candle, 1860) 8 The rate of PM emitted from a pollutant source is a function of the rate fuel is consumed and the fraction of the fuel that is converted to PM in the combustion process. The fraction of fuel carbon emitted as PM2.5 from a kerosene lamp is actually 2-3 times greater than that of a wood fire. However, the lamp fuel consumption rate is roughly one to two orders of magnitude smaller than that of the fire.

30 across households during the 24 hours of the day. The bimodal pattern reflects the pattern shown by the ibuttons, which were used to monitor kerosene lamp use during that period. Similar bimodal patterns of PM2.5 were apparent in both the school pupils’ rooms and the kitchens, although not at the same times of day. Figures showing corresponding PM2.5 patterns for these rooms are presented in Appendix 5.

Figure 10. Four-day baseline PM2.5 concentrations in the main living areas averaged over each hour for each house (grey lines). Large blue circles correspond to the average across houses at each hour of a day.

Figure 11 shows average living area, kitchen and school pupil’s bedroom concentrations separately for each household during both the baseline and follow-up micro-environmental

31 monitoring periods. Houses (1 to 20) are ordered by decreasing baseline concentration in the living room. This order is maintained for all bar charts throughout this report. Although either baseline or follow-up concentrations are missing for a few households, it shows a reduction in average PM2.5 concentration in living rooms and school pupils rooms in almost all households for which both baseline and follow-up data were collected. The one exception is household 18, for which the follow-up concentration in the school pupil’s bedroom was slightly higher than the baseline concentration—but both were low. Kitchens show no obvious pattern of change from baseline to follow-up.

32

3 Figure 11. Paired four-day average PM2.5 concentrations (µg/m ) during baseline (blue) and follow-up (red) in the main living area (top), kitchen (middle) and school pupil’s bedroom (bottom). Houses (1 to 20) are ordered by decreasing baseline concentration in the living room. Note the differences in the vertical scales. Baseline kitchen measurements were not obtained in the first five households recruited into the study (households 4, 6, 11, 12, 20) and follow-up measurements were also not obtained for household 20. In the main living area, baseline data from one house and follow-up data from three houses were lost due to instrument error.

33

Table 5 shows summary PM2.5 data for the three monitored areas. Results indicate that, on average, the main living areas of households experienced a 105 µg/m3 reduction in the 4-day average PM2.5 concentrations after the introduction of the solar lamps--a 61% reduction from baseline (p < 0.05). The pupils’ rooms, which are generally much smaller in size, experienced a larger reduction of 202 µg/m3-- a 79% reduction from baseline (p-value = 0.0003), although this reduction was heavily influenced by two houses with baseline PM2.5 concentrations above 1000 µg/m3. Removal of these two observations results in an average reduction of 66 µg/m3 (55%) in the pupils’ rooms (p-value < 0.05). Changes in the kitchen PM2.5 concentrations were only slightly different between baseline and follow-up (p-value = 1.0).

3 Table 5. Four-day average indoor PM2.5 concentrations (µg/m ) in the main living areas, school pupils’ bedrooms, and kitchens at Baseline and Follow-Up periods. 95% Conf. N Mean (Std. Deviation) Median Min Max Interval Main Room Baseline 17a 174 (143) 111 78 558 100, 247 Follow-Up 68 (25) 60 47 154 55, 81 Difference -105 (144) -44 -11 -508 -179, -31 Percent change -61% p-valuec 0.003 Pupil’s Room Baseline 19a 256 (418) 98.4 64.9 1648.8 54, 457 Follow-Up 54 (13) 50.0 45.0 96.2 48, 60 Difference -202 (421) -51 25 -1602 -405, 0.6 Percent change -79% p-valuec 0.0003 Kitchen Baseline 15b 1008 (1122) 638 268 4631 387, 1629 Follow-Up 968 (745) 766 259 2953 555, 1380 Difference -41 (646) 36 1,096 -1,678 -398, 317 Percent change -4.0% p-valuec 1.0 a Three and one households for the main room and the pupil’s room, respectively, are excluded from analysis due to instrument errors resulting in data loss at either baseline or follow-up. b. Kitchens in the first four households were not measured. C Wilcoxon's Signed Rank test

The World Health Organization (WHO) recommends a 24-hr PM2.5 indoor air quality guideline of 25 µg/m3 (WHO 2010). Thus, average follow-up concentrations in both the main living areas and the school pupils’ rooms still exceed that level. However, the average follow-up indoor PM2.5 levels are around likely ambient concentrations, which might be expected in the absence of indoor sources of smoke. Ambient sources are likely to include drifting smoke from cooking fires in the surrounding area and from traffic emissions. Dust from floors and furniture suspended in the air as a result of indoor movements could also contribute to indoor concentrations.

Kitchens show a proportionately much smaller change in PM2.5. This was expected, as the main source of PM2.5 in the kitchen is the cooking fire and we made no efforts to modify this source. A priori, solar lamps would not be expected to change cooking practices directly. Reduced use

34

of kerosene lamps in the kitchen might account for the small reduction in average kitchen PM2.5 concentrations we observed, but it may just have been a result of random variation. As all kitchens were in separate buildings, as is traditional in much of rural Kenya, we expected only limited impact of the kitchens on living space PM2.5 concentrations.

