Efficacy of interventions to reduce long term opioid treatment for chronic non

Objective To review interventions to reduce long term opioid treatment in people with chronic non-cancer pain, considering efficacy on dose reduction and discontinuation, pain, function, quality of life, withdrawal symptoms, substance use, and adverse events.

Design Systematic review and meta-analysis of randomised controlled trials and non-randomised studies of interventions.

Data sources Medline, Embase, PsycINFO, CINAHL, and the Cochrane Library searched from inception to July 2021. Reference lists and previous reviews were also searched and experts were contacted.

Eligibility criteria for study selection Original research in English. Case reports and cross sectional studies were excluded.

Data extraction and synthesis Two authors independently selected studies, extracted data, and used the Cochrane risk-of-bias tools for randomised and non-randomised studies (RoB 2 and ROBINS-I). Authors grouped interventions into five categories (pain self-management, complementary and alternative medicine, pharmacological and biomedical devices and interventions, opioid replacement treatment, and deprescription methods), estimated pooled effects using random effects meta-analytical models, and appraised the certainty of evidence using GRADE (grading of recommendations, assessment, development, and evaluation).

Results Of 166 studies meeting inclusion criteria, 130 (78%) were considered at critical risk of bias and were excluded from the evidence synthesis. Of the 36 included studies, few had comparable treatment arms and sample sizes were generally small. Consequently, the certainty of the evidence was low or very low for more than 90% (41/44) of GRADE outcomes, including for all non-opioid patient outcomes. Despite these limitations, evidence of moderate certainty indicated that interventions to support prescribers’ adherence to guidelines increased the likelihood of patients discontinuing opioid treatment (adjusted odds ratio 1.5, 95% confidence interval 1.0 to 2.1), and that these prescriber interventions as well as pain self-management programmes reduced opioid dose more than controls (intervention v control, mean difference –6.8 mg (standard error 1.6) daily oral morphine equivalent, P<0.001; pain programme v control, −14.31 mg daily oral morphine equivalent, 95% confidence interval −21.57 to −7.05).

Conclusions Evidence on the reduction of long term opioid treatment for chronic pain continues to be constrained by poor study methodology. Of particular concern is the lack of evidence relating to possible harms. Agreed standards for designing and reporting studies on the reduction of opioid treatment are urgently needed.

Review registration PROSPERO CRD42020140943.

Opioid overprescribing for patients with chronic non-cancer pain, where the harms of opioid treatment outweigh the benefits, has led to the recent promulgation of guidelines recommending the reduction or discontinuation of long term opioid treatment.1234 The publication of these guidelines accords with the acceleration of a pre-existing downward trend in opioid prescribing.567 Yet changing treatment in the context of chronic pain is not straightforward.8910 Tapering is the gradual reduction of opioid drug treatment with the goal to either reduce or discontinue opioids while limiting possible adverse effects such as withdrawal symptoms and increased pain. The US Centers for Disease Control and Prevention recommends against the prescription of more than 90 mg oral morphine equivalent per day in most circumstances.1 However, risks are involved even below this threshold, and perspectives differ as to what constitutes a safe dose.411 Opioid tapering can be complicated by the onset of withdrawal symptoms as well as increased pain, suicidality, and substance use, and such risks could increase when undertaken rapidly or without patient consent.121314

Previous reviews have considered the outcomes of clinical interventions to facilitate opioid tapering, including one Cochrane review and nine other systematic, scoping, and rapid reviews.1516171819202122232425 Included studies evaluate, among other things, pain management programmes, the drug management of withdrawal symptoms, and biomedical procedures. The variety of approaches to opioid tapering reflects the complexity of this process, the differing causes of chronic pain and approaches to treatment (biomedical, biopsychosocial, alternative medicine), and the presence of comorbidities such as substance use disorder. The most comprehensive review to date, published in 2017 by Frank and colleagues,18 found very low quality evidence that several types of intervention might be effective in the reduction or discontinuation of opioid treatment, and that pain, pain related function, and quality of life could improve with opioid tapering. However, the researchers acknowledged a dearth of evidence, especially for adverse events such as overdose and suicide. Owing to a paucity of studies at low risk of bias and the difficulties associated with synthesising clinically heterogeneous interventions, prior reviews have been unable to recommend a particular intervention with better than low certainty (appendix 1).

Clinicians continue to make decisions without strong evidence. In the context of deprescription guidelines,1234 local and regional policy changes,26 and attempts to reduce opioids becoming more common,567 recently published studies could contribute some new evidence. However, effective supporting evidence needs to overcome the clinical heterogeneity that has mired previous systematic reviews by appropriately differentiating between treatment types. Therefore, this systematic review aims to provide a clinically relevant synthesis of up-to-date evidence on the efficacy of interventions to reduce or discontinue long term opioid treatment in patients with chronic non-cancer pain.

