WHAT IS ALREADY KNOWN ON THIS TOPIC

Opioid medications are extensively used in clinical practice for managing postoperative surgical pain.

The effectiveness of interventions in reducing opioid use after surgery is a subject of international interest.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

Clinicians and policymakers can use the identified fundamental behaviour change techniques, such as providing behaviour instructions, goal setting and social support to design more effective postoperative opioid reduction interventions.

The findings emphasise the importance of tailored interventions for different contexts and stakeholders.

Background

Opioids remain the mainstay of postoperative pain management.1 Although these drugs are effective in this acute period, they present a dual challenge with their effectiveness and well-documented adverse effects.2 3 Opioid-related adverse effects range from less severe gastrointestinal side effects, such as nausea and constipation, to more serious adverse effects, such as opioid-induced ventilatory impairment, opioid dependence, addiction and, in extreme cases, opioid-related deaths.2 3 Current best practice promotes the use of multimodal analgesia to avoid over-reliance on opioid medication.1 4 Multimodal analgesia involves using more than one pharmacological class of analgesic medication targeting different receptors along the pain pathway, aiming to provide adequate pain relief while minimising the side effects.3 5

Despite the use of multimodal analgesia and opioid-sparing regimes during the peri-operative period, it has not been possible to reduce the prescribing of opioids postoperatively. Several studies conducted in the USA have shown that opioids are frequently overused, particularly after discharge.6–8 With more complex surgical procedures undertaken as a day case or short stay resulting in a shorter postoperative hospital stay, patients are no longer being fully weaned off their analgesics by the time of hospital discharge.9–11 Therefore, surgery is now considered a risk factor for subsequent opioid dependence.10 11 Furthermore, discharging patients home with opioids poses a risk that there may be unused opioids at home and hence the risk of illicit diversion, misuse and overdose.12 Studies have demonstrated that a substantial proportion of patients may have leftover opioid prescriptions after standard surgical procedures.6

However, a notable gap exists in the literature concerning guidance on optimal types and durations of discharge analgesia medications and effective deprescribing strategies.13 Deprescribing involves tapering or discontinuing drugs to improve outcomes, including advising patients to fill prescriptions only if necessary, thus reducing medication burden and polypharmacy.14 15 To our knowledge, two published systematic reviews have examined effective deprescribing interventions post surgery. Zhang et al reviewed behavioural interventions to decrease opioid prescribing after surgery.16 They found that behavioural interventions, such as cognitive behavioural therapy, mindfulness-based stress reduction and motivational interviewing, were effective in reducing opioid use. However, the review did not synthesise effectiveness or provide a detailed analysis of the behavioural change techniques. Similarly, Wetzel et al also evaluated interventions targeting opioid prescribing after surgery and found diverse interventions targeting patients, healthcare providers and healthcare systems, but did not synthesise effectiveness.17

This systematic review addresses gaps in the current research by providing an updated synthesis and introducing the behaviour change technique taxonomy (BCTTv1) developed by Michie et al to identify effective elements in interventions reducing post-surgery opioid use.18 The BCTTv1 is a comprehensive framework to categorise and describe the different techniques used in behaviour change interventions.18 19 The taxonomy provides a standardised way of identifying the active ingredients of these interventions, making it easier to replicate and compare studies.20 Our review aimed to uncover these active ingredients and provide insights for future strategies, focusing on two key objectives: assessing the effectiveness of interventions aimed at reducing opioid prescribing post surgery and applying the BCTTv1 to identify techniques associated with the effectiveness of these interventions.

Methods

The systematic review protocol was developed following the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) guideline21 and registered with PROSPERO.

Data sources and search strategy

A comprehensive, structured search strategy was devised with a university librarian to include terms and controlled vocabulary related to ‘opioid’, ‘acute pain’, ‘surgery’, ‘tapering’ and ‘intervention’ (online supplemental appendix 1). Databases searched from inception to October 2023 included MEDLINE, Embase, CINAHL Plus, PsycINFO and Evidence-Based Medicine Reviews. Registry databases, such as ClinicalTrials.gov and the International Standard Randomised Controlled Trial Number (ISRCTN) registry, along with the first 10 pages of Google Scholar, were explored with no language restrictions. References from included studies were manually screened to identify additional citations, and citation tracking was undertaken to augment the search further. Articles not in English were screened using online translation tools such as Google Translate.

