Browse the corpus

Liquefied Petroleum Gas or Biomass Cooking and Severe Infant Pneumonia. BACKGROUND: Exposure to household air pollution is a risk factor for severe pneumonia. The effect of replacing biomass cookstoves with liquefied petroleum gas (LPG) cookstoves on the incidence of severe infant pneumonia is uncertain. METHODS: We conducted a randomized, controlled trial involving pregnant women 18 to 34 years of age and between 9 to less than 20 weeks' gestation in India, Guatemala, Peru, and Rwanda from May 2018 through September 2021. The women were assigned to cook with unvented LPG stoves and fuel (intervention group) or to continue cooking with biomass fuel (control group). In each trial group, we monitored adherence to the use of the assigned cookstove and measured 24-hour personal exposure to fine particulate matter (particles with an aerodynamic diameter of ≤2.5 μm [PM2.5]) in the women and their offspring. The trial had four primary outcomes; the primary outcome for which data are presented in the current report was severe pneumonia in the first year of life, as identified through facility surveillance or on verbal autopsy. RESULTS: Among 3200 pregnant women who had undergone randomization, 3195 remained eligible and gave birth to 3061 infants (1536 in the intervention group and 1525 in the control group). High uptake of the intervention led to a reduction in personal exposure to PM2.5 among the children, with a median exposure of 24.2 μg per cubic meter (interquartile range, 17.8 to 36.4) in the intervention group and 66.0 μg per cubic meter (interquartile range, 35.2 to 132.0) in the control group. A total of 175 episodes of severe pneumonia were identified during the first year of life, with an incidence of 5.67 cases per 100 child-years (95% confidence interval [CI], 4.55 to 7.07) in the intervention group and 6.06 cases per 100 child-years (95% CI, 4.81 to 7.62) in the control group (incidence rate ratio, 0.96; 98.75% CI, 0.64 to 1.44; P = 0.81). No severe adverse events were reported to be associated with the intervention, as determined by the trial investigators. CONCLUSIONS: The incidence of severe pneumonia among infants did not differ significantly between those whose mothers were assigned to cook with LPG stoves and fuel and those whose mothers were assigned to continue cooking with biomass stoves. (Funded by the National Institutes of Health and the Bill and Melinda Gates Foundation; HAPIN ClinicalTrials.gov number, NCT02944682.).

Pneumonia is a leading cause of child mortality worldwide, with most deaths in infants1. About 83% of the 808,000 annual child pneumonia deaths occur in sub-Saharan Africa, South Asia, and Latin America1. Observational studies suggest fine particulate air pollution (PM2.5) exposure from incomplete solid fuel combustion is a risk factor for pneumonia1. Nearly ~30% of global pediatric pneumonia deaths are attributed to household air pollution1. About 2.4 billion people – predominantly in low- and middle-income countries (LMICs) – use biomass (wood, charcoal, animal dung, coal) daily to cook or heat their households2. To date, randomized controlled trials (RCTs) of cleaner cooking interventions have not found an effect on primary child pneumonia outcomes3–6. However, it is unclear if lack of benefit stemmed from insufficiently lowered pollutant levels due to inadequate cookstove intervention uptake or performance, lack of specificity in pneumonia case definitions, or low statistical power. The Household Air Pollution Intervention Network (HAPIN) trial was designed to address these limitations in assessing whether cooking with an unvented liquefied petroleum gas (LPG) stove and fuel during pregnancy and the offspring’s first year of life, compared to biomass, reduced severe infant pneumonia incidence and other health outcomes7. We previously reported no evidence of an intervention effect on birthweight8. Here, we report severe pneumonia incidence during the first year of life, one of four primary trial outcomes.

HAPIN was a randomized controlled trial of unvented LPG cookstoves with free, uninterrupted fuel supply, compared to usual cooking practices (primarily or exclusively with biomass fuels), conducted in Tamil Nadu, India; Jalapa, Guatemala; Puno, Peru; and Kayonza, Rwanda from May 2018–September 20217. Sites were selected to cover a range of geographical settings in four continents where biomass is used for cooking. Pregnant women aged 18–34 years with an ultrasound and pregnancy-test confirmed, viable, singleton fetus at 9 to <20 weeks of gestation, biomass stove use at least 4 days a week, and study area residency, were eligible. Pregnant women who smoked tobacco, planned to migrate from the study area during the study, or used or planned to switch to LPG stoves were excluded. One pregnant woman per household could be enrolled. We randomized participants to intervention and control groups on a 1:1 basis. India and Peru used stratified randomization to ensure balance between two and six distinct geographical study areas, respectively. While the intervention assignment could not be blinded to participants and field staff, all investigators were masked to study group at the time of data cleaning, image interpretation, or data analysis.