This review focuses on two key questions: how effective are the interventions to reduce or discontinue long term opioid treatment, and what are their effects on patient outcomes? We followed the PRISMA (preferred reporting items for systematic reviews and meta-analyses) guidelines27 and registered the protocol in PROSPERO (CRD42020140943).

We developed a sensitive search strategy with the help of a research librarian based around the key concepts of "opioids," "tapering," and "pain" (appendix 2) and searched Medline, Embase, PsycINFO, the Cochrane Library, and CINAHL for articles published in English from inception to 28 July 2021. We also examined reference lists from all included studies, sourced studies from known reviews on the topic, and sought input from expert contacts.

Included studies had to answer at least the first key question (regarding the effectiveness of interventions to reduce or discontinue long term opioid treatment) and report the experiences of adults (age ≥18 years) with chronic pain (defined as pain persisting for more than three months) who were prescribed opioid treatment for pain management in randomised controlled trials, non-randomised controlled trials, and uncontrolled studies. Outcomes for the first key question were the number of people who discontinued opioids as a result of treatment and change in opioid dose (in oral morphine equivalent per day). Outcomes for the second key question (regarding the effects of these interventions on patient outcomes) were pain intensity, pain related function, quality of life, opioid withdrawal symptoms, substance use, and adverse events.

Included studies reported original research, clearly described a clinical intervention, and were in the English language. To capture the broad field of literature, interventions were not required to have an explicit goal of opioid tapering. For instance, we included studies in which the intervention might have an auxiliary effect on opioid use. Studies were excluded if they included patients with only acute, surgical, or postoperative pain; patients in hospice or palliative care only; people using opioids for non-medical reasons only; or non-human participants. Studies of patients with cancer and HIV pain were excluded. We included uncontrolled studies alongside controlled studies in order to capture more evidence on infrequent outcomes such as substance use and adverse events. However, case reports and cross sectional studies were excluded.

Titles and abstracts were double screened (by NA, AGM, AG, and PG), with a random sample screened by a third reviewer (NA or PG) to check for inter-rater reliability. Two reviewers (of NA, AGM, AG, or PG) checked full texts against eligibility criteria. Disagreements were resolved by discussion or, if unresolved, through arbitration with a third reviewer selected from the authors.

Two reviewers (of NA, AGM, AGh, or PG) independently extracted data on design, sample, setting, baseline dose, intervention, outcome measures, and results from included studies. Two reviewers (of NA, AM, AGh, PG, CAJ, FS, or FB) appraised the risk of bias of the results of all studies that met inclusion criteria. We used the Cochrane risk-of-bias tools for randomised controlled trials (RoB 2) and for non-randomised studies of interventions (ROBINS-I).2829 For randomised controlled trials with crossover or cluster designs, we used the additional RoB 2 questions.3031 Agreement by consensus was reached on the risk of bias for the results of each study. Disagreements were resolved by discussion with a third reviewer (NA or FS). We applied the GRADE (grading of recommendations assessment, development and evaluation) framework to assess the certainty of evidence on outcomes, using the terms "we are uncertain" to refer to very low certainty, "may" for low certainty, "probably" or "likely" for moderate certainty, and "very likely" or simply the absence of qualification when referring to high level certainty on the effect of an intervention on a particular outcome (eg, improves, reduces).3233

To reduce clinical heterogeneity, we grouped interventions into five categories based on their clinical approach to the problem of long term opioid treatment: pain self-management3435; complementary and alternative medicine36; pharmacological and biomedical devices and interventions3738; opioid replacement treatment394041; and deprescription methods42 (table 1). Where interventions included elements from multiple categories, we assigned the study to the category deemed most applicable to its principal mechanism.

Interventions to reduce long term opioid treatment in people with chronic non-cancer pain

For the non-randomised controlled trials and uncontrolled studies appraised with ROBINS-I, those judged at critical risk of bias for all their included outcomes were deemed too problematic to be included in the evidence synthesis.29 Study results at low, moderate, or serious risk of bias using ROBINS-I and the results of all randomised controlled trials were included in the syntheses.

Controlled studies (that is, randomised and non-randomised controlled trials) and uncontrolled studies are presented separately in the results below. Controlled studies were synthesised in GRADE summary of findings tables for each intervention group. Randomised controlled trials with sufficient clinical homogeneity were synthesised in meta-analyses, and the remaining randomised controlled trials plus non-randomised controlled trials were synthesised narratively. Uncontrolled studies were synthesised narratively.