Eligibility criteria

The Patient/Population, Intervention, Comparator, Outcome (PICO) model22 guided the study question and identified eligible studies (table 1). This review included studies investigating the effectiveness of interventions to deprescribe opioids postoperatively in adult patients (≥18 years) undergoing any elective or emergency surgery in secondary care and subsequently prescribed at least one prescription of opioids. In this context, a single prescription of opioids denotes the provision of at least one dose of an opioid medication. Eligible interventions, whether patient-focused or clinician-led, were delivered at an individual level (psychological, biological, educational, informatics or social care interventions) in any clinical setting. Studies focusing on interventions that did not intend to deprescribe opioids, implemented at the population level and national/controlled drug policies/regulations were excluded. The primary outcome measure was the overall change in morphine milligram equivalent dose post-intervention. Inclusion criteria comprised fully published randomised and prospective/retrospective cohort studies, while case reports, opinions, case-control or cross-sectional studies and conference proceedings were excluded. Language restrictions were not applied to capture the broad field of literature. Additionally, we included studies where patients were on opioids before hospitalisation (table 1).

Table 1

Inclusion and exclusion criteria for selecting studies

Study selection

Two reviewers (NB and ST) independently screened the titles, abstracts and the full text of eligible studies against the inclusion and exclusion criteria to identify articles with disagreements resolved through discussion or arbitration by a third reviewer (L-CC).

Data extraction and risk of bias assessment

Study characteristics were extracted by the first author (NB) and checked by the second reviewer (ST). Outcome data were extracted independently by two reviewers (NB and ST) using the predesigned data extraction form, which obtained characteristics of intervention and comparison and outcome measures, that is, mean or median opioid dose measured in morphine milligram equivalent (MME) dose at baseline and end of intervention.

Two reviewers (NB and ST) independently assessed the risk of bias using the Cochrane risk-of-bias tool V.2 (RoB 2)23 for randomised controlled trials (RCTs) and the risk of bias in non-randomised studies of interventions tool (ROBINS-I) for non-randomised studies.24 Disagreements between reviewers were resolved by discussion to reach a consensus.

Behaviour change technique coding

Data extraction and coding of behaviour change techniques (BCTs) identified from the interventions of included studies, including intervention type, BCT code and label, target group (ie, patient or healthcare professional) and evidence of BCT presence, were conducted by two researchers (NB and REH) using BCTTv1 to define 93 distinct techniques.18 The inter-rater reliability (IRR) was calculated using Cohen’s kappa coefficient to determine consistency between coders using the BCTTv1.25 Discrepancies were resolved by discussion to reach a consensus, with disagreements resolved through discussion or arbitration by a third coder (CJA).

Data analysis

Descriptive statistics were used to summarise study characteristics and outcomes. Due to a wide variation in study aims and outcome measures, meta-analysis was not feasible and this study synthesised effectiveness as follows. The effect size and 95% CI of an intervention’s effectiveness were calculated based on the mean and the SD of MME dose for the intervention and control groups extracted from the included studies and further classified as small (d≤0.2), medium (d≥0.5 or <0.8) or large effect size (d≥0.8) according to Cohen’s formula.26 Any reported median MME dose was converted to the mean and the SD using the formula described by Wan et al 27 to facilitate comparison between studies.

Patient and public involvement

Patients with chronic pain taking opioids long-term after surgery (LJ and AC) were involved in reviewing the final manuscript, providing insights into implementation barriers in clinical practice in the UK and discussing strategies for disseminating findings within the patient community.

ResultsSelection of studies

A total of 6720 records were initially identified from five medical literature databases (MEDLINE, Embase, CINAHL Plus, PsycINFO and Evidence-Based Medicine Reviews), and an additional 174 records were identified from the ClinicalTrials.gov repository, the ISRCTN registry and Google Scholar. After removing duplicates, 3468 unique records underwent screening, leading to the full-text review of 55 articles. Ultimately, 22 articles met the inclusion criteria,28–49 comprising 7 RCTs28–33 49 and 15 non-randomised studies34–48 (figure 1).

