Artículo de revista: VERHULST, Andrea; ROMERO, Julio; ALAM, Nurul; (…) VILLAVICENCIO; Francisco et al. (2022) "Divergent age patterns of under-5 mortality in south Asia and sub-Saharan Africa: a modelling study". Lancet Global Health (Online: September 08, 2022): 1-9 (ISSN: 2214-109X). DOI: https://doi.org/10.1016/S2214-109X(22)00337-0 Articles Divergent age patterns of under-5 mortality in south Asia and sub-Saharan Africa: a modelling study Andrea Verhulst, Julio Romero Prieto, Nurul Alam, Hallie Eilerts-Spinelli, Daniel J Erchick, Patrick Gerland, Joanne Katz, Bruno Lankoande, Li Liu, Gilles Pison, Georges Reniers, Seema Subedi, Francisco Villavicencio, Michel Guillot Summary Background Understanding the age pattern of under-5 mortality is essential for identifying the most vulnerable ages Lancet Glob Health 2022 and underlying causes of death, and for assessing why the decline in child mortality is slower in some countries and Published Online subnational areas than others. The aim of this study is to detect age patterns of under-5 mortality that are specific to September 8, 2022 https://doi.org/10.1016/ low-income and middle-income countries (LMICs). S2214-109X(22)00337-0 Population Studies Center, Methods In this modelling study, we used data from 277 Demographic and Health Surveys (DHSs), 58 Health and University of Pennsylvania, Demographic Surveillance Systems (HDSSs), two cohort studies, and two sample-registration systems. From these Philadelphia, PA, USA sources, we collected child date of birth and date of death (or age at death) from LMICs between 1966 and 2020. We (A Verhulst PhD, Prof M Guillot PhD); computed 22 deaths rates from each survey with the following age breakdowns: 0, 7, 14, 21, and 28 days; 2, 3, 4, 5, 6, Department of Population 7, 8, 9, 10, 11, 12, 15, 18, and 21 months; and 2, 3, 4, and 5 years. We assessed how probabilities of dying estimated for Health, London School of the 22 age groups deviated from predictions generated by a vital registration model that reflects the historical mortality Hygiene & Tropical Medicine, of 25 high-income countries. London, UK (J Romero Prieto PhD, H Eilerts-Spinelli PhD, Findings We calculated mortality rates of 81 LMICs between 1966 and 2020. In contrast with the other regions of the G Reniers PhD); Health Systems world, we found that under-5 mortality in south Asia and sub-Saharan Africa was characterised by increased mortality and Population Studies at both ends of the age range (ie, younger than 28 days and older than 6 months) at a given level of mortality. Observed Division, International Centre for Diarrhoeal Disease mortality in these regions was up to 2 times higher than predicted by the vital registration model for the younger- Research, Dhaka, Bangladesh than-28 days age bracket, and up to 10 times higher than predicted for the older-than-6 months age bracket. This age (N Alam PhD); Department of pattern of under-5 mortality is significant in 17 countries in south Asia and sub-Saharan Africa. Excess mortality in International Health, children older than 6 months without excess mortality in children younger than 28 days was found in 38 countries. (D J Erchick PhD, Prof J Katz ScD, L Liu PhD, S Subedi MPH, In south Asia, results were consistent across data sources. In sub-Saharan Africa, excess mortality in children younger F Villavicencio PhD); than 28 days was found mostly in DHSs; the majority of HDSSs did not show this excess mortality. We have attributed Department of Population, this difference in data sources mainly to omissions of early deaths in HDSSs. Family and Reproductive Health (L Liu), Johns Hopkins Bloomberg School of Public Interpretation In countries with age patterns of under-5 mortality that diverge from predictions, evidence-based Health, Baltimore, MD, USA; public health interventions should focus on the causes of excess of mortality; notably, the effect of fetal growth Population Division, United restriction and infectious diseases. The age pattern of under-5 mortality will be instrumental in assessing progress Nations, New York, NY, USA towards the decline of under-5 mortality and the Sustainable Development Goals. (P Gerland PhD); Institut Supérieur des Sciences de la Population, Université de Funding Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Ouagadougou, Ouagadougou, Institutes of Health. Burkina Faso (B Lankoande PhD); Laboratoire d’Eco-anthropologie, French Copyright © 2022 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND Museum of Natural History, 4.0 license. Paris, France (Prof G Pison ScD); French Institute for Introduction age (which indicates neonatal mortality, and which now Demographic Studies (INED), Paris, France The under-5 mortality rate is the probability that a features alongside under-5 mortality in the SDGs) and (Prof G Pison, Prof M Guillot); newborn baby will die before reaching 5 years of age, and 1 year of age (which indicates infant mortality), but by Centre for Demographic is a key mortality indicator routinely used for tracking weeks, months, and years of age throughout the first Studies, Universitat Autònoma de Barcelona, Barcelona, Spain progress in child health and social development in 5 years of life. (F Villavicencio) populations. As such, the under-5 mortality rate featured Using a fine degree of granularity in the study of age Correspondence to: prominently in the UN Millennium Development Goals, patterns of under-5 mortality is important for identifying Dr Andrea Verhulst, Population and it remains an important component of the UN ages at which children are particularly vulnerable, and Studies Center, University of