used stratified randomization to ensure balance between two and six distinct geographical study areas, respectively. While the intervention assignment could not be blinded to participants and field staff, all investigators were masked to study group at the time of data cleaning, image interpretation, or data analysis. Unvented LPG cookstoves all had ≥2 burners and met local safety standards. Behavioral reinforcement messaging was provided to foster exclusive, safe LPG stove use9. Staff monitored both groups for adherence to group allocation through stove temperature sensor monitoring10. Controls were provided non-monetary compensation to counterbalance the intervention incentive of free fuel provision and mitigate attrition11. As cooking fuel delivery was considered an essential service, the intervention was generally uninterrupted by COVID-19 restrictions with similar delivery times before and during COVID12. Twenty-four-hour personal exposure to PM2.5, carbon monoxide, and black carbon were measured directly using wearable devices for pregnant women at baseline (<20 weeks) and 24-28 and 32-36 weeks’ gestation10. We estimated infants’ exposure to PM2.5, carbon monoxide, and black carbon at 1-3, 6 and 12 months of age using an indirect method13 (Supplementary Appendix).

arbon monoxide, and black carbon were measured directly using wearable devices for pregnant women at baseline (<20 weeks) and 24-28 and 32-36 weeks’ gestation10. We estimated infants’ exposure to PM2.5, carbon monoxide, and black carbon at 1-3, 6 and 12 months of age using an indirect method13 (Supplementary Appendix). We conducted active surveillance of severe pneumonia cases at pre-selected community hospitals and health centers. These facilities were identified during formative work as centers where severe cases receive care14. Passive facility and household surveillance were also conducted to identify missed facility visits, missed hospitalizations, ventilatory support, and deaths. Study staff were trained to evaluate children for severe pneumonia using a standard approach15; in brief, they passed certification examinations and received annual re-trainings. If medical care was needed, mothers could notify HAPIN staff by phone to facilitate appropriate care. In India, Peru, and Rwanda, study staff were available weekdays in-person at sentinel facilities and by phone anytime; in Guatemala, staff were available in-person continuously at the sentinel hospital. We reviewed medical charts of infant deaths and conducted a verbal autopsy to determine whether the death was related to severe pneumonia. Beginning in November 2019, sites in Rwanda increased study staff presence at outpatient clinics as surveillance identified some cases who were not hospitalized. In March 2020, COVID-19-related public health measures commenced at all sites, which limited active in person surveillance and care-seeking during lockdown periods. HAPIN staff also telephoned facility contacts to surveil for possible cases; telephone surveillance was uninterrupted during the study.

e not hospitalized. In March 2020, COVID-19-related public health measures commenced at all sites, which limited active in person surveillance and care-seeking during lockdown periods. HAPIN staff also telephoned facility contacts to surveil for possible cases; telephone surveillance was uninterrupted during the study. The primary outcome was severe pneumonia incidence in the first year of life among participant offspring. The primary case definition was adapted from World Health Organization (WHO) guidelines based on external expert input.16. In July 2019, when follow-up time of infants was <1%, we implemented additional expert recommendations to amend the case definition and improve specificity, objectivity, and to be responsive to formative data we collected (Supplementary Appendix)17,18 The primary definition of severe pneumonia was defined as (1) cough and/or difficult breathing, ≥1 general danger sign (unable to drink or breastfeed, vomiting everything, convulsions, stridor at rest, lethargy or unconscious) or ≥1 neonatal danger sign (unable to feed well, not moving at all or movement only when stimulated, grunting, severe chest indrawing), and pneumonia on imaging, (2) cough and/or difficult breathing with hypoxemia, or (3) a verbal autopsy-confirmed pneumonia death.15 Subsequent symptoms in the same child were considered separate episodes if >14 days after hospital discharge or >30 days from outpatient diagnosis. To be eligible as a case required examination by study staff except for children on ventilatory support or who died.

