International Journal of Epidemiology
O International Epldemlologlcal Association 1996
Vol. 25, No. 3
Printed In Great Britain
The Impact of Primary Health Care
Services on Under-Five Mortality in
Rural Niger3
ROBERT J MAGNANI,* JANET C RICE,* NANCY B MOCK,* AHMED A ABDOH,** DAVID M MERCER1
AND KADRI TANKARI*
Magnanl R J (Tulane University Medical Center, School of Public Health & Tropical Medicine, Department of Internal
Health & Development, 1440 Canal Street, Suite 2200-12, PO Box 13, New Orleans, LA 70112-2737, USA), Rice J C,
Mock N B, Abdoh A A, Mercer D M and Tankari K. The impact of primary health care services on under-five mortality In
rural Niger. International Journal of Epidemiology 1996; 25: 568-577.
Background. Despite large investments in basic primary health care in sub-Saharan Africa over the past two decades,
quantifying the contribution of national programme efforts to the reduction of infant/child mortality in the region has proven
difficult. This study takes advantage of the phased implementation of the national Rural Health Improvement Program in
Niger and conveniently timed survey data to reassess programme impact on under-five mortality during the 1980-1985
period.
Methods. Health service use and under-five mortality rates for children born in the 5 years prior to the 1985 survey are
compared for three groups of villages: villages served by a dispensary, villages served by village health teams (VHT),
and villages without access to modem primary care services. Multi-level regression analyses using both household- and
community-level variables are undertaken in estimating the magnitude of effects.
Results. Children residing in villages proximate to health dispensaries were approximately 32% less likely to have died
during the study period than children without access to modem health services. Village health teams were not, however,
associated with significantly lower mortality probabilities. Formal tests for endogeneity indicated that these effects were
not the result of non-uniform/non-random allocation of resources.
Conclusions. The findings are largely supportive of the key premise underlying selective primary health care interventions—that packages of basic services can be effectively mounted nationally in poor countries and have a significant
impact over a short time period. In Niger, less than optimal implementation of VHT appears to have reduced the
magnitude of the impact achieved.
Keywords: child mortality, primary health care, programme impact, Niger
Infant and child mortality levels have fallen substantially in much of sub-Saharan Africa over the past decade
or so. 1 " 3 While national child survival, maternal-child
health, and related programme efforts have undoubtedly contributed to lower infant-child mortality in the
region, there is surprisingly little conclusive evidence
available on the question of the share of mortality
decline that may be attributed to such programmes.1'4"7
In a recent review of the available evidence, Ewbank
and Gribble1 concluded that while the national health
programmes of most countries include interventions
that have been shown to reduce mortality in small test
programmes, few strong statements could be made
about the overall effectiveness of the large-scale,
primary health care programme efforts in the region.
The inconclusive evidence on this question may be
attributed to several factors. Firstly, most of the research attention has focused on individual interventions
(e.g. Expanded Programme on Immunization, Childhood Diarrhoeal Disease programmes, malaria control,
etc.) as opposed to the combined impact of interventions comprising primary health care programmes.1'7
Determining the net impact of packages of health
services is, however, crucial to the issue at hand in
view of the possibility that gains realized through
single interventions might be offset by continued high
levels of exposure to other risk factors for mortality
• Tulane University Medical Center, School of Public Health & Tropical Medicine, Department of International Health & Development,
1440 Canal Street, Suite 2200-12, PO Box 13, New Orleans,
LA 70112-2737, USA.
•* University of Manitoba, Winnipeg, Manitoba, Canada.
f
Program for Appropriate Technologies in Health, Seattle,
Washington, USA.
• World Health Organization, Dakar, Senegal.
" Revised version of a paper presented at the 1993 Annual Meeting
of the Population Association of America, Cincinnati, Ohio, 1-3 April,
1993.
568
PRIMARY HEALTH CARE AND INFANT MORTALITY IN NIGER
in low-income country settings; that is, through 'replacement mortality'. 7 ' 10
Secondly, many of the studies that have considered
packages of services were undertaken in small geographical areas in a limited number of settings (primarily in Senegal and The Gambia) where long-term
research initiatives have been put in place."" 14 The
generalizability of these findings to national-level
programmes and/or to other settings in the region is
uncertain.
Finally, some of the larger, national-level studies that
have been conducted (e.g. in Liberia and Zaire) 15 ' 16
were unable to use comparison or control groups for
various reasons, making it difficult to attribute
observed changes in mortality to the programmes under
study. Indeed, due to proliferation of limited-scale project efforts and the rapid expansion of basic Maternal
and Child Health (MCH) services in sub-Saharan
Africa, maintaining controlled experimental conditions
in programme impact studies in the region has proven
difficult.4'6
The present study takes advantage of national survey
data collected in Niger in 1985 to re-examine the
question of impact of national-level primary health care
programmes on childhood mortality in sub-Saharan
Africa. In the late 1970s, the Nigerian Ministry of
Health embarked on a national initiative, the Rural
Health Improvement Program (RHIP), to extend the
coverage of primary health care services throughout
rural Niger. The effort focused on upgrading existing
health dispensaries and deploying trained village health
teams (VHT) to unserved villages. During the 1978—
1984 period over 8000 village health workers were
trained, and by 1985 approximately 45% of rural
villages were being provided with primary care services
either through dispensaries or VHT.