Table 6 shows changes in CO concentrations in the main living areas and the pupils’ bedrooms. No CO measurements were taken in the kitchens, which were in separate buildings. Levels in both micro-environments are very low, in most cases never exceeding 1 ppm for more than a minute at any point of the sampling period. This is reflected in Table 6 by the median and minimum levels of CO in both study phases being zero for all but the pupil’s room baseline. The average absolute change, however, is extremely small and within the range of instrument error: 48-hr average CO levels at baseline and follow-up in both locations, were approximately 23-35 times lower than the WHO Indoor Air Guideline levels of 7 ppm over 24-hrs (WHO 2014). Low CO levels were not unexpected given the measured emission rates of CO from kerosene lamps in previous studies (Lam et al. 2012a).

Table 6. Four-day average indoor CO concentrations (ppm) in the main living areas and school pupils’ bedrooms at baseline and follow-up periods. Mean 95% N (Std. Deviation) Median Min Max Conf. Interval Main Room Baseline 20 0.2 (0.4) 0 0 1.8 0.2, 0.4 Follow-Up 0.3 (0.6) 0 0 2.5 0.1, 0.6 Difference 0.1 (0.7) 0 -2 1.2 -0.2, 0.4 Percent change 50% p-value† 0.443 Pupil’s Room Baseline 20 0.3 (0.2) 0.3 0.0 0.9 0.2, 0.5 Follow-Up 0.2 (0.3) 0.0 0.0 0.9 0.04, 0.3 Difference -0.2 (0.4) 0 0 0.8 -0.3, -0.002 Percent change -67% p-value† 0.0463 † Wilcoxon's Signed Rank test

Personal Monitoring At baseline and follow-up, both adults and school pupils were outfitted for 48 hours with a vest containing a MicroPEM PM2.5 monitor and a Lascar CO monitor (Figure 1). School pupils were not required to wear these vests when they were at school and participants were instructed to lay the vests down beside the bed when sleeping at night.

Figure 12 shows, by household, average 48-hour personal monitoring concentrations of PM2.5, at both baseline and follow-up for both school pupils and adult lamp users who wore the vests. As with the micro-environmental monitoring, houses are ordered in decreasing order of micro- environmental PM2.5 monitoring concentration in the kitchen at baseline.

35

Figure 11. Baseline and follow-up average 48-hr personal exposure PM2.5 concentrations (µg/m3) for adult lamp users (top) and school pupils (bottom).

At baseline, adults experienced average PM2.5 exposure levels roughly 60% greater than pupils. This was not unexpected, considering that adults in our study were mostly mothers or caregivers who did the majority of cooking, exposing them to smoke from the open fires. Moreover, pupils were not young children and thus were not present in the kitchen during most of the cooking. Substantial PM2.5 exposure reductions are apparent, particularly for the

36

school pupils. Examples of the PM2.5 personal monitoring profiles, for the same school pupil, at both baseline and follow-up, are shown in Appendix 5.

The data from Figure 12 are summarized in Table 7, showing an average reduction in personal 3 exposure of approximately 100 µg/m PM2.5 for both adults and school pupils, representing a 50% reduction for adults and nearly 75% reduction for pupils. The school pupil data need to be interpreted in the light of the fact that they did not wear their exposure monitoring vests when they were at school. We have no data on exposures at school, but it is not unlikely that they were closer to ambient exposures levels. If that is the case, since the MicroPEM continues to operate even when the vest is not being worn and is likely to be measuring ambient PM2.5 levels, the daily record is likely to be a reasonable reflection of the pupil’s actual exposure over the day. It is also unlikely that unmeasured sources of community pollution affecting pupils would be greater than, for example, the open fire used by adults in our study population, who still experienced a 50% reduction in personal PM2.5 exposure.

3 Table 7. Two-day average personal PM2.5 concentrations (µg/m ) during baseline and follow- up periods. N Mean (Std. Deviation) Median Min Max 95% Conf. Interval Adults Baseline 19† 210 (107) 193 43 443 158, 261 Follow-Up 104 (59) 95 22 205 75, 132 Difference -106 (114) -72 37 -390 -161, -51 Percent change -50% p-value‡ 0.003 Pupils Baseline 19† 132 (99) 105 43 458 83, 179 Follow-Up 35 (27) 29 10 108 22, 48 Difference -97 (105) -80 14 -442 -147, -46 Percent change -73% p-value‡ 0.002 † One individual excluded due to instrument error at either baseline or follow-up ‡ Wilcoxon's Signed Rank test

Personal exposure measurements of CO for adults and pupils both showed less than a 1 ppm reduction in average 48-hr exposure between baseline and follow-up (Table 8). Average exposure to adults was approximately 2-3 times greater than that of the pupils at both baseline and follow-up, which is likely a result of differences in exposure to CO from wood-fire cookstoves. As with micro-environmental CO concentrations, minimal effect of the solar lamps on CO exposure was expected, because of the low CO emission strength of the kerosene lamps. Given these results, it is likely that there were negligible reductions in CO exposure between baseline and follow-up from the kerosene lamps for either the adult or pupil, and what was measured was likely dominated by cookstove exposures. These results are nonetheless useful as an objective confirmation of expectations. The relatively low levels of risk posed by the levels of CO here suggest that CO should not be considered a priority pollutant in future studies assessing the effect of lighting changes.

