Estimating substance abuse treatment need by state.

Abstract

SAMHSA (Substance Abuse and Mental Health Services Administration) and others have used a variety of methods over the past three decades to measure the scope of the substance abuse problem. These efforts have focused on national estimates of heroin addicts, cocaine abusers, and the overall population in need of treatment for any type of substance. There have also been efforts to make these assessments at the state and metropolitan area level. Direct survey estimation, small area estimation models, synthetic estimates, capture-recapture models, and multiple indicator index approaches have been used. Each has its strengths and weaknesses. In this editorial, we describe the method currently used by SAMHSA to estimate treatment need by state, and comment on the index method proposed by McAuliffe & Dunn in this issue of Addiction (McAuliffe & Dunn 2004). A critical first step in assessing treatment need is to define it. SAMHSA uses the DSM-IV criteria for substance dependence or abuse, widely used by policymakers, treatment providers, and researchers. Persons who either had a disorder or received treatment for a disorder within the past 12 months are counted in the need estimates. A hierarchical Bayes estimation method produces state level estimates from the National Survey on Drug Use and Health (NSDUH; formerly called the National Household Survey on Drug Abuse). The estimates are based on the self-report data from respondents in each state, combined with a national model that characterizes the statistical relationship between local area characteristics and rates of treatment need. Because of its requirement for the direct data from each state, this method became feasible with the state-based redesign of the NSDUH sample in 1999. The NSDUH method has several important properties that index methods do not, including the one proposed by McAuliffe and Dunn. First, estimates are based on data collected independently in each state, using probability sampling. Definitions and data collection methods are clear and consistent across all states. Data reflect recent, well-defined time periods. Interpretation of results is aided by measures of reliability (e.g. prediction intervals) for each state estimate, accounting for both sampling error and modeling error. A validation has demonstrated that the state estimates have small biases and much better precision than design-based estimates (OAS 2003). Finally, estimates can be produced for demographic subgroups, for substate areas, and for different levels of problem severity. On the other hand, there are some important limitations and potential biases in the NSDUH method. For determining state differences, the main limitation is small sample sizes, which is most problematic for measures with low prevalence, such as illicit drug treatment need. To improve reliability, SAMHSA has begun to use two or more years of data combined to make state estimates. The other major limitation is the likely underestimation of treatment need due to underreporting and undercoverage, a concern that SAMHSA will continue to address with its ongoing program of methodological studies (Gfroerer, Eyerman & Chromy 2002). The introduction of respondent incentive payments in 2002 appears to have reduced this bias. The index approach proposed by McAuliffe and Dunn provides a useful addition to the research literature on state variations in substance abuse. However, a drawback of this and other index methods is their lack of clarity on what they measure. Because it is based on arrest and mortality data, it covers incarcerated and homeless substance abusers that are not well covered by the NSDUH, and it probably is correlated with heroin and cocaine abuse rates. On the other hand, the index does not adequately capture treatment need among youths or the large population with marijuana problems. We think the small area estimation approach used by SAMHSA for treatment needs assessment is more valid and useful to policy officials. We also recognize the limitations of the NSDUH and regularly evaluate its methods, report the results of these evaluations, and make improvements when necessary. Finally, we believe that NSDUH state estimates ‘are most useful for policy purposes when used in conjunction with other data sources at the state level, recognizing the strengths and limitations of all the data.’ (OAS 2000; p. 169).