The phased implementation of the Nigerian programme provides a convenient natural quasi-experiment
from which to assess programme impact. In the study
reported in this article, data from the 1985 National
Morbidity and Mortality Survey (NMMS) are used to
assess the impact of differential access to health
services through the comparison of service use patterns
and under-five mortality levels among villages
provided different levels of health services.
DATA AND METHODS
The NMMS was a national survey conducted by the
Ministry of Health during April-May 1985. The
primary purpose of the survey was to provide detailed
information on the health status of women and children,
health practices and use of health services. A total of
569
1470 women of reproductive age who had been
pregnant at least once during the 5 years prior to the
survey were chosen for the survey using a stratified,
two-stage cluster sampling procedure. A total of 49
clusters (i.e. villages, city blocks/neighbourhoods or
settlement camps) and 30 eligible respondents residing
in each sample cluster were randomly chosen for the
survey. Clusters were allocated to three sampling strata
(i.e. urban, rural and temporary settlements) on the
basis of level of access to primary care services and
proportionally to the estimated distribution of population across strata. The analyses reported here are based
upon the 953 eligible respondents interviewed in the
35 rural clusters covered in the survey.
Survey interviews were conducted by female interviewers using a standardized questionnaire. Information
was gathered on all pregnancies occurring in the 5-year
period preceding the survey, morbidity and mortality
information for each child, maternal and child health
care practices and use of health services in connection
with the last-born child of each respondent, village and
household environmental conditions, and socioeconomic
and demographic background factors.
The principal outcome considered is the survival
status of children born in the 5 years preceding the
survey. A total 2075 births were reported by survey
respondents during this period, of which 1632 (78.7%)
were reported as having survived to the survey date
and 443 (21.3%) as having died. Because of serious
deficiencies in the recall of dates of births and deaths
on the part of survey respondents, analyses involving survival times were not attempted. The functional
form of the dependent variable used was a binary (0,1)
outcome indicating whether children born during
the 5-year reference period for the study had survived to the survey date. As a sensitivity test, several
survival analyses were run for the subset of cases for
which ages at birth and death were reported to determine whether this specification of the outcome variable would bias the findings. The estimates of the
parameters of primary interest were largely unaffected
and, accordingly, the simple binary outcome was
used in the analyses in order to maximize statistical
power.
Access to health care services was defined in terms
of geographical proximity to different levels of health
services. Villages were classified into three categories:
(1) villages located s 5 km from a dispensary, (2)
villages located >5 km from a dispensary, but in which
a village health team (VHT) had been deployed, and (3)
villages located >5 km from a dispensary and without
a VHT. Nine (n = 9) clusters were randomly chosen to
represent villages in the first category, n = 19 clusters
570
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
TABLE 1 Operational definitions of variables used in the analysis
Literacy
Ethnicity
Maternal age
Marital status
Husband/partner's occupation
Home construction materials
Source of drinking water
Radio
TV
Transport
Meals yesterday
Dispensary
VHT
Primary school
Main road
Store
Machines
Seeds
Individual/Household Level
1 if respondent could read French/local language; 0 otherwise
: 1 if Dzerma; 0 otherwise
1 if age 35 or above at birth of child; 0 otherwise
1 if married/in union; 0 otherwise
: 1 if agriculture; 0 otherwise
> 1 if wood or concrete; 0 otherwise
• 1 if pump or spring; 0 otherwise
1 if household owned a radio; 0 otherwise
1 if household owned a TV; 0 otherwise
: 1 if household owned means of transportation; 0 otherwise
1 if family ate 3 or more meals on prior day; 0 otherwise
Community level
1 if there is a dispensary within 5 km of village; 0 otherwise
: 1 if the village has a village health team (VHT); 0 otherwise
: 1 if a primary school is located in the village; 0 otherwise
• 1 if the village is located on a main road; 0 otherwise
: 1 if a store is located in the village; 0 otherwise
1 if farm equipment is commonly used in village; 0 otherwise
: 1 if there is usually a sufficient supply of seeds available
during planting season in the village; 0 otherwise
the second category, and n = 7 the third. The effectiveness of this operationalization in defining levels of
exposure to modern MCH services is demonstrated
empirically below.
In order to control for the effects of potentially confounding factors, a series of variables with documented
statistical associations with childhood mortality in the
research literature were considered in multivariate analyses. These variables and their operational definitions
are described in Table 1.
The determinants of child mortality are conceptualized as occurring at different levels; that is, as
consisting of factors operating at the level of nations,
regions, villages, households and individual children.