37

Table 8. Two-day average personal CO exposures (ppm) at baseline and follow-up periods. 95% Conf. N Mean (Std. Deviation) Median Min Max Interval Adults Baseline 20 2.1 (1.3) 2 0 5.4 1.5, 2.7 Follow-Up 1.4 (0.9) 1 0 2.7 1.0, 1.8 Difference -0.7 (1.6) 0 -2 4.4 -1.5, 0.04 Percent change -33% p-value† 0.104 Pupils Baseline 20 0.6 (0.5) 0.5 0.0 1.8 0.4, 0.8 Follow-Up 0.5 (0.5) 0.4 0.0 1.7 0.2, 0.7 Difference -0.1 (0.6) 0 -1 2 -0.4, 0.2 Percent change -16% p-value† 0.765 † Wilcoxon's signed rank test

Another way of examining the observed changes in personal exposure is to look at the relationship between baseline levels and their associated reductions at follow-up. This is visualized with the scatterplot in Figure 13. Each point on the plot represents a single participant—either a school pupil or an adult lamp user. Fitted linear regression lines have been added to the scatterplot. Both regressions suggest linearly proportionate PM2.5 exposure reductions associated with the introduction of solar lamps. The line for the school pupils suggests that replacement of kerosene lamps with solar lamps reduces their typical exposure down to an average concentration of about 40 µg/m3, which may be close to the ambient level. The reduction from baseline exposures is less for adults, in part because their baseline exposures are higher, from cooking smoke, and possibly because they spend less time in close proximity to the lamps (e.g., they do not do homework). These factors would result in kerosene accounting for a smaller proportion of total PM2.5 exposure. The adults also exhibit more scatter about the best-fit line, which may be a result of wide variation in cooking exposures, in addition to their lighting-related exposures. Pupils, on the other hand, engage less frequently in cooking activities, and exposure changes are likely a function of fewer factors – probably mainly kerosene lighting.

38

Figure 13. Reduction in 48-hr personal PM2.5 exposure concentrations between baseline and follow-up, relative to baseline concentrations.

39

Discussion

As far as we are aware, this is the first field-based assessment of personal exposure to PM2.5 from kerosene lamps, the first to objectively quantify joint usage patterns of kerosene and solar lamps, and the first to estimate the extent to which such exposures can be reduced by transitioning to solar lamps. Although it has always been obvious that kerosene use, particularly for lighting, is associated with particulate matter--obvious in the form of soot that accumulates on ceilings and is present in nasal discharges—its relative contribution to pollutant exposure has been generally ignored, despite considerable research on household energy’s contributions to air quality and global disease burden. Current estimates in the Global Burden of Disease, for example, do not include the contribution of lighting activities as part of the risk factor “household air pollution” (Fan and Zhang 2001). The overwhelming focus of the relatively recent interest in developing country household exposures to pollutants has been on cooking with biomass (wood, crop waste and charcoal) and coal and, to a smaller extent, heating. There are several reasons why lighting has been absent in the understanding of household energy impacts to date. Cooking and heating fuels are combusted in much larger quantities and usually generate visible and pungent clouds of smoke. They also account for large fractions of national- and household-level energy consumption, making them important components of the inventories used for climate, air quality, and deforestation impact estimates. Kerosene lighting, on the other hand is used mainly at night in more private settings, consuming small quantities of fuel, and the smoke is less visible to the naked eye.

In contrast to solid-fuels, kerosene has historically been viewed as a much cleaner fuel, even though the odor of kerosene burning is distinctive and usually quite obvious to anyone nearby. Studies have shown that when kerosene, or other liquid fuels, are used for cooking, a less common use than lighting, women and their young children are much more likely to stay in the kitchen than if solid fuels are being used for the same purpose (Bates et al. 2013; Saksena et al. 2003). Thus, in relation at least to cooking, a simple comparison of kitchen pollutant concentrations associated with different fuels does not provide a good picture of relative fuel- related exposures. Similarly, when kerosene is used for lighting, room pollutant concentrations may not provide good estimates of exposure. This is particularly so if reading or school work is being done—the light from kerosene lamps is dim at around 1 lux or 1 lumen/m2 (Mills 2005), necessitating that they be kept in close proximity--potentially increasing pollutant exposure above what would be obtained from just being in the same room (Apple et al. 2010).

It is for these reasons, and more generally because people rarely stay in one place in the course of a day or an evening, that personal exposure monitoring is far preferable to micro- environmental monitoring (placing fixed monitors in a room) as an indicator of exposure. However, in practice, personal monitoring has been much less common than micro- environmental monitoring, for the reason that it is more difficult to achieve. To increase compliance and reduce participant burden, the personal monitoring equipment needs to be lightweight, quiet and not likely to cause embarrassment. This need has been particularly difficult to achieve with particulate matter, which does not behave like a gas and requires a pump and batteries to separate out respirable particles, such as PM2.5. Fortunately, in recent

40 years, a few devices have been developed which can fit these requirements. The MicroPEM is one such device. By concealing it in a tailor-made vest, we were able to make it acceptable to users, had good compliance with its wearing9, and collected useful and interpretable data.

