Poor air quality is thought to be an important mortality risk factor globally1,2,3, but there is little direct evidence from the developing world on how mortality risk varies with changing exposure to ambient particulate matter. Current global estimates apply exposure–response relationships that have been derived mostly from wealthy, mid-latitude countries to spatial population data4, and these estimates remain unvalidated across large portions of the globe. Here we combine household survey-based information on the location and timing of nearly 1 million births across sub-Saharan Africa with satellite-based estimates5 of exposure to ambient respirable particulate matter with an aerodynamic diameter less than 2.5 μm (PM2.5) to estimate the impact of air quality on mortality rates among infants in Africa. We find that a 10 μg m−3 increase in PM2.5 concentration is associated with a 9% (95% confidence interval, 4–14%) rise in infant mortality across the dataset. This effect has not declined over the last 15 years and does not diminish with higher levels of household wealth. Our estimates suggest that PM2.5 concentrations above minimum exposure levels were responsible for 22% (95% confidence interval, 9–35%) of infant deaths in our 30 study countries and led to 449,000 (95% confidence interval, 194,000–709,000) additional deaths of infants in 2015, an estimate that is more than three times higher than existing estimates that attribute death of infants to poor air quality for these countries2,6. Upward revision of disease-burden estimates in the studied countries in Africa alone would result in a doubling of current estimates of global deaths of infants that are associated with air pollution, and modest reductions in African PM2.5 exposures are predicted to have health benefits to infants that are larger than most known health interventions.
Access optionsAccess options
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Subscribe to Journal
Get full journal access for 1 year
only $3.90 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank D. Lobell, G. McCord, M. P. Burke and W. Schlenker for useful comments and V. Tanutama for research assistance. We thank the Stanford Environmental Ventures Fund and the National Science Foundation (CNH Award #1715557) for funding.
Nature thanks R. Black, J. Lelieveld, L. Waller and the other anonymous reviewer(s) for their contribution to the peer review of this work.
Extended data figures and tables
Data were obtained from a previous study4. a, Relative risk curve representing the risk from acute lower respiratory infections in infants (obtained from figure 2 of Burnett et al.4). The curve combines point estimates from ambient air pollution (AAP) studies, indoor air pollution (HAP) studies and second-hand smoking (SHS) studies to derive risk responses across the PM2.5 exposure distribution. The histograms show the share of population exposed to different long-run (15-year average) ambient PM2.5 concentrations in North American and Europe where most GBD studies took place, in SSA countries in our sample, and globally. In total, 49% of the overall population in Africa, and 51% globally, live in areas with ambient pollution concentrations exceeding the maximum ambient PM2.5 concentration from the GBD study (25 μg m−3). b, Most studies used to estimate the GBD integrated exposure response4 were carried out in North America or Europe, with the exception of a household air pollution study in Guatemala and second-hand smoking studies in Vietnam, India and South Africa. Median sample size (depicted by marker size in the plot) across these studies is n = 1,250. Country outlines were obtained from Global Administrative Areas, version 2.030.
a, Location of DHS clusters included in our sample. b, The number of births observed in each year in our sample. More births are observed in earlier years because births are recalled in the surveys so each new survey round potentially adds births from all previous years. c, Regional categorization of countries, for regional analysis in Fig. 2c. Sample countries assigned to West Africa region are Benin, Burkina Faso, Ivory Coast, Ghana, Guinea, Liberia, Mali, Nigeria, Senegal, Sierra Leone and Togo. Sample countries assigned to ‘rest of Africa’ are Angola, Burundi, Cameroon, Comoros, DRC, Ethiopia, Gabon, Kenya, Lesotho, Madagascar, Malawi, Mozambique, Namibia, Rwanda, Swaziland, Uganda, Zambia and Zimbabwe. Country outlines were obtained from Global Administrative Areas, version 2.030.
Extended Data Fig. 3 Effect of post-birth PM2.5 exposure is robust under different regression models.
Estimated responses under higher-order polynomials (a), different specifications of the fixed effects (b), restricted cubic spline functions of PM2.5 (c) and additional time controls (d). In each panel, the blue line and shaded region indicate the estimated baseline response shown in Fig. 2a and the bootstrapped 95% confidence interval. Splines in c have knots at 10 μg m−3 (single knot spline) or evenly spaced knots (three- and four-knot splines).
Extended Data Fig. 4 Piecewise linear and cross-sectional relationships between post-birth PM2.5 exposure and infant mortality.
a, Piecewise linear estimates of the effect of PM2.5 exposure below and above the WHO PM2.5 guideline of 10 μg m−3. Shaded regions represent bootstrapped 95% confidence intervals. Slopes above and below the 10 μg m−3 threshold are very similar, although confidence intervals are wider below the threshold due to smaller sample sizes. b, Cross-sectional and panel models give similar estimated effects of post-birth PM2.5 exposure on infant mortality. Blue line shows baseline panel model, orange line shows a cross-sectional model that relates cluster-average mortality to cluster-average PM2.5 exposure. Each response function is centred at sample median exposure (25 μg m−3). Histograms at the bottom show counts of exposure at different PM2.5 levels, for the panel sample (blue) and cross-sectional sample (orange); cross-sectional exposures are slightly narrower given that year-to-year variation has been averaged out.
Effects are estimated by interacting a dummy for each modifying variable with linear PM2.5, and are measured as the percentage change in infant mortality per 10 μg m−3 increase in PM2.5 exposure, relative to baseline mortality rates in each subgroup. Circles indicate point estimates, and whiskers the 95% confidence interval on the point estimate. The last column shows the baseline estimate from the full (uninteracted) linear model.
Panels are the same as Fig. 2e but replicated for different time periods, showing effects in each year independently. Circles indicate point estimates, and whiskers the 95% confidence interval on the point estimate. For each time period 2001 − year t, the sample was restricted to births between 2001 and year t and to surveys that were conducted after year t. These steps help to approximate a consistent geographical sample across the time periods.
a–c, Effect of in utero PM2.5 exposure on low birth weight, low birth size as reported by mothers on a scale from 1 to 5, and neonatal mortality (NMR). d, e, Effect of post-birth PM2.5 exposure on height-for-age and diarrhoeal incidence for living children. In each case, higher PM2.5 concentrations worsen health outcomes. f–h, Placebo tests that relate PM2.5 exposures to child outcomes that should be unaffected: child sex, whether child was born in a multiple birth, and child’s use of a bed net. i, PM2.5 exposure in the 13–24 months after birth has no effect on mortality in the first 12 months after birth. Shaded regions represent bootstrapped 95% confidence intervals in each panel. j, Estimates of the effect of PM2.5 on all-cause infant mortality from published quasi-experimental studies24,25,26,27,28,29, expressed as the percentage change in the infant mortality rate per 10 μg m−3 increase in PM2.5.