Abstract
Background: Long-term opioid use is a known risk factor for opioid-related harms. We aimed to identify risk factors for and predictors of long-term use of prescription opioids in the community-dwelling population of adults without a diagnosis of cancer, to inform practice change at the point of care.
Methods: Using Quebec administrative claims databases, we conducted a retrospective cohort study in a random sample of adult members (≥ 18 yr) of the public drug plan who did not have a cancer diagnosis and who initiated a prescription opioid in the outpatient setting between Jan. 1, 2012, and Dec. 31, 2016. The outcome of interest was long-term opioid use (≥ 90 consecutive days or ≥ 120 cumulative days over 12 mo). Potential predictors included sociodemographic factors, medical history, characteristics of the initial opioid prescription and prescriber’s specialty. We used multivariable logistic regression to assess the association between each characteristic and long-term use. We used the area under the receiver operating characteristic curve to determine the predictive performance of full and parsimonious models.
Results: Of 124 664 eligible patients who initiated opioid therapy, 4172 (3.3%) progressed to long-term use of prescription opioids. The most important associated factors in the adjusted analysis were long-term prescription of acetaminophen–codeine (odds ratio [OR] 6.30, 95% confidence interval [CI] 4.99 to 7.96), prescription of a long-acting opioid at initiation (OR 6.02, 95% CI 5.31 to 6.84), initial supply of 30 days or more (OR 4.22, 95% CI 3.81 to 4.69), chronic pain (OR 2.41, 95% CI 2.16 to 2.69) and initial dose of at least 90 morphine milligram equivalents (MME) per day (OR 1.24, 95% CI 1.04 to 1.47). Our predictive model, including only the initial days’ supply and chronic pain diagnosis, had area under the curve of 0.7618.
Interpretation: This study identified factors associated with long-term prescription opioid use. Limiting the initial supply to no more than 7 days and limiting doses to 90 MME/day or less are actions that could be undertaken at the point of care.
Well-known harms of long-term opioid use include substance use disorder, misuse and overdose, including overdose causing death.1–7 Over time, Canada has seen a steady rise in opioid use disorders.8 Various interventions and policies have been implemented, although they have had limited effect on opioid-related harms.9 The number and dose of opioid prescriptions have been decreasing in Canada since 2016,8 but treatment duration increased from an average of 41.3 days in 2006 to 54.9 days in 2016.10 In the United States, 42% of patients received a supply of 30 days or more in 2017.11
Strategies to identify and prevent unnecessary long-term opioid use, as well as the close monitoring of those receiving long-term therapy, may help to mitigate opioid-related harms. In the literature on opioids, patients with and without cancer are almost always studied separately, because the factors that affect the benefits and risks of opioid use differ between these 2 groups; a recent Canadian guideline also distinguishes between the 2 populations.12
Most published studies on determinants of long-term opioid use in the population without cancer have focused on subpopulations, such as patients with traumatic injuries,13 those who have undergone hip arthroscopy,14,15 those with musculoskeletal disorders16,17 or veterans.18–21 In the current study, conducted in the general community-dwelling adult population without a cancer diagnosis, we aimed to identify risk factors for and predictors of long-term prescription opioid use to inform practice change at the point of care.
Methods
Setting and design
This retrospective cohort study, conducted in the province of Quebec, Canada, involved community-dwelling adults (≥ 18 yr) covered by the Quebec Public Prescription Drug Insurance Plan, which includes the majority (94%) of older people (≥ 65 yr), all welfare recipients and residents whose medication costs are not covered by private insurance (about 43% of residents age < 65 yr).22 We included in our analysis people without a diagnosis of cancer who initiated a prescription opioid in the outpatient setting. The study outcome was progression to long-term use of a prescription opioid over a follow-up period of 12 months.
Data sources
We used 3 administrative claims databases managed by the Régie de l’assurance maladie du Québec: beneficiary, drug dispensing and medical services.22 The beneficiary database includes sociodemographic characteristics, level of copay (which is dependent on income) and coverage periods. The dispensing database lists drugs covered by the plan (excluding in-hospital and over-the-counter medications), with each record specifying the dispensing date, drug code, number of units, dose, days’ supply and prescriber’s specialty; there is no information on indication. The medical services database records claims in inpatient, outpatient and emergency department settings, with diagnoses coded using the International Classification of Diseases, 9th Revision (ICD-9).23 All 3 databases can be linked through deterministic linkage using the person’s health insurance number.
Study population
We obtained data for a random sample of 125 000 adult members (≥ 18 yr) of the public drug plan (the maximum allowed under the data-sharing agreement) who initiated a prescription opioid between Jan. 1, 2012, and Dec. 31, 2016 (where the date of opioid initiation was considered the index date); the last follow-up ended on Dec. 31, 2017.
