1  

Does proximity to health facilities improve child survival? New evidence from a

longitudinal study in rural Tanzania

Malick Kante1,2, Stephane Helleringer1, Mrema Sigilbert2, Francis Levira2, Honorati Masanja2,

Rose Nathan2, John Noronha2, Christine Chung1, Amon Exavery2, Ahmed Hingora2, James F.

Phillips1

Abstract

Distance to health facilities is often cited as a major barrier limiting access to care in sub-Saharan and other developing countries. There are however limited data on the causal effects of distance to facilities on child survival. Existing estimates may be biased because 1) most existing data are on distance to health care facilities are cross-sectional, and 2) existing analyses do not account for the endogeneity of residential choices and health services location. This paper uses unique longitudinal data collected in a rural district of Tanzania to test whether enhanced proximity to health services arising from investment in dispensaries contributed to the rapid decline in under-five mortality recently observed in Tanzania. Data on births, deaths, household socioeconomic characteristics and migrations have been recorded every 120 days since 1999 (n≈85,000). Geographic data on the precise location of households and health facilities have also been collected over time. We use multivariate analysis 1) to measure the causal effects of distance to health facilities on child survival and 2) to test for possible interactions between distance to health facilities and socioeconomic characteristics of households (e.g., educational attainment, wealth). Initial results indicate that, from 2000 to 2010, child mortality declined close to 40% (from 110 to 70 per 1000). The distance to the closest health facility remained a strong determinant of child survival, even after adjusting for endogeneity biases. The development of community-based primary health care in rural communities by posting community health assistants, and conducting regular household visits, can improve health outcomes. It can also increase equity by offsetting the detrimental effects of low maternal education, household poverty and distance to health facilities. Key Words. Proximity, Health facility, Child survival, sub-Saharan Africa, Tanzania,

Demographic Surveillance System

                                                            1 Mailman School of Public Health, Columbia University, New York, USA 2 Ifakara Health Institute, Dar es Salaam, Tanzania 

2  

Background

According to the epidemiological transition theory, progress in medicine and health care not only

reduces the level of mortality, it also changes its structure (Omran 1971): the mean age at death

increases progressively, and the major infectious and parasitic diseases, which are the main

causes of death when mortality is very high, give way to a variety of ‘manmade and degenerative

diseases’ linked both to day-to-day behaviors that present a risk for individuals, and to the

biological ageing process. This long-term and ongoing mortality decline is generally qualified as

the ‘health transition’ as the result of progress in health care, but also of social and economic

development (Lerner 1973; Preston and Nelson 1974; Caldwell and al. 1990).

The fourth Millennium Development Goal aims to achieve a two thirds reduction in child

mortality during 1990-2015 (Word Bank, 2004; United Nations, 2009). In 2010, near to 8

millions of children were died in the Word and most of them occurred in developing countries.

Much of the current burden of child mortality in low-income countries is preventable if effective

coverage of available cost-effective interventions can be achieved (Black et al, 2003). The

average under-five mortality ratio (5q0) in the World has fallen by only third between 1990 and

2010, from 89 to 60 deaths per 1,000 live births (United Nations, 2012). Since 2000, global

health initiatives and resources for health have increased sharply which has increased coverage

of life-saving child health interventions in several countries (Horton, 2006; Masanja et al., 2008).

As the result, child mortality is declining in most of developing countries after decade of

stagnation in the 1990s, due to the mortality attributable to HIV/AIDS and malaria (Murray et

al., 2012; Trape et al., 1999; 2011; Travis et al., 2004). However, sub-Saharan Africa has still

high level of child mortality compared with other developing regions (United Nations, 2012).

Demographic and Health Surveys in Tanzania (TDHS, 1992; 1996; 2005 and 2011) have shown

a decline of 40% of under-five mortality ratio between 1992-1996 and 2006-2010 (from 137 to

81/1,000) with inequalities between age group, residence, mother’s education and wealth

quintile. For example, infant mortality is declined from 71/1000 in 2001–2005 to 51/1000 in

2006–2010. The largest decline is observed in the post-neonatal period (1-59 months), which

rates are dropped from 41 to 25/1,000 against 30 to 26/1000 for the neonatal period (0 -27 days).

At the same time, infant mortality is declining faster in rural Tanzania than urban areas, from 85

to 60/1000 against 73 to 63/1000.

3  

Better understanding the current information on child mortality trends is essential to guide efforts

to scale up child survival. This paper proposes to extend this proposition to include the role of

service proximity. Specifically, we aim to test the proposition that the impact of service

accessibility on mortality declines as the transition progresses. Moreover, we aim to test the

hypothesis that the impact of indicators of health equity, such as maternal education and relative

household wealth, diminish as the health transition progresses.