a, or (3) a verbal autopsy-confirmed pneumonia death.15 Subsequent symptoms in the same child were considered separate episodes if >14 days after hospital discharge or >30 days from outpatient diagnosis. To be eligible as a case required examination by study staff except for children on ventilatory support or who died. Chest imaging was by ultrasound (Sonosite Edge, Bothell, WA, USA)15,19 or radiography if ultrasound was unavailable. The reported sensitivity and specificity of lung ultrasound for diagnosing pneumonia in children are 95.5% and 95.3%, respectively and for chest radiography are 86.8% and 98.2%, respectively. 20. All images were interpreted by adjudication panels blinded to intervention and clinical status15,19,21. Two panelists followed pre-specified interpretation procedures and were required to agree on the presence or absence of pneumonia for the image to be classified as pneumonia. Pneumonia on imaging was a consolidation alone (meeting pre-specified size dimensions), or a pleural effusion near an infiltrate, or pleural abnormalities (ultrasound-specific)15,19,21.

erpretation procedures and were required to agree on the presence or absence of pneumonia for the image to be classified as pneumonia. Pneumonia on imaging was a consolidation alone (meeting pre-specified size dimensions), or a pleural effusion near an infiltrate, or pleural abnormalities (ultrasound-specific)15,19,21. Hypoxemia was defined as a peripheral arterial oxyhemoglobin saturation (SpO2) ≤92% at <2,500 meters altitude (Guatemala, India, Rwanda) or ≤86% at ≥2,500 meters altitude (Peru)15, or receipt of invasive or non-invasive ventilation or high-flow oxygen. To measure SpO2, facility study staff applied a Masimo Rad-G® pulse oximeter (Masimo, Irvine, CA, USA) and pediatric probe to the big toe of infants breathing room air. Staff collected three measurements over two-minutes, and these were averaged. SpO2 measurements were extracted from medical charts when available. Trained, local medical staff performed verbal autopsies with caregivers of deceased infants using a validated protocol22. A physician verbal autopsy panel assigned primary and secondary causes of death using WHO 2016 ICD-10 codes. Two non-study LMIC physicians masked to randomization and other death classifications independently reviewed the autopsy open narrative and closed questions. When the assigned primary cause of death was discordant between the two physicians, a pediatrician panelist arbitrated to achieve consensus. Cases without consensus were undetermined. The final verbal autopsy classification was pneumonia if it was the primary or secondary cause of death.

narrative and closed questions. When the assigned primary cause of death was discordant between the two physicians, a pediatrician panelist arbitrated to achieve consensus. Cases without consensus were undetermined. The final verbal autopsy classification was pneumonia if it was the primary or secondary cause of death. Secondary outcomes were pneumonia per WHO Integrated Management of Childhood Illness (IMCI) guidelines and WHO Pocketbook guidelines23,24, hypoxemia and/or imaging-confirmed pneumonia, and any hospitalized respiratory illness (see Table S1 for secondary outcome definitions).

narrative and closed questions. When the assigned primary cause of death was discordant between the two physicians, a pediatrician panelist arbitrated to achieve consensus. Cases without consensus were undetermined. The final verbal autopsy classification was pneumonia if it was the primary or secondary cause of death. Secondary outcomes were pneumonia per WHO Integrated Management of Childhood Illness (IMCI) guidelines and WHO Pocketbook guidelines23,24, hypoxemia and/or imaging-confirmed pneumonia, and any hospitalized respiratory illness (see Table S1 for secondary outcome definitions). Based on available evidence5,6,25–28, we estimated a sample of 3,200 pregnant women would provide 80% power to detect a 36% reduction in severe pneumonia incidence between study arms assuming a baseline rate of 9/100 infant-years using an α of 0.0125 (multiple hypothesis testing for four trial outcomes)7. The primary analysis was according to intention-to-treat and was conducted independently by two teams. We used Poisson regressions with generalized estimating equations (GEE) to model the incidence of all episodes of severe pneumonia using infant days at risk as the denominator to derive incidence rate ratios (IRRs). The intervention arm was the main covariate and models were adjusted for 10 randomization strata (one in Guatemala and Rwanda, two in India, six in Peru). The threshold for statistical significance for the primary outcome was set a priori at 0.0125 to account for the four primary trial outcomes. When data were incomplete for an outcome classification, we assumed the event did not occur.