Within each category, random effects meta-analysis using Review Manager 5.4 software (Cochrane) was undertaken on groups of randomised controlled trials where interventions, comparator groups, and measures were deemed to be sufficiently similar to enable meaningful meta-analysis.434445 At least two studies were required for each meta-analysis.

When studies included more than two intervention arms, we excluded irrelevant groups or combined relevant groups as recommended in the Cochrane Handbook in order to avoid arbitrary decisions.45 For example, a study comparing inpatient and outpatient versions of the same pain management programme included an additional control group on a waiting list.46 Multiple studies included in the synthesis compare outpatient pain management programmes with a control group without pain management, and as such the inpatient group of this study46 was considered irrelevant in two meta-analyses. In another study, electro-acupuncture was compared with sham electro-acupuncture and no electro-acupuncture.47 Using the formula provided for combining two treatment arms (Cochrane Handbook 6.5.2.10, 23.3.4), we calculated the combined mean difference and standard deviation for sham electro-acupuncture and no electro-acupuncture as a single no acupuncture treatment group.

Effect estimates used included risk ratios for the proportion of people discontinuing opioids, substance use, and adverse events; mean differences for opioid dose; and standardised mean differences (Hedges’ adjusted g) for pain, function, quality of life, and withdrawal symptoms. Standardised mean differences were calculated for the above patient outcomes because we anticipated that multiple different scales would be used to measure the same outcomes. Small, moderate, and large differences between groups were indicated by standardised mean differences of 0.20, 0.50, and 0.80, respectively.48

Study authors were contacted for data if none could be found in the publication and for clarification. If studies included the outcomes of patients who did not receive long term opioid treatment alongside those of patients who did receive long term treatment, we isolated the outcomes of the second group. To obtain these data, we extracted published data and, in one case, used original data supplied by the authors.46 In the case where the original data were supplied, mean differences and standard deviations for each treatment group were calculated in SPSS (IBM). Long term opioid treatment was considered distinct from the opioid treatment used when needed.

Confidence intervals were converted to standard deviations using Review Manager 5.4. When measures of variation were missing for mean differences within each treatment arm of a given study and a test of difference between treatment arms was reported, we converted F statistics, t statistics, and P values to standard errors and standard deviations using the tool in Review Manager 5.4 (Cochrane Handbook 6.5.2.3). Here, t was taken as the square root of F, and it was assumed that the standard deviations of the mean differences in each treatment arm were equal (6.5.2.3). When measures of variation were missing for mean differences within each treatment arm of a given study and the study had no test of difference between groups, we used the highest standard deviation recorded in the same meta-analysis for each treatment arm instead of the study's own data (6.5.2.7).

We assessed heterogeneity by using τ2 and I2 statistics. We also conducted post hoc sensitivity analyses when measures of variance were imputed and when a study reported a clear outlier effect. In each case, the relevant trials were excluded and the meta-analysis was repeated.

Despite no direct patient or public involvement in the development or completion of this review owing to time and funding constraints, the research question was formed as a result of conversations about opioid tapering between clinician authors and patients attending the pain clinic at Royal North Shore Hospital, Sydney, Australia. We have asked a member of the public to read our manuscript after submission in order to solicit feedback on the best way to communicate our findings to the community.

Our search identified 11 420 records from five databases. Another 168 records were identified in systematic reviews, reference lists, and through expert contacts. After removing duplicates, 9999 unique records were screened and the full text of 490 records were reviewed. In total, 166 studies met inclusion criteria and were appraised for risk of bias, including 27 randomised controlled trials,464749505152535455565758596061626364656667686970717273212213 13 non-randomised controlled trials74757677787980818283848586 and 126 uncontrolled studies.87888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211 Data were not synthesised for eight non-randomised controlled trials and 122 uncontrolled studies because of critical risk of bias. The remaining 36 studies contributing to the evidence synthesis included 27 randomised controlled trials,464749505152535455565758596061626364656667686970717273212213 five non-randomised controlled trials,7982838486 and four uncontrolled studies.126146182203 Meta-analysis was conducted on 11 randomised controlled trials4647535456575965687273 (fig 1 and appendices 3-5).

