Dataset Information

Efficacy of deep learning methods for predicting under-five mortality in 34 low-income and middle-income countries.

ABSTRACT:

Objectives

To explore the efficacy of machine learning (ML) techniques in predicting under-five mortality (U5M) in low-income and middle-income countries (LMICs) and to identify significant predictors of U5M.

Design

This is a cross-sectional, proof-of-concept study.

Settings and participants

We analysed data from the Demographic and Health Survey. The data were drawn from 34 LMICs, comprising a total of n=1 520 018 children drawn from 956 995 unique households.

Primary and secondary outcome measures

The primary outcome measure was U5M; secondary outcome was comparing the efficacy of deep learning algorithms: deep neural network (DNN); convolution neural network (CNN); hybrid CNN-DNN with logistic regression (LR) for the prediction of child's survival.

Results

We found that duration of breast feeding, number of antenatal visits, household wealth index, postnatal care and the level of maternal education are some of the most important predictors of U5M. We found that deep learning techniques are superior to LR for the classification of child survival: LR sensitivity=0.47, specificity=0.53; DNN sensitivity=0.69, specificity=0.83; CNN sensitivity=0.68, specificity=0.83; CNN-DNN sensitivity=0.71, specificity=0.83.

Conclusion

Our findings provide an understanding of determinants of U5M in LMICs. It also demonstrates that deep learning models are more efficacious than traditional analytical approach.

SUBMITTER: Adegbosin AE

PROVIDER: S-EPMC7430449 | biostudies-literature | 2020 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Efficacy of deep learning methods for predicting under-five mortality in 34 low-income and middle-income countries.

Adegbosin Adeyinka Emmanuel AE Stantic Bela B Sun Jing J

BMJ open 20200816 8

<h4>Objectives</h4>To explore the efficacy of machine learning (ML) techniques in predicting under-five mortality (U5M) in low-income and middle-income countries (LMICs) and to identify significant predictors of U5M.<h4>Design</h4>This is a cross-sectional, proof-of-concept study.<h4>Settings and participants</h4>We analysed data from the Demographic and Health Survey. The data were drawn from 34 LMICs, comprising a total of n=1 520 018 children drawn from 956 995 unique households.<h4>Primary a ...[more]