for 10 randomization strata (one in Guatemala and Rwanda, two in India, six in Peru). The threshold for statistical significance for the primary outcome was set a priori at 0.0125 to account for the four primary trial outcomes. When data were incomplete for an outcome classification, we assumed the event did not occur. Secondary analyses estimated the intervention effect on the time to first pneumonia incidence via Cox proportional hazards models. Subgroup analyses assessed the influence of outpatient surveillance changes in Rwanda (after November 2019) and the COVID-19 pandemic (after March 2020) using GEE Poisson regression models with indicator variables for relevant time periods, as well as interaction terms between treatment arms and time periods to assess whether the intervention effect changed over time. Given the clustering of deaths very early in life, and that diagnostic accuracy may be lower in neonates, we also conducted sensitivity analyses of the primary analysis that excluded cases <7 and <30 days old. The protocol, available at NEJM.org, was approved by all investigator-affiliated institutional review boards (see Appendix). Participants provided written informed consent. An independent data and safety monitoring board (DSMB) monitored safety and efficacy and received unblinded interim analyses. No pre-defined stopping rules were formulated due to the low intervention risk. Eric D. McCollum, William Checkley, John McCracken, Jennifer Peel, and Thomas Clasen take responsibility for the integrity and completeness of the data and fidelity of the report to the protocol.

HAPIN was a randomized controlled trial of unvented LPG cookstoves with free, uninterrupted fuel supply, compared to usual cooking practices (primarily or exclusively with biomass fuels), conducted in Tamil Nadu, India; Jalapa, Guatemala; Puno, Peru; and Kayonza, Rwanda from May 2018–September 20217. Sites were selected to cover a range of geographical settings in four continents where biomass is used for cooking.

Pregnant women aged 18–34 years with an ultrasound and pregnancy-test confirmed, viable, singleton fetus at 9 to <20 weeks of gestation, biomass stove use at least 4 days a week, and study area residency, were eligible. Pregnant women who smoked tobacco, planned to migrate from the study area during the study, or used or planned to switch to LPG stoves were excluded. One pregnant woman per household could be enrolled.

We randomized participants to intervention and control groups on a 1:1 basis. India and Peru used stratified randomization to ensure balance between two and six distinct geographical study areas, respectively. While the intervention assignment could not be blinded to participants and field staff, all investigators were masked to study group at the time of data cleaning, image interpretation, or data analysis.

Unvented LPG cookstoves all had ≥2 burners and met local safety standards. Behavioral reinforcement messaging was provided to foster exclusive, safe LPG stove use9. Staff monitored both groups for adherence to group allocation through stove temperature sensor monitoring10. Controls were provided non-monetary compensation to counterbalance the intervention incentive of free fuel provision and mitigate attrition11. As cooking fuel delivery was considered an essential service, the intervention was generally uninterrupted by COVID-19 restrictions with similar delivery times before and during COVID12.

Twenty-four-hour personal exposure to PM2.5, carbon monoxide, and black carbon were measured directly using wearable devices for pregnant women at baseline (<20 weeks) and 24-28 and 32-36 weeks’ gestation10. We estimated infants’ exposure to PM2.5, carbon monoxide, and black carbon at 1-3, 6 and 12 months of age using an indirect method13 (Supplementary Appendix).

We conducted active surveillance of severe pneumonia cases at pre-selected community hospitals and health centers. These facilities were identified during formative work as centers where severe cases receive care14. Passive facility and household surveillance were also conducted to identify missed facility visits, missed hospitalizations, ventilatory support, and deaths. Study staff were trained to evaluate children for severe pneumonia using a standard approach15; in brief, they passed certification examinations and received annual re-trainings. If medical care was needed, mothers could notify HAPIN staff by phone to facilitate appropriate care. In India, Peru, and Rwanda, study staff were available weekdays in-person at sentinel facilities and by phone anytime; in Guatemala, staff were available in-person continuously at the sentinel hospital. We reviewed medical charts of infant deaths and conducted a verbal autopsy to determine whether the death was related to severe pneumonia. Beginning in November 2019, sites in Rwanda increased study staff presence at outpatient clinics as surveillance identified some cases who were not hospitalized. In March 2020, COVID-19-related public health measures commenced at all sites, which limited active in person surveillance and care-seeking during lockdown periods. HAPIN staff also telephoned facility contacts to surveil for possible cases; telephone surveillance was uninterrupted during the study.