Sustainable Development Goals (SDGs). This indicator, thus how to target resources.1,2 Childhood causes of death Pennsylvania, Philadelphia, however, conceals important information about the have different age signatures, and accurate detailed age PA 19104, USA verhulst@sas.upenn.edu distribution of mortality within the 0–5-year age range. patterns can help identify these causes.3 Greater precision For guiding and evaluating health policies, it is important in empirical estimates of age-specific mortality is also to examine how the risk of mortality varies within this necessary for modelling mortality in settings in which range; not only by the standard breakdowns at 28 days of empirical data are not available.4,5 www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 1 Articles Research in context Evidence before this study sub-Saharan Africa. These age patterns are characterised by Age patterns of under-5 mortality characterised by higher- high rates of mortality at both ends of the 0–5-year age range than-expected rates of mortality at ages 1–4 years, relative to at a given level of under-5 mortality. These divergent age mortality below the age of 1 year, have been observed in patterns were, to our knowledge, not previously identified. sub-Saharan Africa since the 1960s. Since then, existing This finding was only made possible thanks to a variety of data knowledge has been summarised, and new evidence provided, and the availability of a new reference model based on detailed about the global diversity of age patterns of under-5 mortality, age groups (22 age groups between 0 years and 5 years, including a study from 2019 that focused on sub-Saharan including weeks, months, trimesters, and years). Africa. However, the existing evidence was mostly analysed in Implications of all the available evidence broad age brackets (0 years vs 1–4 years), and there have been The age pattern of under-5 mortality reflects the effect of concerns about the quality of data, meaning that age patterns underlying causes of death at different ages and allows us to of under-5 mortality that are specific to sub-Saharan Africa identify excess and preventable mortality. This study sheds new have not been fully described and rigorously validated. light on the slower decline of child mortality in South Asia and The 2019 study stressed the importance of examining under-5 sub-Saharan Africa compared with the rest of the world. Future mortality by month of age in sub-Saharan Africa in order to progress in under-5 mortality in these two regions will require a monitor the progress of the UN Sustainable Development multi-faceted policy approach that focuses on both neonatal Goals. However, this study did not aim to explore what makes causes of death and risk factors (in particular, foetal growth the sub-Saharan age patterns of under-5 mortality different restriction and low birthweight) and causes of death after from the other world regions. We searched Google Scholar and 6 months or 1 year of age (including malaria and other parasitic Web of Science on Feb 15, 2022 for all articles published since and infectious diseases). Evidence also shows that the age inception, in English, French, and Spanish, using the search pattern of under-5 mortality will be instrumental in monitoring terms ‘age pattern’ and ‘age distribution’ combined with and understanding the effect of public policy that aims to ‘under-5 mortality’ or ‘child mortality’. We did not find other reduce mortality in children younger than 5 years. publications that have studied under-5 mortality by month of age in low-income and middle-income countries. Added value of this study This study identified and validated the existence of age patterns of under-5 mortality that are specific to South Asia and Case studies focusing on rural zones of The Gambia of a different age pattern of mortality, but instead and Senegal between 1961 and 1983 identified an attributed to errors in the data. unusual age pattern of under-5 mortality.3,6–8 A 1985 study8 In this paper, we show that detecting age patterns of summarised these findings and showed that, in these under-5 mortality that are specific to low-income and populations, mortality was 2-to-3 times higher between middle-income countries (LMICs) is only possible when a the ages of 1 year and 5 years than would be expected on fine degree of age granularity is available in the data the basis of Brass’s African Standard (an existing model sources and retained in the modelling approach. We used for mortality in Africa). Supporting this finding, global a model developed in 2022 that is based on high-quality studies of the relationship between infant and child historical and contemporary vital registration data from mortality across a variety of data showed that in most 25 high-income countries sourced from the Under-5 sub-Saharan African countries, mortality after 1 year of Mortality Database.4 Using a multiplicity of retrospective age was higher than expected on the basis of a larger set and prospective data sources, we compared age patterns of standard models.9–11 However, a 2021 study9 observed of under-5 mortality in LMICs with predictions from the that, before 1 year of age, the relationship between vital registration model, which allowed us to assess the neonatal and postneonatal mortality (age between extent to which these age patterns diverge from those 28 days and 1 year) was similar in sub-Saharan Africa seen historically in high-income countries at similar rates and high-income countries. In contrast, a 2006 study12 of under-5 mortality. found that, in India, the ratio between neonatal and postnatal mortality