While persistent opioid use after surgery has been the subject of a large number of studies, it is unknown how much variability in the definition of persistent use impacts the reported incidence across studies. The objective was to evaluate the incidence of persistent use estimated with different definitions using a single cohort of postoperative patients, as well as the ability of each definition to identify patients with opioid-related adverse events.
The literature was reviewed to identify observational studies that evaluated persistent opioid use among opioid-naive patients requiring surgery, and any definitions of persistent opioid use were extracted. Next, the authors performed a population-based cohort study of opioid-naive adults undergoing 1 of 18 surgical procedures from 2013 to 2017 in Ontario, Canada. The primary outcome was the incidence of persistent opioid use, defined by each extracted definition of persistent opioid use. The authors also assessed the sensitivity and specificity of each definition to identify patients with an opioid-related adverse event in the year after surgery.
Twenty-nine different definitions of persistent opioid use were identified from 39 studies. Applying the different definitions to a cohort of 162,830 opioid-naive surgical patients, the incidence of persistent opioid use in the year after surgery ranged from 0.01% (n = 10) to 14.7% (n = 23,442), with a median of 0.7% (n = 1,061). Opioid-related overdose or diagnosis associated with opioid use disorder in the year of follow-up occurred in 164 patients (1 per 1,000 operations). The sensitivity of each definition to identify patients with the composite measure of opioid use disorder or opioid-related toxicity ranged from 0.01 to 0.36, while specificity ranged from 0.86 to 1.00.
The incidence of persistent opioid use reported after surgery varies more than 100-fold depending on the definition used. Definitions varied markedly in their sensitivity for identifying adverse opioid-related event, with low sensitivity overall across measures.
Persistent opioid use after surgery is a matter of great concern
Defining appropriate opioid prescribing practices and policies depends critically on understanding the rate of and reasons for persistent postoperative opioid use
A systematic search of the literature revealed 29 distinct definitions of persistent opioid use employed in 39 different studies
Applying the definitions to a separate study cohort of more than 162,000 surgical patients identified persistent opioid use rates varying more than 100-fold with low sensitivity for the identification of opioid use disorder
Opioids may be indicated for the management of moderate to severe pain in the immediate postoperative period, but concerns have been raised that excessive exposure to opioids may lead to long-term dependence.1–3 In this context, the concept of “new long-term use,” typically defined based on filled pharmacy claims for opioids after surgery, has emerged as an important construct in research on opioid-related outcomes after acute pain treatment.1–3 Moreover, surgeons, and postsurgical opioid prescriptions, have become a large focus of opioid-related policy, research, and intervention.4
The research surrounding persistent postsurgical opioid use has important methodologic limitations, which could have notable implications for patients and policy. Most importantly, the definition of persistent postsurgical opioid use is not consistent between studies, making interpretation of estimates and comparison across studies challenging.5 Notably, it is also unknown whether definitions of persistent postsurgical opioid use reflect the patients at risk for opioid-use disorder or opioid-related toxicity.
The objective of the study was to evaluate whether the definition of persistent postoperative opioid use impacts the reported incidence. In order to achieve this, first, we performed a systematic search of the literature to identify definitions of persistent postoperative opioid use in the medical literature. Second, we evaluated the influence of these definitions on the estimated incidence of persistent opioid use after surgery. Specifically, we applied each definition to the same cohort of opioid-naive patients presenting for surgery to determine the variability in defined persistent use based on the definition employed. Finally, we assessed the sensitivity and specificity of each definition to identify patients with opioid-related adverse events in the year after surgery.
Materials and Methods
Systematic Literature Search for Definitions of Persistent Use
Data Sources and Search Strategy.
The prespecified protocol for this systematic search, which aimed to identify definitions of persistent postoperative opioid use, was developed based on Preferred Reporting Items for Systematic reviews and Meta-analyses Protocols (PRISMA) 2015 guidelines.6 We conducted a computerized search using the OVID versions of MEDLINE, Medline-in-Process, Medline Epubs Ahead of Print (1946–June 2018), and Embase Classic+Embase databases (1946–June 2018). We used both subject headings and textword terms for Opioids AND (surgical procedures or postoperative complications) and (cohort studies or prognosis or risk or time factors). We also searched the reference lists of included studies. The complete search strategy and further details of the systematic review methods can be found in Supplemental Digital Content 1 (http://links.lww.com/ALN/C336). Each title and abstract were screened independently and in duplicate by two authors (N.J. and J.B.). Full-text articles were obtained for any study considered potentially relevant for defining and/or estimating the incidence of persistent opioid use after surgery.