The primary outcome was severe pneumonia incidence in the first year of life among participant offspring. The primary case definition was adapted from World Health Organization (WHO) guidelines based on external expert input.16. In July 2019, when follow-up time of infants was <1%, we implemented additional expert recommendations to amend the case definition and improve specificity, objectivity, and to be responsive to formative data we collected (Supplementary Appendix)17,18 The primary definition of severe pneumonia was defined as (1) cough and/or difficult breathing, ≥1 general danger sign (unable to drink or breastfeed, vomiting everything, convulsions, stridor at rest, lethargy or unconscious) or ≥1 neonatal danger sign (unable to feed well, not moving at all or movement only when stimulated, grunting, severe chest indrawing), and pneumonia on imaging, (2) cough and/or difficult breathing with hypoxemia, or (3) a verbal autopsy-confirmed pneumonia death.15 Subsequent symptoms in the same child were considered separate episodes if >14 days after hospital discharge or >30 days from outpatient diagnosis. To be eligible as a case required examination by study staff except for children on ventilatory support or who died.

Based on available evidence5,6,25–28, we estimated a sample of 3,200 pregnant women would provide 80% power to detect a 36% reduction in severe pneumonia incidence between study arms assuming a baseline rate of 9/100 infant-years using an α of 0.0125 (multiple hypothesis testing for four trial outcomes)7. The primary analysis was according to intention-to-treat and was conducted independently by two teams. We used Poisson regressions with generalized estimating equations (GEE) to model the incidence of all episodes of severe pneumonia using infant days at risk as the denominator to derive incidence rate ratios (IRRs). The intervention arm was the main covariate and models were adjusted for 10 randomization strata (one in Guatemala and Rwanda, two in India, six in Peru). The threshold for statistical significance for the primary outcome was set a priori at 0.0125 to account for the four primary trial outcomes. When data were incomplete for an outcome classification, we assumed the event did not occur.

for 10 randomization strata (one in Guatemala and Rwanda, two in India, six in Peru). The threshold for statistical significance for the primary outcome was set a priori at 0.0125 to account for the four primary trial outcomes. When data were incomplete for an outcome classification, we assumed the event did not occur. Secondary analyses estimated the intervention effect on the time to first pneumonia incidence via Cox proportional hazards models. Subgroup analyses assessed the influence of outpatient surveillance changes in Rwanda (after November 2019) and the COVID-19 pandemic (after March 2020) using GEE Poisson regression models with indicator variables for relevant time periods, as well as interaction terms between treatment arms and time periods to assess whether the intervention effect changed over time. Given the clustering of deaths very early in life, and that diagnostic accuracy may be lower in neonates, we also conducted sensitivity analyses of the primary analysis that excluded cases <7 and <30 days old.

The protocol, available at NEJM.org, was approved by all investigator-affiliated institutional review boards (see Appendix). Participants provided written informed consent. An independent data and safety monitoring board (DSMB) monitored safety and efficacy and received unblinded interim analyses. No pre-defined stopping rules were formulated due to the low intervention risk. Eric D. McCollum, William Checkley, John McCracken, Jennifer Peel, and Thomas Clasen take responsibility for the integrity and completeness of the data and fidelity of the report to the protocol.

3,200 women were randomized, with 1,593 (49.8%) allocated to the LPG arm and 1,607 (50.2%) as controls (Figure 1). Baseline maternal characteristics were similar between groups (Table 1); the pregnant women and their offspring were representative of the broader population of women and infants affected by indoor air pollution from biomass cooking (Table S2). Pregnant women received LPG stoves mid-second trimester (mean 18.1 weeks (SD, 3.3). There were 1,536 livebirths overall in the intervention group and 1,525 in controls. Table 2 reports the characteristics of liveborn children by study group, including vaccination status. Intervention participants used biomass stoves on a median of 0.4% (interquartile range (IQR) 0.0, 2.3) of monitored days12,29. The averaged, post-randomization, 24-hour personal exposures to PM2.5 were overall lower in the intervention arm, compared to controls in the antenatal (median 24.8 μg/m3, interquartile range (IQR) 17.0, 40.5 vs median 77.0 μg/m3, 40.7, 132.8) as well as during postnatal periods (median 24.2 μg/m3, IQR 17.8, 36.4 vs median 66.0 μg/m3, IQR 35.2, 132.0) (Table 2, Table S3)13,30.