Figure 1

Selection of studies. EBM, Evidence Based Medicine; ISRCTN, International Standard Randomised Controlled Trial Number.

Characteristics of studies

All 22 studies included in the evidence synthesis were conducted in the tertiary setting,28–49 involving both RCTs (n=7)28–33 49 and cohort studies (n=15)34–48 with a total of 37 068 patients across studies, ranging from 7628 to 23 29843 per study. The majority of studies were conducted in the USA (n=18),28 30–37 39–46 48 with the remaining studies in Australia (n=3)29 38 47 and Germany (n=1).49 All studies were published in the English language (table 2).

Table 2

Characteristics of included studies

The surgical procedures covered a range of specialties, including gynaecological,28 31 39 orthopaedic and trauma,30 32 33 35 38 43 44 46 49 general surgery,29 34 36 37 40 41 43 47 urology42 45 and other specialist surgical procedures.37 43 Most studies excluded patients with a history of opioid use, opioid tolerance, chronic pain syndromes, sensitivity to study medications, and those who developed postoperative complications (table 2).

Most interventions targeted healthcare professionals (n=11),29 34–37 40–42 44 46 48 with the rest targeting either patients (n=6)28 30–33 49 or both (n=5).37–39 43 45 Notably, none of the studies explicitly reported behaviour change theories underpinning their interventions, nor a systematic and theoretical approach, such as the UK Medical Research Council’s complex intervention framework,50 in their intervention designs. Due to high heterogeneity (I 2=99.93%) across studies, a meta-analysis was not possible. BCTs were identified from 10 intervention elements. All but one study47 reported the behaviour change components used.

Risk of bias

All studies had at least a medium risk of bias, with nine studies showing a high risk (online supplemental appendix 2, appendix 3). The high versus medium risk of bias ratio was 3:4 in RCTs, 4:3 in prospective cohort studies and 2:5 in retrospective studies.

High risk of bias in randomised studies was due to issues in the randomisation process31 32 and outcome measurement.29 31 In cohort studies, high risk was linked to confounding,38 43 48 participant selection,38 missing data43 and outcome measurement.38

Effect sizes of opioid reduction

Among the 22 included studies, the overall effect size ranged from 0.1542 to 12.1541 (figure 2, online supplemental appendix 4). In the seven RCTs, the effect size spanned from 0.1631 to 6.02,48 while in the observational studies, it varied from 0.1542 to 12.15.41 Interventions with a large effect size (Cohen’s d>0.8) were observed in 12 out of 22 studies. The large versus small-to-medium effect size ratio was 2:5 in RCTs,29 49 4:3 in prospective cohort studies34 42 47 48 and 6:2 in retrospective cohort studies.36 39–41 46

Figure 2

Effect sizes associated with the included studies. BCTs, behaviour change techniques; PCS, Prospective cohort study; P, Patient; HP, Healthcare professional; RCS, Randomised controlled study; RCT, randomised controlled trial. a: patient-directed education; c: clinician-directed education; d: guideline implementation; e: individual audit feedback; f: peer comparison performance feedback; h: pharmacist-led intervention.

No apparent association was observed between the effect size and the number of intervention elements or behaviour change techniques used, as only four studies implemented more than one intervention element (online supplemental appendix 5). Additionally, there was no difference in effect size reduction at the subspecialty level among the four studies that delivered interventions across multiple surgical subspecialties.36 40 41 43

Interventions with a large effect size

Cohen’s d indicated a large effect size in 12 studies, with ranges of d<2 (n=4),34 38 40 42 d=2 to 6 (n=6)29 36 46–49 and d>10 (n=2).39 41 Nine of these studies targeted healthcare professionals (mainly clinicians); one focused on patients,49 and two targeted both patients and healthcare professionals.38 39 The most substantial effect (12.15; 95% CI: 11.25 to 13.05) was observed in the study by Howard et al, which implemented prescribing guidelines for healthcare professionals in laparoscopic cholecystectomy patients (general surgery).41 This trend was also seen in other general surgical specialties, though with less prominent effect sizes.40