PMID: 32801191

Similar Datasets

Project description:BackgroundThe progress to achieve the fourth Millennium Development Goal in reducing mortality rate in children younger than 5 years since 1990 has been remarkable. However, work remains to be done in the Sustainable Development Goal era. Estimates of under-5 mortality rates at the national level can hide disparities within countries. We assessed disparities in under-5 mortality rates by household economic status in low-income and middle-income countries (LMICs).MethodWe estimated country-year-specific under-5 mortality rates by wealth quintile on the basis of household wealth indices for 137 LMICs from 1990 to 2016, using a Bayesian statistical model. We estimated the association between quintile-specific and national-level under-5 mortality rates. We assessed the levels and trends of absolute and relative disparity in under-5 mortality rate between the poorest and richest quintiles, and among all quintiles.FindingsIn 2016, for all LMICs (excluding China), the aggregated under-5 mortality rate was 64·6 (90% uncertainty interval [UI] 61·1-70·1) deaths per 1000 livebirths in the poorest households (first quintile), 31·3 (29·5-34·2) deaths per 1000 livebirths in the richest households (fifth quintile), and in between those outcomes for the middle quintiles. Between 1990 and 2016, the largest absolute decline in under-5 mortality rate occurred in the two poorest quintiles: 77·6 (90% UI 71·2-82·6) deaths per 1000 livebirths in the poorest quintile and 77·9 (72·0-82·2) deaths per 1000 livebirths in the second poorest quintile. The difference in under-5 mortality rate between the poorest and richest quintiles decreased significantly by 38·8 (90% UI 32·9-43·8) deaths per 1000 livebirths between 1990 and 2016. The poorest to richest under-5 mortality rate ratio, however, remained similar (2·03 [90% UI 1·94-2·11] in 1990, 1·99 [1·91-2·08] in 2000, and 2·06 [1·92-2·20] in 2016). During 1990-2016, around half of the total under-5 deaths occurred in the poorest two quintiles (48·5% in 1990 and 2000, 49·5% in 2016) and less than a third were in the richest two quintiles (30·4% in 1990, 30·5% in 2000, 29·9% in 2016). For all regions, differences in the under-5 mortality rate between the first and fifth quintiles decreased significantly, ranging from 20·6 (90% UI 15·9-25·1) deaths per 1000 livebirths in eastern Europe and central Asia to 59·5 (48·5-70·4) deaths per 1000 livebirths in south Asia. In 2016, the ratios of under-5 mortality rate in the first quintile to under-5 mortality rate in the fifth quintile were significantly above 2·00 in two regions, with 2·49 (90% UI 2·15-2·87) in east Asia and Pacific (excluding China) and 2·41 (2·05-2·80) in south Asia. Eastern and southern Africa had the smallest ratio in 2016 at 1·62 (90% UI 1·48-1·76). Our model suggested that the expected ratio of under-5 mortality rate in the first quintile to under-5 mortality rate in the fifth quintile increases as national-level under-5 mortality rate decreases.InterpretationFor all LMICs (excluding China) combined, the absolute disparities in under-5 mortality rate between the poorest and richest households have narrowed significantly since 1990, whereas the relative differences have remained stable. To further narrow the rich-and-poor gap in under-5 mortality rate on the relative scale, targeted interventions that focus on the poorest populations are needed.FundingNational University of Singapore, UN Children's Fund, United States Agency for International Development, and the Bill & Melinda Gates Foundation.

Project description:BackgroundThe UN Sustainable Development Goals (SDGs) call for stratification of social indicators by ethnic groups; however, no recent multicountry analyses on ethnicity and child survival have been done in low-income and middle-income countries (LMICs).MethodsWe used data from Demographic and Health Surveys and Multiple Indicator Cluster Surveys collected between 2010 and 2016, from LMICs that provided birth histories and information on ethnicity or a proxy variable. We calculated neonatal (age 0-27 days), post-neonatal (age 28-364 days), child (age 1-4 years), and under-5 mortality rates (U5MRs) for each ethnic group within each country. We assessed differences in mortality between ethnic groups using a likelihood ratio test, Theil's index, and between-group variance. We used multivariable analyses of U5MR by ethnicity to adjust for household wealth, maternal education, and urban-rural residence.FindingsWe included data from 36 LMICs, which included 2 812 381 livebirths among 415 ethnic groups. In 25 countries, significant differences in U5MR by ethnic group were identified (all p<0·05 likelihood ratio test). In these countries, the median mortality ratio between the ethnic groups with the highest and lowest U5MRs was 3·3 (IQR 2·1-5·2; range 1·5-8·5), whereas among the remaining 11 countries, the median U5MR ratio was 1·9 (IQR 1·7-2·5; range 1·4-10·0). Ethnic gaps were wider for child mortality than for neonatal or post-neonatal mortality. In nearly all countries, adjustment for wealth, education, and place of residence did not affect ethnic gaps in mortality, with the exception of Guatemala, India, Laos, and Nigeria. The largest ethnic group did not have the lowest U5MR in any of the countries studied.InterpretationSignificant ethnic disparities in child survival were identified in more than two-thirds of the countries studied. Regular analyses of ethnic disparities are essential for monitoring trends, targeting, and assessing the impact of health interventions. Such analyses will contribute to the effort towards leaving no one behind, which is at the centre of the SDGs.FundingBill & Melinda Gates Foundation, UNICEF, Wellcome Trust, Associação Brasileira de Saúde Coletiva.

