Exposure to Nitrogen Dioxide and Fine Particulate Matter When Cooking with Electricity Compared to Gas, a Randomized Crossover Study in Quito, Ecuador

The contribution of gas cooking to indoor air pollution and health risk is poorly quantified. Although switching to gas cooking could reduce air pollution exposure for those relying on biomass, electric stoves, which produce no in-use emissions, may be a promising “leapfrog” technology.^1,2 Elevated nitrogen dioxide (${\mathrm{NO}}_2$), associated with poor respiratory outcomes,³ is a main concern with gas cooking.^3,4

In this study of households with both electric induction and gas stoves, we assessed ${\mathrm{NO}}_2$ exposures when the same individual used each stove type. Participants served as their own controls, eliminating time-invariant confounders like kitchen characteristics and other factors that drive both pollution differences and stove choice. Participants also were familiar with both stoves, alleviating the concern that households must adapt to new technologies.

Data collection occurred March–December 2021 in peri-urban and urban Quito, Ecuador. Participants ($n=38$) were recruited through radio, newspaper, social media, and bulletin board announcements and emails through Universidad San Francisco de Quito (USFQ) newsletters. Liquified petroleum gas (LPG) stoves had 4 burners (20 households), 6 burners (11), or 2–3 burners (7); 27 gas stoves had an oven. Induction stoves were tabletop 2- or 4-burner models. Three-fifths of households had a kitchen separate from other rooms in the house. Three-quarters of kitchens had $\ge 1$ window. Median kitchen size was $24{\mathrm{m}}^2$.

Primary cooks were randomly assigned to cook with only gas or only induction in the first 48-h period, then use only the other stove in a subsequent 48-h period. Our primary outcome was 48-h personal ${\mathrm{NO}}_2$ exposure, measured using passive badges (OGAWA PS-100) affixed near the breathing zone of a vest to be worn except when bathing and sleeping. Vests also had a time-resolved, light-scattering personal exposure monitor (PATS+) for detection of fine particulate matter (${\mathrm{PM}}_{2.5}$, fine particulate matter with aerodynamic diameter $\le 2.5\mathrm{\mu }\mathrm{m}$). Twelve randomly selected participants also had personal gravimetric ${\mathrm{PM}}_{2.5}$ (Ultrasonic Personal Air Sampler) and time-resolved kitchen area ${\mathrm{NO}}_2$ concentrations (AeroQual Series 500) measured; six of these individuals wore duplicate passive badges. Duplicate personal ${\mathrm{NO}}_2$ measurements were averaged in analyses ($r=0.87$). Kitchen samplers sat on countertops $\ge 1\mathrm{m}$ from the gas stove and, to the extent possible, equidistant between stove types.

Stove use was determined using temperature loggers (LPG) and current-voltage meters (induction). LPG cooking was based on highly positive slopes over short periods (doubling) and when $>{40}^{°}\mathrm{C}$ (rarely exceeded absent cooking). Induction cooking was identified when $>2.5$ A. Cooking and noncooking events lasted $\ge 5$ and 30 min, respectively.

We estimated the effect of stove randomization (reference: induction) on exposure in panel fixed effects regressions via ordinary least squares (OLS). The outcome was, separately, natural log-transformed 48-h average personal ${\mathrm{NO}}_2$ exposure, kitchen ${\mathrm{NO}}_2$ concentration, and personal ${\mathrm{PM}}_{2.5}$ exposure; we evaluated nontransformed models to estimate absolute changes. We included fixed effects for participant, month of year, and day of week. This intention-to-treat (ITT) analysis is a lower bound of the estimate of the effect of gas cooking on pollution; any deviation from stove assignment would attenuate the true effect. We excluded one kitchen area ${\mathrm{PM}}_{2.5}$ concentration estimate in the LPG group based on implausibility ($\text{mean}=513{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$).

To account for background pollution variations, we controlled for average 48-h ambient ${\mathrm{NO}}_2$ or ${\mathrm{PM}}_{2.5}$ concentrations from the nearest central site monitor (typically in the same neighborhood). To account for variation in vest wearing, we controlled for the proportion of time between 0600–2200 hours where PATS+ movement was detected. We estimated the effect of treatment on the treated by dividing our ITT estimate by the average fraction of total minutes cooked on the assigned stove when both stoves were concurrently monitored.

We estimated the effect of cooking events on short-term changes to natural log-transformed and nontransformed kitchen ${\mathrm{NO}}_2$ concentrations and personal ${\mathrm{PM}}_{2.5}$ exposure (mean and maximum of 5-min rolling windows) using panel fixed effects regressions estimated via OLS, where the exposure was whether LPG or induction stove use, modeled separately and jointly. We included fixed effects for participant, month of year, day of week, hour of day, and, for kitchen ${\mathrm{NO}}_2$, monitor identifier fixed effects. We top-coded the highest 2% of 5-min ${\mathrm{PM}}_{2.5}$ estimates to reduce outlier influence and improve model performance.