Study Selection and Outcome Definition.
Two reviewers assessed the full text of each retrieved citation independently and in duplicate. We included studies of adults (18 yr or older) undergoing any surgical procedure. We required that studies have a cohort study design and measure a patient’s opioid use at least 30 days after discharge from surgery and provided a definition for persistent opioid use in the postoperative period. We excluded studies that (1) were not performed in humans, (2) were not written in English, (3) were review articles, (4) were randomized controlled trials as these are likely to be highly selected (selection bias) and were unlikely to estimate the incidence of persistent postoperative opioid use, or (5) where we were unable to extract the relevant data. Our outcome of interest was the definition(s) of persistent postoperative opioid use employed by the authors.
Two reviewers extracted data independently and in duplicate using a standard form (N.J., F.R., J.B.). Information extracted included patient characteristics, study design, setting, surgical procedures, the definition of persistent postoperative opioid use employed, and the type of opioids measured.
Population-based Cohort Study
After approval by Sunnybrook Health Sciences Center Research Ethics Board (Toronto, Canada), we then used a population-based sample of opioid-naive patients undergoing surgery to assess the estimate of persistent opioid use with each definition. We used health administrative data from the province of Ontario, Canada. All residents of Ontario, which has a population of approximately 14 million individuals, obtain their healthcare services from a single payer and provider (the Government of Ontario). Each individual is assigned a unique, encoded identifier that permits linkage deterministically across provincial and national administrative databases and analysis at ICES. For details of the databases, see Supplemental Digital Content 2 (http://links.lww.com/ALN/C337). These databases have high levels of completeness and have been validated for many outcomes, exposures, and comorbidities.7,8
Our population of interest was all patients who were 18 yr or older between July 1, 2013, and March 30, 2017. Our exposure of interest was 1 of 18 surgical procedures, using Canadian Classification of Health Intervention codes (to identify procedures) and associated International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes (to identify diagnoses; Supplemental Digital Content 3, http://links.lww.com/ALN/C338). This enrollment window was selected to provide a minimum 1-yr look-back period and 1-yr follow-up period for all patients. The index date was defined as the date of admission (inpatient surgery) or date of surgery (outpatient surgery). The specific procedures included those listed in Supplemental Digital Content 3 (http://links.lww.com/ALN/C338). These surgical procedures were chosen as they are common,9 and have been previously used to evaluate postoperative opioid consumption.10 The validity and reliability of codes to identify surgical procedures have been confirmed through reabstraction studies.11
We excluded patients if they (1) did not have valid provincial health insurance; (2) had a surgical procedure in the year before their index surgical procedure; (3) sought palliative care services, as subsequent opioid use was likely for end-of-life care12 ; and (4) had a prolonged hospitalization defined as a length of stay greater than 7 days. Finally, we excluded patients who filled one or more opioid prescription(s) in the year before surgery, omitting the first 30 days preoperatively (e.g., –365 days to –31 days, preoperatively). This definition was chosen as it represented the most inclusive definition of opioid-naive in our literature review and facilitated the operationalization of the greatest number of persistent postoperative opioid use outcome definitions.
The primary outcome was persistent postoperative opioid use. We operationalized each definition of persistent postoperative opioid use obtained from our systematic literature review (table 1). If persistent opioid use was defined as a prescription filled beyond 1 yr postoperatively, these definitions were not assessed. However, for opioid consumption definitions that used patient-reported measures (in-person, survey, or telephone), we interpreted these definitions to be operationalized using prescription information available in the administrative data. Finally, a recent consensus statement proposed a new definition of persistent opioid use.5 As this definition has not yet been evaluated in a surgical population, we included this definition post hoc in our analysis.
Opioids included were the most commonly prescribed outpatient opioids, and captured the greatest number of opioids used in our review: morphine, hydromorphone, fentanyl, codeine, oxycodone, tramadol, and meperidine.13 Oral formulations, buccal strips, and transdermal patches were the only formulations included. Opioids primarily used as cough suppressants in Ontario such as hydrocodone and normethadone were not included. Opioid prescriptions were identified using Drug Identification Numbers in the Narcotic Monitoring System. Since July 1, 2012, pharmacist data entry into the Narcotic Monitoring System is mandatory for any opioid dispensed in Ontario irrespective of age or insurance coverage.14
As a secondary analysis, we evaluated whether each definition of persistent opioid use reflected the patients at risk for opioid-related events, defined as development of opioid-use disorder or opioid-related toxicity. Opioid-use disorder was defined as either (1) presentation to an emergency department, an acute care facility or mental health institution in the year after discharge with a diagnosis of opioid use disorder, using International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Canadian Modification (ICD-10-CA) diagnosis code F11.XX, or (2) filling a new prescription of buprenorphine or methadone in the year after surgery as identified using Drug Identification Numbers in the Narcotic Monitoring System. Opioid-related toxicity was defined using ICD-10-CA diagnosis codes T40.0, T40.1, T40.2, T40.3, T40.4, and T40.6.15
All available data were used for each analysis, and as such, formal sample size calculations were not performed.
