Analyzing the Impact of Palliative Care Interventions on Cost of Hospitalization: Practical Guidance for Choice of Dependent Variable.

CONTEXT: Multiple cost analyses of hospital-based palliative care have been published in recent years, but there are important differences between studies in their choice of dependent variable, complicating interpretation of results.
OBJECTIVES: The purpose of this article was to compare three different established approaches to estimating treatment effect on hospital costs, to highlight that different approaches yield different results, and to provide some practical guidelines for investigators performing hospital cost analysis in future.
METHODS: A simple example is developed using simulated cost data for four hospitalized patients, one of whom receives usual care only and three of whom receive different interventions. The impacts of the interventions are calculated and compared for three different dependent variables: cost of hospitalization, mean daily costs, and "before-and-after" costs.
RESULTS: Both the magnitude of an intervention's cost-saving effect and the relative impact of different interventions vary according to which dependent variable is used. Cost of hospitalization provides the most useful results of the three options for evaluating an intervention's impact on resource use. Alternative approaches visible in the literature can be misleading with respect to cost effects. Where the intervention is first administered to different patients at different points in a hospital admission, incorporating intervention timing is essential to maximize accuracy of cost-effect estimates.
CONCLUSION: Investigators evaluating the impact of palliative care programs on hospital costs ought to use cost of hospitalization as the dependent variable in primary analysis unless the research question specifically justifies an alternative approach. Mean daily costs and "before-and-after" costs should be used only to address relevant research questions, and results must be interpreted carefully. Analyses should also incorporate timing of the intervention where appropriate.