ObjectiveThis paper measures the level and analyses trends and determinants of child mortality in rural

Tanzania from 2000 to 2010, a period of rapid transition in childhood survival. The analysis aims

to clarify the role that access to nearest primary healthcare points has played in explaining the

relative pace of under-five mortality decline over 15 years of surveillance and also to elucidate

interaction of proximity with social determinants of child survival as the transition progresses.

HypothesisThis paper examines hypotheses about (1) the relative contribution of social and economic

change versus health service proximity as determinants of the rapid decline of mortality in under-

five year olds in rural Tanzania and (2) the scale up of health program by opening in each village

a dispensary has reduced social inequalities of child mortality between subgroup (poor – rich,

educated – no educated parents and easy – difficult access to health facilities).

Dataandmethod

StudysiteEver since 1998, the Ifakara Health Institute has maintained an integrated health and

demographic surveillance system (HDSS) in a rural district of the Coastal (Pwani) region

(Rufiji). The HDSS is a population based platform for research and evaluation of health

strategies and interventions. The Rufiji surveillance area extends between latitude 7.47° and

8.03° S and longitude 38.62° and 39.17° E, about 180 km south of Dar es Salaam. The district

experiences an annual precipitation ranging between 800 and 1000 millimeters, with generally

hot climatic conditions throughout the year, but seasonally heavy rains from February to May

4  

and less intensive precipitation from October to December. The annual average temperature is

26°C. The Rufiji River transects the district, producing a large deltaic flood plain.

The surveillance area covers approximately 1,800 square kilometers and 33 villages (IHI, 2011).

As of 2010, the population under surveillance was about 84,741 individuals, representing half of

the District population. The HDSS area is largely rural; although clustering of households around

townships exists. Local economies are dominated by subsistence farming, fishing and petty

trading (IHI, 2011).

HealthcareprovisionThe Rufiji surveillance area has a total of 18 health facilities (one hospital, two health centers

and 15 dispensaries) and nearly 90% of the population lives within 5 km of a formal health

facility.

DatacollectionThe Rufiji HDSS was established in November 1998. Since then all the households registered

have been visited thrice a year, once every four months. Basic demographic events including

births, deaths, migration and marital status changes are updated during these visits (rounds). To

ensure completeness in reporting births and deaths of newborns, pregnancies are also

documented and outcomes are solicited in subsequent household visits. Once annually, data on

social and economic indicators that define health equity have been compiled in HDSS areas over

time. They include level of education, occupation, and household social economic status, as

measured with proxy markers such as household condition, ownership of durable assets and

availability of services such as water and toilet facilities. As well, geographic data on the precise

location of households and health facilities have also been determined by a global positioning

system (GPS) and mapped in a geographic information system (GIS) database (Schellenberg et

al., 2001; Mwageni et al., 2002; Shabani et al., 2010).

[Figure 1 about here]

5  

Methodofanalysis

DefinitionofvariablesDepending on data availability, our analysis uses information about all 38,069 children born

between 1 January 2000 and 31 December 2010. For these children, we will analyze survival and

study mortality inequalities between subgroups. Survival analysis is concerned with studying the

time between entry to a study and a subsequent event, “the death”. HDSS dataset allow to follow

each individual during the study period.

Our dependent or outcome variable is defined by the vital status known for each individual in 31

December 2010 (end date) in the HDSS dataset. The vital status has two modalities: “DEAD” if

individual submitted the event or “Not DEAD (alive or migration)” if the individual is censored.

Our independent or explanatory variables included several characteristics of:

a. children, i.e. age, sex (male/female) and birth order (first, second-third and more than third),

b. mothers, i.e. age at birth of child (less than 20, 20-35 and more than 35 years old) and

education attainment (not educated, primary level and secondary level or more),

c. households, i.e. socioeconomic status, defined by asset score (1:poorest, 2:second, 3:middle,

4:fourth and 5:richest) and travel time to the nearest health facility, defined by less than 30

minutes of walking hour, between 30 minutes and less than 1 walking hour and more than 2

hours.

