Determinants of Under-five Mortality in Vietnam: Evidence From a Nationwide Survey

Study Design and Data Source

We conducted a secondary analysis of nationally representative survey data from Vietnam to examine associations between maternal, child, and household characteristics and under-five mortality. Data were obtained from the Vietnam Multiple Indicator Cluster Survey (MICS) 2020–2021, also referred to as the Survey measuring Sustainable Development Goal indicators on Children and Women (SDGCW) 2020–2021, implemented by the General Statistics Office (GSO) with support from UNICEF. ¹⁰

Survey Setting and Sampling

The SDGCW 2020–2021 used a multi-stage, stratified cluster sampling design intended to provide estimates at national and subnational levels (including urban and rural areas and major geographic domains). Sampling weights are provided in MICS because the design is generally not self-weighting. For this study, however, all descriptive and analytical analyses were conducted using unweighted data.

Study Population

The analytic cohort included children identified from women’s Fertility/Birth History records linked to maternal and household characteristics available in the women’s and household questionnaires.Children were included if they were born alive within the five years preceding the survey and had sufficient information to classify survival status and age at death (if deceased) or age at interview (if alive). Children with critical missing data on survival status or maternal identifiers were excluded. Public-use microdata and documentation were obtained through the World Bank Microdata Library and the MICS dissemination platform. ¹¹ The final analytic sample included 4,475 children under five years of age (unweighted count). To accurately capture familial and biological risks, all eligible children born to the surveyed women were included. Consequently, multiple births (twins) and various birth orders originating from the same mother were fully retained and independently coded in the analysis.

Outcome Definition

The primary outcome was under-five mortality, defined as death of a live-born child before 60 months of age. For regression analyses, under-five mortality was operationalized as a binary outcome (died before age 60 months vs alive at the time of the survey interview).

Candidate Covariates

Candidate determinants were prespecified based on the MICS content framework and epidemiologic plausibility, and grouped into four domains.

Child characteristics included child age (in months), sex, plurality, and birth order.

Household sociodemographic context included urban or rural residence, region, ethnicity (Kinh/Hoa vs other), and household wealth quintile.

Maternal characteristics and reproductive history included maternal education, health insurance coverage, self-reported overall happiness, maternal age at childbirth, preceding birth interval, and age at first sexual intercourse. Selected perinatal care practice indicators included prenatal care provider type, delivery assistance, and umbilical cord care practices.

Information environment and health awareness included mass media exposure (newspaper or magazine, radio, television), digital connectivity (computer or tablet use; mobile phone ownership), and awareness of HIV/AIDS and HPV vaccination.

Statistical Analysis

All data management and statistical analyses were conducted in R. Participant characteristics were summarized using unweighted counts and proportions for categorical variables, and appropriate summary measures for continuous variables. To ensure the robustness of the statistical models, candidate covariates were strictly required to have a missing data rate of less than 10%. Variables exceeding this threshold of missingness (e.g., maternal recall of birth weight) were excluded from the multivariable modeling. To reduce redundancy and identify a parsimonious set of predictors, we used Bayesian Model Averaging (BMA) as a variable-selection step based on posterior inclusion evidence. Associations between candidate covariates and under-five mortality were evaluated using univariable and multivariable logistic regression, reporting crude odds ratios (cOR) and adjusted odds ratios (aOR) with 95% confidence intervals. Variables not selected by BMA were evaluated in univariable models and reported in Supplementary materials (Table S1). Statistical significance was defined as P<0.05.

Ethics Approval and Informed Consent

This study analyzed publicly available, de-identified microdata from the Vietnam Multiple Indicator Cluster Survey (MICS), implemented by the General Statistics Office (GSO) of Vietnam with support from UNICEF. Under its duties and powers as defined by Vietnam’s laws, the GSO is authorized to conduct official national statistical surveys and is required to protect the confidentiality of information provided by respondents under the Statistics Law 2015 (Law No. 89/2015/QH13). In the original survey, verbal informed consent was obtained from each respondent, and all participants were informed of the voluntary nature of participation and the confidentiality and anonymity of the information collected. All maternal, child, and socioeconomic characteristics were inherently linked within the fully de-identified microdata provided by UNICEF. Because the dataset strips all personal identifiers prior to public release, no individual could be traced. Because this secondary analysis used anonymized data and involved no direct contact with participants, no additional ethical approval was required. Access to the public-use dataset was granted by UNICEF through the MICS registration and approval process (https://mics.unicef.org/).