Literature flowchart. Key questions refer to (1) how effective interventions are to reduce or discontinue long term opioid treatment, and (2) what their effects are on patient outcomes

The 36 studies contributing to the evidence synthesis were conducted in the following settings: outpatient (n=17), primary care (n=6), inpatient (n=5), unclear (n=5), outpatient and community (n=2), and community only (n=1). The baseline opioid dose for participants on long term opioid treatment was reported in 24 studies, with participants taking a mean daily dose of ≤100 mg oral morphine equivalents in 14 studies (58%) and >100 mg oral morphine equivalents in 10 studies (42%). Twenty six studies reported a programme goal of opioid dose reduction or discontinuation for all or some patients. Studies often reported that patients had chronic non-cancer pain without providing more detail. Otherwise, studies reported patients with diverse chronic pain syndromes including various back, spine, and neck disorders; musculoskeletal pain; sacroiliac joint pain; osteoarthritis; headaches; neuropathy; and fibromyalgia. In 12 studies, some or all patients had chronic pain and comorbid prescription opioid use disorder, opioid dependence, or previous substance use.495560616467707179146182203

The 27 randomised controlled trials and five non-randomised controlled trials contributing to the evidence synthesis are described in table 2. Two articles212213 were secondary analyses of a randomised controlled trial,71 and thus the three articles were considered together as one study. Most randomised controlled trials had a high risk of bias overall, with only two studies appraised as have a low risk of bias overall (fig 2 and appendix 4). Meta-analysis was possible in three categories (pain self-management, complementary and alternative medicine interventions, and pharmacological and biomedical devices and interventions) where multiple studies were found with comparable treatment arms (fig 3, fig 4, and appendix 6). In the pain self-management group, meta-analysis was possible for opioid discontinuation, opioid dose, pain intensity, and function from six studies.465456656872 In the complementary and alternative medicine group, meta-analysis was possible for three studies on acupuncture for opioid dose and pain intensity.475773 In the pharmacological and biomedical group, meta-analysis was possible for two studies of spinal cord stimulation on opioid discontinuation.5359 The remaining outcomes and studies in the above three groups and all studies in the opioid replacement treatment and deprescription groups were too clinically heterogenous to allow meaningful meta-analysis and so were described narratively. GRADE analysis of the 32 controlled studies showed that none of the outcomes had a high level certainty, with three outcomes of moderate certainty and the remainder of low and very low certainty (table 3 and appendix 7).

Characteristics of randomised and non-randomised controlled trials investigating interventions to taper long term opioid treatment for chronic non-cancer pain

Risk-of-bias summary for randomised controlled trials included in the evidence synthesis, using Cochrane risk-of-bias tool RoB 2

Meta-analyses of randomised controlled trials investigating interventions to taper long term opioid treatment for chronic non-cancer pain, according to opioid discontinuation and opioid dose. IV=inverse variance; M-H=Mantel Haenszel test; SD=standard deviation; df=degrees of freedom; OME= oral morphine equivalent

Meta-analyses of randomised controlled trials investigating interventions to taper long term opioid treatment for non-cancer chronic pain, according to pain intensity and function. IV=inverse variance; SD=standard deviation; df=degrees of freedom

Certainty of evidence and summary effect estimates assessed by GRADE (grading of recommendations, assessment, development, and evaluation) of controlled trials investigating interventions to taper long term opioid treatment for non-cancer chronic pain

Meta-analyses were possible for six studies in which patients in the intervention arm were provided with non-pharmacological techniques to manage their pain.465456656872 Techniques were based on the principles of cognitive behavioural therapy (CBT) and mindfulness, and primarily took place within an outpatient multidisciplinary pain programme. In one study, the programme was delivered online.65 Control groups included patients on waiting lists receiving usual care, which typically involved opioid treatment management with patients’ regular clinicians and limited restrictions for other treatment.46656872 In two instances, the control groups participated in a support group and discussed their experiences of pain and opioids (without receiving pain self-management training).5456 Each study evaluated pain self-management versus no pain self-management.

Pain self-management programmes compared to no pain self-management probably moderately reduced opioid dose (mean difference –14.31 mg oral morphine equivalent, 95% confidence interval –21.57 to –7.05, τ2=0.00, I2=0%, moderate level certainty), based on five studies of 428 participants. Pain self-management might have had a moderate effect on pain intensity (standardised mean difference –0.59, 95% confidence interval –1.02 to –0.16, τ2=0.00, I2=0%, low level certainty) and might have had no effect on function (–0.27, –0.69 to 0.15, τ2=0.00, I2=0%, low level certainty), based on three studies of 92 participants. We were uncertain with the estimate of participants being twice as likely to discontinue opioids than controls (risk ratio 2.15, 95% confidence interval 1.02 to 4.53, τ2=0.00, I2=0%, very low level certainty), based on two studies of 238 participants.

