Trends in socioeconomic disparities in a rapid under-five mortality transition: a longitudinal study in the United Republic of Tanzania

Abstract Objective To explore trends in socioeconomic disparities and under-five mortality rates in rural parts of the United Republic of Tanzania between 2000 and 2011. Methods We used longitudinal data on births, deaths, migrations, maternal educational attainment and household characteristics from the Ifakara and Rufiji health and demographic surveillance systems. We estimated hazard ratios (HR) for associations between mortality and maternal educational attainment or relative household wealth, using Cox hazard regression models. Findings The under-five mortality rate declined in Ifakara from 132.7 deaths per 1000 live births (95% confidence interval, CI: 119.3–147.4) in 2000 to 66.2 (95% CI: 59.0–74.3) in 2011 and in Rufiji from 118.4 deaths per 1000 live births (95% CI: 107.1–130.7) in 2000 to 76.2 (95% CI: 66.7–86.9) in 2011. Combining both sites, in 2000–2001, the risk of dying for children of uneducated mothers was 1.44 (95% CI: 1.08–1.92) higher than for children of mothers who had received education beyond primary school and in 2010–2011, the HR was 1.18 (95% CI: 0.90–1.55). In contrast, mortality disparities between richest and poorest quintiles worsened in Rufiji, from 1.20 (95% CI: 0.99–1.47) in 2000–2001 to 1.48 (95% CI: 1.15–1.89) in 2010–2011, while in Ifakara, disparities narrowed from 1.30 (95% CI: 1.09–1.55) to 1.15 (95% CI: 0.95–1.39) in the same period. Conclusion While childhood survival has improved, mortality disparities still persist, suggesting a need for policies and programmes that both reduce child mortality and address socioeconomic disparities.


Introduction
Despite evidence that social, demographic and residential disparities in the mortality of children younger than 5 years have declined, in many low-income countries -especially in sub-Saharan Africa -such disparities still exist. 1 Measures of socioeconomic status such as poverty, 2-7 maternal educational attainment, 5,8-10 social class, 11,12 geographical setting, 13,14 and rural and remote residence 15,16 have all been shown to be associated with under-five mortality rates. However, less is known about whether socioeconomic disparities decrease as a result of a reduction in child mortality.
In the United Republic of Tanzania, the under-five mortality rate has declined by 40% from 137 deaths per 1000 live births in 1992-1996 to 81 in 2006-2010. [17][18][19][20] However, there are still disparities in risk of death among children younger than 5 years. Mortality is highest among infants, boys, children of uneducated mothers, children of youngest or oldest mothers and children from relatively poor households.
Here we describe the mortality transition between 2000 and 2011 in three rural districts of the United Republic of Tanzania and show the longitudinal associations with indicators of health equity, i.e. maternal and household factors.

Surveillance system characteristic
We used data from the Ifakara Health Institute's integrated health and demographic surveillance system that continu-ously collected data on mortality levels in conjunction with data on social and economic factors. 21,22 The data cover the entire population of the surveillance sites, their household relationships and demographic events such as births, deaths and migration in and out of households.
The surveillance system collects data in three rural districts in the United Republic of Tanzania. The Rufiji surveillance system is located in Rufiji district in the eastern part of the country while the Ifakara surveillance system is located in Kilombero and Ulanga districts in the south central part of the country. The surveillance sites cover approximately 4200 km 2 (1800 in Rufiji and 2400 in Ifakara) and 58 villages (33 in Rufiji and 25 in Ifakara). The sites are largely rural with economies that are dominated by subsistence farming, fishing and petty trading. 21,22 Baseline census data were obtained in September 1996 for the Ifakara surveillance population and in November 1998 for the Rufiji surveillance population. Since then, the data have been updated in 120-day intervals by interviewers revisiting households to register the occurrence of all births, deaths, household migrations and marital status changes. Pregnancies are documented and pregnancy outcomes are solicited in subsequent household visits. In annual cycles, data on socioeconomic indicators are collected such as level of education, occupation and household socioeconomic status measured with proxy markers -such as household possessions, household water and sanitation and building materials.
In the beginning of 2011, the total surveillance population was about 190 000 persons (82 012 persons in Rufiji and Objective To explore trends in socioeconomic disparities and under-five mortality rates in rural parts of the United Republic of Tanzania between 2000 and 2011. Methods We used longitudinal data on births, deaths, migrations, maternal educational attainment and household characteristics from the Ifakara and Rufiji health and demographic surveillance systems. We estimated hazard ratios (HR) for associations between mortality and maternal educational attainment or relative household wealth, using Cox hazard regression models.

