Published on 14 September 2026
Stimulant use disorder: do psychosocial treatments keep patients in care without making them abstinent?
In brief
This is the first update of a Cochrane review published in 2016, and it brings together 64 randomised trials and 8,241 adult participants with stimulant use disorder. The primary comparison, labelled against no intervention, in fact sets a psychosocial intervention added to usual care against usual care alone, across 44 trials: it is the comparison that answers the question of adding a psychotherapy to what is already being done, and it carries five of the six high-certainty results of the review. Psychosocial interventions reduce dropout from treatment (RR 0.82, 95% CI 0.74 to 0.91; 30 studies, 4,078 participants; high certainty). On point abstinence they make little to no difference, at the end of treatment (RR 1.15, 95% CI 0.94 to 1.41; 12 studies, 1,293 participants) as at the longest follow-up (RR 1.22, 95% CI 0.91 to 1.62; 9 studies, 1,187 participants), and in both cases at high certainty. Between those two poles sit the intermediate results: frequency of drug intake falls (standardised mean difference, SMD, of −0.35, 95% CI −0.50 to −0.19; high certainty), the longest period of abstinence lengthens (SMD 0.54, 95% CI 0.41 to 0.68; high certainty), and continuous abstinence at the end of treatment probably increases (RR 1.89, 95% CI 1.20 to 2.97; moderate certainty). Of those three outcomes, only continuous abstinence is measured again at the longest follow-up, and the gain is not found there; the other two were not measured at a distance, which is not the same thing as an absence of effect. The second comparison, 12 trials, sets psychotherapy against usual care instead of adding it: only the reduction in dropout holds (RR 0.79, 95% CI 0.65 to 0.97; high certainty), the rest resting on one to four trials where the outcome was measured at all, with certainty ranging from very low to moderate. The authors conclude that the most studied and most promising approach is contingency management, and that the others have been too little explored to rule out imprecision.
The context
No medicine is approved for the treatment of stimulant use disorder. The review opens on that statement, and it governs everything else: here the psychosocial intervention is not an adjunct to treatment, it is the treatment. The question of its efficacy is therefore not academic, it is the only question available.
The body of trials reflects a precise epidemiology. Of the 64 trials, 73% concern cocaine or crack cocaine use disorder, 10.9% methamphetamine, 3.1% amphetamine and 12.5% stimulants without further distinction. In 18 trials, all participants were in methadone maintenance treatment, which describes a population of polydrug users followed in a service, and not the patient who consults for isolated cocaine use.
A note on method
High certainty that makes a large effect unlikely
It is worth pausing on the exact wording the authors use, because it is uncommon and because it is often misread. On point abstinence they do not write that an effect has not been demonstrated. They write that psychosocial treatments “make little to no difference to point abstinence at the end of treatment”, and they attach a high GRADE certainty to that sentence. What this licenses is precise: a large effect is unlikely. It amounts neither to proof of no effect, nor to a demonstration of equivalence.
Two situations that have nothing in common nevertheless look alike on a forest plot, since in both cases the interval crosses the line of no effect.
| Situation | What is observed | What may be said |
|---|---|---|
| Absence of evidence | Wide interval, few trials, low or very low certainty. Example in this review: point abstinence against usual care, RR 1.67, 95% CI 0.64 to 4.31; 1 trial, 128 participants | We do not know. The interval is compatible with a fourfold increase as much as with a loss of benefit |
| A large effect made unlikely | Interval compatible at most with a modest effect, substantial body of trials, high certainty. Example: point abstinence against no intervention, RR 1.15, 95% CI 0.94 to 1.41; 12 trials, 1,293 participants | A large effect is unlikely, which is not the same thing as an absence of effect. Running further trials on this outcome and this comparator has little chance of revealing a large benefit |
The distinction is the one GRADE was built to make visible, and it changes what should be done. An absence of evidence calls for trials. A large effect made unlikely by a substantial body of trials calls instead for a change of question.