was about one-third higher in the Methods Indian National Family Health Surveys (ratio between Study design and data sources 0·8 and 0·6 in the surveys of 1992–93 and 1998–99) than In this modelling study, we used data from 277 standard For more on the Demographic in historical data of England and Wales (ratio between Demographic and Health Surveys (DHSs). DHSs are and Health Surveys see http:// 0·6 and 0·4 at the same levels of infant mortality). the most comprehensive and publicly available source of www.dhsprogram.com However, this deviation was not considered an indication data for the study of under-5 mortality in LMICs. We 2 www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 Articles selected all the publicly available standard surveys Cumulative probabilities of dying were compared with few exceptions; exclusion criteria are listed in the with those predicted by the vital registration model, appendix (p 16). DHSs retrospectively collect data from controlling for a given level of under-5 mortality See Online for appendix reproductive-age women, including the date of birth and (appendix p 2). We used ratios between observed and the age at death of their liveborn children (age at death is predicted probabilities of dying to detect age patterns of recorded by day for the first month of age and by month under-5 mortality in LMICs that diverge from the vital for the first 2 years of age, otherwise in years). We used registration model. For each DHSs and each age group, the date of birth and the age at death from all records we calculated 95% uncertainty ratios using a standard collected between 1986 and 2020 to estimate death rates jackknife variance estimation method.14 by detailed age with a standard event–exposure method.13 To quantify how different age ranges contributed There were no missing data in DHS datasets and no separately to deviations from the vital registration exclusion criteria. model’s predictions, we used the probability of dying To validate the divergent age patterns found in between age 4 months and 6 months instead of the DHSs, we searched for all publicly available alternative under-5 mortality rate as a predictor. Using the probability prospective data providing the same detailed age of dying between age 4 and 6 months as a predictor has breakdowns over similar periods of time in a given the advantage of bringing out the role of underlying country. The largest alternative source of data that we causes of death: perinatal conditions and congenital identified is the Health and Demographic Surveillance anomalies before 28 days of age and infectious diseases Systems (HDSSs), which is a population-based health after 6 months of age. Another advantage of using the and vital event registration system that is publicly probability of dying between age 4 months and 6 months available on the INDEPTH data repository. We also is the robustness to potential omissions of early deaths, For more on the INDEPTH data relied on bilateral agreements to access some add­ an issue of concern in data from LMICs. Therefore, repository see https://www. indepth-ishare.org itional non-public data, including case studies and this predictor enables us to better understand whether sample-registration systems (appendix p 16). divergent age patterns found are real or biased by data As a result, the prospective data include 58 HDSSs, errors. By controlling for the probability of dying between two cohort studies, and two sample-registration systems. 4 months and 6 months of age, we are able to compare HDSS and cohort studies include similar variables age patterns across different geographical areas and to as DHSs (date of birth and date of death) and allowed verify the existence of divergent age patterns in a given a similar mortality estimation method. The sample- country. registration systems data required a different procedure (appendix p 16). Using each data source, we computed Role of the funding source 22 deaths rates with the following age breakdowns: 0, 7, The funder of the study had no role in study design, 14, 21, and 28 days; 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 15, 18, data collection, data analysis, data interpretation, or the and 21 months; and 2, 3, 4, and 5 years. This study writing of the report. was approved by the Institutional Review Board of the University of Pennsylvania. Results We analysed 277 DHSs across 80 countries, covering the Vital registration model period 1976–2020. We classified the DHSs results into The vital registration model summarises how under-5 five categories depending on the ratio of observed to mortality is distributed across the same 22 age groups in predicted cumulative probabilities of dying between 25 high-income countries between 1920 and 2016. The the ages of 0 years and 5 years, and assessed the degree model is based on two parameters: the under-5 mortality of overlap or deviation from the range of predictions rate as an indicator of the overall level of mortality and a allowed by the model (figure 1). A representation of the parameter, k, that determines the earliness or lateness of corresponding cumulative probabilities of dying (q[x]) the age pattern of mortality relative to the model’s used to compute the ratios is provided in the average. Predictions can be obtained from any single appendix (p 2). Three categories correspond to age mortality rate within the age range 0 years to 5 years patterns that fall within the scope of the vital registration (appendix p 2).4 model, and two categories correspond to age patterns that diverge from it. The comparison of the blue and red Data analysis areas in figure 1 shows the extent to which age patterns For each data source from LMICs, we cumulated the of under-5 mortality observed by DHSs correspond to, death rates in order to obtain 22 probabilities of dying or deviate from, those observed in the vital registration from birth to age x (represented by q[x]) or probabilities data, which represent under-5 mortality in high-income of dying from age z to age x (represented by q[z, x]). countries historically. Results for each individual survey Cumulative probabilities of dying include standard are in the appendix (p 21). mortality indicators such as the neonatal mortality and Surveys for which ratios are more than 1 (37 [13%] of under-5 mortality rates. 