In the present analysis, we consider factors measured at
two levels: villages and children. We ascertained
through the testing of alternative statistical models that
it was unnecessary'to adjust for the 'nesting' of children
within households (i.e. that a three-level model was
unnecessary). Conventional logit regression procedures
were used in the analysis. 17
Several statistical estimation problems must be
addressed in order to model child survival outcomes as
a function of access to health care services. One problem concerns the possible endogeneity of the location
of health services and health outcomes. While an
observed statistical relationship between measures of
health service access and health outcomes in empirical
studies might signify a positive impact of health
services, such a relationship might also be explained by
the location of health services near sub-populations that
were predisposed to better health e.g. in areas with
populations of higher socioeconomic status. It may be
demonstrated that where programme resources are
distributed neither uniformly nor randomly with respect
to factors that are important determinants of outcomes
under study, estimates of programme impact will be
inconsistent and possibly biased.18"20 As part of the
analysis, we formally test for endogeneity following
procedures proposed by Guilkey el a/.21
A second problem concerns the estimation of standard errors for regression parameters where variables
have been measured at two or more levels. Because
community-level factors are common to all subjects
within a given community, observations within each
community are not independent. The failure to compensate for this lack of independence results in incorrect
estimates of standard errors of regression coefficients
for community level variables, and thus biased statistical tests. 22 ' 23 In the analyses presented below, we correct the standard errors of community-level parameters
in the final model for lack of independence following
the procedure proposed by Huber.23
FINDINGS
Sample Characteristics
Table 2 provides summary socio-demographic information on the sample. The profile of households is typical of a rural, West African population. Agriculture
571
PRIMARY HEALTH CARE AND INFANT MORTALITY IN NIGER
TABLE 2 Distribution of independent variables, by level of access to health services
Level of Access to Health Services
Number of villages
Number of births
Literacy (% literate)
Ethnicity (% Hausa)
Maternal age (% age 35+)
Marital status (% married)
Husband's occupation (% agriculture)
Home construction (% wood/concrete)
Source drinking water ( * pump/spring)
Radio (% owning)
TV (% owning)
Transport (% owning)
3 meals yesterday (9b yes)
Primary school (% yes)
Majn road (% yes)
Store (% yes)
Machines (% yes)
Seeds (% yes)
Total
Dispensary
VHT*
None
35
2075
5.0
80.3
6.6
97 3
77.1
58.6
51.3
19.4
19.9
41.8
33.3
51.4
20.0
28.6
74.3
51.4
9
517
8.7
89.9
7.0
96.9
54.0
48.9
56.9
30.9
31.1
56.9
51.8
89.9
33.3
55.6
77 8
55.6
19
1123
3.7
68.3
6.9
97.6
87 1
62.2
54.8
14.9
14 7
35.5
20.9
42.1
5.3
15.8
73.7
42.0
7
435
4.1
100.0
7 1
96.8
78 9
60.9
35 6
17.5
19.8
40.2
43 4
28.6
42 9
28.6
714
71.4
1
Village health team.
NB: denominators for computation of percentages for individual/household level variables was the number of live births and for community-level
variables the number of villages.
provides the primary means of subsistence. Households
possess relatively few durable assets, with less than
20% of households owning even a radio. Women in this
population have relatively low levels of education, with
less than 5% of sample women reporting being literate.
Fertility levels are also typical of the region, with an
estimated total fertility rate (TFR) of 6.4 during the
5 years prior to the survey. Births to women less than
20 years of age accounted for about 8% of all births
occurring during the reference period for the survey,
and those to women 35 years of age or older 18%.
Nearly all respondents (97%) reported being currently
married or in union.
An important observation from Table 2 is that the
characteristics considered vary in a systematic manner
across access-to-health-care categories. In general, the
populations of villages proximate to dispensary-based
health services were of higher socioeconomic status
and villages with VHT of lower socioeconomic status
than villages without access to modern health services.
This general pattern holds for most of the variables
considered.
Thus, it is clear that dispensaries and VHT were
placed in communities neither uniformly nor randomly.
For logistical, infrastructural, and possibly political
reasons, dispensaries appear to have been located in
villages that were predisposed to lower child mortality.
The VHT, on the other hand, appear to have been
deployed in communities that were predisposed to
higher mortality, although level of health risk was not a
formal criterion used in site selection (community
support and location within 15 km of a dispensary were
the primary criteria used in selecting sites for VHT
deployment). One challenge in measuring the impact of
primary health care improvements in Niger during the
period studied adequately is to take into account the
non-random deployment of programme resources.
Use of Health Services
To what extent are differentials in physical access to
health care services reflected in service use patterns?
