Study population characteristics
Of 220 group 1 houses that were randomly selected, 102 control and 109 intervention households were enrolled. An additional 38 (19 intervention and 19 control) group 2 households that were purposefully selected based on their cooking location at baseline were also enrolled. Inability to contact households to schedule sampling visits was the primary reason for non-enrollment. All enrolled houses received at least one round of entomological sampling. Five intervention and six control houses did not receive a second round of sampling because they completed and exited the HAPIN trial before the second visit. Sampling was interrupted during the third round due to the COVID-19 pandemic and 26 intervention houses and 29 control houses did not receive a third visit. A total of 567 sampling visits were conducted among group 1 houses, and 103 visits were conducted among group 2 houses (Fig. 1).
Flow diagram of household selection from HAPIN trial population.
Fidelity was high; intervention houses received LPG stoves and fuels a median of 9.5 days after randomization (IQR = 6–16 days), and all houses received the intervention prior to the start of entomological sampling. Maternal age and gestational age at baseline were similar between participants in each arm, whereas fewer mothers in the control arm had completed secondary or further education (Table 1). Wood was the most common fuel source among control and intervention groups at baseline, followed by charcoal. Most participants in both groups reported cooking outdoors or in a separate cooking structure at baseline. Open/3-stone fires were the most common primary stove type in both arms, followed by simple wood-burning stoves called ronderezas and portable charcoal-burning stoves called imbaburas; fewer households in the control arm reported using imbaburas compared to the intervention arm.
Intervention adherence was high; 97% of intervention houses reported using LPG as their primary cooking fuel at follow-up (the day of each visit) and > 99% of control houses reported using biomass as their primary cooking fuel at follow-up. Nearly 90% of control houses reported cooking outdoors or in a separate cooking structure at follow-up, whereas 91% of intervention houses reported cooking inside the main house (Table 2). Mean PM2.5 concentrations were 30.7 µg/m3 (SD = 26.1) in control bedrooms, which was slightly higher than 25.3 µg/m3 (SD = 30.2) measured in intervention bedrooms.
Entomological outcomes
Anopheles mosquitoes
We collected 356 Anopheles mosquitoes during 567 sampling nights; 336 (94%) were collected via CDC light traps in participant bedrooms and an additional 20 (6%) were collected by Prokopacks, mostly from kitchens (Supplementary Table 1). An. gambiae s.l. accounted for 82% of all Anopheles collected (Supplementary Table 2). Mean Anopheles density was 0.5 (SD = 2.4, median = 0, range 0–31) per sampling night in control houses and 0.7 (SD = 3.0, median = 0, range 0–31) in intervention houses (Table 3; Fig. 2). In both unadjusted and adjusted models, Anopheles densities were similar in the intervention compared to the control group (unadjusted rate ratio (RR) = 0.92, 95% CI: 0.33–2.55; adjusted RR = 1.23, 95% CI: 0.51–2.99). Covariate effect estimates are available in Supplementary Table 3.
Densities of Anopheles mosquitoes (a), culicine mosquitoes (b) and synanthropic flies (c) by intervention status. Medians are shown with solid horizontal lines at the center of each box plot.
We then assessed potential effect modification by cooking location, including the group 2 houses. Among houses that cooked indoors, Anopheles densities were higher among intervention houses compared to control houses, although confidence intervals were wide and included the null (unadjusted RR = 1.86, 95% CI: 0.18–19.28; adjusted RR = 3.66, 95% CI: 0.92–14.59) (Supplementary Table 4). In contrast, Anopheles densities were similar in intervention and control houses that cooked outdoors (unadjusted RR = 0.73, 95% CI: 0.10–5.47; adjusted RR = 1.09, 95% CI: 0.31–3.83). We observed similar effects when we restricted the analysis to just the randomly selected group 1 houses, although the effect estimates were less precise (data not shown). Mean PM2.5 concentrations were the highest in control houses which cooked indoors at follow-up (Supplementary Table 5). We observed a negative but not statistically significant association between each standard deviation increase in PM2.5 and Anopheles densities (RR = 0.65, 95% CI: 0.15–2.82) (Supplementary Table 6).
A single Anopheles mosquito from a control household was P. falciparum sporozoite-positive, whereas no Anopheles from intervention households were positive. Blood-fed status and bloodmeal composition of blood-fed mosquitoes were similar in both groups (Supplementary Table 7).