s to PM2.5 were overall lower in the intervention arm, compared to controls in the antenatal (median 24.8 μg/m3, interquartile range (IQR) 17.0, 40.5 vs median 77.0 μg/m3, 40.7, 132.8) as well as during postnatal periods (median 24.2 μg/m3, IQR 17.8, 36.4 vs median 66.0 μg/m3, IQR 35.2, 132.0) (Table 2, Table S3)13,30. We identified 85 severe pneumonia episodes in the intervention group and 90 in the control group (Figure 2) from 1,243 healthcare facility visits and 55 verbal autopsies (Figure S1, Tables S4–S8). Among these, there were 12 deaths attributed to pneumonia (6.9% of pneumonia outcomes), eight in the control group and 4 in the intervention group.) (Table S9). The severe pneumonia incidence rate in the first year of life was 5.67 (95% CI 4.45, 7.07) per 100 child-years in the LPG group and 6.06 (95% CI 4.81, 7.62) per 100 child-years in the control group (incidence rate ratio (98.7% CI) 0.96 (0.64, 1.44; p=0.81) (Figure 2). No evidence of an intervention effect was observed for secondary outcomes (Figure 2, Table S10) or when stratified by study site or other subgroups (Figure S2). Although the observed incidence of severe pneumonia across all study sites decreased by 77% (95% CI 61%, 86%) during the COVID-19 pandemic period (Figure 3, Figure S3, Table S11), there was no appreciable change in the IRR when our models accounted for the pandemic period and child’s age (IRR 0.96, 95% CI 0.70, 1.31). The IRRs before (0.71 IRR, 95% CI 0.12, 4.23) and after (IRR 0.80, 95% CI 0.49, 1.31) surveillance changes (November 2019) in Rwanda were also similar (Table S12).

S3, Table S11), there was no appreciable change in the IRR when our models accounted for the pandemic period and child’s age (IRR 0.96, 95% CI 0.70, 1.31). The IRRs before (0.71 IRR, 95% CI 0.12, 4.23) and after (IRR 0.80, 95% CI 0.49, 1.31) surveillance changes (November 2019) in Rwanda were also similar (Table S12). Burns were reported by three infants (0.2%) in the intervention group and seven infants (0.5%) in the control arm. No burn was classified as a serious adverse event Table S13).

3,200 women were randomized, with 1,593 (49.8%) allocated to the LPG arm and 1,607 (50.2%) as controls (Figure 1). Baseline maternal characteristics were similar between groups (Table 1); the pregnant women and their offspring were representative of the broader population of women and infants affected by indoor air pollution from biomass cooking (Table S2). Pregnant women received LPG stoves mid-second trimester (mean 18.1 weeks (SD, 3.3). There were 1,536 livebirths overall in the intervention group and 1,525 in controls. Table 2 reports the characteristics of liveborn children by study group, including vaccination status.

Intervention participants used biomass stoves on a median of 0.4% (interquartile range (IQR) 0.0, 2.3) of monitored days12,29. The averaged, post-randomization, 24-hour personal exposures to PM2.5 were overall lower in the intervention arm, compared to controls in the antenatal (median 24.8 μg/m3, interquartile range (IQR) 17.0, 40.5 vs median 77.0 μg/m3, 40.7, 132.8) as well as during postnatal periods (median 24.2 μg/m3, IQR 17.8, 36.4 vs median 66.0 μg/m3, IQR 35.2, 132.0) (Table 2, Table S3)13,30.

We identified 85 severe pneumonia episodes in the intervention group and 90 in the control group (Figure 2) from 1,243 healthcare facility visits and 55 verbal autopsies (Figure S1, Tables S4–S8). Among these, there were 12 deaths attributed to pneumonia (6.9% of pneumonia outcomes), eight in the control group and 4 in the intervention group.) (Table S9). The severe pneumonia incidence rate in the first year of life was 5.67 (95% CI 4.45, 7.07) per 100 child-years in the LPG group and 6.06 (95% CI 4.81, 7.62) per 100 child-years in the control group (incidence rate ratio (98.7% CI) 0.96 (0.64, 1.44; p=0.81) (Figure 2).

No evidence of an intervention effect was observed for secondary outcomes (Figure 2, Table S10) or when stratified by study site or other subgroups (Figure S2). Although the observed incidence of severe pneumonia across all study sites decreased by 77% (95% CI 61%, 86%) during the COVID-19 pandemic period (Figure 3, Figure S3, Table S11), there was no appreciable change in the IRR when our models accounted for the pandemic period and child’s age (IRR 0.96, 95% CI 0.70, 1.31). The IRRs before (0.71 IRR, 95% CI 0.12, 4.23) and after (IRR 0.80, 95% CI 0.49, 1.31) surveillance changes (November 2019) in Rwanda were also similar (Table S12).