Meta-analyses were possible for three studies involving a total of 158 participants that evaluated the efficacy of either electro-acupuncture or general acupuncture compared to no acupuncture in the context of opioid tapering in an outpatient pain clinic.475773 In one study, two control arms were combined into one group of no acupuncture (appendix 6). We were uncertain in the estimate that, in the context of clinician guided opioid reduction, additional acupuncture had little or no effect on opioid dose compared to no additional acupuncture (mean difference –1.56 mg oral morphine equivalent per day, 95% confidence interval –19.03 to 15.92, τ2=155.05, I2=69%, very low level certainty), and had no effect on pain intensity (standardised mean difference 0.02, –0.29 to 0.34, τ2=0.00, I2=0%, very low level certainty).

Meta-analyses were possible for two studies involving 97 participants that evaluated the efficacy of spinal cord stimulation compared to conventional medical treatment.5359 We were uncertain in the estimate that those individuals who received spinal cord stimulation were six times more likely to discontinue opioids than those who received conventional medical treatment (risk ratio 6.07, 95% confidence interval 1.16 to 31.77, τ2=0.00, I2=0%, very low level certainty).

Nine randomised controlled trials and one non-randomised controlled trial evaluated inpatient and outpatient programmes comprising CBT or mindfulness combined with exercise, education, and management of drug treatment.46545662636568697283 Certainty of evidence for all outcomes not included in meta-analyses was low or very low.

In two studies, patients who undertook three and four week pain management programmes incorporating opioid discontinuation achieved greater improvements in pain, function, and opioid use than those who undertook physical therapy in one study (all P<0.001)69 and wait listed controls in the other study.46 Less intensive programmes were also successful. Patients who undertook eight sessions of group mindfulness achieved greater reductions in opioid dose at three and four month follow-ups than patients who undertook eight sessions of support group therapy (P=0.006, P=0.02).5456 However, six months of peer delivered pain management training did not achieve significant differences in opioid dose, pain, and quality of life compared with one session of training.62

Three studies evaluated pain self-management programmes delivered online or by telephone.636583 A post hoc analysis65 of a randomised controlled trial215 reported that eight weeks of online pain self-management increased rates of opioid discontinuation (33% v 14%, P=0.27) and led to greater dose reduction (46% v –3%, P=0.003) than wait listed controls. No difference in opioid discontinuation rates was observed in a comparison of in-clinic versus web delivered programmes of pain self-management .83 Both groups achieved clinically significant improvements, while the in-clinic group had significantly greater improvement in pain and disability.83 Another study found that patients who received CBT reminders by an automated telephone service had better opioid discontinuation, dose, pain, and function outcomes than control participants.63

We had low or very low certainty in the evidence on the outcomes of the four controlled studies that did not contribute to the meta-analyses in this group. Of the three acupuncture studies, one reported no effect on function,47 another reported no difference between groups for withdrawal symptoms,57 and two reported no effect on quality of life and serious adverse events, with few mild adverse events due to electro-acupuncture.4773 A non-randomised controlled trial evaluating a herbal medicine, Kampo, found significant differences favouring the treatment group over controls who did not receive Kampo on opioid discontinuation and dose.82 Substance use was not measured.

Two main biomedical approaches were evaluated in controlled trials: the pharmacological management of pain and withdrawal symptoms and invasive procedures (eg, surgery, device implantation). For all outcomes not included in the meta-analysis, certainty in the evidence was low. Substance use was not measured.

Regarding pharmacological interventions, no significant difference on withdrawal symptoms was reported between groups treated with varenicline (primarily used to treat nicotine withdrawal) and placebo in the context of an outpatient pain programme.55 No significant difference in pain scores and opioid use at six months was observed between patients receiving intravenous ketamine infusions and those not receiving ketamine, and two adverse events occurred (supraventricular arrhythmia and anxiety).79 In a placebo controlled study of ibudilast for patients with headache from the overuse of drug treatment, researchers found that opioid use and quality of life did not differ between groups throughout follow-up, whereas patients in the intervention group reported significantly more adverse events, including nausea, pruritis, and diarrhoea (P=0.02).58

Regarding invasive procedures, studies of sacroiliac joint arthrodesis52 and spinal cord stimulation5359 versus conventional treatment reported better outcomes in opioid dose,52 pain,525359 disability,52 and quality of life525359 in the intervention group. Patients with knee osteoarthritis who received transcutaneous electrical nerve stimulation reported better opioid discontinuation rates, pain, function, and quality of life than controls at follow-up.50 Adverse events were common in patients receiving sacroiliac joint arthrodesis52 and spinal cord stimulation (electrode migration, wound infection),5359 and were not observed in patients receiving electrical nerve stimulation.50 Lastly, a double blinded trial of patients on intrathecal morphine comparing stable dose with 20% weekly dose reduction found that 70% (7/10) of the tapering group dropped out due to worsening pain.66

