|Jensen, Henrik. 2002. Analysis of Multivariate Survival Data from Longitudinal Epidemiological Studies - with special reference to the impact of routine immunisation in infancy. Bandim Health Project, Danish Epidemiology Science Centre, Department of Epidemiology Research, Statens Serum Institut. University of Copenhagen. 162 p.
|Collection of survival times from children in low-income countries is most often of such a character that children are related at different levels. Levels include sampling cluster, geographical unit, village, house, and biological mother. Inclusion of such interdependence in survival analysis is needed either to estimate the interdependence or to test whether the interdependence has any consequences for the results. In the present thesis different statistical models for multivariate survival data are studied and compared in real life examples from Guinea-Bissau, West Africa. The focus is on possibly dependent survival times of children, left truncated survival data, and the association between routine vaccinations and childhood survival.
Two approaches, the marginal and the conditional, to the analysis of multivariate survival data, are discussed and contrasted. The major statistical contribution in this thesis is the study of the shared frailty models in which left truncated survival times are more complicated than in the rest of the models covered. An approximate likelihood function is proposed and a simulation study is presented comparing the proposed model with an earlier studied "naive" model. The simulation study showed that both the proposed and "naive" models are appropriate, but under different conditions.
Five observational studies from Guinea-Bissau are presented. One study of crowding and health, one study of death clustering, and three studies with the main objective to test the association between BCG, DTP, polio, and measles vaccines and childhood survival. Furthermore, an illustration is given of how to use vaccine information in the analysis of survival. The survival data from the five studies are all multivariate survival data at different levels. The analyses showed only minor numerical differences between the regression parameter estimates from the two statistical approaches. When the level or related individuals is a confounder we recommend a conditional approach. However, the interdependence in most of the data sets did not change results significantly.
The analyses of an association between BCG, DTP, polio, and measles vaccines and childhood mortality all point in the same direction of a positive nonspecific effect of BCG and measles vaccines on survival, and a negative nonspecific effect of DTP/polio on survival. In conclusion, there is clearly a need to critically review the different routine vaccines to clarify their role on childhood survival, particularly in low-income communities.