In addition to the prescribing guidelines, the other seven studies targeting healthcare professionals implemented various educational or pharmacist-led interventions. Bohan et al reported a large effect size (2.56; 95% CI: 2.08 to 3.04) with the delivery of a thirty-minute lecture to surgical residents, with the largest effect size in patients undergoing both open and laparoscopic inguinal hernia repairs (4.97; 95% CI: 4.119 to 5.733).36 Three studies in orthopaedic surgeries showed large effect sizes (2 to 3) with a group education session,28 pharmacist-led pain regimen guidance45 and procedure-specific opioid prescribing guidelines.37 Similarly, two general surgery studies reported large effect sizes with pharmacist assisting in preparing discharge prescriptions (2.29; 95% CI: 2.09 to 2.49)47 and interactive educational sessions (1.44; 95% CI: 1.31 to 1.57).33 Jacobs et al’s study in urology demonstrated large effect sizes with multifaceted approaches, including formal education, individual audit feedback and peer comparison performance feedback, especially in prostatectomy (5.59; 95% CI: 5.07 to 6.08) and nephrectomy procedures (7.53; 95% CI: 6.78 to 8.24).42

Regarding interventions targeting other recipients, Glaser et al reported a substantial effect size (10.01; 95% CI: 9.63 to 10.45) in gynaecological surgery, with preoperative educational sessions for patients and healthcare professionals and tiered guidelines with written discharge instructions.39 Bui et al reported an effect size of 0.82 (95% CI: 0.53 to 1.11) for pharmacist-led opioid de-escalation targeted at patients.38 Stuhlreyer et al observed a significant effect of 6.02 (95% CI: 4.7 to 7.35) in orthopaedic surgery with a digital application combined with physician rounds for patients.49

Interventions with a small to medium effect size

Interventions in 10 studies with a small or medium effect size (Cohen’s d ranged from 0.16 to 0.51) mainly targeted patients (n=5),28 30–33 followed by three studies targeting patients and healthcare professionals37 43 45 and two studies targeting healthcare professionals only.35 44

Additionally, all five studies targeting patients with a small effect size were RCTs, with three having a negative lower limit of 95% CI, indicating no significant effect in reducing opioid use. These three RCTs included a single handout for breast surgery patients (0.22; 95% CI: −0.23 to 0.68),27 an educational handout for gynaecological patients (0.29; 95% CI: −0.03 to 0.59),30 and advice at discharge to orthopaedic surgery patients to fill opioids only if necessary (0.16; 95% CI: −0.24 to 0.56).32 The other two RCTs involved preoperative education: a 2 min narrated video, a handout on narcotic overuse risks (0.36; 95% CI: 0.02 to 0.7)30 and an 11-minute video on opioid risks (0.48; 95% CI: 0.11 to 0.84)33 for orthopaedic surgery patients, both showing small effect sizes.

The two studies targeting healthcare professionals demonstrated consistent findings: a small effect size (0.25; 95% CI: 0.08 to 0.42)35 with mandatory education lectures and a medium effect size (0.51; 95% CI: 0.43 to 0.59)44 with guideline implementation for orthopaedic doctors. The three studies targeting both patients and healthcare professionals also demonstrated small effect sizes. These interventions included promoting opioid guidelines via posters, brochures and educational seminars (0.15; 95% CI: 0.05 to 0.25),43 a 20 min educational lecture (0.35; 95% CI: 0.15 to 0.55),37 and providing an information sheet, education by nurses at discharge and prescribing guidelines (0.21; 95% CI: 0.28 to 0.4).45

Studies reporting other patient-related outcomes

14 of the 22 studies28 30 31 33 36–39 41–44 47 48 reported patient outcomes of the intervention. Six studies30 33 38 41 42 44 reported pain intensity following the intervention as an outcome and found either no difference33 38 41 42 44 or lower pain scores in the intervention group.30 One study39 evaluated patient satisfaction with pain relief following the intervention and refill rates and found no difference. Additionally, six studies28 31 36 37 43 48 measured refill rates only and found either no difference31 37 43 48 or a reduction in refill rates following the intervention.28 36 One study47 measured the proportion of patients admitted to the hospital within 10 days of discharge due to insufficient analgesia and found no admission in the intervention group.

