Quantification of Cost Determinants in Health Care

Piotr Swierkowski

doi:10.25907/01104

Effective planning and delivery of health care requires accurate estimation of the value of health services. Yet, reliable health care costing methods have often proven elusive. There are known limitations of commonly used costing methods, which this study aims to remedy. The outcomes significantly advance the accuracy of health cost predictions. Important tools are developed for health care funders and administrators to improve financial custodianship of health care system, on both local and system levels. Some of the main sources of inaccurate costings can be remedied by focusing the analysis on the underlying cost drivers. Using this approach, the chief drivers of patient health care costs are determined. The cost driver identification takes place in two key health care settings - in a mid-size hospital and in general practice. Large data sets are used. Data are first pre-processed with dimension-reduction techniques, to eliminate statistical noise. Then, the analysis crucially involves elimination of effect overlaps between potential cost drivers. This is made possible through the choice of suitable, rigorous techniques that enable the influence of all potential cost drivers to be quantified concurrently, and not in isolation from each other. In relation to the hospital data, penalised regression techniques are used to conduct variable selection and also as a bias–variance trade-off estimator to minimise the overall error. The sparse-group LASSO (SGL) method is specifically selected and used. In relation to the primary care data, a hybrid use of penalised regression with Artificial Neural Networks is successfully carried out. This innovative approach combines the variable selection property of the former with the ability to effectively handle multiple interactions of the latter. Computation is supplemented by cluster analysis and random forest analysis, where necessary. The main cost drivers are reported for each setting, in the order of importance. Then, the cost drivers are compared across the settings. Commonalities of cost driver patterns are observed using the principles of context theorising, to facilitate meaningful comparisons in different contexts. It turns out that the main drivers appear to follow the patient journey through the health system and are not necessarily setting-specific. The two-stage approach addresses challenges related to uniquely structured, setting-specific data sets and the different nature of care provision in each setting. Additionally, conducting a literature review to support the above-mentioned cost driver analysis has enabled scoping of current and alternative provider payment methods in Australian general practice. While providing invaluable perspective on the above-mentioned cost driver analysis, these findings have also led to recommendations regarding future optimal payment mechanisms. The identification of a relatively small set of main cost drivers improves accuracy of cost predictions. This, in turn, enables evidence-based financial provisioning. Such an approach highlights the need to undertake costing in health care on the level of cost drivers, which facilitates comparability. The alternative approach of analysing different expenditure streams is less reliable. The tools described in this study enable accurate patient-level costings in health care. These methods also serve as a template for conducting accurate evaluation of the efficiency of a particular health service, using benchmarking with other services across the health sector. Given the large sizes of health care budgets in Australia and indeed around the world, these outcomes constitute significant public policy findings.

Quantification of Cost Determinants in Health Care

Files and links (1)

Abstract

Details

Metrics