Five randomised controlled trials compared various protocols of opioid maintenance treatment using buprenorphine/naloxone and methadone. Certainty in the evidence was either low or very low for all outcomes, with quality of life not measured. Each study showed no significant difference between treatment groups reported for opioid dose,64 function,64 and withdrawal symptoms.6467 Withdrawal symptoms including headache, nausea, and diarrhoea, were reported in both treatments of a crossover trial comparing the tolerability of full μ opioid agonist dose at 50% of participants’ usual doses versus a similar dose of buccal buprenorphine.70 This study reported no differences between treatments on pain scores,70 however, another study reported greater analgesic effects for participants receiving higher doses of buprenorphine/naloxone.67 Two studies reported use of heroin, benzodiazepine, and alcohol.4964 Two of 43 participants trialling various buprenorphine/naloxone doses dropped out owing to nausea and heavy sedation,67 and a comparison of buprenorphine/naloxone tapering (n=6) with maintenance (n=6) failed to retain any participants in the tapering arm.49 Counselling did not affect the completion rate of those individuals tapering use of buprenorphine/naloxone.71212213

We included one randomised controlled trial of a prescriber focused deprescription intervention and three trials of patient focused deprescription interventions. Certainty in the trial outcomes was moderate for the prescriber focused intervention, and low or very low for the patient focused interventions.

The randomised controlled trial of a prescriber focused intervention evaluated changes to primary care practices in relation to opioid prescribing, including nurse care management, one-to-one academic detailing, and an electronic decision tool for clinicians.61 Based on this study (n=964), prescriber focused deprescription interventions probably had a small effect on opioid dose (mean difference –6.8 mg (standard error 1.6) oral morphine equivalent, P<0.001; adjusted odds ratio for dose reduction 1.6, 95% confidence interval 1.1 to 2.4; moderate level certainty) and probably had no effect to small effects on opioid discontinuation (adjusted odds ratio 1.5, 95% confidence interval 1.0 to 2.1; moderate level certainty). This study did not measure non-opioid patient outcomes.

Of the patient focused trials, one compared patient controlled with clinician controlled opioid reduction during a four week pain management programme.84 Researchers found no differences between groups on opioid discontinuation, dose, pain, function, quality of life, and benzodiazepine use at six months. The second trial found that periods of abrupt opioid cessation caused significant increases in pain, disability, and withdrawal symptoms (including diarrhoea, anxiety, muscle twitching, and rhinorrhoea) when compared to opioid maintenance.51 The third trial compared six months of a weekly dose reduction by 10% of baseline dose with six months of a stable dose, and results were inconclusive owing to high a dropout rate.60

Of 126 uncontrolled studies meeting inclusion criteria, four were at serious risk of bias and thus were eligible for narrative synthesis (fig 5 and appendix 4). Primary care prescriber education and an opioid tapering referral programme was associated with a mean dose reduction of 111 mg oral morphine equivalent (P value not reported) over 12 months.146 Prescriber education and a dose reduction policy was associated with a mean dose reduction of 64 mg oral morphine equivalent (95% confidence interval 32 to 96, P<0.001) over 16 months.203 For patients with a new diagnosis of depression, adherence to antidepressants increased the odds of opioid cessation (hazard ratio 1.24, P=0.007).182 Lastly, patients experienced a 16% reduction in opioid dose after total knee arthroplasty (adjusted incidence rate ratio 0.84, 95% confidence interval 0.78 to 0.90, P<0.001) and 7.2% (52/720) discontinued long term opioid treatment.126

Risk-of-bias summary for non-randomised studies investigating interventions to taper long term opioid treatment for non-cancer chronic pain, using Cochrane risk-of-bias tool ROBINS-I

The remaining 122 uncontrolled studies were at critical risk of bias, including studies from all five intervention categories. Risk of bias varied between low, moderate, serious, and critical across all ROBINS-I domains except for confounding where the risk of bias was uniformly critical.

This review evaluated 166 studies on the efficacy of interventions to reduce or discontinue long term opioid treatment in people with chronic non-cancer pain and the effect of these interventions on patient outcomes: pain, function, quality of life, withdrawal symptoms, substance use, and adverse events. Only 36 studies were at lower than critical risk of bias, including 27 randomised controlled trials, five non-randomised controlled trials, and four uncontrolled studies. Meta-analyses were conducted on 11 randomised controlled trials, and the remaining 25 randomised controlled trials, non-randomised controlled trials, and uncontrolled studies were narratively synthesised. For the 32 controlled studies contributing to the GRADE summary of findings tables, certainty in the evidence was low or very low for all outcomes aside from three which were moderate.