The evidence presented in Table 3 indicates that
proximity to a dispensary is associated with significantly higher use of MCH services. Women residing
near a dispensary were between two and five times
more likely to have received antenatal care services in
connection with their most recent birth, to have had the
delivery attended by trained health personnel, to have
received nutrition and health education consultation,
and to have been able to explain how to mix oral
rehydration solution (ORS) than women residing in
villages with neither a dispensary nor a VHT. Children
572
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
TABLE 3 Use of health services in cnnection with most recent births, by level of access to health services
Level of Access to Health Services
Number of births
Attended deliveries (%)
(physician/nurse/matron)
Women making at least one
antenatal visit (%)
Women receiving at least
one nutritional consultation (*)
Women receiving at least
one health education session (%)
Women able to describe preparation
of ORS (sugar-salt solution) (%)
Children with health card (%)
Children at least partially immunized (%)
Last episodes of diarrhoea referred to
health facilities/personnel11 (%)
Total
Dispensary
VHT*
None
983
53.4
245
604
535
66.2
203
11.3
22.1
46.5
138
14.3
14.6
38.0
5.8
9.9
22.4
39.2
187
11.8
18 1
15.5
34.4
38.7
20.0
36.3
44.1
59 7
20.1
7.5
37.0
33 3
103
11.3
15.8
27 6
•Village health team.
b
Health facilities defined as hospitals and dispensaries, health personnel defined as physicians, nurses, midwives, or village health volunteers
residing in such villages were approximately three
times more likely to have been at least partially immunized and to have had a health card, and twice as likely
to have had their most recent episode of diarrhoea referred to a health worker than children in 'control' villages.
The magnitude of differentials between villages with
VHT and villages with neither a dispensary nor a VHT
are less pronounced. Only for birth attendance by
trained personnel, immunization, and maternal knowledge of ORS were sizeable differences observed. It
should be noted, however, that for these indicators the
levels observed for villages with VHT approximated
those found in villages proximate to dispensaries.
These figures might reflect a tendency of VHT to
have concentrated on certain activities and/or programmes (e.g. birth attendance and EPI) in the early stages
of the rural health service expansion effort in Niger.
Nevertheless, that the deployment of dispensaries and
VHT is associated with higher service use rates (for at
least some services) is evident from these data. Equally
clear, however, is that there is considerable room for
improvement in programme coverage levels, as only in
the case of birth attendance by trained personnel did
coverage levels exceed 50% in villages covered by
programme health services.
Infant/Child Mortality Levels
The unconditional probability of a child born during the
5 years prior to the NMMS having died by the survey
date was 0.213, or just over 1 in 5. In the absence of
sufficient sample size and birth history data that would
permit the direct estimation of infant/child mortality
rates, the 'preceding birth technique' was used to convert these probabilities into conventional mortality
24 25
measures.
Unconditional probabilities of last and penultimate
births surviving to the survey date were 0.8940 and
0.7228, respectively. The estimated probabilities of
dying by exact ages one and 5 years (lqO and 5q0) under
four different sets of assumptions regarding the underlying age pattern of mortality using the 'preceding birth
technique' are shown in Table 4, along with 'direct'
retrospective estimates derived from 1992 Niger
Demographic and Health Survey (DHS) birth history
data for the 1980-1985 reference period.26
The indirect estimates of the infant mortality rate
range from 129/1000 live births under the Brass African
Standard model to 159/1000 for the UN General
model.27 Taking into account the direct estimate from
the 1992 DHS, we believe that an estimate in the middle
to upper part of this range is reasonable for rural Niger
during the reference period for this study.
Except for the UN West African model, the indirect
estimates of the under-five mortality rate are in
somewhat closer agreement across the different model
mortality schedules at around 220 per 1000 live births.
The estimate from the UN West African (hypothetical)
model is considerably higher (289 per 1000). However,
573
PRIMARY HEALTH CARE AND INFANT MORTALITY IN NIGER
TABLE 4 Estimated probabilities of dying by selected ages, rural Niger, 1980-1985
Indirect estimates, by assumed age pattern of mortality
Age
UN
Brass
General
Standard
Brass
African
Standard
UN
General
Pattern
W African
Pattern
Direct estimate,
1980-1985 period,
1992 DHS 1
0 143
0.221
0.129
0 227
0.159
0.220
0.152
0 289
0.138
0.321
1
Source: Direction de la Slatistique et des Comptes Nationaux/Macro International, 1993, Tableau 9.1. DHS estimate for the 1980-1985 reference
period obtained by averaging the published esumates for the 1978-1982 and 1983-1988 periods.
in view of the tendency of systems of model life table
to underrepresent the high levels of childhood (i.e. ages
1—4 years) relative to infant mortality found in many
sub-Saharan African populations28 and the estimates
from the Niger DHS, we are inclined to assign greater
weight to the estimate from the West African model.
Thus, our 'preferred' estimates would put the infant
mortality rate in the 140-150 range and the under-five
mortality rate in the 290-320 range during the period
studied. The vagaries of indirect estimation procedures
aside, it may be concluded with virtual certainty that
infant/child mortality levels in rural Niger were quite
high during the period studied. The drought and famine
that struck the Sahelian region during 1984-1985 and a
measles outbreak in Niger in 1985 undoubtedly contributed to high infant-child mortality during this period.
