What works in drug addiction?

Substance dependence, or ‘addiction’, is diagnosed taking several factors into consideration (Box 1). Substance misuse refers to the non-therapeutic use of drugs in a manner that is potentially harmful, but does not meet criteria for dependence. Guidelines for drug addiction treatment have been published by the Department of Health (1999). Many trials report significant benefits of addiction treatments (National Consensus Development Panel, 1998). However, only 20% of subjects report abstinence from all illicit substances for at least 1 year, despite receiving treatment. Furthermore, drop-out rates of nearly 50% are common. It is notable that only half of patients with other chronic disorders (such as hypertension or diabetes) fully adhere to medication schedules, and high drop-out rates are common in many forms of psychotherapy. Trials of treatment for drug addiction are liable to all the common methodological flaws seen in clinical trials in psychiatry, including failure to use intention-to-treat analysis, failure to randomise results, lack of socio-demographically matched control groups and confounding due to unplanned variations in contact with treatment services. A US government report recently concluded that ‘results derived from self-selected patients who remain in treatment optimistically skew findings in favour of effectiveness’ (National Research Council, 2002). There is no consensus on outcome measures of trials of addiction treatments. Urine (and saliva) analysis can provide objective measures of drug use. However, many trials report subjective ratings, such as scores on the Addiction Severity Index (McLellan et al, 1980), a 45-minute semi-structured interview based on psychosocial functioning and drug use. Meta-analysis results are often expressed as an effect size: the difference in mean scores divided by the pooled standard deviation. This statistical technique allows the direct comparison of the results of trials that have used different outcome measures. A trial comparing 50–100 users and controls is usually sufficient to identify a treatment with a modest effect size (conventionally 0.25–0.5) that is likely to be clinically significant.