This study was designed primarily as a proof of concept—firstly, to show that personal exposures to pollutants from kerosene lamps were indeed substantial and we had the means to quantify them. Without having done that, we would have no basis for designing a larger study intended to quantify health impacts. Basic to epidemiologic statistical power and sample size calculations is the need for a good idea of the magnitude of exposures that are to be expected and some reasonably well-grounded expectations of the likely associated risks for health outcomes. Although the present study collected only limited data on health outcomes—by way of symptom self-reports—there are other data that can be used to estimate plausible risk reductions associated with PM2.5 exposure reductions of the magnitude found in this study— around 100 µg/m3 as an average across 24 hours for both adults and school pupils. Although this may not seem like a major reduction, particularly for those who are exposed to cooking smoke, there is increasing evidence that PM2.5 from all sources is not the same in terms of health impact. In particular, there is evidence that PM2.5 from kerosene may be a more potent cause of health effects than PM2.5 from biomass burning.

Two studies from Nepal, where kerosene has been used for cooking and lighting, highlight the potential differences in kerosene- and biomass-derived PM2.5 impacts on health. In one case- control study of household air pollution and pulmonary tuberculosis (TB) in the city of Pokhara, cooking with biomass was associated with an odds ratio (OR)10 of 1.21 (95% confidence interval [CI]: 0.48, 3.45), but kerosene used for cooking had an OR of 3.36 (95% CI:1.01, 11.22) and kerosene for lighting an OR of 9.43 (95% CI: 1.45, 61.3) (Pokhrel et al. 2010). The authors interpreted the higher OR for lighting as possibly reflective of high exposure from close-up work with lamps.

The second study, a case-control study of acute lower respiratory infection (ALRI or child pneumonia), in the city of Bhaktapur, Nepal, had the unusual situation that approximately equal numbers of the 917 participating families had electricity, gas, kerosene or biomass as their primary household cooking fuels. Thus, very unusually and possibly uniquely for a developing country study, it was possible to use electricity as the baseline fuel and get separate relative risk estimates for ALRI associated with gas, kerosene and biomass cooking fuels (Bates et al. 2013). ORs for kerosene and biomass primary stoves were 2.33 (95% CI: 1.40, 3.86) and 2.13 (95% CI: 1.34, 3.41) respectively. In a separate publication from the same study that focused on kitchen PM2.5 concentrations, mean kitchen concentrations, after the subtraction of

9 We estimate a good compliance based on several factors: staff visited households daily during personal monitoring to conduct surveys and noted compliance levels upon arrival. Pupils logged the times they arrived home from school and put the vest on. Finally, the MicroPEM itself measures movement, which provides an indication of whether it was worn. 10 An odds ratio is an estimator of the relative risk—the risk in the exposed group divided by the risk in the unexposed group.

41 estimated outdoor concentrations, were 649 and 135 µg/m3 for biomass and kerosene, respectively (Pokhrel et al. 2015). The implication of this is that PM2.5 from kerosene may be several times more potent as a cause of at least some health conditions than equal concentrations of PM2.5 from biomass. Supporting this hypothesis is a growing body of literature suggesting that disease risk estimates based on black carbon, a component of particulate matter that varies from source to source, result in higher effect estimates than undifferentiated particulate matter (Baumgartner et al. 2014; Janssen et al. 2011). Particles emitted from kerosene lighting, and to a lesser extent kerosene cooking, contain some of the highest fractions of black carbon among measured pollutant sources (Lam et al. 2012b). This may have particular implications for the lamp-using adults in this study, whose PM2.5 exposure reduction associated with solar lamps might be seen as relatively small in the context of their major exposure to PM2.5 from cooking with biomass.

Finally, in providing context to the levels of exposure reductions we observed, it is important to consider what is already known about exposure-response relationships between PM2.5 levels and health outcomes. Recent work on the relationship between particulate exposure concentrations and various health outcomes suggests non-linear relationships are common, as opposed to the often-assumed linear relationships. Specifically, the non-linearity implies that the incremental health benefit from a unit reduction in PM2.5 exposure is likely to be greater when baseline levels are less than approximately 100-200 µg/m3, depending on the health outcome (Pope et al. 2009; Burnett et al. 2014). Thus, while there will be some benefit from exposure reductions anywhere along the exposure-response curve, the expected benefit of an identical absolute incremental change in exposure (e.g., 10 µg/m3) on health will not always be the same. Thus, considering the baseline and eventual follow-up levels of exposure is important when evaluating the potential benefits of a health intervention. One caveat in this situation, however, is that the existing evidence on non-linear exposure-response relationships has been derived from non-kerosene exposures – a combination of outdoor ambient concentrations (e.g., from gasoline and diesel emissions), cooking exposures (e.g., from biomass) and tobacco-related second-hand smoke, and tobacco smoking, to produce what are known as integrated exposure-response curves. It is possible that different exposure-response relationships apply to PM2.5 from kerosene sources, particularly if it is more toxic.

The baseline and follow-up personal exposure PM2.5 concentrations that we observed fell within the exposure window where some of the greatest incremental benefits from exposure reductions might be expected. School pupils, for example, showed 48-hr average PM2.5 personal exposure reductions from 132 µg/m3 to 35 µg/m3. Adults also had substantial reductions, but follow-up exposures still averaged in excess of 100 µg/m3, due probably to cooking-related exposures. These results suggest that solar lamp benefits are likely to differ across individuals in the same home and achieving “healthy” levels of exposure means considering not just lighting, but other aspects of the household energy system, particularly cooking.