Despite high LPG intervention uptake and substantial reductions in air pollutant exposure, we found no significant difference in the incidence of severe infant pneumonia between the intervention and control arms in this multi-country trial. Our findings are consistent with null findings from a cluster randomized trial in Ghana of a similar cookstove3, indicating that unvented LPG cookstoves are unlikely to reduce severe infant pneumonia. Our trial also found no difference between study arms in the other primary endpoints of birthweight8 and stunting (reported in another article in this issue of the Journal,) 31.

ter randomized trial in Ghana of a similar cookstove3, indicating that unvented LPG cookstoves are unlikely to reduce severe infant pneumonia. Our trial also found no difference between study arms in the other primary endpoints of birthweight8 and stunting (reported in another article in this issue of the Journal,) 31. There are several potential explanations for our null findings for severe infant pneumonia. First, evidence suggests household air pollution is more closely linked with bacterial than viral nasopharyngeal carriage32,33. While nasopharyngeal carriage is considered a prerequisite for the development of invasive or mucosal bacterial diseases like pneumonia34, populations vaccinated against Haemophilus Influenzae type B (Hib) and Streptococcus pneumoniae (pneumococcus) are well protected from nasopharyngeal carriage progressing to disease35. Our study population had high rates of vaccination against Hib and pneumococcal pneumonia, making severe bacterial pneumonia less likely. As observed in this trial (Figure 3, Figure S3) and elsewhere, the fact that mitigation efforts during the COVID-19 pandemic dramatically reduced both respiratory virus circulation and pediatric hospitalizations provides indirect evidence on the central role of viruses in severe childhood respiratory disease36,37. However, definitively determining the etiology of severe childhood pneumonia is challenging, and we do not have information on respiratory pathogens in these infants.

virus circulation and pediatric hospitalizations provides indirect evidence on the central role of viruses in severe childhood respiratory disease36,37. However, definitively determining the etiology of severe childhood pneumonia is challenging, and we do not have information on respiratory pathogens in these infants. Second, the PM2.5 levels we achieved were lower than levels in other trials3–6 but remained above WHO recommendations38. Although uncertain, it is possible that lower PM2.5 exposure levels than were achieved may be required to reduce the risk of severe pneumonia and greater reductions may require broader community interventions, rather than household strategies as we employed. Third, even though unvented LPG cookstoves produce nitrogen dioxide (NO2) at levels lower than biomass cookstoves, these levels are nevertheless above recommendations39. Elevated NO2 concentrations have associations with asthma in children40, and may have contributed to our null results.

hold strategies as we employed. Third, even though unvented LPG cookstoves produce nitrogen dioxide (NO2) at levels lower than biomass cookstoves, these levels are nevertheless above recommendations39. Elevated NO2 concentrations have associations with asthma in children40, and may have contributed to our null results. Limitations of our study should be noted. Incomplete assessments at facility visits may have led to missed cases, although this is unlikely to have impacted our results because missingness among screened children was low (Figure S1). It is also possible that incomplete case ascertainment occurred due to children seeking care at clinics outside of the surveillance area or failing to seek care at all. Missed cases may have been more common during the COVID-19 pandemic period, particularly in the first months during lockdowns. We accounted for the pandemic in our analysis but did not find evidence of differential effects of the pandemic on our results (Table S11). The wide confidence intervals around our effect estimates mean that we cannot exclude clinically important reductions or increases in severe pneumonia risk with the use of unvented LPG cookstoves compared to biomass cookstoves. Also since there is no gold standard for pneumonia diagnosis, the accuracy of our primary case definition for severe pneumonia is undetermined. However, we sought and incorporated external expert recommendations intended to optimize the definition’s objectivity and specificity. The results for outcomes using alternative pneumonia definitions were also consistent with results for the primary outcome.

primary case definition for severe pneumonia is undetermined. However, we sought and incorporated external expert recommendations intended to optimize the definition’s objectivity and specificity. The results for outcomes using alternative pneumonia definitions were also consistent with results for the primary outcome. In conclusion, in this multicenter trial involving four LMICs, unvented LPG cookstoves did not reduce the incidence of severe infant pneumonia compared to biomass cookstoves.