Behaviour change techniques

In total, 23 unique BCTs were identified across the studies and coded 140 times (online supplemental appendix 6). Definitions of these BCTs are listed in online supplemental appendix 7. There was good agreement between researchers involved in the coding process, with IRR ranging from 0.49 to 1.00 per study. The number of unique BCTs in the intervention group ranged from 1 to 10 (median: 3 BCTs per study). The most frequently employed BCTs were instructions on performing the behaviour (n=35),29–32 34–41 43 44 46 48 49 behaviour substitution (n=22),28 33 36 37 42 43 45 information about health consequences (n=20)28–31 33 35 37 38 42 43 45 and pharmacological support (n=14)31 33 37 45 49 51 (figure 3).

Figure 3

Frequency of behavioural change techniques identified from the included studies. The Behavioural Change Techniques Taxonomy (v1) groups 93 hierarchically clustered techniques into 16 groups; this study did not identify any behavioural change techniques (BCTs) in groups 14 (Scheduled consequences, n=10), 15 (Self-belief, n=4) and 16 (Covert learning, n=3). * The number of BCTs in each group defined by the BCTTv1 was listed in the denominator.

The BCTs identified from the interventions of the 12 studies’ with a large effect size included instructions on how to perform the behaviour, behaviour substitution, goal setting (outcome), social support (practical), social support (unspecified), pharmacological support, prompts/cues, feedback on behaviour, adding objects to the environment, graded tasks, review outcome goal(s), information about health consequences, action planning, social comparison, credible source, feedback on outcome(s) of behaviour and social reward.29 35 36 38–42 46–49

Interventions with a large effect size primarily targeted healthcare professionals, except for the study by Stuhlreyer et al, which focused on patients.49 The healthcare professionals included specialty/resident doctors,29 35 36 38–42 48 with a few studies targeting other professional groups such as pharmacists.29 46 47 These interventions typically involved educational initiatives or guideline implementation. In contrast, interventions aimed at pharmacists involved activities such as assisting in writing prescriptions or providing pharmacist-led postoperative multimodal pain regimen guidance and order-set standardisation.

Discussion

This systematic review critically assessed 22 studies that implemented diverse interventions across various surgical procedures, all aimed at mitigating opioid use post surgery. We planned to include all settings where the intervention took place, with a focus on patients who had been prescribed opioids following a surgical procedure in secondary care. However, on conducting the review, we found that all interventions in the included studies were conducted within the secondary care setting. Given substantial heterogeneity, meta-analysis was not feasible. Instead, Cohen’s d was calculated to categorise the effect size into large, medium and small groups.26 Effect sizes ranged widely (0.15 to 12.15), generally smaller with narrower 95% CI in RCTs versus prospective and retrospective cohort studies, likely due to differences in sample sizes, intervention recipients, intervention elements (including BCT types) and study biases, with all studies exhibiting medium to high risk of bias.

No direct correlation was observed between the number of intervention elements, the number of behaviour change techniques employed in individual studies, and the effect size. Except for the study by Stuhlreyer et al, which targeted patients alone and demonstrated a large effect size, with a wide range of 95% CI,49 other studies focusing solely on patient interventions28 30–33 demonstrated a smaller effect size compared with those involving both patients and healthcare professionals37–39 43 45 or healthcare professionals alone.29 34–37 40–42 44 46 48 Coincidently, five of the seven RCTs targeting patients also demonstrated small effect sizes,28 30–33 with three studies indicating no significant reduction in effect size due to a negative lower limit of 95% CI.31–33

Although it is challenging to draw a direct correlation between the effect size and the type of intervention and recipient of intervention, 9 of the 12 studies with a large effect size implied that interventions targeted directly at healthcare professionals were generally effective and mainly focused on implementing prescribing guidelines with various educational or pharmacist-led interventions to facilitate the implementation.29 34 36 40–42 46–48 In addition, studies with a large effect size targeted patients and healthcare professionals (n=2)38 39 or patients alone (n=1)49 also demonstrated the benefit of educational sessions (on patients and healthcare professionals),39 facilitated by either a written instruction in the chart,37 digital application combined with physician rounds targeted towards patients.49 The interventions in studies with a large effect size were often more comprehensive and tailored, implementing multifaceted approaches. In contrast, studies with a small/medium effect size were generally simpler, such as educational handouts or lectures.

