Rate and Risk Factors Associated With Prolonged Opioid Use After Surgery

This systematic review and meta-analysis assess the rate and patient-level factors associated with increased risks of prolonged use of opioid medications after surgery.

and Meta-analyses (PRISMA) reporting guidelines. 35,36 The study protocol is available in the PROSPERO database (CRD42019129239).

Literature Search
Relevant studies were identified through an initial literature search of MEDLINE, Embase, and Google Scholar from inception of these databases to August 30, 2017, with an updated search performed on June 30, 2019. Eligible studies were identified from electronic databases using search terms and keywords such as opioid analgesics, general surgery, surgical procedures, persistent opioid use, and postoperative pain. The full search strategy is available in the eAppendix in the Supplement. We also searched bibliographies of relevant articles to identify additional eligible publications.

Study Selection
Two of us (J.G. and A.M.) independently assessed all titles and abstracts of studies to determine studies eligible for full-text review. Eligible studies were restricted to published observational studies evaluating opioid use after surgery. Studies were included if they (1) were published in the English language; (2) enrolled participants 18 years or older; (3) included a minimum of 50 patients; (4) involved a noninjectable opioid prescription fill at least 3 months after the index surgical procedure; and (5) reported the rate and adjusted outcome estimates for patient-level risk factors associated with prolonged opioid use after surgery. Given differences with opioid use in cancer vs noncancer pain management, we excluded studies evaluating cancer pain. In addition, eligible studies needed to have accounted for opioids dispensed in the perioperative period or to have incorporated a lag period for at least 1 month after the index surgical procedure. This criterion was included to account for opioids prescribed as part of routine management of postoperative pain before assessing prolonged opioid use after surgery.
Currently, there is not an accepted definition of prolonged opioid use in the medical literature.
Therefore, all studies that met the inclusion criteria were considered regardless of variations in the operational definition of prolonged opioid use within and among studies. However, because opioids are often prescribed preemptively to manage peri-or postoperative pain during the few days or months, in some instances, after surgery, we set a 3-month threshold after the index surgical procedure before assessing prolonged opioid use. As such, prolonged opioid use in this study refers to any opioid use pattern reported by the included studies occurring at least 3 months after surgery.
An exception is use of the term chronic opioid use. Based on previous literature, 2,11,[37][38][39][40] we defined chronic opioid use as the receipt of at least 10 opioid prescription fills, at least 90 consecutive days' supply of opioids, or 120 cumulative days in the first year after surgery, excluding the initial 90 postoperative days. Because we expected substantial between-study variation in prolonged opioid use definitions, in sensitivity analyses, we repeated our primary analysis to assess the pooled rate of prolonged opioid use by aggregating evidence across studies involving comparable definitions for opioid use after surgery (eTable 1 in the Supplement).

Quality Assessment
The quality of included studies was assessed by 2 independent reviewers (O.D.L. and J.G.) using the Newcastle-Ottawa Scale, 41 and disagreements were resolved by discussion. Studies with a Newcastle-Ottawa Scale score greater than 7 were considered to be high in quality.

