Neural Network-Based Drug Abuse Treatment Optimization

In 2009, 2.6 million people in the United States over 12 years of age received treatment for substance abuse at a “specialty facility”. The direct cost of substance abuse treatment was estimated to be $22 billion in 2005, up from $11 billion in 1991. With recent federal health care reform, more clients may seek treatment, but it is unclear how this would affect clinics’ costs per client. Although past studies on substance abuse costs find some evidence of economies of scale, they also find wide variation in the average costs of treatment, the reasons for which require further investigation. As a result, state substance abuse agencies and other payers face difficulty in determining how to pay providers, in fact, very often funding is inadequate or wasted. The drug abuse treatment process can be viewed as a complex adaptive system in which system inputs such as workers, physical capital and other materials are converted into outputs such as treatment completions and outcomes. Treatment clinics desire the most cost efficient input mix to generate desired outputs. Although substance abuse researchers are ahead of many other fields in developing and refining instruments such as the Drug Abuse Treatment Cost Analysis Program, many of these instruments are restricted by the fact that databases have many omissions making it difficult to answer research questions that might otherwise be informed by economic analysis. Additionally, due to the large numbers of variables involved, trade-offs are difficult to explore. Questions such as “what are the relationships between costs and case-mix?” or “what are reasonable reimbursement rates for substance abuse treatment and how do they differ for specialized populations such as heroin users?” are difficult to answer. The work reported in this paper provides significant insight that can help policymakers to decide where to focus funding through 1) prediction of missing treatment data and 2) identification of critical variables.