The BCTs associated with large effect sizes offer a valuable foundation for developing novel interventions, drawing from a well-established base of effective BCTs. We did not identify any BCTs under the following groups of the taxonomy: scheduled consequences, self-belief and covert learning, which may be attributed to reporting bias. It is possible that studies included in this systematic review might have used those BCTs but did not specifically document them; hence, those BCTs were not explicitly identified.

Although not focusing on postsurgical opioid deprescribing, a previous systematic review and meta-analysis by Hansen et al examined the BCTs in deprescribing interventions, highlighting that the number of individual BCTs used did not consistently correlate with intervention effectiveness on medication use and inappropriate prescribing.52 Similarly, Hansen et al identified BCTs such as goal setting, action planning, social support, use of a credible source, instructions on performing a behaviour, feedback on behaviour and information about health consequences as effective in deprescribing interventions.52 Building on these findings, our study identified additional BCTs, including behaviour substitution, pharmacological support, providing prompts and cues, incorporating objects into the environment, graded tasks, regularly reviewing outcome goals, social comparison and introducing social rewards. The expansion of identified BCTs contributes to a more comprehensive understanding of effective strategies for behaviour change in the context of opioid prescribing practices.

The diversity in the number and types of BCTs employed across interventions suggests a lack of standardisation in applying behaviour change principles.53 The absence of explicit behaviour change theories or a systematic approach in intervention design signals an area for improvement in future research. Integrating behavioural science expertise in intervention development and employing structured frameworks such as the BCTTtv1 are essential steps to ensure comprehensive documentation and enhance study replicability.54 These efforts are advancing the field and identifying effective strategies to curb opioid use in postsurgical patients.

The strengths of our study include a comprehensive literature search and using a validated taxonomy to describe intervention content that facilitates behaviour change. Including all types of interventions and surgical procedures in this comprehensive review provides results that can be widely applicable to different surgical procedures. Interventions targeting healthcare professionals, particularly those incorporating guidelines and educational components, demonstrated substantial impact. Most included studies were undertaken in the United States, emphasising the geographical focus of the current literature on opioid prescribing reduction in surgical settings.

We acknowledge some limitations of this systematic review. Since the studies lacked predesigned specifications for BCTs, our reliance on the information provided to identify the present BCTs may have introduced the potential for under-coding or over-coding. In addition, the high risk of bias across studies warrants caution in generalising these findings.

Moreover, the inconsistent documentation of clinical outcomes (such as reductions in pain, improvements in function and enhancements in quality of life) associated with the interventions constrain our ability to definitively ascertain the clinical ramifications of the reported interventions. This underscores the imperative for standardised outcome reporting to facilitate meaningful comparisons and augment the applicability of the study findings. Another limitation of our review was the heterogeneity among the studies, particularly in reporting the time points of interest; hence, it was difficult to standardise across the different studies. For this reason, we have reported the reduction in MME based on the results of the individual studies.

Future studies aimed at designing deprescribing interventions should prioritise targeting healthcare professionals, as studies targeting this group have consistently demonstrated a large effect size. Additionally, identifying specific BCTs associated with intervention effectiveness offers valuable insights for designing tailored interventions to optimise opioid prescribing practices.

Conclusion

This systematic review provides a critical review of interventions targeting the reduction of opioid prescribing postsurgical procedures. The observed variations in study design, characteristics of target groups, risk of bias, interventions, BCTs, and outcomes highlight the intricate nature of addressing opioid prescribing practices. We illuminate a subtle understanding of the BCTs’ effectiveness, spotlighting crucial strategies with substantial impact. Our findings, which delineate effective interventions and identify research gaps, inform future initiatives to enhance postoperative pain management and mitigate opioid-related risks.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

Not applicable.