Variable definitions
We collected information on all children younger than 5 years living in the surveillance sites between 1 January 2000 and 31 December 2011. Beginning with all children younger than 5 years on 1 January 2000, we added children who entered the surveillance system through birth or migration into the study area. Children were observed until the end of the study period, 31 December 2011, unless they were censored due to death, migration out of the study area or reaching the age of 5 years. Variables included sex of child, birth order (first, second, third and fourth or higher), maternal age at birth of child (younger than 20, 20-34 and older than 35 years), maternal educational attainment (none, incomplete primary level, complete primary level and secondary level or more) and household wealth quintile (first, second, third, fourth and fifth, where first represent the poorest and the fifth the richest). Time period was included into the analysis as a continuous variable from 2000 to 2011 to account for changes in mortality over time.

Classifying relative wealth
We constructed an asset index as a proxy for wealth using the principal components analysis (PCA). 24 PCA has been shown to be a reliable method for classifying relative socioeconomic status in rural households in developing settings. 25,26 Information used to create the assets index included ownership of means of mobility, possession of electronic and electric devices, ownership of household equipment, landholding, livestock, household construction, and type of energy used for cooking or lighting.

Data limitations
Household socioeconomic data were collected each year in Ifakara from 1997. In Rufiji however, data were collected in 2000, 2004 and in every year since 2007. For the years lacking these data, we used the most recent estimate available. This limitation to our data involved the assumption that the most recent assessment of household socioeconomic status was applicable to subsequent periods and therefore, that household socioeconomic status did not change We did not control for possible confounding factors related to child characteristics, such as birth type (single or multiple), or maternal characteristics, such as marital status changes, presence or absence of household heads or the geographic remoteness of households because these data were only collected from 2003 onwards.

Regression models
The overall trends in under-five mortality for the period 2000-2011 were calcu-lated using Kaplan-Meier analysis. 27 We used the Cox proportional hazard model 28 to test for associations between the risk of child death and characteristics of children, mothers and households with multivariate controls. The efron method was used to handle tied failures. 29 A multivariate model of child survival was created to measure the relationship between mortality and maternal, child and household characteristics. This multivariate hazard regression model (model 1) was constructed as follows: (1) where h defines the hazard of childhood death for exact ages in days from t = 0 to 60 months and x defines i indicators of maternal or child characteristics that are unrelated to equity and y defines j indicators of social and economic disparity and z defines time in periods of two calendar years each.
To determine whether mortality disparities by wealth quintiles and educational status were changing during the mortality transition, a model was created that assessed for interactions between time and maternal educational attainment and between time and household wealth. This multivariate hazard regression model (model 2) was constructed as follows: (2) for both model 1 and model 2, three analyses were conducted, one using the Ifakara data set, one using the Rufiji data set and a third using the combined data set. For each main effects variable coefficient, a P-value of 0.05 was defined as statistically significant. For model 2, because of low statistical power for interaction tests, [30][31][32] we assessed the interaction coefficients with an error rate of 20% rather than the traditional 5%. 32 For statistical analyses, we used Stata version 13.1 (StataCorp. LP, College Station, United States of America).

Ethical statement
Following an explanation of the health and demographic surveillance system, each participant in the surveillance system is asked to provide written informed consent (parents, guardians or next of kin, on behalf of all minors/children enrolled in our study). Local leaders at village and district levels were informed about the survey and the use of data. Respondents remained anonymous. The data collection procedure was reviewed and approved by the ethical review committees of the Ministry of Health and Social Welfare of the United Republic of Tanzania, the National Medical Research Coordination Committee of the National Institute for Medical Research and the Ifakara Health Institute's Institutional Review Board. We sought authorization for use of data from the Ifakara Health Institute through the Connect project (IHI/IRB/No. . Ethical clearance for the analysis was accorded by the Columbia University Medical Center Institutional Review Board (Protocol-AAAF3452).

Research
Socioeconomic changes and childhood mortality in United Republic of Tanzania Almamy Malick Kanté et al.

Trends
The under-five mortality rate, measured as a probability of dying before 5 years of age, declined by 43

Model 1
Boys, children of low birth order, children of youngest or oldest mothers, children of uneducated mothers and mothers who had only attended primary school and children of poor households had a statistically significant higher risk of dying compared to others. Overall, children living in the poorest quintiles of households had a risk of dying that was 1.28 (95% CI: 1.18-1.38) times greater than children living in the richest households. Children of uneducated mothers had a risk of dying that was 1.31 (95% CI: 1.12-1.54) times greater than children whose mothers had received education beyond primary school. Over the course of the study period, the risk of dying was significantly reduced by more than one-third in Ifakara and in Rufiji. Overall, in 2010-2011 the risk of dying was 0.65 (95% CI: 0.60-0.71) times lower than in 2000-2001 (Table 2).