We described the baseline characteristics of opioid-naive patients presenting for surgery; see Supplemental Digital Content 2 (http://links.lww.com/ALN/C337) for details of variables. Normally distributed patient characteristics are described using means and SDs, while nonnormally distributed characteristics are described using a median and interquartile range. While 0.5% of patients had missing data for Neighborhood Income Quintile, they were not excluded from the cohort, and all patients were included in each analysis.
We then computed the incidence of persistent postoperative opioid use based on each outcome listed in table 1. Each definition provided a binary outcome (persistent postoperative opioid use or no persistent postoperative use). The incidence was calculated as the total number of patients classified as persistent users using each definition divided by the total number of patients in the cohort.
Next, to measure the degree to which commonly used definitions identified the same individuals as persistent postoperative opioid users, we did pairwise comparisons of the five most commonly applied definitions in the literature. We computed the level of agreement between each pair of definitions using Cohen’s κ coefficient, where greater than 0.61 was defined as substantial agreement.16 Akin to interrater reliability used in qualitative research,16 this enabled us to calculate the “interdefinition” reliability between each pair of definitions. Cohen’s κ was calculated based on methods described previously.16
Finally, we calculated the sensitivity and specificity of each definition of persistent postoperative opioid use to identify patients with opioid-related events. Sensitivities and specificities were calculated based on 2 × 2 tables that were constructed for each definition of persistent opioid use. True positive represented an individual classified with persistent postoperative opioid use based on the definition used and had an opioid-related adverse event in the year after surgery. Conversely, true negative represented an individual who was classified without persistent postoperative opioid use and did not have an opioid-related adverse event in the following year.
All analyses were conducted using SAS software (Enterprise Edition, Version 9.4, SAS Institute Inc., USA).
As patients who filled a prescription for opioids in the 30 days before surgery were not excluded from our study, it is possible that these individuals had preexisting pain and, as a result, were more likely to meet definitions of persistent postoperative opioid use. Therefore, in a secondary analysis, we excluded all patients with an opioid prescription in the year before index date and calculated the proportion of patients who met each definition of persistent postoperative opioid use.
Definitions of Persistent Postoperative Opioid Use
After removal of duplicate studies, the search retrieved 9,431 articles, and 39 articles were eligible for inclusion (fig. 1).5,17–54 Postoperative opioid use was evaluated by measuring filled or written opioid prescriptions (32 studies) or patient-reported opioid consumption (seven studies; table 1). The majority of studies (22 studies) classified persistent opioid use as a patient who filled one or more prescriptions, or reported consuming opioids, at a distinct time point. Second most frequent was the classification of persistent use based on the duration of filled or written opioid prescriptions (15 studies). Finally, studies classified persistent opioid use based on the number of prescriptions written or filled, or their associated duration or dose (six studies). The majority of studies (24 studies) did not define which type or formulation of opioids were used in the classification of persistent opioid use. Of the remaining studies (15 studies), three studies excluded opioids primarily used for the management of cough, three studies did not include tramadol, and all studies included methadone (see Summary of Definitions, Supplemental Digital Content 4, http://links.lww.com/ALN/C339).
Estimated Rates of Persistent Opioid Use
After exclusions (see Supplemental Digital Content 5 for flow diagram, http://links.lww.com/ALN/C340), a total of 162,830 (mean age ± SD, 50.1 ± 15.7; 67.9% women) opioid-naive patients underwent one of the chosen surgical procedures during the study period (table 2), and 66% (n = 107,496) of patients filled an opioid prescription within the first 7 days of discharge. Estimates of persistent opioid use in the year after surgery ranged from 0.01% (n = 10) to 14.7% (n = 23,442) of patients, depending on the definition of persistent use employed (fig. 2). In a sensitivity analysis excluding individuals who filled an opioid prescription in the 30 days before surgery, the incidence of persistent opioid use after surgery remained qualitatively unchanged (Supplemental Digital Content 6, http://links.lww.com/ALN/C341).
