Usefulness of US cost-of-illness studies in healthcare decision making.

OBJECTIVE: Cost-of-illness studies have been completed on scores of diseases over the past 30 years. The goal of this study was to review published cost-of-illness studies on US populations in order to evaluate the potential usefulness of the results in decision making.
METHODS: Medline and related databases were searched using diagnosis and economic terms. The bibliographies of the articles found were reviewed visually to identify further studies. Inclusion criteria required a specified diagnosis, the study to be published between 1 January 1985 and 30 April 1999 in an English-language peer-reviewed journal, a clearly defined US sample or national population, available and recent epidemiological data on prevalence and incidence of diagnosis, and money estimates of direct and/or indirect costs. Three readers reviewed each study. The senior reviewer settled all differences.
RESULTS: Searches found 1725 published studies; only 110 (6.4%) met all inclusion criteria. Main reasons for rejection were insufficient cost data (80%), insufficient information on data sources and aggregation or estimation methods (56%), inadequate sector data e.g. hospitalisations or work loss (48%), study of value, not cost, of illness (44%), not a US population (30%) and insufficient population detail (19%). There were 80 diagnosis categories, 28 of which had more than one study. Only 5 diagnoses had > or = 5 studies--Alzheimer's dementia, depression, diabetes mellitus, mental illness and stroke. Multifold cost variations were found among studies within diagnosis categories, even with the same method and data sources. The more narrowly defined diagnoses, depression and stroke, had the smallest cost variation, 41.7 and 17.2%, respectively. A generalised linear regression model found that a significant portion of total and direct cost variance could be explained only for Alzheimer's dementia.
CONCLUSIONS: The wide variation of cost estimates for the same diagnosis raises serious questions of comparability, accuracy, validity and usefulness of all studies. Implementing guidelines to standardise methods and study design for cost-of-illness studies would be a worthwhile first step. The advantages and disadvantages of using money or another metric such as disability-adjusted life-years as the prime outcome measure should also be publicly discussed.