Our study area, Jianzha County in Qinghai province, China (35°52’N 101°51’E, altitude: 3500 m), is located in the Himalayas where heating is an important energy requirement throughout the year. The primary livelihood of study participants was herding, and participants migrated multiple times per year at altitudes up to 5400 m. Among our study population, income was earned primarily from selling animal products, engaging in seasonal labor, and collecting local medicinal herbs, and was below the international poverty line at $0.91 per person per day [20].

We discussed potential participation with all 147 households in the local community. An initial selection was conducted based on whether or not the heads of household were both present during our initial visit to provide informed consent. Among these households, 100 households were randomly selected in this study. All study protocols were reviewed and approved by the Harvard T.H. Chan School of Public Health Institutional Review Board and Chinese Ministry of Science and Technology.

CO monitors ran continuously and simultaneously for 44 sampling days in all 100 participating households from March to April 2012, yielding a total of 4400 household-days. Samples were logged at 5-min intervals. Monitors were located approximately one meter from the stove and one meter above the ground—near breathing level for people sitting around the stove. The sensors employed were NAP-505 electrochemical CO gas sensors (Range: 0–1000 ppm; Accuracy: ±2%; Nemoto, Burgermeester Haspelslaan 53, 1181NB Amstelveen, The Netherlands) embedded in EL-USB-CO Data Loggers (Lascar Electronics, Module House, Whiteparish, Salisbury, SP5 2SJ, United Kingdom). Duplicate measurements were conducted in 2% of homes. Background CO levels were measured outdoors and found to be below the limit of detection. Outdoor temperature was monitored continuously outside the home of the lead field assistant throughout the study period. The outdoor daily average temperature ranged from −3 to 9 °C. The effects of temperature and pressure on sensor output were assessed at altitude and corrected for in the calculation of the CO concentrations.

For each 5-min of the day, boxplots of CO measurements from all 100 households across all 44 days were generated to illustrate diurnal variation in median and interquartile CO concentration. The average of CO when active cooking stopped was calculated to determine the contribution from combustion for heating.

Field assistants visited each study home three times per day (morning, afternoon, evening) on three consecutive days in March-April 2012. During these visits, field assistants recorded time-activity budgets of 100 male and 100 female heads of household and administered daily surveys characterizing smoking and other sources of personal exposure aside from the stove (incense, kerosene, candles, etc.). Each field assistant and each participant were provided a watch to facilitate detailed time recordings; time syncing was checked by field assistants before the first visit of each day. Field assistants recorded the present activity of the participant (cooking, herding, other outdoor, other indoor, resting, collecting water, or collecting fuel). Participants were asked for a summary of their activities since waking up (if a morning measurement) or since the last visit (if an afternoon measurement). Location of activity was described as indoors or outdoors, and time spent on each activity daily was reported by subjects. Subjects also reported number of cigarettes smoked during activity.

We measured CO in exhaled breath (pCO) for practical reasons of feasibility, cost and participant burden. Laboratory and field validation studies found pCO exhaled breath methods compared well to gold standard blood-gas methods with r-square correlation values between 0.88 and 0.98 [21,22,23]. We measured pCO with a ToxCOTM Breath CO Monitor (Range: 0–600 ppm, Accuracy: ±5%, Bedfont Scientific Ltd, Station Rd, Harrietsham, Maidstone, Kent ME17 1JA, UK). This instrument employs an electrochemical sensor to measure CO (reported in ppm) in exhaled breath and then calculates COHb using the Haldane equation and physiological parameters consistent with healthy males at sea level [24]. The Haldane equation is a well-established and validated approach to correlate exhaled CO and COHb levels, and since the participants in our study are fully acclimated to the altitude of their location, it is reasonable to assume the key parameters in the equation (e.g., blood O2 level and the ratio of the affinity of blood for CO to that for O2) used for the sea-level population can be also applied to our study population.

During each visit, assistants administered the ToxCOTM breath test. The test was conducted outdoors to avoid contamination from indoor CO. Participants were asked to inhale and hold their breath for 15 s before exhaling completely into the ToxCOTM mouthpiece, in accordance with ToxCOTM standard operating procedures. This was repeated three times and values were averaged. All breath monitors were calibrated weekly in the field using the calibration kit supplied by the manufacturer. Exposures during each activity were compared by averaging COHb (calculated from pCO) within activity and gender.

A baseline survey was conducted to collect information on house types, stove types, smoking status, number of stoves, socio-economic status (SES) rank, number of family members, age of each child, estimated time spent collecting fuel, and smoking status. Specifically, SES ranking was conducted by the council of village elders. Each elder separately ranked all study households from 1 to 100 based on ownership of animals, land, and material items as well as on number of children, education level, and other factors influencing future opportunity. The elders then debated these rankings to arrive at a final ranking by consensus. Women were asked whether they carried children on their backs while cooking (always, usually, sometimes, seldom, never), whether they let their children play outside in winter (always, usually, sometimes, seldom, never), and whether they ask their children to assist with their father or mother.

We fitted linear mixed-effects models with subject-specific random intercepts to investigate the relationship between COHb and activity patterns. COHb was log transformed to account for its positive skewness and regressed against self-reported activity type prior to COHb measurement (i.e., cooking, other outdoor, other indoor, and resting), mean 24-h CO concentration (ppm) in home, age, weight, sex, and smoking within 4 h prior to measurement. Contributions from incense, butter lamps, and other indoor combustion sources were analyzed based on observed and reported usage in religious ceremonies but found to be insignificant and excluded from the model. All statistical analyses were performed in R version 4.1.2.