An asset index as a proxy for wealth has been constructed using principal component analysis

(PCA). The HDSS collected data annual since 2007. Before that year, data are collected only in

2000 and 2004. Information on own-ship of mobility (bicycle, motorcycle, car), of electronic and

electric materials (radio, television, fridge, clock, video, phone, fan, iron), of house equipment

(sofa, bed, net, mattress, sew, …), of land, animals, livestock (rice, millet) and material for

agriculture (hoe, machete,…), of source of water supply (tap in the household, pump, well,…),

on sanitation facility (latrines, WC, …), on type of floor, wall and roof materials and type of

energy used for cooking or lighting (gas, wood, electricity, kerosene…). Studies have found that

such indexes are relatively robust to classify households in rural developing countries (Filmer

and Pritchett, 1998 and 2001; Gwatkin et al., 2000a; 2000b; Montgomery, 2000; Schellenberg et

6  

al., 2003; Wagstaff and Watanabe, 2003; Vyas, 2006; Ickowitz, 2011). Studies are shown that

the SES are not changing yearly (REF?) we will make an assumption by keeping the same SES

for each household between 2000 and 2003 and between 2004 and 2006, i.e. if a household x in

2000 was in the “poorest” category we assume that between 2000 and 2003 it remained in that

category.

We have collected information on the nature of the ground (mud, sand, laterite, …) and also

tracked all roads available in each village using GPS. A network analysis has been conducted to

to calculate the travel time between household and nearest health facility (dispensary, health

center or hospital).

ModelWe firstly draw the general trends of child mortality and examine separately the existence of

relationships between each characteristic of children, parents, households and the risk of child

death (univariate analysis). The statistical significance is defined as a p-value <0.05. Chi-square

and Wilcoxon’s tests were used to examine possible associations between risk of child death and

characteristics of children, parents, households, villages, etc.

Then, using multivariate logistic regressions we control effects of significant variables in the

univariate variables to determine factors of child mortality decline. One of models allowing

multivariate analysis of survival data is the Cox Model. It measures the risk of death differentials

by characteristics of children (sex, age, and birth order), parents (education attainment, age)

households (socioeconomic status and travel time to the nearest health facility) in taking into

account the time in the regression. The Cox model allows also the analysis of time varying

covariates as here the socioeconomic status. It also tests for possible interactions between

distance to health facilities and socioeconomic characteristics of households (e.g., educational

attainment, wealth quintile).

 

7  

PreliminaryResults

Childmortality:level,trendsanddifferentialsUnder-five mortality rates (5q0) have been reduced close to 40% in eleven years, from 115 per

1,000 live births in 2000 to 67/1,000 in 2010. From 2000 to 2005, the average percentage of

annual decrease was slow, only 1.9%; then it accelerated during the next 5 years with 6.4% of

annual decrease.

[Figure 2 about here]

Childmortality:theCoxmodelModel1 (Annex2):

All other things being equal, boys, low birth order, not educated, youngest and oldest mothers,

poorer and households located far from health facilities have higher child mortality compared to

others. However, the model does not show how these factors interact between them and

contribute to the child survival.

Model2 (Annex3):

All other things being equal, mothers’ education doesn‘t matter if household is near to health

facilities but the SES has an impact on child mortality differentials. However, the proximity to

health facility remained a strong determinant of child survival.

[Annex 3 about here]

DiscussionandConclusion

The health equity has improved as mortality has declined but that educational attainment and

household relative wealth remain as significant covariates of childhood survival. Children of

mothers with low education, living in poor households and located more than 5km of health

facilities have a high mortality comparing to others children.

8  

Findings respond to debates at the core of primary healthcare policy in Tanzania that concern the

relative effect of geographic health service inequalities vis-à-vis the effect of social and

economic change as determinants of rapid improvement in the survival of Tanzanian children.

Although the geographic density of services remains high, residual service proximity effects

suggest that plans to launch a large scale community health services initiative could have

incremental health equity effects. Under-five year old mortality in Tanzania reduced from 137 to

81 deaths per 1000 live births between 1992-1996 and 2006-2010 , placing the country on target

to achieve Millennium Development Goals (MDG 4) (TDHS 2011; Masanja and al. 2008).

The Connect project, a randomization trial funded by Doris Duke Charitable Foundation, will

test the hypothesis that the development of community-based primary health care in rural

communities by posting community health assistants, and conducting regular household visits,

can improve health outcomes. It can also contribute to increased equity by offsetting the

detrimental effects of low maternal education, household poverty and distance to health facilities.

Caldwell J.C., Findley S., Caldwell P., Santow G.M., Cosford W.H., Braid J. and Broers-

Freeman, D. (eds.). (1990). What we know about health transition: The cultural, social and

behavioural determinants of health; The proceedings of an International Workshop, Canberra,

May 1989. Canberra: Health Transition Centre. (Health transition series; 2).