Sociodemographic and Child Determinants Associated With Under-five Mortality

Our study exposes profound survival disparities driven by geographic and socioeconomic stratification. In our unadjusted analyses, mortality was disproportionately concentrated among disadvantaged groups, particularly rural residents and children from the poorest households. Furthermore, residing in the remote Northern midlands and mountain region was associated with a substantial increase in mortality risk. These findings mirror global patterns of disadvantage, where geographic isolation and economic barriers limit healthcare access.^4,12 Crucially, our results align with recent modeling in Vietnam indicating that poverty exerts a direct effect on under-five mortality that supersedes the influence of undernutrition alone. ¹³ These converging inequalities, based on both ethnicity and economic status, highlight the urgent need for targeted equity-focused strategies. Interventions must prioritize ethnic minority communities by addressing socioeconomic deprivation, enhancing food and nutrition security and ensuring culturally sensitive access to quality health services to meaningfully reduce preventable child deaths.

Beyond socioeconomic factors, birth-related characteristics played a pivotal role. Compared to singletons, children from multiple births exhibited a significantly higher odds of under-five mortality, which is in agreement with previous research carried out across diverse geographical settings.^14-16 Multiple births were significantly more vulnerable, attributable to inherent biological disadvantages such as prematurity, low birth weight and birth asphyxia. ¹⁷

Maternal Sociodemographic and Reproductive Determinants With Under-five Mortality

Our multivariable analysis revealed surprising insights regarding maternal age. While previous research consistently identifies adolescent childbearing (<20 years) as a major driver of under-five mortality, our adjusted model showed no significant difference in mortality risk between adolescent mothers and those aged 20–34. Instead, advanced maternal age (≥35 years) emerged as a strong protective factor. This finding contrasts sharply with several studies that associate advanced maternal age with elevated risks of stillbirth, early neonatal, and under-five mortality.^18-21 This aligns with the U-shaped pattern described by Finlay et al, who demonstrated that adolescent childbearing markedly increases the risk of neonatal and under-five mortality, largely due to biological immaturity, limited access to antenatal care, and socioeconomic disadvantage. ²² In the Vietnamese context, however, this protective effect likely reflects a socio-cognitive advantage. Older mothers may possess better financial stability, extensive child-rearing experience, and greater empowerment to seek healthcare.

In addition, our findings align with prior research showing that short birth intervals (<2 years) substantially increase child mortality risks. In Bangladesh, children born within short intervals had up to a 63% higher risk of neonatal death and nearly threefold higher under-five mortality compared to longer intervals. ²³ Similar patterns were observed in West Africa, where short spacing was linked to an 82% higher risk of under-five mortality. ²⁴ In our study, optimal spacing conferred substantial protection, with intervals of ≥4 years reducing mortality odds by over 70% (aOR = 0.278, 95% CI: 0.196–0.390, P<0.001). Biologically, closely spaced pregnancies contribute to prematurity, growth restriction, and maternal nutritional depletion, underscoring the importance of promoting optimal birth spacing as a key child survival strategy. The subgroup analysis of single births provides deeper insight into the interaction between birth spacing and birth order. While short preceding intervals consistently elevated mortality risk, this detrimental effect was particularly pronounced among children of higher parities. Although birth order was not selected in our final BMA model, likely because its statistical variance was largely captured by the more proximal birth interval variable, the combination of high parity and rapid, sequential childbearing remains a significant clinical risk, often driven by maternal nutritional depletion. These findings suggest that public health interventions and family planning programs should specifically target multiparous women to encourage adequate birth spacing.