One reservation, all the same, about the second case. The upper bound of the interval reaches 1.41, that is a relative increase of 41% in the proportion of patients abstinent at a given moment. Calling that little to no difference presupposes a threshold of clinical relevance that the authors nowhere state, and such a threshold is a judgement, not a measurement. The hurried reader will take away that the effect is nil; the attentive reader will take away that a modest effect remains compatible with these data, and that twelve trials and 1,293 participants carry this estimate, against thirty trials and 4,078 participants for dropout.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Adults with stimulant use disorder. 64 randomised trials, 8,241 participants. Cocaine or crack in 73% of trials, methamphetamine 10.9%, amphetamine 3.1%, stimulants without distinction 12.5%. In 18 trials, all participants were in methadone maintenance treatment | |
| Intervention | |
| Any psychosocial intervention. Comparison 1, labelled against no intervention: psychosocial intervention added to usual care against usual care alone, 44 trials: contingency management 27, cognitive behavioural therapy 12, motivational interviewing 3, psychodynamic therapy 1, CBT plus contingency management 1. Comparison 2, psychotherapy against usual care, 12 trials: CBT 7, contingency management 2, motivational interviewing 2, CBT plus motivational interviewing 1. To these are added 7 trials comparing contingency management reinforcement related to abstinence against reinforcement not related to abstinence, and 7 trials comparing two psychosocial approaches with one another | |
| Comparator | |
| No intervention, treatment as usual, or a different intervention | |
| Outcomes | |
| Dropout from treatment, point abstinence at the end of treatment and at the longest follow-up, continuous abstinence at the end of treatment and at the longest follow-up, frequency of drug intake, longest period of abstinence, harms | |
| Design | |
| Cochrane systematic review with meta-analyses of randomised trials, first update of a review published in 2016. Cochrane Drugs and Alcohol Group Specialised Register, CENTRAL, MEDLINE, Embase, three other databases and two trials registers, searched in September 2023, non-English language literature included, handsearching of the references of topic-related systematic reviews and of the included studies · CEBM 1a · Cochrane risk of bias tool and GRADE |
Quality control
| Criterion | Status |
|---|---|
| Completeness of the search | Sound |
| FindingSpecialised register, CENTRAL, MEDLINE, Embase, three further databases, two trials registers, non-English language literature, handsearching. What is expected of a Cochrane review, and here it is done | |
| Grading of confidence | Sound |
| FindingGRADE applied outcome by outcome and comparison by comparison, with levels running from very low to high. Every estimate is accompanied by the number of trials and of participants | |
| Calibration of the conclusions | Sound |
| FindingThe authors write explicitly that “the most studied and promising psychosocial approach is contingency management”, and that “relatively few studies explored the other approaches, so we cannot rule out the possibility that the results were imprecise due to small sample sizes”. They do not generalise to psychotherapy as a whole | |
| Sequence generation | Reservation |
| Finding65.6% of trials judged at low risk of bias | |
| Allocation concealment | Weakness |
| FindingOnly 19% of trials judged at low risk. This is the weak point of the body of evidence, and it is not attributable to the review | |
| Blinding | Reservation |
| FindingImpossible with this type of intervention. All trials are judged at high risk of performance bias for subjective outcomes, and at low risk for objective outcomes. Biological verification of drug use limits the reach of the problem on the outcomes that matter most here | |
| Detection bias, subjective outcomes | Weakness |
| FindingOnly 22% of trials judged at low risk | |
| Missing data | Sound |
| Finding69% of trials judged at low risk of attrition bias, which is high for this field | |
| Collection of harms | Weakness |
| Finding5 trials out of 64 report harms related to psychosocial interventions, and 4 of them state that no adverse events occurred. There is therefore nothing on which to discuss tolerability, which does not license the assumption that it is perfect | |
| Independence | Reservation |
| FindingThe interests declared by the authors are exclusively editorial. Silvia Minozzi is Joint-Coordinating Editor of Cochrane Drugs and Alcohol Group, Laura Amato and Roberta Agabio are editors of the same group, and all three state that they were not involved in the editorial process of the present review; Rosella Saulle and Francesco Traccis declare none known. No link with industry is declared, and an absence of mention is not an absence of link | |
The findings
Comparison 1, against no intervention (psychotherapy added to usual care against usual care alone). This is the primary comparison of the review, 44 trials, and it is the one that concerns the clinician: it asks what a structured psychotherapy adds to what is already being done.