Project description:BackgroundWeather and climate-related disasters, including floods, impact undernutrition through multiple pathways, including food security, inadequate child care practices, and water and sanitation. This review aimed to provide systematic evidence of the impact of floods on undernutrition in children under five years of age in Low and Middle-income countries (LMICs).MethodsWe searched PubMed, Web of Science, Embase, MEDLINE, CINAHL and Scopus for peer-reviewed articles. Popline, WHO Library database (WHOLIS), the International Disaster database (EM-DAT), Food and Agriculture Organisation (FAO), UNICEF and Eldis were searched for grey literature articles. Database searches were first conducted in 2016 and updated in 2020. We included English language articles that reported the effect of floods on undernutrition outcomes in children under 5 years of age in LMICs, without limitation to study design and year of publication. The quality of selected studies was assessed using the National Institutes of Health (NIH) tool for Observational Cohort and Cross-Sectional Studies.ResultsOf the 5701 articles identified, 14 met our inclusion criteria. The review noted stunting as the most frequently reported significant form of undernutrition in flood-affected areas. Severe and recurrent floods showed the greatest impact on undernutrition. Due to weak and limited evidence, the study is inconclusive on the most significant forms within the short-term and intermediate periods following floods. On the other hand, stunting was noted as the most frequently reported significant form of undernutrition in the long-term period following floods. There was generally little evidence of the effect of floods on micronutrient deficiencies. Factors associated with child undernutrition in the flood-affected areas included age, gender, diarrhoea, maternal and paternal education, maternal age, household size, land ownership and socioeconomic status. Overall, the quality of the evidence was fairly weak, with the main challenge lying in the inability of the studies to establish causal pathways for the observed effects.ConclusionsThe review suggests clear plans and strategies for preventing and reducing the long-term impact of floods on undernutrition in children under five years. Future research utilising long-term prospective data is indispensable to provide more robust evidence to guide better prevention measures, response decisions and interventions.

Project description:Undernourishment is a persistent public health problem contributing to increased mortality in children under five in low-income countries, likely exacerbated by socio-economic disparities within communities. This paper aimed to examine the effect of wealth-related inequality on undernutrition in children under five in low, lower-middle, and upper-middle-income countries (LMICs). We analyzed cross-sectional data from the demographic and health survey program collected between 2017 and 2022 from 24 LMICs. Children born within 5 years preceding the survey were included in the analysis. Child undernutrition was the dependent variable (measured by stunting, wasting, and underweight) and country-level wealth-based inequality was the independent variable assessed by concentration index values stratified by the World Bank's income categories. Within country inequality of child undernutrition was determined by concentration index (C) values with 95% confidence intervals (95% CI) and sub-group analysis by place of residence and sex of the child. We then fit bootstrapped meta-regression to check the variation in inequality of child undernutrition across different income category countries. The analysis was controlled by potential confounding variables. From the total sample size of 334,502 children included in the study, 35% were undernourished. Wealth-related inequality in child undernutrition was observed in 11 countries, consistently across income categories. Child undernutrition was highly concentrated among the poor households of Türkiye [C: - 0.26, 95% CI - 0.31 to - 0.20], and Cameroon [C: - 0.19, 95% CI - 0.22 to - 0.17], and relatively it was less concentrated among the poor in Liberia [C: - 0.07, 95% CI - 0.11 to - 0.04], and Gambia [C: - 0.07, 95% CI - 0.11 to - 0.04]. There is no difference in undernutrition associated with inequality between the three broad LMIC categories. The wealth-related inequality in child undernutrition within many of the included countries is still very significant. However, the economic category of countries made no difference in explaining wealth-related inequality in child undernutrition. Inter-sectoral collaboration to fight poverty and render special attention to the disadvantaged population segments would potentially help to address the observed inequity.

Dataset Information

Efficacy of deep learning methods for predicting under-five mortality in 34 low-income and middle-income countries.

Objectives

Design

Settings and participants

Primary and secondary outcome measures

Results

Conclusion

Publications

Efficacy of deep learning methods for predicting under-five mortality in 34 low-income and middle-income countries.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets