Examining the Influence of Antenatal Care Visits and Skilled Delivery on Neonatal Deaths in Ghana

Abstract

Background

Many Sub-Saharan African countries may not achieve the Millennium Development goal of reducing child mortality by 2015 partly due to the stalled reduction in neonatal deaths, which constitute about 60 % of infant deaths. Although many studies have emphasized the importance of accessible maternal healthcare as a means of reducing maternal and child mortality, very few of these studies have explored the affordability and accessibility concerns of maternal healthcare on neonatal mortality.

Objective

This study bridges this research gap as it aims to investigate whether the number of antenatal visits and skilled delivery are associated with the risk of neonatal deaths in Ghana.

Methods

Using individual level data of women in their reproductive years from the 2008 Demographic and Health Survey, the study employs an instrumental variable strategy to deal with the potential endogeneity of antenatal care visits.

Results

Estimates from the instrumental variable estimation show that antenatal care visits reduce the risk of neonatal death by about 2 %, while older women have an approximately 0.2 % higher risk of losing their neonates than do younger women.

Conclusion

Findings suggest that women who attend antenatal visits have a significantly lower probability of losing their babies in the first month of life. Further, results show that women’s age significantly affects the risk of losing their babies in the neonatal stage. However, the study finds no significant effect of skilled delivery and education on neonatal mortality.

Appendix: The Probit Model

The actual probit estimation is as follows:

Let N be the probability of death of child i, and P _i represent the number of ANC visits for each woman, X _i is a vector of the child’s birth-specific characteristics such as gender of the baby and birth interval. It also includes place of delivery, presence of skilled assistance during delivery, and whether or not the mother delivered via caesarian section. The household’s socio-economic characteristics are captured by the term Y _i . The terms Z _i and T _i capture the region and place of residence, respectively. ε is an independent and identically distributed residual with ε ~ n (0, σε ²).

$$ N_{i} = \beta_{0} + \beta_{1} P_{i} + \beta_{2} X_{i} + \beta_{3} Y_{i} + \beta_{4} Z_{i} + \beta_{5} T_{i} + \upsilon_{i} $$

(1)

Given the binary nature of the response variable, the equation above is estimated using a probit model. Theoretically, the probit model is given as:

$$ \Pr \left( {N = 1} \right) = \Pr \left( {N = {\text{probability of child's death}}|X} \right) = F \left( X \right) $$

(2)

where P (.) and F (.) represent the conditional probability of obtaining the variable of interest and the Normal distribution function, respectively. The term X is a vector of variables that are likely to affect the health of the child and subsequently the probability of the child’s death.

N is assumed to depend on a set of M explanatory variables X _m , where m = 1, …, M. These explanatory variables are assumed to explain the variation in the probability of the outcome variable N. This is given by:

$$ P = P(N = 1 |X_{1} , \ldots , X_{m} ) $$

(3)

The probit estimation is assumed to satisfy the following assumptions:

The data are randomly generated, which implies that the observations on N are independent of each other.

Linear independence between the explanatory variables, implying that each X _m has some variation across observations and none should be perfectly correlated.

The probit model is estimated using the maximum likelihood procedure. This estimation is given as follows. Given that the probability that an outcome occurs is given as $P_{i} = P(N_{i} = 1 | X_{i} )$, then the probability that it does not occur can be expressed as $(N_{i} = 0|X_{1} ) = 1 - P_{i}$ . If ρ is assumed to be the set of parameter estimates, the likelihood function can be defined as $L(N|X, \rho ) \equiv P(N|X) .$

Therefore, estimating the above function using the sample observations on N and X and the set of m values for ρ would yield a number between 0 and 1, which would represent the probability of observing N if ρ were indeed the ‘true’ value. The point of the maximum likelihood procedure is to choose an estimate of ρ that would make the likelihood of observing that particular N as large as possible. Therefore, the set of estimates of ρ, given as $\bar{\rho }$, are the set of m numbers that yield the largest value $L(N|X, \bar{\rho }) = \max_{b} L(N|X, \rho )$.

The probit estimation aims to provide a measurement of the relationship between the set of explanatory variables, X, and dependent variable, N. The coefficients obtained are asymptotically unbiased, and the standard errors derived provide a measure of the variation in the estimated coefficients. The reported t-statistic tests the null hypothesis that the explanatory variable has no effect on the dependent variable. To test the joint hypothesis that all the explanatory variables have no effect in explaining the variation in the dependent variable, the F statistic is calculated.