The extent to which survival probabilities were
mediated by primary health care services is addressed
in the analysis below.
Programme Effects on Childhood Mortality
The unadjusted proportions of children born in the
5 years prior to the survey who had died by the survey
date were 0.191 in villages served by a dispensary,
0.203 in villages served by VHT, and 0.267 in villages
with neither dispensaries nor VHT. A series of regression models using different combinations of the independent variables described in Table 1 (and interactions
among variables) were tested. The results of the most
parsimonious regression model that included the two
access to health care dummy variables (Dispensary and
VHT) are shown in Table 5.
Three individual-level and three community-level
factors emerged as significant determinants of child
mortality when the effects of other factors considered
were controlled. Of primary importance for the purposes
of the present study is that the presence of a dispensary
in a village was associated with significantly lower
probabilities of mortality (odds ratio [OR] = 0.68; 95%
confidence interval [CI] : 0.49-0.94). The effect of
the presence of a VHT was not, however, significant
(OR = 0.80; 95% CI : 0.58-1.09).
The effects of the other variables included in the final
model were more or less as anticipated. Among
individual-level variables, maternal literacy was associated with lower probabilities of child mortality, an
effect observed in many prior studies both in subSaharan Africa and elsewhere.29*30 Older maternal age
was associated with higher mortality, a finding that is
also consistent with previous studies.31 Interestingly,
however, adjusted under-five mortality rates were
lower among Dzerma children, a minority ethnic group.
At the village level, the presence of farm machinery/
equipment and access to seeds to plant the next
seasons' crops, used as measures of community wealth,
were associated with lower levels of child mortality.
The interaction between these variables was the only
significant interaction effect observed, with the presence
of both increasing the likelihood of child mortality.
This result probably reflects the fact that either factor
alone operates to reduce child mortality levels, but the
presence of both is redundant and does not serve to
further reduce mortality levels.
The final step in the analysis was the conduct of tests
for endogeneity to determine whether the non-random
allocation of programme resources might have biased
the regression results reported in Table 5. Following the
procedure proposed by Guilkey et al.2x two logit
regression equations predicting location of a dispensary
and a VHT, respectively, in sample villages were estimated using the community-level variables described in
Table 1 as independent variables. The predicted values
from these equations were then added to the regression
equations used to generate the results shown in Table 5
as additional independent variables and t-tests performed to assess the significance of the added terms.
574
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
TABLE 5 Multi-level regression results, final model
Variable
Coefficient
P-value
Odds ratio
(95% confidence interval)
Literacy
-1.478
(-3.30)
0.639
( 3.69)
-0.738
(-4 64)
-0.393
(-2.33)
-0.230
(-141)
-0.295
(-2.41)
-0.766
(-3.10)
0.745
(2.72)
-0.717
0.001
0.23
(0.10-0.29)
0.000
1.90
(1.35-2.66)
0.000
0.48
(0.35-0.65)
0.020
0.68
(0.49-0.94)
0.157
0.80
(0.58-1.09)
Maternal age
Ethnicity
Dispensary
Village health team
Machines
Seeds
Machines x seeds
Constant
0.016
0002
0.006
Hosmer-Lemeshow y 2 «> 9.87, d.f. = 2, P m
NB: t-statistics shown in parentheses beneath regression coefficientt
* Odds ratios (OR) and 95% confidence intervals (CI) for these variables adjusted for interaction were as follows:
Agricultural machines and seeds: OR » 0.75 (95% CI : 0.59-0.95)
Agricultural machines and no seeds: OR = 1.57 (95% CI : 1.25-1.97)
Seeds and agricultural machines: OR » 0.47 (95% CI : 0.29-0.77)
Seeds and no agricultural machines: OR = 0.98 (95% CI : 0 69-1.38)
Neither parameter emerged as significant, indicating
that although programme resources clearly were distributed neither uniformly not randomly, this does not
account for the programme effects observed in Table 5.
This test result also indicates that more complex randomeffects models are not necessary in the present analysis.
DISCUSSION
Given the scarcity of resources in developing country
settings and the multiple competing demands for them,
information on the relative effectiveness of programmes
competing for resources is needed to guide resource
allocation decisions. In line with initiatives in other
developing regions, sizeable investments have been
made in sub-Saharan African countries over the past
two decades in national public health programmes
aimed at reducing high levels of infant-child mortality.
Direct evidence as to the impact of these programmes
is, however, relatively scarce, and much of the available
evidence comes from small-scale studies with limited
geographic scope and uncertain generalizability to
national-level programmes. The present study attempted,
at least partially, to fill this information gap.