Many uncontrolled before-and-after studies met inclusion criteria but were at critical risk of bias because they were unable to deal with confounding. Studies that accounted for underlying trends in opioid use before the intervention and observed changes during and after the intervention were deemed at serious risk of bias.126146203 Their results were synthesised, but because these studies did not perform time series comparisons of projected versus actual observations, they could not be considered equivalent to randomised controlled trials. One study with serious risk of bias that did not assign patients to intervention and control groups was considered to approximate a non-randomised controlled trial.182

We had moderate certainty in the efficacy of interventions to support prescriber adherence to opioid reduction guidelines and pain self-management programmes in the reduction or discontinuation of opioids. The 6.8 mg oral morphine equivalent per day difference found in a multi-component intervention encouraging prescribers to follow guidelines recommending against high dose opioids was probably a true effect, and the higher opioid discontinuation rate associated with this intervention was also probably a true effect.61 Pain self-management programmes probably achieved a moderate reduction in dose (14.31 mg oral morphine equivalent per day) when compared to no training in pain self-management, but we were uncertain in the evidence that participants are twice as likely to discontinue opioids. According with previous evidence,2122 the pooled results for studies of acupuncture found no effect on opioid dose when compared to no acupuncture, and the pooled results for studies of spinal cord stimulation found that intervention patients were six times more likely to discontinue opioids than patients receiving conventional care. However, owing to blinding issues, small samples, statistical heterogeneity, and confidence intervals of effect estimates spanning contradictory clinical decisions, we were uncertain in the evidence for both outcomes.

The certainty of evidence for the effect of interventions on patient outcomes was uniformly low or very low. The pooled data on the effect of pain self-management programmes on pain and function accorded with previous evidence that pain self-management might be an effective alternative to opioids;19216217 however, certainty in our evidence was low. Nearly all randomised controlled trials were at high risk of bias in their reporting of patient outcomes, owing to the use of self-report measures. Thus, we were not able to evaluate the effect of interventions on pain, function, quality of life, and withdrawal symptoms. In a subset of studies, researchers blinded participants to sham, placebo, or other interventions,515558677073 but small samples precluded us from assessing their findings with anything other than low or very low certainty.

The lack of evidence on adverse events and substance use was a concern, and capturing their occurrence is outside the scope of most trials, which are usually brief. However, adverse events were reported by patients receiving electro-acupuncture,4773 ketamine,79 ibudilast,58 sacroiliac joint arthrodesis,52 and spinal cord stimulation.5359 Two studies of opioid replacement treatment reported instances of alcohol, heroin, and benzodiazepine use,4964 although these findings did not contribute to evidence on outcomes at anything better than low certainty. One observational study reported the hazard ratio for death of patients who discontinued long term opioid treatment in primary care, but it was excluded from the evidence synthesis owing to critical risk of bias.142

Several studies that were ineligible for this review have considered the association of adverse events with opioid discontinuation. Their findings were mixed. One study found that opioid discontinuation was associated with fewer overdoses and injuries than maintenance.218 Another study found that a significant increase in the risk for overdose death in patients on opioids for longer periods before discontinuing,219 indicating that tapering should not be delayed. Yet evidence also suggests that changes in opioid treatment might be dangerous. High rates of suicidal ideation and self-harm have been found in US veterans with and without substance use disorder who discontinued opioids.220 Moreover, the initiation of opioids, tapering, and the three months after discontinuation have been associated with higher risk of overdose, suicide, mental health crises, and heroin use.219221222223

A lack of good quality evidence remains a barrier to more conclusive findings. While some randomised controlled trials were at low risk of bias, imprecision because of small samples was the main reason for reducing our certainty in the outcomes. Meta-analyses included the few studies where pooling results was clinically meaningful but were limited by the exclusion of studies where event rates were zero in each group, by missing data, and by the necessity to impute missing measures of variance. Effect sizes might have been overestimated owing to the small-study effect.224225 Funnel plots and meta-regression were not performed because of the small number of studies contributing to the meta-analyses. Lastly, while the outcomes synthesised in the present study reflect standard practice in the literature, their applicability and meaningfulness to patients’ lived experiences needs further consideration. For any protocol deviations, see appendix 8.