| Outcome | Published result |
|---|---|
| Dropout from treatment | RR 0.82 (95% CI 0.74 to 0.91); 30 studies, 4,078 participants; high certainty |
| PEB readingEstablished the best supported result of the review. Relative reduction of about 18% in the risk of dropout | |
| Point abstinence, end of treatment | RR 1.15 (95% CI 0.94 to 1.41); 12 studies, 1,293 participants; high certainty |
| PEB readingLittle to no difference at high certainty, which is not the same thing as a lack of data. Reservation: the upper bound stands at 1.41 | |
| Point abstinence, longest follow-up | RR 1.22 (95% CI 0.91 to 1.62); 9 studies, 1,187 participants; high certainty |
| PEB readingLittle to no difference the picture does not improve with time | |
| Continuous abstinence, end of treatment | RR 1.89 (95% CI 1.20 to 2.97); 12 studies, 1,770 participants; moderate certainty |
| PEB readingProbable increase the only frankly positive result on abstinence, and it is downgraded by one level. It concerns abstinence maintained during treatment, which is precisely what contingency management rewards | |
| Continuous abstinence, longest follow-up | RR 1.14 (95% CI 0.89 to 1.46); 4 studies, 295 participants; low certainty |
| PEB readingUncertain four trials and 295 participants. What was gained during treatment is not found afterwards | |
| Frequency of drug intake, end of treatment | SMD −0.35 (95% CI −0.50 to −0.19); 10 studies, 1,215 participants; high certainty |
| PEB readingEstablished reduction a small to moderate effect size by the usual conventions. Using less is not using rarely enough to be abstinent | |
| Longest period of abstinence | SMD 0.54 (95% CI 0.41 to 0.68); 17 studies, 2,118 participants; high certainty |
| PEB readingEstablished lengthening the largest effect in the review. Patients hold out longer, without more of them being abstinent at the moment they are measured | |
Comparison 2, against usual care. This one, 12 trials, sets psychotherapy against usual care instead of adding it to usual care: it asks whether a structured psychotherapy does better than routine management, not what it adds to it. It is almost empty.
| Outcome | Published result |
|---|---|
| Dropout from treatment | RR 0.79 (95% CI 0.65 to 0.97); 9 studies, 735 participants; high certainty |
| PEB readingEstablished retention survives the change of comparator | |
| Point abstinence, end of treatment | RR 1.67 (95% CI 0.64 to 4.31); 1 study, 128 participants; low certainty |
| PEB readingNot conclusive one trial, an interval running from 0.64 to 4.31. The label of little to no difference chosen by the authors does not describe the point estimate, which would correspond to a large effect: it comes from the interval crossing the line of no effect, and here it reflects imprecision, not a demonstrated equivalence | |
| Point abstinence, longest follow-up | RR 1.31 (95% CI 0.86 to 1.99); 2 studies, 124 participants; very low certainty |
| PEB readingWe do not know the authors themselves write that they are uncertain | |
| Continuous abstinence, end of treatment | RR 1.18 (95% CI 0.92 to 1.53); 1 study, 128 participants; low certainty |
| PEB readingNot conclusive a single trial | |
| Continuous abstinence, longest follow-up | No trial assessed this outcome in this comparison |
| PEB readingData absent this is not a negative result, it is a hole | |
| Frequency of drug intake | SMD −1.17 (95% CI −2.81 to 0.47); 4 studies, 479 participants; moderate certainty |
| PEB readingEstimate unusable the interval runs from a very large effect to a modest unfavourable one. Describing that as little to no difference is not a reading of the point estimate, which corresponds here to a large effect: the label comes from the interval crossing the line of no effect | |
| Longest period of abstinence | SMD −0.16 (95% CI −0.54 to 0.21); 1 study, 110 participants; low certainty |
| PEB readingNot conclusive one trial, 110 participants | |
A point of reporting, for honesty. On the dropout outcome in comparison 1, the figures used here are those of the publication: 30 trials and 4,078 participants. Neighbouring but different values circulate, notably 29 trials and 4,049 participants; they do not appear in the abstract of the review, which is the source of the figures used in this analysis. The gap would in any case bear on one trial and 29 participants, and it would change neither the estimate nor its certainty.