Although limited due to a relatively small sample
size, the findings of the study are largely supportive of
the key premise underlying selective primary health
care interventions in low-income countries; namely that
packages of basic primary health care services can be
effectively mounted at the national level so as to have
a significant impact on infant-child mortality over a
fairly short period of time. Our estimates indicate that
during the 1980-1985 period, children residing in
Nigerian villages proximate to a dispensary were 32%
less likely to have died than children residing in
villages without access to modern health care services.
The inclusion of individual- and village-level control
variables in multivariate analyses, as well as formal
tests for endogeneity, suggest that these findings were
not the result of dispensaries having been located in
villages that appear to have been predisposed to lower
mortality.
The findings with respect to the impact of deployment of VHT are somewhat less encouraging. Adjusted
under-five mortality levels for children residing in
villages served by a VHT were not significantly different from children in villages without access to modern
health services. The failure to observe an impact of
PRIMARY HEALTH CARE AND INFANT MORTALITY IN NIGER
VHT deployment is not due to insufficient statistical
power of the study (power is estimated at 0.9). It is
worth noting, however, that although mortality levels
for villages served by VHT were not significantly different from villages without access to modem primary
care services, neither were they significantly different
from villages served by dispensaries.
The observation of a significant effect of services
provided by dispensaries but not through the deployment of VHT in the case of Niger is seemingly
accounted for by differences in levels of programme
coverage attained. In villages served by a dispensary, it
would appear that a core 40-60% of women availed
themselves of MCH care services during the period
studied, while service coverage exceeded 40% only for
birth attendance in villages with VHT, although
immunization coverage for childhood vaccine preventable diseases approached 40%. However, under the
assumption that differentials in non-programme determinants of mortality among villages were adequately
accounted for in the multivariate analyses, the findings
are not inconsistent with the notion that reductions in
mortality levels of a magnitude comparable to that
observed in villages served by a dispensary could be
obtained through the deployment of VHT given comparable levels of programme coverage.
Is the level of programme impact inferred from the
present study plausible? In the context of the Nigerian
Rural Health Improvement Program in the early 1980s,
the major contributors to reduced mortality are likely
to have been (1) reduced incidence of neonatal tetanus
due to birth attendance by trained health workers, (2)
reduced incidence of childhood vaccine preventable
diseases due to EPI control efforts, and (3) improved
case-management of childhood diarrhoeal disease.
Tetanus is a major cause of neonatal death in many
African countries, and tetanus mortality rates in Africa
are apparently among the highest anywhere in the
world. lt32 Death rates due to neonatal tetanus in the
range of 10-20 per 1000 live births are apparently not
unusual in African countries,1'32 with some estimates
being as high as 70 per 10OO.33 In one of the few available intervention impact studies in Africa, Dan et al.34
estimated that the training of traditional birth attendants
in aseptic delivery procedures in a small number of
villages in Senegal resulted in a reduction in the
neonatal death rate of over 50% in comparison with the
rates observed in control villages. Higher estimates of
intervention effectiveness have been obtained in studies
elsewhere.35 Even at lower levels of effectiveness, a
significant number of infant deaths are likely to have
been averted in Niger during the period covered by the
present study given the estimated coverage of attended
575
births of over 60% in villages served by dispensaries or
VHT.
With regard to vaccine preventable diseases, programme efforts to control measles alone might account
for a large share of the mortality reduction implied by
the present study. Previous research suggests that, for a
variety of reasons, case-fatality rates for measles are
especially high in West Africa.1'36 Small-scale studies
in Benin, Guinea Bissau, and Zaire provide estimates of
mortality reductions associated with measles immunization efforts of between 30% and 50%.l4-37-38 Given
this and estimates of EPI programme coverage in the
40-50% range (although the specific coverage rate for
measles vaccination could not be reliably ascertained
from the available data), it is entirely plausible to
conclude that a significant number of childhood deaths
were averted during the period studied, even if the
Nigerian programme operated at lower levels of
efficiency than the programmes observed in the studies
cited above. Measles immunization may have been
especially important in rural Niger during the period
studied in view of the measles epidemic that struck the
country in early 1995. Control efforts for the other
vaccine preventable diseases during the study period,
pertussis and diphtheria in particular, are likely to have
averted additional deaths, although quantifying the
magnitude of the potential impact is difficult given the
limited data available from the region.
Finally, diarrhoeal disease is thought to be an
especially pernicious public health problem in subSaharan Africa. A recent review of empirical evidence
from the region suggests a median incidence of 4.9
episodes per year among children under 5 years of age
and the region and a median point-prevalence of 10%,
among the highest rates found anywhere in the world.39
Although data for large populations in the region are
scarce, global estimates indicate a median of 38% of all
deaths among children under 5 years of age as being
associated with diarrhoea;39 a figure that may well be
higher in sub-Saharan Africa given the relatively high
levels of childhood malnutrition found in the region.1
Hospital-based studies of ORT interventions in Angola,
Malawi, and Nigeria found reductions in case-fatality
rates for diarrhoeal patients of between 39-95%,"*° and
population-based studies in Bangladesh have observed
declines in infant mortality rates of between 1-8% and
in childhood mortality rates of between 4-14% thought
to be attributable to hospital- and treatment centrebased ORT interventions.41'42 In view of this, and
although again we have no basis from which to assess
the relative effectiveness of diarrhoeal disease control
efforts in Niger during the period studied, it is quite
likely that a substantial number of childhood deaths
576
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
were averted in villages proximate to a dispensary
given that nearly 60% of last diarrhoeal episodes had
been referred to a hospital or dispensary.