The clinical implications of this review are modest. No intervention stands out for recommendation. Nevertheless, owing to the risks associated with long term opioid treatment, clinicians should discuss with patients the prospect of opioid tapering when it is safe to do so. Close follow-up is important given the possibility that severe adverse events are associated with changes in treatment.226 Comorbidities such as depression and substance use disorder need specific attention, and the risks involved with forced tapers (that is, when patients are not involved in decision making) are substantial and must not be ignored.182 For patients at increased risks when tapering because of complex persistent opioid dependence, transition to buprenorphine should be considered, although the evidence for this approach, like others, remains limited.23227 Multidisciplinary pain management programmes are probably effective at helping to reduce opioid dose. However, access remains an issue for people who are unable to take time off work, in regional areas where services are limited, and for people from culturally and linguistically diverse communities.

Researchers should design studies with replicability in mind,225 deal with the problem of dropout rates,132211228 place a stronger emphasis on outcomes relevant to both clinical practice and patients’ lived experiences,229230 and use longer follow-up periods to reflect the scale of patients’ experiences in treatment.229

Systematic reviewers and guideline authors must consider the clinical heterogeneity of interventions in this field, as well as variation in risk of bias. Future reviews should consider the limited value of including uncontrolled studies, and perhaps exclude in the first pass those clearly at critical risk of bias. Refined research questions and eligibility criteria are required to isolate uncontrolled studies at lower than critical risk of bias.

The evidence to guide patients and clinicians on the efficacy of interventions to reduce or discontinue long term opioid treatment in patients with chronic non-cancer pain continues to be constrained by poor study methodology. Of particular concern is the lack of evidence regarding possible harms associated with these interventions and the reduction of opioids. Agreed standards for designing and reporting studies on the reduction or discontinuation of opioids are urgently needed.

Opioid tapering is the gradual reduction of opioid treatments with the goal of either reducing or discontinuing opioid use while limiting possible adverse effects, including withdrawal symptoms and increased pain

Guidelines recommend that people on long term opioid treatment for chronic non-cancer pain should consider opioid tapering when it is safe to do so and when the risks of opioid treatment outweigh the benefits

Reviews to date are inconclusive as to the most effective approach for tapering opioid treatment and the effect of such interventions on patient outcomes (eg, pain, function, and quality of life)

This review indicates that interventions supporting prescribers’ adherence to opioid guidelines and participation in pain self-management programmes are probably effective in reducing opioid use by small and moderate amounts, respectively

Psychosocial support should be provided to patients tapering opioid use owing to the lack of evidence regarding the effect of opioid tapering interventions on adverse outcomes

Studies at critical risk of bias dominate this topic; agreed standards for designing and reporting studies on the reduction or discontinuation of opioids are urgently needed

Not applicable.

See appendices 1-8 for all data available.

We thank the authors of the meta-analysis studiesfor providing original data and clarification of findings, and Daniel Costa and Michael Nicholas for their guidance on methodological and clinical questions.

Contributors: NA, PG, and CEA-J conceived the study idea and designed the search strategy. NA, AGM, AG, and PG screened studies for eligibility and extracted data. NA, AGM, AG, CEA-J, FS, FMB, and PG double appraised the results of studies for risk of bias. NA, PG, AGM, FS, and FMB undertook the data synthesis and analysis. NA wrote the first drafts of the figures, tables, and appendices. NA wrote the first draft of the manuscript with PG. PG, AGM, CEA-J, FS, FMB, RM, and AG interpreted the data analysis and critically revised the manuscript. NA and PG are guarantors. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Funding: This study was funded by philanthropic gifts from the Salteri Family Foundation, Perpetual Foundation, Pain Foundation, and Ernest Heine Family Foundation. None of these parties had any role in the design or conduct of the study, in the collection, management, analysis or interpretation of the data, or in the preparation, review or approval of the manuscript.

Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/disclosure-of-interest/ and declare: support from the Salteri Family Foundation, Perpetual Foundation, Pain Foundation, and Ernest Heine Family Foundation for the submitted work; PG has received grants from the Australian National Health and Medical Research Council and the Medical Research Futures Fund (Australian government), is a board member and director of Pain Australia and Pain Foundation, and at the time of undertaking the review was a medical adviser to Cymra Life Sciences, an Australian medical cannabis company; FB has received grants from the Sydney Medical Foundation (University of Sydney) and the Medical Research Futures Fund and is the deputy chair of the board of the ANZAC Research Institute (Sydney Local Health District); no other relationships or activities that could appear to have influenced the submitted work.

The lead author affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Dissemination to participants and related patient and public communities: We plan to disseminate the findings of this review to relevant patient and clinician groups at conferences and other meetings.

Provenance and peer review: Not commissioned; externally peer reviewed.

This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.