Appendix 6 shows integrated exposure-response curves for several health outcomes, along with the measured personal exposure concentrations of adults and pupils in this study. It

42 provides some indication of the benefits that might be achieved if solar lamps displace kerosene lamps. It should be interpreted with great caution however, as this study contained only 20 pupils and 20 lamp-using adults and, therefore, confidence intervals around the exposure point estimates are wide.

A second important outcome of this study was clear support for our hypothesis that the solar lamps were acceptable to the families and that they would displace the kerosene lamps previously used. From our lamp use monitoring data, supported by PM2.5 concentration monitoring profiles, both from personal exposure and micro-environmental monitoring, this hypothesis appears to have been confirmed (Figure 9 and Table 5). Also, increased hours of lighting in the group of households with 2 kerosene lamps at baseline indicates that these houses found additional use from a third solar lamp, and that there were unmet lighting service needs at baseline. Households with 3-5 kerosene lamps at baseline simply replaced most of their baseline lighting service needs with solar, without increasing the number of hours of lighting. It is unclear whether there exist unmet lighting needs in the 3-5 lamp group that would have been met with a fourth solar lamp. What can be inferred, however, from the low level of kerosene use during follow-up periods, is that any remaining needs are relatively modest and not critical enough to necessitate significant continuing use of kerosene. The frequent use of the solar lamps for events lasting less than 1 minute also indicates the possible opening up of new tasks/services that might not be met with kerosene lamps because of the time and effort required for lighting them.

The small amount of continuing use of kerosene after solar lamp distribution could also indicate that selected tasks are preferentially met with kerosene – for example, exterior lighting or in the kitchen when the stove is being used. In such instances, adding lamps may do little to curb kerosene use. Supporting this, one respondent reported the head of household would not allow use of a solar lamp in the kitchen during cooking. This may be due to fears that the smoke from the fire would damage the lamp, an issue also observed by clean lighting practitioners in Nepal11. While evidence from this study suggests that the conditions where kerosene lamps continued to be used contributed only modestly to exposure, it remains important to understand drivers of continued kerosene usage.

The 90% displacement in terms of the average hours of usage of traditional kerosene lamps found in this study is much higher than the displacement levels documented from improved cookstove distribution programs, which generally have an upper-bound of 40-60 % (Ruiz- Mercado et al. 2015). There may exist lessons from lighting that can enhance other programs aimed at improving residential energy services. As a critical household energy service need, the high displacement measured here is encouraging for the household energy sector as a whole, but merits further examination to determine whether these benefits and use trends are sustained. Ideally, demonstration of long term and sustained exposure benefits should be one objective of any study that follows this one.

11 Personal correspondence: Sandeep Joshi, Center for Rural Technology, Nepal.

43

Of importance to kerosene lamp displacement and exposure reductions, we believe, was the provision of 3 solar lamps per household. Most households have multiple kerosene lamps and lamp users and, if we had provided only one lamp per household, we would have had less control over who used it and children doing their homework might not have benefited at all.12 In any case, substantial kerosene lamp use would likely have continued within the households. Other related studies of which we are aware, although few, have focused on the economic implications of solar lamp introduction (Kudo et al. 2015; Grimm et al. 2014) or reduction in burns (Chamania et al. 2015) and have mostly provided only one solar lamp per household. This may have limited the measured benefits.

A third objective of this study was to test some basic symptom questions to see if they might be useful in a questionnaire for a larger study with more of a focus on health impacts of solar lamp provision. We did not expect to be able to detect any changes in health status with any degree of statistical confidence. However, what we found was a complete remission of symptoms reported at baseline.

This reduction in ocular and respiratory symptoms after transitioning to solar lamps is striking, but difficult to interpret. It is likely that the reduction in exposure to kerosene lamp emissions was at least partly responsible. However, usually symptoms of the type we inquired about have multiple causes and we would not expect replacement of the kerosene lamps with solar lamps to completely eliminate all such symptoms. Therefore, it is possible that the reported symptom reduction was, at least in part, a manifestation of the so-called “Hawthorne effect”, in which knowledge of the investigation and assumptions about what the investigators hoped to see influenced symptom reporting (Chen et al. 2015; McCambridge et al. 2014). In any case, there is no obvious way to distinguish this possibility from a real reduction in symptom etiology. Despite this uncertainty, we believe that the questions were properly understood and, with some modifications, will be useful in a further study, if that takes place. Such a study would likely include more objective measures of health status, less susceptible to a Hawthorne effect, such as spirometry.

We conclude that this study has been successful in almost all of the ways that we had hoped. The intention to obtain some illuminance and visual acuity data was unachievable, for safety reasons, but we believe with some study design modifications it will be possible to safely implement it in a future study. Most importantly, this study has shown (i) that kerosene lamp use in Busia county, Kenya, is associated with substantial measureable exposure to PM2.5, both in adult and school pupil lamp users; (ii) these exposures are of such a magnitude that they have high potential to cause adverse health effects; and (iii) provision of at least 3 solar lamps per household provides a potentially very successful means of reducing these exposures and likely mitigating health impacts of household air pollution. Results also indicate that there is minimal value in monitoring CO and in any future study the focus should be on PM2.5.

12 Though there is some evidence from SolarAid research that school-going children are prioritised for use of the solar light in the household.