277 surveys; appendix p 6) display an early age pattern www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 3 Articles A Early pattern of under-5 Late pattern of under-5 Average pattern of B Early and late pattern of Very late pattern of mortality mortality under-5 mortality under-5 mortality under-5 mortality 1·50 1·25 1·00 Ratio 0·75 Data DHS Vital registration model 0·50 0 12 24 36 48 60 0 12 24 36 48 60 0 12 24 36 48 60 0 12 24 36 48 60 0 12 24 36 48 60 Age (months) Age (months) Age (months) Age (months) Age (months) Figure 1: Ratio of observed to predicted cumulative probabilities of dying from birth to age x (q[x]), controlling for the level of the under-5 mortality rate, including DHS age patterns that are within the range predicted by the vital registration model (A) and DHS age patterns that diverge from the vital registration model (B) Ratios were computed by dividing observed by predicted probabilities of dying from birth to age x values for 22 age groups between age 0 years and age 5 years. Ratios are presented on a log scale. The blue area covers the range of DHS estimates. The blue line represents the mean ratios across DHS. The red area represents the range of predictions of the vital registration model. Curves were smoothed with splines. The unsmoothed curves and spread of the raw data are shown in the appendix (p 4). DHS=Demographic and Health Surveys. of under-5 mortality, which is suggestive of a higher rate 277 surveys with a very late age pattern that falls outside of mortality before 28 days of age in the DHS data than the range of values predicted by the model. This category the average age pattern of the vital registration model at is potentially an amplified version of the early and late a given level of under-5 mortality. Surveys with ratios of category in which the high mortality after 6 months of age less than 1 (59 [21%] of 277 surveys) captured a late tends to conceal excess mortality at younger ages. age pattern, which suggests a higher rate of mortality These categories of age patterns of under-5 mortality after 28 days in the DHS data than the average age are clustered regionally (figure 2). The large majority pattern of the vital registration model (appendix p 2). (138 [97%] of 142 surveys) of divergent age patterns Nevertheless, the ratios for the early and late categories (ie, early and late and very late categories) are in remain within the range of predictions of the model. We south Asia and sub-Saharan Africa (appendix p 6). South classified the average category as surveys that also Asian divergent surveys (12 [52%] of 23) are mostly remain within the limits of the model but that have comprised of the early and late divergence (10 [83%] of several crossovers more than 1 or less than 1 (35 [13%] of 12 surveys). These divergent patterns are observed in 277 surveys). These crossovers reflect random variation three countries: Bangladesh, India, and Nepal (appendix around the model average. Populations that fall into p 21). Sub-Saharan African divergence (126 [89%] of these first three categories are well represented by the 142 surveys) mainly comprises very late divergence vital registration model. (114 [90%] of 126 surveys). However, this finding is not We then identified 144 surveys (52%) with an age the case for southern Africa, for which only two (22%) of pattern of under-5 mortality that clearly falls outside the nine surveys captured a divergent age pattern. The other range of predictions of the model, indicating divergent world regions have patterns that are within the range of age patterns. In the fourth category, early and late, we prediction of the vital registration model. identified 22 (8%) of 277 surveys that remain within the Regional clustering has remained stable over time, limits of the model but that have a single crossover. These but some divergent age patterns have converged surveys have ratios more than 1 at the younger- towards the age patterns represented by the vital than-6 months age range and ratios less than 1 at the registration model. In south Asia, countries with an older-than-6 months age range. This crossover reflects a early and late age pattern have transitioned towards the higher concentration of mortality than predicted at both early pattern over time, especially after 2012 (appendix ends of the age range between 0 years and 5 years p 21). In sub-Saharan Africa, the change in age patterns (appendix p 2). The fifth category includes 120 (43%) of has been slower than in south Asia; however, few 4 www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 Articles A Early Average Late Early and late Very late No DHS B Figure 2: Clustering of categories of age patterns of under-5 mortality by region, from the first collected DHS in a given country (average year of collection 1993; A) to the last collected DHS in a given country (average year of collection 2014; B) For countries with only one survey, the result was