Culicine mosquitoes
We collected 2,145 culicine mosquitoes, 2,048 (96%) of which were Cx. quinquefasciatus (Supplementary Table 2). Among all culicines, 1920 (90%) were collected via CDC Light traps, 164 (8%) were collected by Prokopacks, and 61 (3%) were collected by sticky fly traps (Supplementary Table 1). Mean culicine density was 3.3 (SD = 5.4, median = 0, range 0–41) per sampling night in control houses, compared to 4.23 (SD = 8.7, median = 0, range 0–41) in intervention houses (Table 3; Fig. 2). In both unadjusted and adjusted models, the intervention did not affect culicine densities (unadjusted RR = 1.17, 95% CI: 0.83–1.63; adjusted RR = 1.12, 95% CI: 0.79–1.68) (Table 3 & Supplementary Table 3). Cooking location did not modify the effect of the intervention, and neither PM2.5 nor cooking location were associated with culicine densities (Supplementary Tables 4 & 6). Approximately 1% of Cx. quinquefasciatus mosquitoes collected in both control and intervention houses were blood-fed. Bloodmeal composition was similar between control and intervention houses (Supplementary Table 7).
Synanthropic flies
We collected 1,022 synanthropic flies, of which 475 (46%) were Muscidae and 436 (43%) were Fanniidae. Mean synanthropic fly density was 2.77 (SD = 4.84, median = 1, range 0–28) in control houses compared to 0.91 (SD = 2.69, median = 0, range 0–10) in intervention houses (Table 3; Fig. 2, & Supplementary Table 2). In both unadjusted and adjusted analyses, the intervention was associated with a ≥ 65% reduction in synanthropic fly densities (unadjusted RR = 0.31, 95% CI: 0.22–0.45; adjusted RR = 0.35, 95% CI: 0.24–0.51) (Table3 & Supplementary Table 3).
Cooking indoors at follow-up was associated with a 62% reduction in fly densities (adjusted RR = 0.38, 95% CI: 0.27–0.55) (Supplementary Table 6), whereas densities were similar in intervention houses compared to control houses that cooked indoors (adjusted RR = 1.09, 95% CI: 0.53–2.25) (Supplementary Table 4). Indoor cooking at endline was also much higher among intervention households compared to control households (91% vs. 11.2%), suggesting that cooking location mediated much of the observed intervention effect on fly densities. Intervention kitchens were a mean of 2–3 m further away from latrines and rubbish pits, which are common fly breeding sites, compared to control kitchens (Table 2). PM2.5 in bedrooms was not associated with fly densities (Supplementary Table 6).
Secondary outcomes: malaria and diarrheal disease
A total of 69 malaria cases (39 control, 30 intervention) were reported among mothers. Mean longitudinal prevalence of malaria was 8.1% (SD = 13.0) among mothers in the control group and 5.8% (SD = 12.1) among mothers in the intervention group. After controlling for potential confounders, the intervention was not associated with malaria prevalence among mothers (adjusted longitudinal prevalence ratio (LPR) = 0.92, 95% CI: 0.56–1.47) (Table 4). The results were similar when we restricted the analysis to only confirmed malaria cases. PM2.5 and cooking location were not associated with reported malaria risk in mothers (Supplementary Table 8).
Participants reported 12 malaria cases in children (9 control, 3 intervention) during 754 periods of observation. Mean longitudinal prevalence of malaria was 2.7% (SD = 9.8) among children in the control group and 0.8% (SD = 4.6) among children in the intervention group. After controlling for potential confounders, malaria prevalence was lower children in the intervention arm compared to the control, although this effect estimate was imprecise and included the null (adjusted LPR = 0.42, 95% CI: 0.09–1.44) (Table 4). We did not evaluate the effects of PM2.5 on malaria prevalence in children because there were too few cases for which PATS + measurements were available (n = 5).
A total of 62 diarrhea episodes (32 control, 30 intervention) were reported among children during 756 follow-up periods of observation. Mean longitudinal prevalence of diarrhea was 8.8% (SD = 0.14) among children in the control group and 7.4% (SD = 14.27) among children in the intervention group. After controlling for potential confounders, the intervention was not associated with diarrhea prevalence in children (adjusted LPR = 0.96, 95% CI: 0.59–1.56) (Table 4). We observed a positive but not statistically significant association between each standard deviation increase in PM2.5 and diarrhea (adjusted LPR = 1.59, 95% CI: 0.86–2.82) (Supplementary Table 8).
