Eligible studies enrolled patients with the diagnosis of cocaine dependence of any duration or severity randomly allocated to either acupuncture, sham or other control. Acceptable outcomes measures included: self-reported frequency of cocaine use, self-reported amount of cocaine use, or biochemical confirmation of cocaine abstinence. Biochemical confirmation of cocaine abstinence is defined as the absence of the cocaine metabolite benzoylecognine in the urine. We excluded trials of acupuncture methods and trials enrolling patients with polysubstance use or dependence.

Databases searched included: AMED (1985–November 2004), Campbell Collaboration (2001–January 2005), CINAHL (1982–January 2005), Cochrane Library (1998–January 2005), Cochrane Controlled Trials Registry (January 2005), E-Psyche (1993–January 2005), HTA (1988–January 2005), and MEDLINE (1966–January 2005). We additionally searched the Chinese literature through Wanfang (1997–January 2004) and the Chinese Hospital Knowledge Database (CHKD, 1994–2004). Unpublished studies were also sought using Clinicaltrials.gov and the UK National Research Register. We supplemented this search by hand-searching key journals and searching bibliographies of retrieved trials and reviews. We additionally contacted 5 authors to identify additional published or unpublished studies and to clarify methodological issues. There were no language restrictions.

Two reviewers (EM, PW) working independently and in duplicate, reviewed the abstracts and full text versions of identified reports and adjudicated their inclusion.

Three reviewers (PW, EM, JG) working independently extracted data from the included studies using a standardized form which included: patient characteristics, treatment and control descriptions, types of outcomes measured, adverse events, and study results.

Two reviewers (EM, PW) working independently and in duplicate assessed the reporting quality of the reports. Items collected included randomization procedure, allocation concealment, blinding of patients, care providers, and outcome assessors, adequate description of loss to follow-up, and co-interventions. These items were treated as single items and were abstracted for the purpose of sensitivity analyses in our meta-analysis, but were not used as a weighting application in our analysis.

We measured chance-adjusted inter-rater agreement for eligibility using the kappa statistic (κ). Where reported, we extracted each trial's outcome data from the intention-to-treat analyses. If biochemical confirmation of cocaine abstinence was not reported, we contacted the authors for details. When a study had more than one control arm, we used the sham acupuncture arm as the control group. We pooled the rates of biochemically-confirmed cocaine abstinence in intervention and control groups across included trials using a random-effects meta-analysis and report the results using a forest plot describing the odds ratios (OR) and 95% confidence intervals (CI) at last reported time point. A priori explanations of between-study differences beyond chance included reporting of allocation concealment, use of intention-to-treat analyses, and large loss to follow-up (>50%). We tested for heterogeneity using the Zalen test and measured the proportion of between-study differences not attributable to chance with the I² statistic12. Our secondary analysis considered all drop outs as treatment failures. All statistics were performed using StatsDirect (Manchester, 2003).

The 9 RCTs were conducted in the USA and included 1747 participants: 488 participants in active groups and 821 assigned to control groups (One RCT did not describe group sizes25).

One RCT included only crack cocaine users27, 5 RCTs included samples with mixed forms of cocaine abuse (eg. intravenous, inhaled, or intranasal) and 3 RCTs did not describe the type of cocaine or the route of administration262931. RCTs enrolled participants with different rates of anti-craving medication use and 3 RCTs included only patients using methadone in addition to cocaine232631. Three RCTs enrolled some patients using methadone242728, 2 RCTs excluded patients who used methadone2529 and 1 did not report the use of anti-craving medication among enrolled subjects30.

All 9 trials employed auricular acupuncture, 4 employed a specific auricular acupuncture regimen (National Acupuncture Detoxification Association: NADA) and 2 used a combination of auricular and body points. Five trials had more than 1 control group 2425262831or randomized subjects to receive methods including relaxation262831, anti-craving medication and brainwave modification24, or psychosocial treatment25.

Eight trials used urine assays for cocaine metabolites (benzoylecgonine) for biochemical confirmation of abstinence at follow-up; we were able to obtain results from 7 of them. Eight trials examined the likelihood of retaining patient participation in the trial, and 5 trials examined cocaine cravings; no trials reported participant follow-up or relapse.