Thus, the estimated mortality impact of dispensarybased primary health care services in rural Niger
appears to be plausible. It should be borne in mind,
however, that the apparent mortality reductions were
accomplished in an environment of very high initial
mortality levels. Effects of the magnitude observed in
the present study may not be generalizable to new initiatives in many countries of the region, where substantial reductions in infant-child mortality have already
taken place. It is also possible that the effects of access
to health services were exaggerated in the present study
by the 1984—1985 famine and the severe measles
outbreak in Niger in 1985. However, insofar as infantchild mortality levels remain high in many parts of the
region, the findings of this study suggest that initiatives
to mobilize additional resources for the expansion and
improvement of basic primary care services in the
region are likely to result in demonstrable gains in child
survival.
REFERENCES
1
2
3
4
5
6
7
8
Ewbank D C, Gribble J N. Effects of Health Programs on Child
Mortality in Sub-Saharan Africa Washington, DC: National Research Council, National Academy Press, 1993.
Feachem R G, Jamison D T, Bos E R. Changing patterns of
disease and mortality in sub-Saharan Africa. In: Feachem
R G, Jamison D T (eds). Disease and Mortality in SubSaharan Africa New York: Oxford University Press for
the World Bank, 1991, pp. 3-25.
Hill A. Trends in childhood mortality in sub-Saharan Africa. In:
Pison G, van de Walle E, Sala-Diakanda M (eds). Mortality
and Society in Sub-Saharan Africa. Oxford: Clarendon
Press, 1992, pp. 10-31.
Chen L C, Hill A G, Murray C J L, Garenne M. A critical analysis of the design, results and implications of the mortality
and use of health services survey. Int J Epidemiol 1993;
22(Suppl. 1): S73-S81.
Ewbank D C. A strategy for evaluating the mortality impact of
child survival programmes in Africa: combatting childhood
communicable diseases project. Int J Epidemiol 1993;
22(Suppl. 1):S8-S14.
Ewbank D C. Impact of health programmes on child mortality in
Africa: evidence from Zaire and Liberia. Int J Epidemiol
1993; 22(Suppl. 1): S64-S72.
Gadomski A, Black R. Impact of the direct interventions. In:
Hill K (ed.). Child Survival Programs: Issue for the 1990s.
Baltimore, MD: Johns Hopkins University, Institute for
International Programs, 1990. pp. 85-128.
Ewbank D, Zimicki S. Implications of differences in the causestructure of mortality for the design of health interventions.
In: Hill K (ed.). Child Survival Programs: Issue for the
1990s. Baltimore, MD: Johns Hopkins University, Institute
for International Programs, 1990, pp. 141-56.
9
Koenig M A. Mortality reductions from measles and immunizations: a review of the evidence. In: Hill K (ed.). Child
Survival Programs: Issue for the 1990s. Baltimore, MD:
Johns Hopkins University, Institute for International
Programs, 1990, pp. 43-72.
10
Mosely H, Becker S. Demographic models for child survival and
implications for health intervention programmes. Health
Policy Plann 1991; 6: 218-33.
" Billewicz W Z, McGregor I A. The demography of two West
African (Gambian) villages, 1951-1975. J Biosoc Sci
1982; 13: 219-40.
12
Lamb W H, Foord F A, Lamb C M B, Whitehead R G. Changes
in maternal and child mortality rates in three isolated Gambian villages over ten years. Lancet 1984; i:
912-13.
13
Pison G, Trape J F, LeFebvre M, Enel C. Rapid decline in child
mortality in a rural area of Senegal. Int J Epidemiol 1993;
22: 72-80.
14
Valema J P, Alihonou E M, Gandaho T, Hounye F H. Childhood
mortality among users and non-users of primary health care
in a rural West African community. Int J Epidemiol 1991;
20: 474-79.
13
Becker S B, Thornton J N, Holder W. Infant and child mortality
in two counties of Liberia: results of a survey in 1988 and
trends since 1984. Int J Epidemiol 1993; 22(Suppl. 1):
S42-S49.
16
Chahnazarian A, Ewbank D, Makani B, Ekouevi K. Impact of
selective primary care on childhood mortality in a rural
health zone of Zaire. Int J Epidemiol 1993; 22(Suppl. 1):
S32-S41.
17
Cox D, Snell E Analysis of Binary Data (2nd Edn). London:
Chapman and Hall, 1989.
18
Wonnacott T H, Wonnacott R J. Econometrics. New York' John
Wiley, 1970.