44

The present study used a purposive selection of participating families, based on certain selection criteria. Therefore, we do not consider it necessarily to be a representative sample, even among families that fit our selection criteria. Importantly, the conditions that led to the exposure reductions observed in this study may not be the same across all kerosene-using communities. In rural contexts, where kerosene is most likely to be used, the effectiveness of solar lamps may be reduced if the open cooking fire is in or close to the room in which household members gather or sleep. In Kenya, this may be less of a concern since over 70% or rural houses report cooking in separate buildings or outdoors (USAID 2015). Nonetheless, we used multiple methods to collect our data and there was sufficient internal consistency (e.g., between use monitor and PM2.5 monitor results) and plausibility of our findings that we believe that the results are sufficient to provide prima facie evidence of likely health harm from kerosene lamp use and benefits of providing solar lamps to displace kerosene lamp use.

Demonstrating such health improvements and the sustainability of any such solar lamp intervention would need to be the objective of a much larger and longer study, for which the present study provides a basis for design and sample size calculation. We consider the recruitment of participating families via a high school to have been a very successful approach, which we highly recommend for any such future study.

45

Acknowledgements

This study was funded by award number 20150370 from Google Ireland Limited and SolarAid (London).

The authors gratefully acknowledge the assistance of the study participants and their families. We are also very appreciative of the assistance with many aspects of this study by numerous organizations and individuals, including St. Peter’s Budokomi Mixed Secondary School, Busia County, Kenya, with particular thanks to head teacher Leonard Oduor, head of science Geib Osinde, and the teaching staff; the Kenya Medical Research Institute (KEMRI), particularly Director Dr. Solomon Mpoke and Everlyne Wambai; the Busia County Ministry of Education Office; the Center for Integrated Research and Community Development Uganda (CIRCODU), particularly Director-General Joseph Arineitwe and Archibald Mutaremwa; the SunnyMoney Kenya office and field staff; and the New Zealand Translation Centre, Wellington, NZ. Other individuals who helped this study immensely are Nancy Smith of the School of Public Health, U.C. Berkeley; Adina Rom and Yael Borofsky of the Swiss Institute of Technology, Zurich; Tracy Allen of EME Systems, Berkeley; Ryan Chartier of RTI; Tami Bond of the University of Illinois; Samantha Dalapena, David Pennise and Dana Charron of the Berkeley Air Monitoring Group; and Alberto Calatroni of Bonsai Systems GmBH, Zurich, Switzerland.

46

References

Apple J, Vicente R, Yarberry A, Lohse N, Mills E, Jacobson A, et al. 2010. Characterization of particulate matter size distributions and indoor concentrations from kerosene and diesel lamps. Indoor Air 20(5): 399-411. Apte JS, Marshall JD, Cohen AJ, Brauer M. 2015. Addressing Global Mortality from Ambient PM2.5. Environ Sci Technol 49(13): 8057-8066. Bates MN, Chandyo RK, Valentiner-Branth P, Pokhrel AK, Mathisen M, Basnet S, et al. 2013. Acute lower respiratory infection in childhood and household fuel use in Bhaktapur, Nepal. Environ Health Perspect 121(5): 637-642. Baumgartner J, Zhang Y, Schauer JJ, Huang W, Wang Y, Ezzati M. 2014. Highway proximity and black carbon from cookstoves as a risk factor for higher blood pressure in rural China. Proceedings of the National Academy of Sciences 111(36): 13229-13234. Burnett R, Pope CA, Ezzati M, Olives C, Lim S, Mehta S, et al. 2014. An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure. Environmental health perspectives 122(4): 397-403. Chamania S, Chouhan R, Awasthi A, Bendell R, Marsden N, Gibson J, et al. 2015. Pilot project in rural western Madhya Pradesh, India, to assess the feasibility of using LED and solar-powered lanterns to remove kerosene lamps and related hazards from homes. Burns 41(3): 595-603. Chen LF, Vander Weg MW, Hofmann DA, Reisinger HS. 2015. The Hawthorne Effect in Infection Prevention and Epidemiology. Infect Control Hosp Epidemiol 36(12): 1444-1450. Epstein MB, Bates MN, Arora NK, Balakrishnan K, Jack DW, Smith KR. 2013. Household fuels, low birth weight, and neonatal death in India: the separate impacts of biomass, kerosene, and coal. Int J Hyg Environ Health 216(5): 523-532. Fan C-W, Zhang J. 2001. Characterization of emissions from portable household combustion devices: particle size distributions, emission rates and factors, and potential exposures. Atmos Environ 35(7): 1281-1290. Fine PM, Cass GR, Simoneit BR. 1999. Characterization of fine particle emissions from burning church candles. Environmental science & technology 33(14): 2352-2362. Grimm M, Munyehirwe A, Peters J, Sievert M. 2014. A first strep up the energy ladder? Low cost solar kits and household's welfare in rural Rwanda. (IZA Discussion Paper).Institute for the Study of Labor. Janssen NAH, Hoek G, Simic-lawson M, Fischer P, Bree LV, Brink H, et al. 2011. Review Black Carbon as an Additional Indicator of the Adverse Health Effects of Airborne Particles Compared with PM 10 and PM 2 . 5. 119(12): 1691-1699. Kudo Y, Shonchoy A, Takahashi K. 2015. Impacts of Solar Lanterns in Geographically Challenged Locations: Experimental Evidence from Bangladesh. (IDE Discussion Paper).Institute of Developing Economies.

