Death clustering in families in Kenya: implications for child survival and maternal Health

Abstract

A number of demographic studies in recent years have focused on the problem of death clustering or inter-family heterogeneity in child survival (Das Gupta, 1990, 1997; Guo and Rodriguez, 1992; Curtis, Diamond and McDonald, 1993; Guo, 1993, Zenger, 1993, Ronsmans, 1995; Zaba and David, 1996). Death clustering refers to the concentration of infant and child deaths among certain mothers and is a consequence of siblings sharing many of the same mortality risks. It has been noted that ignoring the effects of death clustering in the estimation of parameters of covariates of infant and child survival may bias the results. Obtaining unbiased estimates of parameters relating to various covariates is of great value to policy makers because of the need to distinguish real from spurious effects of various covariates in order to design appropriate intervention measures. Context This study examines factors affecting infant and child survival in families in Kenya without and with taking into consideration unmeasured and/or unobserved factors. The second aspect of the study has explored the implications of child survival on maternal health in Kenya. Death clustering can have considerable implications for reproductive health and child survival programmes. It is posited that the risk of infant and child survival is a function of known covariates and plus those due to unmeasured/unobserved factors. Xl Data and Methods The data for the study is drawn from the Kenya Demographic and Health Survey (KDHS, 1993). The study has not utilized information from all the districts in which the survey was carried but rather only 15 rural districts. These are generally the larger districts in the provinces which show wide disparities in infant and child health and maternal mortality. These districts were also oversampled in the survey to allow for production of estimates for certain variables at district level. The information used in the study are obtained from the woman's individual questionnaire schedule. The information was based on retrospective birth histories collected from woman aged 15-49at the time of the survey. The eligible women for were identified through the household schedules. In all, 7952 women were identified as being eligible for the individual interviews and out of these, 7540 or 95 % were successfully interviewed. The rural sample consisted of 6379 eligible women and out of these, 5370 women were drawn. from 'the 15 focus districts. The quality of the data used was also assessed and was found to yield reasonable results. A variety of statistical methods have been used 10 the analysis. These include frequency distribution and comparison of observed and theoretical. distribution of number of deaths in families. Standard logistic and generalized logistic regression (random effects model) and log linear models. Results The results of frequency and theoretical distributions indicate that there is some evidence of death clustering at every level of childhood mortality and that at any parity, there are women who contribute higher than average deaths to the mortality risk. A number of factors have been shown theoretically to influence infant and child survival at various ages. From the results of regression analysis, in the neonatal and post neonatal periods, the variables that appear to be highly associated with survival are non receipt of BeG vaccination, immunization against (polio, DPT, measles), prematurity, maternal education and length of preceding birth interval (postneonatal period). This therefore indicates that survival in these periods is influenced by both biological and behavioral factors. At infancy, biological, behavioral and ecological zone of residence factors such as, prematurity, immunization, breastfeeding and region of residence are found to be highly associated with survival in this period. Similarly in childhood, behavioral and household environmental factors as well as ecological zone of residence are highly associated with survival in this period. The effect of maternal age which was-not important in the earlier periods, but becomes important at the childhood stage. This could be as a result of the inexperience of younger women in child rearing practices. Further, it is observed that in the neonatal period, infants of women with no education have higher survival chances than those of women in the other educational categories while in the post neonatal period children of women who have not completed primary education have lower survival chances than for children of women in the other educational categories. The results do not change when the generalized logistic regression model is used instead of the standard logistic regression. The results of examining the relationship between child survival and maternal health indicate that most of the variables that are associated with low BMI (indicator of maternal health), are the same as those which were highly associated with infant/child survival namely high parity, region of residence, maternal education and use of health technology for intervention. Conclusion The indication of death clustering in families in Kenya, may be as a result of replacement and/or hoarding strategies of family building, however before controlling for other factors this may not be the sole explanation. In examining factors that have been shown to theoretically influence infant and child survival at various ages without and with inclusion of the random effects term, there is no evidence of excess variation in parameter estimates. 'This findings are however not consistent with those of other similar studies which used similar models but are consistent with some which used proportional hazards .models. The results of the present study does not invalidate the hypothesis of existence of death clustering in families, but calls for further research using appropriate study designs and data sets. If indeed some families experience higher than average risks of child deaths, then interventions should be more focused to target such families.