Resource allocation in a patient population exhibiting bimodal and logarithmic cost behavior

Budgeting and forecasting is a part of the resource allocation process. Statistical models used in these processes often use data based on unimodal or normal probability distributions using mean values for analysis. However, health care data are frequently skewed and stratified, demanding careful analyses. In this study, financial data was examined over a 10-month period at an overseas federal health care facility in a patient population of 70 with a primary diagnosis of post-traumatic stress disorder without a concurrent concussive injury. There were 823 direct provider contacts incurring a total cost of $286,917 that included allocated fixed overhead. The data were stratified and highly variable as two distinct groups based on chronicity. Group A (acute) consisted of 62 cases costing $117,612. Group B (chronic) consisted of 8 cases costing $169,755. This data, presented in descending order, fit a logarithmic equation with an r value of 0.95. Using this equation, a decreasing linear budget model is developed that represents an alternative nonparametric approach to allocate resources for this population. The model predicted an expenditure of $284,880, within 0.70% of the amount actually spent, and addresses the subcomponent simplification issue raised by a 2008 Rand study and mitigates problems associated with data stratification or transformations.