shown in A if the data were collected before 2000 and in B if after. As an exception, the data for China come from the National Maternal and Child Health Surveillance System (2001–15; appendix p 5). DHS=Demographic and Health Surveys. countries, including Kenya, Namibia, and Zimbabwe, studies collected in Tamil Nadu, India, and Sarlahi, Nepal; transitioned from the very late to the late category after appendix p 8). It is notable that the ratios for neonatal 2000 (appendix p 21). mortality are consistent across data sources, at about 2, We compared DHS data from south Asia with HDSS despite the variety of geographical areas and time periods data from Matlab, Bangladesh, and sample-registration represented in the study (figure 3; appendix p 8). This systems data from India across similar periods, controlling finding is particularly striking for the case of Matlab, for the level of mortality between 4 and 6 months. where data have been collected since 1966. In contrast, the Consistently, the three data sources depict the same clear different sources of data support the decreased ratio of U-shaped pattern reflecting excess mortality at both early observed to predicted mortality in children aged between and late ages relative to the model prediction; in other 6 months and 5 years. However, the regional estimates words, higher-than-predicted mortality at both ends of the from the Indian DHS (2005–06) and sample-registration 0–5 year age range (up to 2 times higher before 28 days of systems (2004–13) show that in some regions excess age and more than 5 times higher after 2 years of age; mortality after 1 year can remain high (particularly in the figure 3). We also provide the mean values of the deviation central and eastern regions; appendix pp 8, 11). ratios and supply additional results for three other sites In contrast, the DHS and HDSS findings from sub- (HDSS data from Chakaria, Bangladesh, and two case Saharan Africa are not consistent (figure 3; appendix p 9). www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 5 Articles of age; appendix p 10). In sub-Saharan Africa, the excess South Asia Sub-Saharan Africa 10·00 mortality at neonatal ages tends to be concentrated in west Africa (average ratio of 1·28 before 28 days of age; appendix p 10). Results for individual surveys show that the excess mortality at both ends of the 0–5-year age 5·00 range is significantly higher than 1 in most south Asian 4·00 and west African countries, but also in Ethiopia (appendix 3·00 p 12). Excess mortality after 2 years of age is significant in almost all sub-Saharan countries (appendix p 12). 2·00 Discussion 1·25 To our knowledge, studies of under-5 mortality have Ratio 1·00 mostly focused on early versus late age patterns; that is, 0·75 higher concentration of mortality either before or after 28 days (or 1 year) of age relative to a reference age 0·50 pattern. Using detailed age groups, we showed that there were excesses of mortality at both ends of the 0–5-year age range; namely an excess of mortality before 28 days of age and after 6 months of age. Populations with these 0·25 Data Data age patterns of under-5 mortality are primarily located in DHS (1982−2011) DHS (1976−2020) south Asia and sub-Saharan Africa. These results were HDSS−Matlab (1966−2009) HDSS (1972−2017) Sample registration system−India found consistently across data sources except for the (2004−13) excess mortality before 28 days of age in sub-Saharan 0·10 Africa. In contrast with DHSs, a majority of HDSSs s s s ar s o2 s s s s ar s rs did not show excess of mortality before 28 days of age. ay ay th th ar r th th ar ea ea ye ye ye ye on on on on 8d 8d 5y 5y o1 o1 o2 o2 m m m m o2 Later, we suggest that this difference is probably due to to st st 4 o6 4 o6 rt rt st st th th rs rs to to ea ea ay ay st st ea ea on on ys ys to omissions of early deaths in HDSSs. Improved 1y 1y 0d 0d th th 2y 2y m m da da on on 6 6 28 28 m m understanding and appreciation of this pattern of under-5 4 4 Age Age mortality will be instrumental for designing interventions Figure 3: Ratio of observed versus predicted mortality using DHS data with divergent age patterns and to reduce mortality rates. alternative sources of data, by age group and controlling for the level of mortality between 4 and 6 months With the decline in mortality, deaths are increasingly Ratios were computed by dividing observed by predicted probabilities of dying between ages z and x for six age concentrated in the first weeks of life,15 and the SDGs groups between age 0 years and age 5 years. Ratios are presented on a log scale. The shaded areas cover the range aim to reduce neonatal mortality to 12 deaths or fewer of estimates. The lines represent the mean ratios across estimates. The spread of the raw data is shown in the appendix (p 7). Alternative sources of data include HDSS and sample registration system. DHS=Demographic and per 1000 livebirths in all countries by 2030. In south Asia Health Surveys. HDSS=Health and Demographic Surveillance Systems. and sub-Saharan Africa, we found that, along with the decline of neonatal mortality, the excess mortality at Both sources of data support the existence of a strong age 0–28 months has remained stable. Therefore, this excess mortality after 2 years, up to 10 times higher than specific age pattern of mortality at early ages could predicted by the vital registration model (figure 3). explain why south Asia and sub-Saharan Africa have However, we found poor consistency between DHS and neonatal mortality rates twice or thrice higher than in the HDSS data for neonatal mortality; at a given level of other regions of the