Educational disparities strongly shaped child survival. Completing high school or above was associated with a striking reduction in mortality odds (aOR = 0.086, 95% CI: 0.052–0.137, P<0.001) compared to mothers with pre-primary or no schooling. Mandal et al. consistently showed that higher levels of maternal education are associated with significantly reduced under-five mortality rates in India. ²⁵ This aligns with robust global evidence of Balaj et al. demonstrating the protective effect of maternal education. Completing secondary schooling is associated with a striking 31% reduction (95% CI: 29.0–32.6) in under-five mortality compared with no schooling. Additionally, each extra year of maternal education corresponds to an approximate 3% decrease (95% CI: 2.8–3.2) in under-five mortality. ²⁶ These findings underscore maternal age and education as key, modifiable determinants of child mortality, highlighting the need for targeted public health interventions.

Crucially, our study highlights the life-saving impact of professional healthcare. Receiving prenatal care from a medical doctor and delivery assistance by a doctor significantly reduced mortality odds. This strongly reaffirms that clinical expertise is vital for managing obstetric complications and providing early neonatal interventions. Our descriptive analysis of care pathways further highlights the critical importance of skilled care continuity. As observed, the absence of doctor-led prenatal care combined with reliance on non-skilled delivery assistance (e.g., traditional birth attendants or relatives) creates a highly vulnerable scenario for child survival. Conversely, ensuring continuous doctor involvement throughout pregnancy and childbirth confers the greatest protective advantage. While prenatal and delivery care were evaluated as separate variables in our multivariable model to maintain statistical power and avoid sparse categories, these pathway observations emphasize the practical need for an uninterrupted chain of professional maternal care.

Other Determinants Associated With Under-five Mortality

Mothers of the deceased children experienced significant disadvantages in accessing information and digital technology. In our multivariable model, prior use of a computer or a tablet emerged as a significant independent protective factor, reducing the odds of under-five mortality by nearly 45% (aOR = 0.553; 95% CI: 0.341–0.883, P=0.015). As observed in our study, households experiencing child mortality suffered from a profound digital and media divide, which strongly aligned with markedly lower health literacy levels, such as a severely reduced awareness of infectious diseases like HIV and HPV. Access to digital technologies likely serves as a proxy for broader information access. This aligns with evidence from rural African settings showing that technology access, such as mobile phones, enhances antenatal care utilization, exclusive breastfeeding, and health-seeking behaviors, thereby supporting better child health outcomes. ²⁷ Furthermore, web-based tools have been shown to positively influence maternal and neonatal outcomes by empowering mothers with essential health knowledge. ²⁸

However, in our study, these physical environmental factors and socio-economic proxies lost significance and were not selected by the BMA algorithm for the final multivariable model. This strongly suggests that their influence operates indirectly through more proximal sociodemographic determinants. Rather than the physical structure of the house or the fuel used, it is the maternal education, adequate birth spacing, professional healthcare utilization, and digital connectivity that directly dictate child survival pathways. By addressing multicollinearity, our model emphasizes that public health interventions must prioritize empowering women through education and digital access, alongside providing skilled medical professionals, to effectively overcome environmental disadvantages.

Strengths and Limitations

This study’s primary strength is the use of a large, nationally representative dataset, ensuring high statistical power and broad generalizability. Methodologically, combining BMA with multivariable logistic regression robustly resolved multicollinearity, effectively isolating the strongest independent predictors of mortality from complex, overlapping variables. Furthermore, integrating diverse determinants which are ranging from biological vulnerabilities to socio-cognitive factors like maternal education and digital connectivity, provides a holistic, multidimensional perspective on child survival.

Our findings should be interpreted in light of some limitations. First, as a secondary analysis, the sample size was predetermined by the MICS design rather than calculated a priori for our specific multivariable models, although the cohort remains sufficiently powered. Second, the cross-sectional and retrospective design precludes causal inference, introduces recall bias, and captures environmental conditions that may not accurately reflect those present at the exact time of death. Third, the lack of longitudinal clinical registries prevents tracking precise medical causes of mortality or continuous healthcare referral pathways. Finally, critical clinical metrics like gestational age and birth weight were excluded due to substantial missingness (>10%); however, we mitigated this gap by utilizing ‘multiple births’ as a robust proxy for perinatal biological vulnerability.