Longitudinal methods for analysis of early child health in Pakistan

Abstract

Statistical methods to analyse some aspects on early child health in Lahore, Pakistan are discussed. The aim of this thesis is to choose, examine and use statistical methods suitable to analyse different aspects of early child health in a developing country when the child is measured repeatedly over time (longitudinal data). This kind of data are very informative but require care in the choice of statistical method, since seemingly similar analyses can give quite different answers. This licentiate thesis consists of two parts, which will be referred to in the text by their Roman numerals: I: Anders Carlquist, Valdemar Erling, Rifat Ashraf, Marianne Frisen, Fehmida Jalil, Lars A Hanson, Shakila Zaman. The impact of season and climate on growth during early childhood in different socio-economic groups in Lahore, Pakistan. Research Report, Department of Statistics, GOteborg University 1999:10. II: Anders Carlquist, Valdemar Erling and Marianne Frisen. Longitudinal Models for Analysis of the Influence of Breastfeeding on Early Child Health in Pakistan Research Report, Department of Statistics, Goteborg University 1999:11. In report I, an explorative cross-sectional analysis is first made to examine the main features of some relations between variables. Then, the method of derived variables is used to investigate the longitudinal patterns. The method of derived variables (also called the method of summary statistics or two-stage method) is a simple and often very effective method. The first step is to summarise the repeated values into a summary statistic, which then, in the second step is analysed. This method has been advocated for it's many attractive features. The results are readily interpretable. No assumptions are needed about the covariance structure among the repeated measures (but it is useful to take it into account when choosing the summary statistic). These methods are used to study the impact of temperature on health, measured as growth. In Lahore, Pakistan temperature was shown to be a factor that influences child health. It is demonstrated that the impact is quite different in different ages and areas of living. In report II, we aim to elucidate the causal effects of breastfeeding on early child health, measured as occurrence or not of diarrhoeal episodes. Here it is not possible to use the method of derived variables. We use generalised linear mixed models to construct a model for the binomial response variable and use both fixed and random explaining effects. In order to elucidate the causal effects of breastfeeding on early child health we use the two-step approach recently advocated in the statistical literature but modify the procedure to be practicable for the present longitudinal study. The selection effects of breastfeeding are examined and variables with major effect on the breastfeeding pattern are included in the final model. For some, but not all, areas of living the analysis then motivates the conclusion that breastfeeding helps prevent the occurrence of diarrhoea