Model 2
Mortality disparities according to maternal educational attainment slightly decreased over time at both surveillance sites. However, reduction in the risk of dying over time for children of uneducated mothers compared to children of mothers who had received education beyond a primary school was not significant in Ifakara and Rufiji, HR: 0.97 (95% CI: 0.84-1.11) and HR: 0.93 (95% CI: 0.83-1.05), respectively (Table 3).
Disparities in mortality due to household wealth showed opposite patterns between the two surveillance sites. In Ifakara, mortality disparities between children living in the richest households and those living in poorer quintiles of households slightly declined over time.
Only children living in households that belonged to the second quintile showed a statistically significant reduction over time (HR: 0.97; 95% CI: 0.94-1.00). In Rufiji, wealth-based disparities in mortality widened over time. There was a significant increase between children living in the richest households and those living in households belonging to the second, third and fourth quintiles. Over time, the risk of dying for children living in the second quintile was 1.04 times (95% CI: 1.00-1.08) that of children living in the fifth quintile, for example (Table 3). Table 4 and

Discussion
Many developing countries are currently undergoing rapid demographic, economic and under-five mortality transitions. In these countries, research on socioeconomic health inequalities usually focuses only on childhood mortality and its determinants. 33 Few studies have explored trends in social and economic disparities associated with rapid mortality decline in sub-Saharan Africa.
Here we show a decline in the under-five mortality rate between 2001 and 2011 in Rufiji and Ifakara surveillance sites. However, variables such as the sex of the child, maternal age and educational attainment and the relative household wealth are still associated with an increased risk. During the under-five mortality transition, indicators of social and economic disparities have not narrowed. Thus, while our research confirms other studies in the improvement in childhood mortality, we demonstrate that equity problems remain despite this progress. [34][35][36] It has been shown that there is a differential improvement in proximate mortality determinants across socioeconomic groups, with slower and later improvements among more disadvantaged groups. 37 National estimates in the United Republic of Tanzania have also shown declines in under-five mortality rates during 2001 and 2011, particularly after 2005 and most prominently in rural areas. 20 This decline in rural areas is probably related to health system improvement and other factors, such as increased coverage of immunization, of insecticidetreated nets and access to clean and safe water. [38][39][40][41] Since 2000, public expenditure on health has increased twofold in the country. More resources may enable district health teams to allocate funds for cost-effective interventions in line with the local burden of diseases. 42 The WHO Integrated Management of Childhood Illness approach, vitamin A supplementation, immunization and insecticidetreated nets were introduced and scaled up during this period. 38 Our findings on the associations between disparities in the under-five mortality rate and social and demographic characteristics of children and their mothers are consistent with previous studies in sub-Saharan Africa. 6,8,10,25,43 It has been demonstrated that in sub-Saharan Africa, boys have a higher risk of dying compared to girls. [44][45][46] Higher maternal educational attainment has been shown to be associated with lower under-five mortality rates at a national level in the United Republic of Tanzania 20 and elsewhere. 8,10 Our findings that childhood mortality disparities by maternal educational attainment narrowed a HRs were calculated using multivariate analysis with interaction between time and mother's education attainment or household's socioeconomic status, Cox model. b In the analysis 73 047 children and 3744 events were included. c In the analysis 58 366 children and 3019 events were included. d In the analysis 131 413 children and 6763 events were included. e First quintile represents the poorest households and fifth quintile represents the richest. f We used two-year period averages to stabilize rates differentials due to fewer numbers of child deaths in some years. Note: We used Cox proportional hazard analysis with interaction between time and mother's education attainment or household's socioeconomic status.

Research
Socioeconomic changes and childhood mortality in United Republic of Tanzania Almamy Malick Kanté et al.
slightly over time are consistent with findings from south Asia. 47,48 In rural areas of the United Republic of Tanzania 43,49 and other low-and middle-income countries 3,4 household wealth is a predictive factor of disparities in under-five mortality. During our study period, mortality disparities between the poorest and richest worsened in Rufiji and remained stable in Ifakara. Therefore, improved child survival does not necessarily indicate improved equity. These findings are consistent with previous studies from Malawi, 34 Bangladesh 35 and Chile 36 but inconsistent with an Indonesian study showing that socioeconomic inequities in childhood mortality declined during a period of economic growth and improved child survival. 47 In conclusion, our findings suggest that children from all socioeconomic strata are benefitting from improved child survival. We also provide evidence that mortality disparities by maternal educational attainment and household wealth have not significantly narrowed. These results suggest that the longstanding equity problems of rural areas of the country still persist. Policies and health programmes are needed to reduce mortality, and offset social and economic disparities between population subgroups. ■ Funding: The study was financed by grants to Columbia University from the Doris Duke Charitable Foundation (DDCF) and support to the Ifakara Health Institute from Comic Relief. Conclusión A pesar de que la supervivencia infantil ha mejorado, las desigualdades en las tasas de mortalidad siguen presentes, lo que sugiere una necesidad de que las políticas y los programas reduzcan la mortalidad infantil y aborden las desigualdades socioeconómicas.