Agreement between Definitions of Persistent Opioid Use
The five most common definitions found in our literature search were further examined to evaluate whether they identified the same patients, namely their level of agreement (see Supplemental Digital Content 7 for full results, http://links.lww.com/ALN/C342). While most definitions had relatively low levels of agreement (Cohen’s κ < 0.50), more stringent definitions of opioid use in the year after surgery such as 90 days of continuous prescribing or 120 nonconsecutive filled prescriptions in 90 to 365 days postoperatively, or 10 more prescriptions or 120 nonconsecutive days of filled prescription in the 90 to 365 days postoperatively, had high levels of agreement (Cohen’s κ = 0.84; 95% CI, 0.82 to 0.87).
Identification of Patients with Opioid-related Events
In the cohort of 162,830 patients, 0.1% (n = 164) had an opioid-related adverse event. Specifically, the proportion of patients who had an opioid-related overdose in the year of follow up was 0.01% (n = 20). Furthermore, 146 patients (0.1%) were classified as patients with opioid use disorder in the same time period. We assessed the sensitivity and specificity of each operationalized definition to identify patients with the composite measure of opioid-use disorder or opioid-related toxicity (table 3). The sensitivity of each operationalized definition ranged from 0.01 to 0.36, while specificity ranged from 0.86 to 1.00.
In our systematic search of the literature, we identified 29 unique definitions of persistent postoperative opioid use employed in 39 studies. Using a standardized population of opioid-naive adults who had surgery, the incidence of estimated persistent opioid use varied more than 100-fold, depending on the definition used. There was similar variability in the sensitivity of these definitions for detection of opioid-related events, which was low in absolute terms across all measures used.
The high variability in definitions of persistent use, and subsequent estimates using these definitions, have important implications for patients, prescribers, researchers, and policy makers.
First is the concern of misinterpretation of results. A definition that requires a prescription for opioids filled 90 to 180 days after surgery has been described by the mainstream media as a patient who is “still taking opioids three to six months later”56 or who “continues to take drugs for three to six months after surgery.”57 However, one prescription filled during this time frame may not represent continued opioid use, albeit being interpreted as such. For example, while 4.7% of patients filled one prescription of opioids in the 90 to 180 days after surgery, less than 0.3% filled prescriptions that represent 90 days of continuous use or at least 120 days of noncontinuous use in the year after surgery. As persistent opioid use often implies and is interpreted as continuous use, we suggest that definitions with one filled prescription are likely inadequate to represent persistent use and may overestimate the risk associated with surgery. Moreover, as nearly 14% of Ontario residents fill a prescription for opioids in a given year,58 it is possible that some definitions of persistent use approximate the baseline rate of use in the general population.
When the reported rate of persistent opioid use is high, this can create anxiety in patients, who may refuse opioids in the treatment of acute pain for fear of developing opioid use disorder.59 Moreover, policy efforts to curb opioid use have focused on the supply of opioids, such as opioid prescribing guidelines, dose-limit laws, and law enforcement.60 These efforts alongside the high reported rates of persistent use, while successfully preventing excess opioid prescribing, may also contribute to a fear of opioid prescribing in physicians. Importantly, inadequate postoperative pain control can be associated with worse outcome for patients such as increased length-of-stay, reduced likelihood of mobilization, and sleep disturbance.61 Conversely, when the reported rate of persistent opioid use is low, physicians and patients may underestimate the risk associated with a single, short, initial prescription. Finally, research that aims to evaluate the effect of interventions on opioid use over the long term will benefit from standardization of definitions.
We have demonstrated that defining the time frame and degree of opioid prescribing that adequately characterizes persistent opioid use is a challenge of balancing sensitivity and specificity, while maintaining face validity. When persistent use was defined as at least one opioid prescription filled between 90 and 365 days after surgery, we identified patients at risk for opioid use disorder or toxicity more often. However, this definition likely identified patients who obtained opioids for reasons other than surgery. Similarly, classification of persistent use required 10 or more opioid prescriptions or greater than 120 days’ supply within 90 to 365 days postoperatively represented patients with significant opioid use but missed patients at risk for opioid-related overdose or use disorder. In our assessment of agreement among the most frequently used definitions, the highest agreement was noted with definitions that required some duration of prescriptions rather than relying on prescriptions filled within a certain time period. Our work demonstrates that, akin to concurrent criterion validity, definitions that represented patients with more severe opioid use in the year after surgery such as 120 days of noncontinuous prescribing or 10 or more filled prescriptions had high levels of agreement. Definitions with requirements of continuous opioid use are likely to reflect the patients with true persistent use. Moreover, prolonged opioid therapy may be associated with respiratory depression, addiction, and accidental death.62 Thus, more stringent definitions of persistent use likely reflect patterns of use that place the patient at risk for harm.
