Can the costs of the World Health Organisation antenatal care programme be predicted in developing countries?

The aims of this thesis are to identify and test alternative methods for analysing and predicting health care costs, to construct a framework for guiding analysts in making better cost predictions, and to identify future areas of research in this area. The thesis uses costs collected from a multi-country trial measuring the cost-effectiveness of an evidence-based programme of antenatal care. Detailed costing studies of maternity services (antenatal care, childbirth and postpartum care) were done in two trial countries (Cuba and Thailand), and also a nontrial country, South Africa. Costs are broken down and reviewed by cost components: prices, resource use, and health service use. The review initially considers the application of economic theory to public health care institutions, to identify factors likely to cause cost variation between setting. Then the review seeks empirical evidence proving or disproving the existence of these factors from the health care literature, as well as a review of the methods for analysing health care costs. The empirical analysis first compares health service use, unit costs and cost per pregnancy between settings (between: women with different case-mix, health facilities, trial arms and 'study countries) and examines the causes of variation, before testing alternative cost prediction methods. Variations in unit cost are found to be due to several factors, including different levels of resource productivity, occupancy levels, staffing patterns, prices and exchange rates (between country), input mix and health facility size. Also, uncertainty and measurement error are considered likely to cause some variation in unit costs. Variations in health service use are due to case-mix, clinical practice, and accessibility differences. Again, not all variation is explained. Finally, a range of different cost predictions methods are tested, and their results compared with observed costs in each country. The most accurate cost prediction method is to build costs based on expected changes in resource use, health service use and morbidity rates (called the incremental cost impact approach). The direct and adjusted cross-country transfer methods (transfering costs between countries), although accurate on occasions, are less reliable. Cost predictions using predictors from a regression analysis are highly unreliable for cross-country predictions. Methodological issues and policy implications in relation to cost prediction and generalisability are discussed, including the choice of cost-prediction approach, the valuation methods (opportunity cost and currency conversion methods for cross-country predictions), the measures used for comparing the performance of cost prediction methods, and the limitations· of cost analyses to understand costs. It was concluded that caution is needed in predicting costs both within study countries due to cost variability, and in lower-resourced settings where u,nit costs and health service use are lower. Further cost analyses and testing of cost prediction methods are needed in other areas of health care to compare with the results from this thesis, and build a fuller picture of cost behaviour as well as strengths and weaknesses of alternative cost prediction methods.