world. mortality between 4 months and 6 months of age, the Neonatal deaths are mainly related to preterm birth, HDSS estimates of neonatal mortality are generally intrapartum events, congenital malformation, and lower than the DHS estimates (1·00 average ratio for respiratory infection.16 However, the usual cause-specific HDSSs versus 1·26 for DHSs). This finding is observed mortality fractions do not explain the concentration in most countries separately, with very low ratios in some of deaths during the first weeks of life. In south Asia, cases (approximately 0·50 on average in Burkina Faso, where an early age pattern of under-5 mortality was Côte d’Ivoire, and Mozambique [appendix p 9]). found to be the highest, none of the main causes of Further comparisons across all DHSs (appendix p 10) neonatal death have been more prominent than in the in south Asia and sub-Saharan Africa show that excesses other world regions. In contrast, the factor that stands of mortality before 28 days and after 6 months of age, out in this region is the exceptionally high prevalence of and particularly from 1 year of age, are unique in their fetal growth restriction, measured as the prevalence magnitude. This finding holds even when examining of small for gestational age (defined as birthweight in these two age groups separately (eg, in the comparison of the lowest 10th percentile of a reference population for a divergent patterns in south Asia with early patterns in given gestational age and sex), which in 2010 was 45% in the other regions, there were average ratios of 1·82 vs south Asia and 24% in sub-Saharan Africa.17 However, 1·31 before 28 days of age, and 2·73 vs 0·96 after 2 years in most sub-Saharan countries in which we found high 6 www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 Articles concentration of neonatal mortality, particularly in west at 3, 6, 9, 12, 18, or 24 months of age and could have Africa, the prevalence of small for gestational age was biased our estimates. However, the triangulation of data above average and as high as 36% (appendix p 15). Alone collected retrospectively (ie, based on the mother’s recall or combined with prematurity, previous studies found of their entire reproductive history) as in the DHSs, and that small for gestational age was associated with a 2-to- prospectively (ie, collecting vital events longitudinally 15 times increased risk of neonatal death in populations in households over recurrent periods of time) as in from south Asia and Africa.17–20 Access to, and quality of, the HDSSs, means that some part of these potential health care are also important factors that contribute displacements can be controlled for: prospective data to the specific excess of neonatal mortality in these reduces the recall period of the respondent, and so the two regions.21 risk of event displacement is substantially reduced. For At the other end of the 0–5-year age range, excess example, in contrast with retrospective survey data mortality from 6 months of age reflects the large effect of collected in the same area, the HDSS data from Matlab infectious and parasitic diseases. In a case study of the in Bangladesh (with monthly visits) and Dabat in village of Ngayokhem in rural Senegal (1963–81), two age Ethiopia (with visits twice a year) did not show evidence periods of high risks were identified.3 The first period of age heaping.28 Nonetheless, in both countries, we was from 6 months to 17 months and was associated with found an excess of mortality on both ends of the the loss of the maternal passive immunity; mortality 0–5-year age range across all sources of data. during this period was strongly associated with malaria. Omissions of deaths can also affect the age pattern of Results from 2001 to 2005 showed that this age pattern under-5 mortality. In particular, omissions of early deaths of malaria mortality was similar across endemic areas of are a major concern in all sources of data.9 However, Africa.22 The second period (18 months to 60 months) these omissions would make our results conservative, was marked by the age-specific rise of diarrhoeal diseases underestimating the true amount of excess mortality and measles due to the weaning of children. In 2019, before 28 days of age. Evidence of such omissions in the parasitic and infectious diseases including malaria, HDSSs of sub-Saharan Africa is shown by the lower- diarrhoeal diseases, measles, and lower respiratory than-predicted rate of neonatal mortality, and the infections were causing 73% of deaths occurring between divergence with the DHSs that display excess mortality at 1 month and 5 years of age in west and central Africa, this age. Although lower-than-predicted mortality could compared with less than 9% in children of the same age be associated with the effect of health interventions from in high-income countries.16,23 In east and southern Africa, which the HDSSs have benefited,30 the absolute rate the percentage of deaths generated by these causes of neonatal mortality points to the quality of data. In was 52% in 2019, and 4% of deaths were attributed to 25% of the HDSSs from sub-Saharan Africa, the rates of HIV/AIDS. neonatal mortality are as low as 2–13 deaths per 1000 live In south Asia, the decline of parasitic and infectious births. These low mortality rates are implausible given diseases (mainly diarrhoea and respiratory infections) that over the considered period (1995–2020), neonatal probably explains the convergence of the age patterns mortality declined from 44 to 27 deaths per 1000 livebirths after 6 months of age towards the age patterns of the in sub-Saharan Africa, from five to