Statistical Analysis
The primary outcomes of interest were the pooled rate and magnitude for individual risk factors of prolonged opioid use after surgery. No restrictions were made in the type of effect estimates extracted; therefore, studies reporting odds ratios (ORs), risk ratios, or hazard ratios were eligible for inclusion. Based on the overall low prevalence of the risk factors in the individual studies (ie, <10%), we regarded risk ratios and ORs as equivalent risk measures. However, we pooled studies reporting hazard ratios in a separate analysis. We calculated the pooled rate of prolonged opioid use after surgery weighted by the sample size of each eligible study. When 2 or more studies reported adjusted estimates for the same risk factor, a pooled OR and the corresponding 95% CI were estimated using the inverse variance method with a random-effects model. 42 Based on an a priori assumption of substantial between-study variation, we prespecified to use the random-effects model for all metaanalyses. Between-study heterogeneity was tested using the Cochran Q statistic 43 and quantified by the I 2 value. 44 We considered heterogeneity to be significant at P < .10 and I 2 >50% to indicate substantial between-study variation that was beyond chance. 45 Heterogeneity was further assessed in sensitivity analyses. Small-study effect, commonly referred to as publication bias, was examined using a funnel plot and the Egger regression test. Except for heterogeneity, statistical significance was set at P < .05; all tests were 2-sided. Statistical analyses were conducted using Comprehensive Meta-Analysis Software, version 3.0 (Biostat).
We performed additional analyses to evaluate the potential sources of heterogeneity and robustness of the primary findings. First, we restricted our analyses to studies enrolling only opioidnaive patients before surgery. We accepted definitions of opioid naive from each eligible study. The definitions of opioid naivety in studies enrolling only opioid-naive participants are presented in eTable 2 in the Supplement. Second, we examined whether the rate of prolonged opioid use after surgery differed by source population or insurance plan, follow-up period (Յ6 months vs >6 months), type of surgery (major vs minor surgery and orthopedic vs nonorthopedic surgery), and studies conducted in the US vs non-US countries to assess whether our main finding was moderated by potential differences in prescribing patterns across countries. Classification of major or minor surgery was based on previously published reports 11,12,46 and expert opinion (E.B.) (eTable 3 in the Supplement). We then aggregated evidence across studies involving comparable definitions for chronic or prolonged opioid use after surgery. Lastly, for each risk factor reported by at least 3 studies, we recalculated the pooled effect by omitting 1 study at a time. This leave-one-out analysis was performed to determine the influence of an individual study on the pooled effects.
With the exception of anxiety, we were unable to find comparable risk factors across the 2 studies 50,55 that used hazard ratios. Therefore, our analyses on risk factors for prolonged opioid use were derived from studies reporting risk ratios or ORs.
Increased risk of prolonged opioid use was associated with use of antidepressants, opioids, benzodiazepines, alcohol, cocaine, or tobacco before surgery ( Medical comorbidities were differentiated into 3 specific categories: psychological, painassociated conditions, and a broader category composed of disorders such as diabetes, pulmonary disease, and obesity ( Table 2). When evaluating the association between psychological disorders

Sensitivity and Additional Analyses
Our primary findings remained largely unchanged in leave-one-out analyses (eTable 4 in the Supplement). No evidence of publication bias was found with the Egger regression-based test (intercept, -20.99; 95% CI, -46.04% to 4.07%; SE, 12.28; P = .10) (eFigure in the Supplement).
Studies involving only opioid-naive participants before surgery had lower pooled rates of prolonged opioid use after surgery (1.2%; 95% CI, 0.4%-3.9%). In the restricted analysis assessing chronic Squares indicate event rates, with horizontal lines representing 95% CIs. The diamond represents the pooled total, with the points of the diamond representing 95% CIs. The data show substantial between-study heterogeneity (P < .001; I 2 = 99.96%).  Table 3.