47

Lam NL, Smith KR, Gauthier A, Bates MN. 2012a. Kerosene: a review of household uses and their hazards in low- and middle-income countries. J Toxicol Environ Health B Crit Rev 15(6): 396-432. Lam NL, Chen Y, Weyant C, Venkataraman C, Sadavarte P, Johnson MA, et al. 2012b. Household Light Makes Global Heat: High Black Carbon Emissions From Kerosene Wick Lamps. Environmental Science & Technology 46(24): 13531-13538. Lam NL, Pachauri S, Purohit P, Nagai Y, Bates MN, Cameron C, et al. 2016. Kerosene subsidies for household lighting in India: what are the impacts? Environmental Research Letters 11(4): 044014. Lozier MJ, Sircar K, Christensen B, Pillarisetti A, Pennise D, Bruce N, et al. 2016. Use of Temperature Sensors to Determine Exclusivity of Improved Stove Use and Associated Household Air Pollution Reductions in Kenya. Environ Sci Technol 50(8): 4564-4571. McCambridge J, Witton J, Elbourne DR. 2014. Systematic review of the Hawthorne effect: new concepts are needed to study research participation effects. J Clin Epidemiol 67(3): 267-277. Mills E. 2005. The specter of fuel-based lighting. Science 308(5726): 1263-1264. Pillarisetti A, Vaswani M, Jack D, Balakrishnan K, Bates MN, Arora NK, et al. 2014. Patterns of stove usage after introduction of an advanced cookstove: the long-term application of household sensors. Environ Sci Technol 48(24): 14525-14533. Pokhrel AK, Bates MN, Acharya J, Valentiner-Branth P, Chandyo RK, Shrestha PS, et al. 2015. PM 2.5 in household kitchens of Bhaktapur, Nepal, using four different cooking fuels. Atmos Environ 113: 159-168. Pokhrel AK, Bates MN, Verma SC, Joshi HS, Sreeramareddy CT, Smith KR. 2010. Tuberculosis and Indoor Biomass and Kerosene Use in Nepal: A Case-Control Study. Environ Health Perspect 118(4). Pope CA, Burnett RT, Krewski D, Jerrett M, Shi Y, Calle EE, et al. 2009. Cardiovascular mortality and exposure to airborne fine particulate matter and cigarette smoke shape of the exposure-response relationship. Circulation 120(11): 941-948. Ruiz-Mercado I, Canuz E, Smith KR. 2012. Temperature dataloggers as stove use monitors (SUMs): Field methods and signal analysis. Biomass Bioenergy 47: 459-468. Ruiz-Mercado I, Canuz E, Walker JL, Smith KR. 2013. Quantitative metrics of stove adoption using Stove Use Monitors (SUMs). Biomass Bioenergy 57: 136-148. Saksena S, Singh P, Prasad RK, Prasad R, Malhotra P, Joshi V, et al. 2003. Exposure of infants to outdoor and indoor air pollution in low-income urban areas: A case study of Delhi. J Expo Anal Environ Epidemiol 13(3): 219-230. Schare S, Smith KR. 1995. Particulate emission rates of simple kerosene lamps. Energy For Sustain Dev 2(2): 32-35. USAID. 2015. Demographic and Health Surveys.

48

WHO. 2006. WHO Air quality guildelines for particualte matter, ozone, nitrogen dioxide and sulfur dioxide: Global update 2005. Geneva World Health Organization WHO. 2010. WHO Guidelines for indoor air quality: Selected pollutants. Geneva World Health Organization WHO. 2014. Indoor air quality guidelines: household fuel combustion.: World Health Organization. Zai S, Zhen H, Jia-song W. 2006. Studies on the size distribution, number and mass emission factors of candle particles characterized by modes of burning. Journal of aerosol science 37(11): 1484-1496.

49

Appendices

Appendix 1: Solar lamps and voltage loggers

Households were given three Sun King Eco solar lamps (Greenlight Planet Inc., U.S.A.). On a full charge, lamps provide 4 to 30 hours of light, depending on the light intensity setting. Each lamp comes with a small photovoltaic panel, and can be detached from the stand for use as a portable lamp/torch.

Figure A1.1. Sun King Eco solar lamp (left). Three lamps were given to each household following baseline measurement. Usage loggers were installed inside of the lamp casing (right)

50

Appendix 2: Micro-environmental measurements in the kitchens and pupils’ rooms

Figure A2.2. Four-day baseline PM2.5 concentrations in the kitchens (top) and school pupils’ rooms averaged over each hour for each house (grey lines). Large blue circles correspond to the averages across houses at each hour of a day.