Standard errors were clustered at the participant level. We used $\mathrm{\alpha }=0.05$ to determine statistical significance and R statistical software (version 4.2.2; R Development Core Team) for analyses.

The institutional review boards at the Columbia University Medical Center and the Bioethics Committee at Universidad San Francisco de Quito approved this research and COVID-19 safety protocols. Participants provided informed consent online prior to visits or written consent on the day of visits.

Air pollution measurements, detected cooking events, and detected monitor wearing are summarized in Table 1. Participants generally used the assigned stove during the designated period, and minutes cooked were comparable across randomization.

Mean personal ${\mathrm{NO}}_2$ exposure was 51% higher [95% confidence interval (CI): 31%, 71%; $9.9\text{\hspace{0.17em}ppb}$ higher (95% CI: 4.5, 15.3)] during the 48-h periods (cooking and noncooking) when households were randomized to LPG (Figure 1). Half (19/38) of induction period exposure estimates fell below the World Health Organization (WHO) 24-h ${\mathrm{NO}}_2$ guideline ($13.29\text{\hspace{0.17em}ppb}$),⁵ in comparison with 10% (4/38) in the LPG period; all exposure estimates but two (induction) exceeded the WHO annual ${\mathrm{NO}}_2$ guideline ($5.3\text{\hspace{0.17em}ppb}$).⁵ Mean kitchen ${\mathrm{NO}}_2$ concentrations were 15% higher [95% CI: $-5\%$, 35%; $2.5\text{\hspace{0.17em}ppb}$ higher (95% CI: $-0.5,5.4$)] and mean personal ${\mathrm{PM}}_{2.5}$ exposure was 70% higher [95% CI: $-46\%$, 186%; $11.3{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ higher (95% CI: $-0.3$, 23.0)] when randomized to LPG in comparison with induction.

LPG cooking was associated with a 20% increase [95% CI: 14%, 26%; $4.4\text{\hspace{0.17em}ppb}$ increase (95% CI: 2.9, 5.9)] in 5-min average ${\mathrm{NO}}_2$ kitchen concentrations and a 40% increase [95% CI: 26%, 54%; $12.5{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ increase (95% CI: 6.7, 18.4)] in 5-min average personal ${\mathrm{PM}}_{2.5}$ exposure, in comparison with noncooking periods. Induction cooking was not associated with changes to short-term ${\mathrm{NO}}_2$ kitchen concentrations but was associated with a 23% increase [95% CI: 14%, 32%; $7.5{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ increase (95% CI: 3.9, 11.1)] in personal ${\mathrm{PM}}_{2.5}$ exposure. Similarly, LPG cooking was associated with higher rolling 5-min maximum ${\mathrm{NO}}_2$ kitchen concentrations [26% increase (95% CI: 19%, 34%)] and personal ${\mathrm{PM}}_{2.5}$ exposure [58% increase (95% CI: 41%, 76%)] in comparison with noncooking periods; induction was associated with higher 5-min maximum personal ${\mathrm{PM}}_{2.5}$ exposure [36% increase (95% CI: 25%, 47%)] but not 5-min maximum kitchen ${\mathrm{NO}}_2$ concentrations.

Results were robust to inclusion of monitor wearing and ambient air pollution as controls. The effect of treatment on the treated for 48-h personal ${\mathrm{NO}}_2$ exposure was $10.3\text{\hspace{0.17em}ppb}$ higher when households were randomized to LPG.

We assumed that within-household variation in stove use was uncorrelated with other behaviors that affect air pollution concentrations; randomization and high adherence suggest this assumption holds. Although our study has strong internal validity, our findings may not generalize to other settings with different stove, cooking, kitchen, and ventilation characteristics. Future studies could benefit from more participants, repeating measurements, and measuring cooking-related behaviors (e.g., window opening, vent hood use, cooking method).

Our results align with previous cross-sectional studies that found higher ${\mathrm{NO}}_2$ concentrations with gas relative to electric stoves,^3,4 a trial that found reductions in ${\mathrm{NO}}_2$ when replacing gas with electric stoves,⁶ and studies that found increased ${\mathrm{PM}}_{2.5}$ exposures with both gas and induction cooking.^7,8 Our study strengthens arguments for measuring personal air pollution exposures, which differed across stove types, instead of kitchen area measurements, which did not.

The authors acknowledge funding support from the US National Institutes of Health Common Fund through the Clean Cooking Implementation Science Network. The authors are grateful to Sam Heft-Neal for helpful figure edits and thoughtful comments from Misbath Daouda, Minghao Qiu, and members of the Stanford Environmental Change and Human Outcomes Lab, as well as research support from Alan García, Andrea Yánez, and Iván Nolivos.

The authors declare they have nothing to disclose.