Gwatkin DR, Rustein S, Johnson K et al. 2000a. Socio-economic differences in Ethiopia. Health,

Nutrition, and Population in Ethiopia. Washington, DC: HNP/Poverty Thematic Group of the

World Bank.

Gwatkin DR, Rustein S, Johnson K et al. 2000b. Socio-economic differences in Nigeria. Health,

Nutrition, and Population in Nigeria. Washington, DC: HNP/Poverty Thematic Group of the

World Bank.

Honorati M., Don De S., Paul S., Joanna S., Theopista J., Conrad M., Gabriel U., Ties B., Cesar

V., Tom S., Hassan M. (2008). Child survival gains in Tanzania: analysis of data from

demographic and health surveys. Lancet, 371: 1276–83.

Lerner M. (1973). Modernization and health: A model of the health transition. Annual Meeting

of the American Public Health Association (APHA), San Francisco, November 1973.

9  

National Bureau of Statistics (NBS) [Tanzania] and ICF Macro. 2011. Tanzania Demographic

and Health Survey 2010. Calverton, Maryland: NBS and ICF Macro: 478.

Omran A. (1971). The epidemiologic transition: a theory of the epidemiology of population

change. Milbank Memorial Fund Quarterly 29(4): 509-538.

Preston S.H. and Nelson V.E. (1974). Structure and change in causes of death: an international

summary. Population Studies 28(1): 19-51.

Schellenberg JA, Victora CG, Mushi A et al. 2003. Inequities among the very poor: health care

for children in southern Tanzania. The Lancet 361: 561–6.

Vyas S, Kumaranayake L (2006) Constructing socio-economic status indices how to use

principal component analysis. Health Policy and Planning vol.21(no.6): 459–468.

Filmer D, Pritchett LH (2001) Estimating wealth effects without expenditure data-or tears: an

application to educational enrolments in states of India. Demography 38(1): 115–132.

Montgomery MR, Gragnolati M, Burke KA, Paredes E (2000) Measuring living standards with

proxy variables. Demography 37(2): 155–174.

Ickowitz Amy (2011), Wealthiest Is Not Always Healthiest: What Explains Differences in Child

Mortality in West Africa? Journal of African Economies, Vol. 21, number 2, pp. 192–227

doi:10.1093/jae/ejr035 online date 20 September 2011

Filmer, D. and L. Pritchett (1998) ‘The Effect of Household Wealth on Educational Attainment:

Demographic and Health Survey Evidence’, World Bank Policy Research Working Paper No.

WPS1980.

Almamy M Kante and Gilles Pison. Maternal Mortality in Rural Senegal. The Experience of the

New Ninéfescha Hospital. Population-E, 65 (4), 2010, 653-678.

Figure 1: Map of Ifakara and Rufiji HDSS, Tanzania

10  

Figure 2: Child mortality trends (5q0, with 95% CI) in Rufiji HDSS-Tanzania, 2000-2010

11  

  

Annex 2: Influence of child, mother and household’s characteristics on mortality, 2000-

2010: multivariate analysis (Model 1)

Variable Type Hazard Ratio

Gender Boy 1.09*

Girl Ref.

Birth order

1 - 2 1.34***

3 - 4 1.15*

More than 5 Ref.

Group age of mother

Under 20 years 1.18**

20 – 35 years Ref.

More than 35 years 1.29***

Education of mother

No education 1.14*

Primary 1.12*

Secondary/college Ref.

0

50

100

150

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Pro

bab

ility

of

dyi

ng

(5q

0, p

er 1

,000

)

Source: Database of Rufiji HDSS, 2012

5q0

CI 95%

Trend line

12  

SES of household

Poorest 1.22**

Second 1.18**

Middle 1.06.

Fourth 1.08

Least poor Ref

Travel time from household to

nearest health facility (hospital or

health center or dispensary)

(minutes)

Less than 30 Ref.

30 – 60 1.22**

More than 60 1.28**

Number of children 38069

Number of events 2682

*** (p<0.001); ** (p<0.01); * (p<0.05);

Annex 3: Interaction between “Travel time” and “Mother’s education” and

“Socioeconomic status”, 2000-2010: multivariate analysis (Model 2)

Travel time from household to nearest health facility

Less than 30 min More than 30min

Mother’s education attainment

Not educated 1.08 1.16**

Educated Ref. 1.12*

Household’s socioeconomic status

Poorer 1.18** 1.16**

Poor 1.05 1.14*

Less Poorer Ref. 0.93

N = 38069; Event = 2682; ** p<0.01; * p<0.05