two deaths per other world regions. However, at the subnational level, 1000 livebirths in west Europe, and from 34 to 17 deaths areas such as the central and east regions of India have per 1000 livebirths at the global scale.31 Although once maintained strong divergent age patterns of under-5 recorded in HDSSs a birth will probably not be omitted mortality; pneumonia and diarrhoea are still the leading over the following visits, these low rates of neonatal causes of under-5 mortality in these two regions.24,25 With mortality corroborate previous evidence that a pregnancy the decline in mortality, it is expected that preventable followed by an early death can easily go unnoticed.9 mortality will be increasingly concentrated in subnational Therefore, we think that the age pattern of under-5 areas and hotspots.26 The age patterns of under-5 mortality is also an important tool to identify data quality mortality, as shown by our study, are important for issues. Despite these limitations associated with some identifying geographical areas that are susceptible to of the HDSSs, the triangulation of retrospective and high, preventable mortality rates. prospective data provides key evidence that the divergent One limitation of our study is the potential errors age patterns that we have identified are true rather than affecting the data. Two principal sources of errors can data artifacts. affect the variables in this study and potentially bias the Another limitation of our study is the sparsity of age pattern of under-5 mortality: displacements and direct evidence linking the age patterns we observe to omissions of deaths.27 First, the displacement of deaths underlying explanations. Even though we provide (through the misreporting of the age at, or date of, possible explanations on the basis of existing literature, death) is reflected by digit preference or age heaping. further research is needed to better understand the Transference of deaths across the 28th day of life factors causing the divergent age patterns that we iden­ can either underestimate or overestimate neonatal tified, including proper quantification of the burden of mortality.28,29 In DHSs, age heaping can potentially occur diseases and the risk factors of mortality in narrow age www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 7 Articles groups between 0 years and 5 years. Some local sites 4 Guillot M, Romero Prieto J, Verhulst A, Gerland P. Modeling age with divergent age patterns such as the HDSS of Matlab patterns of under-5 mortality: results from a log-quadratic model applied to high-quality vital registration data. Demography 2022; have benefited from evidence-based public health 59: 321–47. interventions aiming to reduce under-5 mortality.32 In 5 Romero Prieto J, Verhulst A, Guillot M. Estimating the infant these sites, the age pattern of under-5 mortality should mortality rate from DHS birth histories in the presence of age heaping. PLoS One 2021; 16: e0259304. be used to track progress and understand the effect of 6 Cantrelle P, Leridon H. Breast feeding, mortality in childhood and these interventions. For example, further work should fertility in a rural zone of Senegal. Popul Stud (Camb) 1971; investigate how maternal nutrition interventions affect 25: 505–33. 7 McGregor IA, Williams WZ. Mortality in a rural west African the excess mortality before 28 days of age. Thereby, the village (Keneba) with special reference to deaths occurring in the age pattern of under-5 mortality could potentially first five years of life. London, UK: Medical Research Council, become an indicator of the effect of such interventions 1979. 8 Pison G, Langaney A. The level and age pattern of mortality in when specific data such as birthweight are not available. Bandafassi (eastern Senegal): results from a small-scale and Availability of mortality estimates by detailed age intensive multi-round survey. Popul Stud (Camb) 1985; 39: 387–405. between age 0 years and age 5 years is essential for inter­ 9 Eilerts H, Prieto JR, Eaton JW, Reniers G. Age patterns of under-5 ventions aiming to improve child health. In situations in mortality in sub-Saharan Africa during 1990-2018: a comparison of estimates from demographic surveillance with full birth histories which data are scarce, mortality estimates by detailed age and the historic record. Demogr Res 2021; 44: 415–42. can be inferred on the basis of models representing age 10 Guillot M, Gerland P, Pelletier F, Saabneh A. Child mortality patterns of under-5 mortality. Therefore, the divergent age estimation: a global overview of infant and child mortality age patterns in light of new empirical data. PLoS Med 2012; patterns that we identified in south Asia and sub-Saharan 9: e1001299. Africa call for an important update of existing models, 11 Hill K. Age patterns of child mortality in the developing world. which tend to be heavily based on data from high-income Popul Bull UN 1995; 39: 112–32. 12 Hill K, Choi Y. Neonatal mortality in the developing world. countries. Demogr Res 2006; 14: 429–52. Contributors 13 Hill K. Direct estimation of child mortality from birth histories. PG, JK, LL, GP, GR, and MG conceptualised the project. AV, JRP, In: Moultrie T, Dorrington R, Hill A, Hill K, Timæus I, Zaba B, eds. and MG designed the study. AV, JRP, HE-S, DJE, BL, SS, and FV Tools for Demographic Estimation. Paris, France: International prepared the data for analysis. AV and MG directly accessed and verified Union for the Scientific Study of Population, 2013. the underlying data reported in the manuscript. AV and JRP developed 14 Pedersen J, Liu J. Child mortality estimation: appropriate time the software. AV and MG did the analysis and wrote the manuscript. periods for child mortality estimates from full birth histories. All authors reviewed results, discussed interpretations, and contributed PLoS Med 2012; 9: e1001289. to the revision of the manuscript. All authors had full access to all the 15 Karlsson O, Kim R, Hasman A, Subramanian SV. Age distribution of data in the study and had final responsibility for the decision to submit all-cause mortality among children younger than 5 years in low- and middle-income countries. JAMA Netw Open 2022; 5: e2212692-e. for publication. 