" Briscoe J, Akin J, Guilkey D People and not passive acceptors
of threats to health: endogeneity and its consequences. Int
J Epidemiol 1990; 19: 147-53
M
Cebu Study Team. Underlying and proximate determinants of
child health: the Cebu longitudinal health and nutrition
study. Am J Epidemiol 1991; 133: 185-201.
21
Guilkey D. Correcting for Community Effects in Demographic
and Health Survey Data. Columbia, MD: Macro International, Inc., Demographic and Health Surveys, 1992.
22
Guilkey D , M r o z T A , T a y l o r L . Estimation
and Testing in
Simultaneous
Equations Models with Discrete
Outcomes
Using Cross Section Data. Chapel Hill, NC: Carolina
Population Center, The EVALUATION Project, 1992.
23
Huber P. The behavior of maximum likelihood estimates under
non-standard conditions. Proceedings of Fifth Berkeley
Symposium on Mathematical Statistics and Probability.
1967; 1: 221-33.
24
Hill A G, David P. Monitoring changes in child mortality: A
new method for use in developing countries. Health Policy
Plann 1988; 3: 214-26.
23
Hill A G, Aguirre A. Childhood mortality estimates using the
preceding birth technique: some applications and extension. Poput Stud 1990; 44: 317-40.
26
Direction de la Statistique et des Comptes Nationaux/Macro
International. Enquete Demographique et de Same, Niger.
1992. Columbia, MD: Macro International, Inc., 1993.
27
United Nations. Model Life Tables for Developing Countries.
New York: Department of Economic and Social Affairs,
1982.
PRIMARY HEALTH CARE AND INFANT MORTALITY IN NIGER
19
30
31
32
33
34
35
United Nations. Indirect Methods for Demographic
Estimation.
New York: Department of Economic and Social Affairs.
1980.
Tabutin D, Akoto E. Socio-economic and cultural differentials
in child mortality in Sub-Saharan Africa. In: van de Walle
E, Pison G, Sala-Dialcanda G (eds). Mortality and Society
in Sub-Saharan
Africa. N e w York: Oxford University
Press, 1992, pp. 3 2 - 6 4 .
United Nations. Socio-economic
Differentials in Child Mortality
in Developing
Countries.
N e w York: Department of
Economic and Social Affairs, 1985.
National Research Council. Contraception
and
Reproduction:
Health Consequences
for Women and Children in the
Developing
World. Washington, DC: National Academy
Press, 1989.
Rodrigues L C. EPI target diseases: measles, tetanus, polio,
tuberculosis, pertussis, and diphtheria. In: Feachem R G,
Jamison D T (eds). Disease and Mortality in Sub-Saharan
Africa. N e w York: Oxford University Press for the World
Bank, 1991, pp. 1 7 3 - 8 9 .
Kandeh B S. Causes of infant and early childhood deaths in
Sierra Leone. Soc Sci Med 1986; 2 8 : 2 9 7 - 3 0 3 .
Dan V, Debroise A, Zucker J M, Satge P. Importance du tetanos
neo-natal au Senegal. Afr J Med Sci 1971; 2: 3 9 1 - 9 7 .
Rahman S. T h e effect of traditional birth attendants and tetanus
toxoid in reduction of neonatal tetanus mortality. J Trop
Pediatr 1982; 28: 163-65.
36
37
38
39
40
41
42
577
Aaby P. Malnourished
or Ovennfecled:
An Analysis of the
Determinants
of Acute Measles Mortality.
Copenhagen:
Laegeforeningens Forlag, 1988.
Aaby P, Pederson I R, Knudsen K et al. Child mortality related
to seroconversion or lack of seroconversion after measles
vaccination. Ped Infect Dis J 1989; 8: 197-200.
Kasongo Project Team. Influence of measles vaccination on
survival patterns of 7 - 3 5 month-old children in Kasongo,
Zaire. Lancet 1981; I: 7 6 4 - 6 7 .
Kirkwood B R. Diarrhea. In: Feachem R G, Jamison D T (eds).
Disease and Mortality in Sub-Saharan Africa. N e w York:
Oxford University Press for the World Bank, 1 9 9 1 ,
pp. 134-^5.
World Health Organization. Impact of oral rehydration therapy
on hospital admissions and case-fatality rates for diarrheal
disease: results from 11 countries. Wldy Epidemiol Rec
1988; 8: 49-52.
Oberle M W, Merson M H, Islam M S, Rahman A S M , Huber
D H, Curlin G. Diarrhoeal disease in Bangladesh: epidemiology, mortality averted and costs at a rural treatment
centre. Int J Epidemiol 1990; 9: 341^18.
Chen L C, Rahman M, D'Souza S, Chakraborty J, Sardar A M,
Yunnis M. Mortality impact of an MCH-FP program in
Matlab, Bangladesh Stud Fam Plann 1983; 1 4 : 1 9 9 - 2 0 9 .
(Revised version received October 1995)