Although all trials were randomized, only 4 trials described the randomization technique26283031. Two trials employed restriction to balance the groups2628. Allocation concealment was not adequately reported in any study. Information regarding acupuncture technique needle depth and needle type was present in 3 trials232629. Methods of inserting a sham needle as a control were used in 8 trials2325262728293031. Five trials reported more than 20% loss to follow-up (mean loss to follow-up across all trials was 50% [range, 0–63%])2526282930. Only 4 trials described reasons for withdrawals25262829.

We pooled results from 7 trials that reported biochemical confirmation of cocaine abstinence. The pooled odds ratio estimating the effect of acupuncture on cocaine abstinence was 0.76 (95% CI 0.45–1.27, P = 0.3, I² = 30%, Heterogeneity P = 0.19) at the last reported time point (range 4–12 weeks) (see figure 2). Our a priori hypotheses failed to explain the heterogeneity that was present. That is, for each a priori explanation, the magnitude of the effect differed little irrespective of the level of the hypothesized explanatory factor. Our secondary analysis, considering all dropouts as treatment failures, resulted in a pooled odds ratio of 0.76 (95% CI, 0.54–1.08), P = 0.12, I² = 0%, Heterogeneity P = 0.5).

Adverse effects reported included pain and fear of needles2326282931 No trials reported the proportion of participants suffering adverse effects.

This systematic review and meta-analysis does not support the use of acupuncture for the treatment of cocaine dependence. However, most trials were hampered by large loss to follow up and the strength of the inference is consequently weakened.

We minimized publication bias by conducting an extensive search through multiple databases, including Chinese-language databases. We successfully received original data from several authors. We also sought to minimize selection and ascertainment bias by including only randomized controlled trials with biochemical confirmation of abstinence in our meta-analysis. However, large loss to follow-up and unexplained inconsistency across the included trials weakens our inferences.

The conduct of explanatory trials to ascertain the efficacy of acupuncture for cocaine addiction is challenging. We do not know whether there is a group of patients more likely to respond to this intervention who will agree to take part and stay enrolled in a clinical trial with adequate duration of follow-up. Pragmatic trials enrolling heterogeneous patients need to be large enough to detect small treatment effects and should be conducted according to the intention-to-treat principle to avoid large loss to follow-up. However, we recognize the difficulty of enrolling and maintaining patients who use or are dependent on cocaine in an RCT. Design strategies to maintain patients enrolled in trials of cocaine addiction treatment represent a research frontier.

Blinding of participants in a clinical trial is important because knowledge of group assignment may influence responses to treatment32. Use of sham controls and blinding of patients may prevent bias. The sham acupuncture method has been an area of debate as it is difficult to determine if a needle inserted into the skin away from designated acupuncture points is inert2033. In this systematic review, 8 trials applied sham acupuncture in the control groups2325262728293031, 3 of which used the NADA technique for acupuncture protocol and these trials found inconsistent results262831. Three trials also used a similar 5-point acupuncture protocol and did not find a significant effect252729.

We identified one additional systematic review of acupuncture for cocaine addiction which is in the process of data collection for a Cochrane review, and worked collaboratively with the primary investigator [S. Gates, personal communication]. We found systematic reviews assessing the efficacy of acupuncture in patients abusing other substances (nicotine10, alcohol and heroin69). Our findings are consistent with these other reviews in that they included trials with large loss to follow-up and were unable to draw strong inferences about the efficacy of acupuncture as a single intervention in the management of substance abuse.

At present, no proven pharmacotherapies exist for cocaine dependence83435. Current psychosocial strategies may only be effective in select patients with stable living conditions and social support and meta-analyses of these interventions are ongoing636. However, a lack of proof of efficacy does not indicate a lack of efficacy, and the data are dramatically hampered by dropouts. In treatment environments wishing to use multimodality therapy, acupuncture may be used for treatment of cocaine dependence since it is a relatively safe intervention and may be in line with patient values and cultural experiences. However, in resource-constricted environments, the use of acupuncture for cocaine dependence is not justified with the current evidence.