51

Appendix 3: Development of the Light Type Detector (LTD)

Two prototype LTDs were produced in collaboration with Dr. Tracy Allen of EME Systems (Berkeley, CA, USA). The intended purpose of this effort was to design a low-cost monitor that could objectively distinguish between an LED or fuel-based light sources during operation in a room. It would also log these readings in real-time, providing a timestamped record of light use. It was decided that the current version (v. 1.0) and future iterations should meet the following design and application criteria:

 Provide measurements sufficient for discernment between any use of LED and flame- based light sources in a room.  Be discrete and unobtrusive (e.g. silent) to household members  Be robust  Permit unattended operation for a minimum of 24-hrs on battery power  Have low cost (< $500 per unit)

The current prototype incorporates two sensors for measuring light: a low-level lux sensor and an infrared (IR) sensor. The lux sensor provides a general measure of illuminance while the IR sensor can be used to distinguish the presence of open flames in the evenings when there is no sunlight. Levels of both lux and IR are integrated over the focal window and logged at a user- specified time interval, with a minimum resolution of 1 second. Version 1.0 of the monitor weighs approximately 200 grams. The device is powered by a rechargeable lithium-ion battery (Figure A3.1b) which, when fully charged can be deployed for 1-3 weeks, depending on the programmed scanning interval. Data are logged to a removable SD card as a text file, which can then read easily into most analysis and spreadsheet software.

Figure A3.1a. Version 1.0 light meter Figure A3.1b. Version 1.0 light meter with top showing the sensor lens, charging port and removed, showing the lithium-ion battery SD card port. A small hole to the left of the resting atop the datalogger. power input blinks to indicate that the device is logging. The other small hole is the manual reset.

52

Controlled Testing

In June 2015, a series of controlled lab experiments were performed in the city of Busia, Kenya to provide initial characterization of:

1. The ability to distinguish between a LED and a flame light source 2. Characterize the monitor (sensors) response to distance to the light sources, in order to determine the current maximum operational distance.

Figure A3.2 shows the lux and IR response from a single set of experiments in which an LED light source or kerosene lamp light source was placed at various distances from the sensor. The height of the sensor port and center of each light source were approximately constant across all tests. As expected, the open flame sources showed a distinctly higher IR to lux ratio relative to the LED source, although IR was not entirely absent from the LED. The IR to lux ratio, however, is approximately constant within each light source across tested distances. Both the IR and lux sensor response decayed exponentially with distance from the light source. As a result, v1.0 showed an operational distance of at least 4 feet in these early experiments. Future experiments are planned to establish whether the angle of the sensor affects performance and whether it is possible to increase the operational distance in order to allow for the monitor to be placed on a wall of a room.

Figure A3.2. Real-time output of the light meter lux and IR response.

53

Based on these initial field tests, the prototype light detector showed a promising ability to distinguish between kerosene and LED light sources. In its current state, however, limitations in its operational effectiveness would restrict its usability in actual houses, since the device would need to be mounted very close to any light source. Next steps in development will need to emphasize increasing the detectable range of the light detector so the device can be mounted to a wall overlooking a larger indoor space.

54

Appendix 4: Visual acuity measurement protocol We use the Colenbrander Mixed Contrast Card Set for reading distance 40 cm (16 inches).

1. First encounter with participant.

1.1. Ensure the room is dark; if necessary by using shades to block light from entering the room.

1.2. With only the kerosene lamp operating, ask the participant to arrange the chart so that he/she can see it optimally and then read (both eyes with normal correction—that is, usual glasses used for reading) high contrast numbers (black numbers). Any participants who cannot read numbers will use the tumbling Es on the other side of the card. Say: “I want you to try to read the smallest numbers that you can. Show me where you see them best. Now begin reading from the top of the chart. Only read the black letters.”

1.3. Participant reads the numbers until they can read no more. Rule for stopping is if the participant reads 3 or more letters correctly on a row, the tester encourages an attempt at the next row. Read: “Can you read anything at all on the next row? It is OK to guess if you are not quite sure”.

1.4. We record on the grading sheet: 1.4.1. Number of numbers or Es successfully read. 1.4.2. The card distance from the eyes when they are reading or attempting to read the smallest print they can read. (measure in cm from the eyes to the smallest print) 1.4.3. The lux at the card surface at the time of reading

1.5. Procedures 1.1 to 1.4 are repeated with the low contrast (grey) numbers or E’s. We record the same data.

1.6. Participant is then asked to use the cord to position the card 40 cm from his/her cheek, square on. Participant positions him/herself as best they can in relation to the lamp.

1.7. We record the same data for this set distance, as in 1.4.

1.8. The participant is provided with a solar lamp, which is set to maximum brightness.

1.9. Procedures 1.1 to 1.7 are repeated with the solar lamp.

2. Second meeting with participant after they have had a period of 2-4 weeks to use the solar lamp.

2.1. We repeat procedures 1.1 to 1.7, but only with the solar lamp.

55

Appendix 5: Personal exposure profile examples

Figure A5.1. Example of changes in personal exposure profile for PM2.5 between baseline (top) and follow-up (bottom) for one school pupil (House 20).

56

Appendix 6: Integrated exposure-response relationships

Figure A6.1 shows integrated exposure-response curves for several health outcomes, along with the measured personal exposure concentrations of adults and pupils in this study. The figure shows that follow-up PM2.5 levels are well within the “steep” portion of the curves for at least some outcomes, including lower respiratory infections. These data should be interpreted with caution, as they are based on results from only 20 school pupils and 20 adults, and exposure confidence intervals (not shown) are wide.

Figure A6.1. Published concentration-response functions (Burnett et al. 2014, using tables provided by (Apte et al. 2015) for selected health outcomes compared to average personal PM2.5 exposure concentrations measured at baseline and follow-up in pupils and adult lamp users. IHD (60-65) = Ischemic heart disease (60-65 years of age), COPD = Chronic obstructive pulmonary disease.