Three further precise values circulate in the same way and are not taken up here, because they appear neither in the abstract nor in the plain language summary: an absolute effect of 64 dropouts avoided per 1,000 patients treated, heterogeneity of 98% for frequency of drug intake against usual care, and a downgrade of continuous abstinence for funnel plot asymmetry. The published abstract reports moderate certainty on that last outcome without giving the reason.
Critical appraisal
| Domain | Judgement |
|---|---|
| Level of evidence of the synthesis | Sound |
| FindingThe top of the hierarchy. Meta-analyses of randomised trials exclusively, standard Cochrane methodology, GRADE outcome by outcome, first update of a review maintained since 2016 | |
| What the body of trials actually measures | Reservation |
| FindingTwenty-seven of the 44 trials in the primary comparison concern contingency management, that is the material reward of a biologically verified abstinence. Twelve concern CBT, three motivational interviewing, one a psychodynamic therapy. Reading these results as a verdict on psychotherapy in general would be an error, and the authors take care not to commit it | |
| Retention and abstinence come apart | Central finding |
| FindingRetention improves at high certainty in both comparisons. On point abstinence, against no intervention, a large effect is unlikely, also at high certainty. Between the two, frequency of drug intake falls and the longest period of abstinence lengthens. These are not contradictory results: they describe a patient who stays, who uses less and who holds out longer, without being abstinent at the moment of assessment | |
| Durability | Weakness |
| FindingThe only frankly positive result on abstinence concerns the end of treatment. At the longest follow-up, continuous abstinence returns to RR 1.14 with an interval crossing the line of no effect and low certainty. When the reinforcement stops, the effect fades. This is the known limit of contingency management, and this review does not lift it | |
| Comparator relevant to practice | Weakness |
| FindingIn comparison 2, the one that sets a psychotherapy against usual care instead of adding it, six outcomes out of seven support no conclusion: five rest on one to four trials, and one, continuous abstinence at the longest follow-up, was never assessed. Only dropout is solidly documented. Whether a structured psychotherapy advantageously replaces routine management therefore remains unanswered | |
| Gap between the label and the interval | Reservation |
| FindingTwo estimates are described as little to no difference although their point estimates correspond to large effects and their intervals are very wide, RR 1.67 (0.64 to 4.31) and SMD −1.17 (−2.81 to 0.47). The label therefore does not come from the point estimate, it comes from the interval crossing the line of no effect: the authors apply a convention founded on significance, not on the size of the effect. It does not say that equivalence is demonstrated, and the reader must look at the interval | |
| External validity | Reservation |
| FindingA body of trials dominated by cocaine and crack, methamphetamine at 10.9% while its place is growing in several European countries, and 18 trials in which all participants were in methadone maintenance treatment. The geographical spread weighs too: most trials were conducted in the United States, the others being distributed between Spain (4), Australia (3), the United Kingdom (3), Switzerland (2), Brazil (2), Iran (2), the Netherlands (1) and South Africa (1). Contingency management, which carries most of the evidence, is first of all a North American model of service organisation, which bears on the transferability of the whole. Programmes range from a single session to twelve months, for an average duration of about four months | |
| Safety | Weakness |
| FindingFive trials out of 64 document effects related to the intervention. A psychotherapy has potential adverse effects, and a conditional reward scheme has specific ones. This field is not explored | |
What is demonstrated, what is suggested and what is a matter of judgement must be named. Demonstrated: in comparison 1, psychosocial interventions reduce dropout, reduce the frequency of drug intake and lengthen the longest period of abstinence, and a large effect on the proportion of patients abstinent at a given moment is unlikely there, which does not amount to a demonstrated absence of effect. These four statements rest on high certainty. Suggested: continuous abstinence increases at the end of treatment, at moderate certainty, and that gain is not maintained. Judgement: that the result is largely carried by contingency management is a fact about the body of trials, but inferring from it that the other approaches are ineffective would be an inference the data do not support. The authors say the opposite, namely that they have been too little studied.