Our study has a number of strengths. First, we performed a rigorous, systematic search of the literature to identify the full range of definitions of persistent opioid use after surgery. While the incidence of persistent opioid use after surgery has been extensively addressed in the literature, many of these studies highlight that comparing estimates across studies is not possible due to differing definitions of persistent use or surgical procedures. This study addresses both of these issues, and identifies the significant influence outcome classification has on estimates of persistent use. The majority of research evaluating persistent opioid use is limited to specific populations, such as veterans, those who are insured by specific insurance plans, or specific age groups. This study included population-level data of all residents of Ontario presenting for one of our chosen procedures, irrespective of age.
The results of this study need to be interpreted in the context of its limitations. First, we used prescription claims data that cannot capture actual opioid consumption or, more importantly, the reason for the opioid fills. We note that in most studies reviewed, authors attribute longer-term opioid prescriptions to the surgery. However, it is unclear whether persistent opioid use in some individuals is the result of a subsequent unrelated pain complaint. Additionally, these data only captured outpatient prescriptions. As a result, we were unable to evaluate whether the surgical patients were exposed to opioids as inpatients. Second, while we attempted to accurately operationalize each definition of persistent opioid use, for definitions that required in-person follow-up, we may have over- or underestimated rates. Third, in order to evaluate each definition quantitatively, we calculated the sensitivity and specificity of each definition to identify patients with opioid use disorder or toxicity. This definition was defined based on ICD-10-CA codes and therefore only identifies patients with severe opioid use disorder who present to an acute care center for diagnosis. In this context, it is likely that we underestimated the incidence of opioid use disorder in the population and, as a result, the sensitivity of each definition of persistent opioid use. Similarly, as opioid-related adverse events were rare, the reported sensitivity and specificity estimates had broad CIs, making interpretation of the specific performance of each definition difficult. Furthermore, our follow-up period may be inadequate to capture opioid use disorder that emerges as a consequence of opioid exposure after surgery. Finally, many studies of surgical patients involve different types of surgeries. While our study isolated the impact of the definition used for the outcome of persistent opioid use, other sources of variability exist, such as the individual procedure performed or the system of care, that were not addressed in this study.
This systematic literature search and population-based cohort study demonstrated the variability of definitions of persistent opioid use employed in the current medical literature. More importantly, this study highlights that among a population of opioid-naive patients, the estimated incidence of persistent opioid use is highly dependent on the definition employed. The development of a clinically meaningful definition of persistent opioid use will require consensus from patients, providers, and policy makers. Widespread adoption of a single definition is important not only for comparability of studies but also to ensure accurate communication of risk.
The authors thank IMS Brogan Inc. (Mississauga, Canada) for use of their Drug Information Database.
Funding was provided by the National Institute on Drug Abuse (North Bethesda, Maryland; R01 DA042299–01A1 to Drs. Neuman, Bateman, and Wunsch and R01 DA044293–01A1 to Dr. Bateman). Dr. Wunsch received support from the Canadian Institutes of Health Research (Ottawa, Canada; grant No. 365432). Dr. Ladha is supported by a Merit Award from the Department of Anesthesia at the University of Toronto (Toronto, Canada). Dr. Wijeysundera is supported in part by a New Investigator Award from the Canadian Institutes of Health Research (Ottawa, Canada), an Excellence in Research Award from the Department of Anesthesia at the University of Toronto (Toronto, Canada), and the Endowed Chair in Translational Anesthesiology Research at St. Michael’s Hospital and University of Toronto. This study was supported by ICES (Toronto, Canada), which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (Toronto, Canada).
Dr. Bateman is an investigator on grants to Brigham and Women’s Hospital (Boston, Massachusetts) from the U.S. Food and Drug Administration (Silver Spring, Maryland), Baxalta (Bannockburn, Illinois), GSK (Brentford, United Kingdom), Lilly (Indianapolis, Indiana), Pacira (Parsippany-Troy Hills, New Jersey), and Pfizer (New York, New York) for unrelated work; and a paid consultant to Aetion, Inc. (New York, New York), and the Alosa Foundation (Boston, Massachusetts). The other authors declare no competing interests.