Discussion
This systematic review and meta-analysis of observational studies 11,12,16,27,[29][30][31][32][33] extend the results of a previous meta-analysis 73 reporting prolonged opioid use among approximately 1 in 10 individuals undergoing a major or minor surgical procedure. Our analyses indicated that approximately 7% of patients filled opioid prescriptions at 3 months and more than 1 year after surgery, a time beyond the normal postoperative recovery period. 74 A higher rate was observed when prolonged opioid use was defined as the filling of at least 1 prescription for opioids within 91 to 180 days after surgery. However, our primary pooled rate was attenuated when we restricted our analyses to patients considered as opioid naive before surgery or to studies involving a more conservative definition of prolonged use that is commonly used in the medical literature to characterize chronic opioid use. Although these rates may appear to be relatively low, the negative  of surgical procedures, implies that up to 5.7 million Americans may potentially become persistent opioid users annually after surgery. Of note, individuals with prolonged opioid use after surgery constitute a group with potentially significant risk of chronic use. Therefore, prioritizing strategies that mitigate the transition of patients undergoing surgery to persistent opioid use while still optimizing the management of postoperative pain is of importance.
A possible approach to reducing the burden of prolonged opioid use is to characterize the underlying mechanisms, including patient-level risk factors, that may be associated with prolonged and/or chronic use of opioids after surgery. This approach, in part, rests on the assumption that patient-level risk factors associated with prolonged opioid use may be modifiable and can be used in screening for at-risk individuals. 74 Our results indicate that preoperative exposure to medications, such as opioids, antidepressants, benzodiazepines, or cocaine; demographic factors, such as sex; and presence of medical comorbidities, including chronic pain, back pain, substance abuse, mood disorders, or depression before surgery, had some of the strongest associations with prolonged opioid use after surgery. Congruent with previous reviews, 73,78 the strongest association in the current study was observed with preoperative opioid use, wherein individuals who had filled at least 1 opioid prescription in the year before surgery had a 5.3-fold risk of prolonged opioid use after surgery (pooled OR, 5.32; 95% CI, 2.94-9.64). These findings of increased risk of preoperative opioid use and prolonged use after surgery was further corroborated when we restricted our analyses to studies enrolling opioid-naive participants at baseline; the pooled rate of prolonged opioid use after surgery decreased more than 5-fold. Appropriate prescribing of the dose and quantity of opioids after surgery, the evaluation of opioid use in patients before surgery, and attempts to modify patientlevel risk factors when possible or to treat underlying medical conditions before surgery may be included as part of a comprehensive strategy to reduce prolonged opioid use after surgery.
Multimodal analgesia, psychobehavioral management of pain, and regional and neuraxial anesthesia have also been listed in the literature [78][79][80] as strategies associated with reducing the transition to prolonged opioid use after surgery.
Although our analyses suggest that surgery may be associated with long-term opioid use, it is possible that the observed association was enhanced by confounding from an underlying chronic pain condition, the developing of persistent postsurgical pain, or surgical procedures exacerbating preexisting conditions and thus warranting long-term opioid management. Persistent postsurgical pain is a recognized complication of surgery and has been reported after common surgical procedures, including cesarean delivery or hip replacement. 38,81 Several studies 38,[81][82][83][84] suggest that between 20% and 60% of individuals who undergo surgical procedures may transition from acute to persistent or chronic postsurgical pain. Because opioids were considered the standard of care for chronic noncancer pain management for studies included in this meta-analysis, 2,10 the findings suggest that a high rate of prolonged opioid use after surgery may reflect the expected opioid use patterns among individuals with persistent postsurgical pain or underlying chronic pain. Because of a lack of information in the included studies, we were unable to assess the association between these confounding factors and opioid use after surgery in our analyses.
Of note, other mechanisms not associated with surgical pain before or after undergoing the procedure could have explained the findings of increased prolonged opioid use with surgery.
Because major surgical procedures are likely to be associated with higher frequencies or intensities of postoperative pain and perhaps with a longer recovery time compared with minor surgical procedures, we expected significant differences in the pooled rate of prolonged opioid use in major vs minor surgical procedures. However, we found no such evidence in our subgroup analysis.
Although a similar finding was recently reported in a large retrospective study of US adults undergoing minor or major surgical procedures, 12

Limitations
This study has limitations. Because the studies included in our analyses were observational by design, our findings may be prone to several forms of systematic bias, including selection bias and measurement errors. Of importance, our findings may have been subject to confounding by the underlying indication and inadequate bias adjustment. Second, although we performed several sensitivity analyses to explore the sources of heterogeneity, we were unable to explain the substantial heterogeneity present in most of our analyses. We used a random-effects model for our analyses, with the a priori assumption that the included studies would be heterogenous in their design, sample size, definitions of prolonged opioid use and risk factors, and adjustment of covariates. Third, because of a paucity of eligible studies and suboptimal reporting, we were unable to exclude studies involving participants with chronic opioid use at baseline, participants with preexisting pain disorders, or participants with a diagnosis of cancer before surgery-conditions that are frequently managed with opioids. Of note, the inclusion of these individuals may have led to an overestimation in the magnitude of prolonged opioid use after surgery. 85 In addition, although less likely to be substantial, it is unknown the extent to which some of the eligibility criteria (eg, requiring studies to have reported the rate and risk factors for prolonged use) or not contacting authors may have affected the magnitude of observed association.
Despite these limitations, confidence in our findings is perhaps reinforced because of the absence of small-study bias and consistent results from study-level factors that might have