From evidence to service
Contingency management consists in rewarding, in material terms, an abstinence verified by biological testing, according to a reinforcement schedule defined in advance. It is the intervention that carries most of the evidence in this review, and the European Union Drugs Agency classifies it among beneficial interventions on its best practice portal, while noting that the effect is not maintained at the longest follow-up.
Its place in routine services is another matter, and it is worth being precise about the status of what we assert. We found no inventory, national or international, of how widely it is available outside research settings. What the review itself shows is that its evidence base was built mainly in the United States, in a model of service organisation that is not the one most readers work in, and that the programmes studied ran from a single session to twelve months.
We therefore present the limited availability of contingency management as a field observation, not as a measured datum. The consequence, on the other hand, is clear: the most solid evidence we have in this disorder concerns the intervention least accessible to our patients, and what we actually provide has almost never been compared with usual care.
Level of evidence
PEB appraisal: high confidence in the reduction of dropout, in the reduction of the frequency of drug intake and in the lengthening of the longest period of abstinence, high confidence too in the fact that a large effect on point abstinence is unlikely, moderate confidence in the increase of continuous abstinence at the end of treatment, and very low to moderate confidence, with intervals that are often unusable, in everything that sets a psychotherapy against usual care beyond dropout. The methodological apparatus is that of a Cochrane review conducted by the book, with conclusions that do not outrun the data. What limits the reach of the work lies in the body of trials: allocation concealment documented in only 19% of trials, blinding impossible, harms almost never reported, and a concentration of the evidence on a single approach.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ Sixty-four trials, eight thousand patients, and the solid result is that they stay in care. On abstinence at a given moment, no large benefit to expect, and this time it is not for want of data. Fine. It is still worth keeping them, they use less and they hold out longer. But what works best in these trials is paying for abstinence, and that I cannot offer. ”
What this means in practice: what changes on Monday morning is not the prescription, it is what we promise the patient. Retention is a defensible objective in itself, provided it is named as such.
What you can do with this on Monday morning. What should be done does not change. What changes is the way we talk about it and what we decide to measure.
- Reword the objective announced to the patient. What these data support is that psychotherapeutic follow-up helps a person stay in care, use less often and hold out longer between episodes of use. What they do not support is the promise of stabilised abstinence. Announcing the first objective rather than the second is not giving up, it is preventing the patient from reading a lapse as a failure of treatment, which is a frequent reason for dropping out of follow-up.
- Change the indicator used in consultation. Rather than scoring abstinent or not each time, record the longest period of abstinence since the last consultation and the frequency of use over the past four weeks. These are the two outcomes on which the review finds an effect at high certainty, and they give the patient a visible trajectory where the binary outcome shows only a relapse.
- Document retention as an outcome. An attended consultation is a treatment outcome in this disorder, not a mere precondition. Saying so to the patient, and writing it in the record, helps at the point of transitions and referral letters.