16 Perin J, Mulick A, Yeung D, et al. Global, regional, and national Declaration of interests causes of under-5 mortality in 2000-19: an updated systematic GR reports grants from Bill & Melinda Gates, from the Evidence Fund, analysis with implications for the Sustainable Development Goals. and from UNICEF during the conduct of the study. All other authors Lancet Child Adolesc Health 2022; 6: 106–15. declare no competing interests. 17 Lee ACC, Katz J, Blencowe H, et al. National and regional estimates of term and preterm babies born small for gestational age in Data sharing 138 low-income and middle-income countries in 2010. The majority of data used in this study are publicly available Lancet Glob Health 2013; 1: e26–36. (appendix p 16). For the additional data that are not publicly available, 18 Black RE, Allen LH, Bhutta ZA, et al. Maternal and child the authors will share the estimated aggregated death rates on demand. undernutrition: global and regional exposures and health An R package to apply the model used in the study is available on a consequences. Lancet 2008; 371: 243–60. GitHub repository (https://github.com/verhulsta/logquad5q0). 19 Black RE, Victora CG, Walker SP, et al. Maternal and child Acknowledgments undernutrition and overweight in low-income and middle-income countries. Lancet 2013; 382: 427–51. Research reported in this article was fully supported by the Eunice Kennedy Shriver National Institute of Child Health and Human 20 Katz J, Lee ACC, Kozuki N, et al. Mortality risk in preterm and small-for-gestational-age infants in low-income and middle-income Development of the National Institutes of Health (award R01HD090082). countries: a pooled country analysis. Lancet 2013; 382: 417–25. Early versions of this paper were presented in 2021 at the Population 21 Lawn JE, Cousens S, Zupan J. 4 million neonatal deaths: when? Association of America Annual Meeting as well as at the International Where? Why? Lancet 2005; 365: 891–900. Population Conference. The views expressed in this paper are those of 22 Abdullah S, Adazu K, Masanja H, et al. Patterns of age-specific the authors and do not necessarily reflect the views of the United mortality in children in endemic areas of sub-Saharan Africa. Nations. The authors would like to thank participants of these seminars, Am J Trop Med Hyg 2007; 77 (suppl): 99–105. as well as Jere Behrman, Robert Black, and Ken Hill for their feedback. 23 Black RE, Liu L, Hartwig FP, et al. Health and development from References preconception to 20 years of age and human capital. Lancet 2022; 1 Mejía-Guevara I, Zuo W, Bendavid E, Li N, Tuljapurkar S. 399: 1730–40. Age distribution, trends, and forecasts of under-5 mortality in 24 Liu L, Chu Y, Oza S, et al. National, regional, and state-level 31 sub-Saharan African countries: a modeling study. PLoS Med all-cause and cause-specific under-5 mortality in India in 2000–15: 2019; 16: e1002757. a systematic analysis with implications for the Sustainable 2 Burstein R, Wang H, Reiner RC Jr, Hay SI. Development and Development Goals. Lancet Glob Health 2019; 7: e721–34. validation of a new method for indirect estimation of neonatal, 25 Bassani DG, Kumar R, Awasthi S, et al. Causes of neonatal and infant, and child mortality trends using summary birth histories. child mortality in India: a nationally representative mortality survey. PLoS Med 2018; 15: e1002687. Lancet 2010; 376: 1853–60. 3 Garenne M. The age pattern of infant and child mortality in 26 Burke M, Heft-Neal S, Bendavid E. Sources of variation in under-5 Ngayokheme (rural west Africa). Philadelphia, PA, USA: University mortality across sub-Saharan Africa: a spatial analysis. of Pennsylvania, 1982. Lancet Glob Health 2016; 4: e936–45. 8 www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 Articles 27 Pullum TW, Becker S. Evidence of omission and displacement in 30 Deribew A, Ojal J, Karia B, Bauni E, Oteinde M. Under-five DHS birth histories. DHS Methodological Reports No. 11. Rockville, mortality rate variation between the Health and Demographic MD, USA: ICF International, 2014. Surveillance System (HDSS) and Demographic and Health Survey 28 Nareeba T, Dzabeng F, Alam N, et al. Neonatal and child mortality (DHS) approaches. BMC Public Health 2016; 16: 1118. data in retrospective population-based surveys compared with 31 UNICEF. Child mortality estimates. Regional and global neonatal prospective demographic surveillance: EN-INDEPTH study. mortality rate. 2021. https://data.unicef.org/resources/dataset/ Popul Health Metr 2021; 19 (suppl 1): 7. neonatal-mortality-data/ (accessed June 22, 2022). 29 Helleringer S, Liu L, Chu Y, Rodrigues A, Fisker AB. Biases in survey 32 Alam N, Ali T, Razzaque A, et al. Health and Demographic estimates of neonatal mortality: results from a validation study in Surveillance System (HDSS) in Matlab, Bangladesh. Int J Epidemiol urban areas of Guinea-Bissau. Demography 2020; 57: 1705–26. 2017; 46: 809–16. www.thelancet.com/lancetgh Published online September 8, 2022 https://doi.org/10.1016/S2214-109X(22)00337-0 9