- Explain the gap with what the patient may have read. A patient who has searched online will find contingency management presented as the reference treatment. You can explain in one sentence what it is, why it is not on offer here, and what in its underlying principles remains transferable: making the benefit of abstinence immediate and concrete rather than deferred and abstract, explicitly acknowledging every negative test, bringing consultations closer together in high-risk periods.
- Report the gap upwards. The distance between the available evidence and the treatment on offer is a field observation that belongs in exchanges with addiction services and in service planning, rather than in individual resignation.
- Take the reading grid with you. High certainty that makes a large effect unlikely, against low certainty on a wide interval: the distinction serves in every file, well beyond stimulants.
- The course of action is set out in the NICE decision tree for depression in adults.
Frequently asked questions
Should we stop offering psychotherapy in stimulant use disorder?
No, and the review suggests it at no point. It establishes at high certainty that these interventions reduce dropout, reduce the frequency of drug intake and lengthen the longest period of abstinence. Since no medicine is approved in this indication, they remain the available therapeutic offer. What is revised is the objective announced, not the indication.
How can one judge that a large effect is unlikely?
By having a sufficient body of trials and a confidence interval compatible at most with a modest effect. That is what a high GRADE certainty paired with the phrase little to no difference conveys: it does not say that the treatment does nothing, it says that a large benefit is improbable on that outcome. The judgement also presupposes a threshold of clinical relevance, which the authors do not state. Conversely, a wide interval with two or three trials and low certainty means only that we do not know. Both situations coexist in this review, and confusing them reverses the reading.
Why does continuous abstinence increase when point abstinence does not follow?
The two outcomes do not measure the same thing. Point abstinence is a state observed on a given date. Continuous abstinence is a series of consecutive negative tests during treatment, which is exactly what contingency management rewards. A scheme that pays for a succession of negative tests mechanically shifts that second outcome. Whether the gain persists after the reinforcement stops remains an open question, and the data at the longest follow-up do not show it.
Is contingency management available in routine practice?
We found no inventory that would allow an answer supported by a figure. What the review documents is where its evidence comes from: most of the trials were conducted in the United States, and contingency management is first of all a North American model of service organisation. The European Union Drugs Agency classifies the intervention as beneficial on its best practice portal while noting that the effect is not maintained at the longest follow-up, and its summary carries no information on implementation in member states. PEB therefore presents the limited availability of this approach as a field observation, and not as a figure.
Do these results hold for methamphetamine?
With caution. Trials concerning methamphetamine use disorder represent 10.9% of the body of evidence, against 73% for cocaine and crack. The review reports no separate analysis by substance in its abstract. Transposing the overall estimate to methamphetamine is an extrapolation, not a reading.
What is known about the harms of these interventions?
Almost nothing. Five trials out of 64 report effects related to the psychosocial intervention, and four of them state that no adverse events occurred. A collection this incomplete allows neither a conclusion of safety nor the identification of a risk. It is a missing datum, not a reassuring one.
Annotated bibliography
Minozzi S, Saulle R, Amato L, Traccis F, Agabio R (2024). Psychosocial interventions for stimulant use disorder. Cochrane Database of Systematic Reviews, 2, CD011866. DOI 10.1002/14651858.CD011866.pub3 · PMID 38357958 · PMCID PMC10867898. Source study analysed here. First update of a review published in 2016, evidence current to 26 September 2023. Authors’ declarations of interest: Silvia Minozzi, Joint-Coordinating Editor of Cochrane Drugs and Alcohol Group, Laura Amato and Roberta Agabio, editors of the same group, all three stating that they were not involved in the editorial process of the present review; Rosella Saulle and Francesco Traccis, none known. No link with industry is declared, which does not amount to an absence of link. The figures cited come from the published abstract and from the plain language summary of the review.
European Union Drugs Agency (EUDA). Contingency management to reduce stimulant use, best practice portal. euda.europa.eu. Evidence summary classifying the intervention as beneficial, with the statement that the effect is not maintained at the longest follow-up. The summary carries no information on the implementation of this approach in member states, which was checked.