We applied the following inclusion criteria: absence of claims with a cancer diagnosis (ICD-9 codes 140–239.9), continuous drug plan coverage in the 12 months before and 12 months after opioid initiation, absence of methadone dispensing in the previous 12 months (which would indicate opioid use disorder) and patient alive at the end of the 12-month follow-up. Apart from cancer, we applied no exclusions based on diagnoses.
Following receipt of the sample data, we excluded patients who were admitted to hospital or long-term care for 70% or more of the follow-up period, because no inpatient dispensing data were available. We were unable to determine the flow of patients through the phases of the sample selection.
Variables
We included in the analysis all prescription opioids covered by the public drug plan, specifically morphine, codeine, hydromorphone, oxycodone, butorphanol, fentanyl, meperidine and pentazocine. We excluded methadone (because it is an indicator of an opioid use disorder), and suboxone was not covered. Because the combination product acetaminophen–codeine is available over the counter, we excluded this medication from the index opioids but assessed it in terms of the patient’s history of using prescription pain medication. We used the most frequent definition of long-term opioid use found in the literature:15,24–27 90 or more consecutive days or 120 or more cumulative days over 12 months, regardless of switches. We considered prescription overlap to represent concomitant use.
We considered the following sociodemographic characteristics: age group, sex, region of residence and level of copay. As is the case for many claims databases, alcohol usage and smoking history were not available, and we were unable to perform linkage with complementary data sources. We determined overall health status using the Charlson Comorbidity Index,28 with diagnoses weighted according to the probability of death over 1 year. We considered claims in the 12 months before the index date with diagnoses of known risk factors for opioid use disorder29–33 (codes listed in Appendix 1, available at www.cmajopen.ca/content/9/1/E96/suppl/DC1), including mood disorder, schizophrenia, anxiety, other psychiatric conditions, substance use disorder and dementia, as well as previous dispensings of benzodiazepines, antipsychotics, antidepressants, analgesics (prescribed acetaminophen, acetaminophen–codeine, corticosteroids, nonsteroidal anti-inflammatory drugs, gabapentinoids) and antidepressants used for pain (tricyclic antidepressants, venlafaxine, duloxetine, paroxetine, fluoxetine). Duration of past use was categorized as short-term (< 90 d) or long-term (≥ 90 d).
Given that indications were not available, we cross-linked initial opioid dispensing with diagnoses and procedures to identify those likely to be associated with acute or chronic pain. Based on expert clinical assessment (by coauthors C.T. and P.V.) and a study conducted in Ontario,34 we created the following mutually exclusive categories of inferred indications for opioid use (codes listed in Appendix 2, available at www.cmajopen.ca/content/9/1/E96/suppl/DC1): claims for surgical procedures or diagnosis related to accident or fracture made in the emergency department in the 30 days before the index date, or initial opioid prescription by a dentist; diagnosis associated with chronic pain in the 12 months before the index date (arthritis, spinal stenosis, fibromyalgia, osteoporosis, deformation of the spine, neuropathy, limb pain, neck pain); diagnosis of conditions associated with other types of pain in the 12 months before the index date (abdominal, thoracic, tendon or digestive system pain; abscess); or, in the absence of the above criteria, a hospital or emergency department claim in the 7 days before the index date, to account for in-hospital procedures not billed on a fee-for-service basis.
Characteristics of the index opioid dispensing were product code, concomitant opioid dispensings, mode of action (short- or long-acting), duration of initial supply (< 15, 15–29 or ≥ 30 d) and prescribed daily dose. Using established conversion factors (Table 1),10,25,34,35 we converted the initial daily dose prescribed into daily morphine milligram equivalents (MME/d),12,26,35 categorized as follows: < 30, 30 to < 50, 50 to < 90, and ≥ 90 MME/day. Opioid dosage intensity was the total MMEs divided by treatment duration, excluding the number of days in hospital.
Statistical analysis
We identified factors associated with progression to long-term prescription opioid use through multivariable logistic regression analysis, initially including all variables. Sex was considered a priori to modify the effect of age; as such, we included an interaction term.
We used the area under the receiver operating characteristic curve to determine the predictive performance of the model,36 using a threshold of greater than 0.70 to define good performance.37 We developed a parsimonious predictive model using backward selection of covariables. The backward selection of variables, including the variables associated with collinearity, was automated. The aim was to identify which patients were likely to progress to long-term use of prescription opioids, based on the simplest model in a clinical setting. The final reduced model that we retained was the one with the least number of variables resulting in area under the curve greater than 0.70.
We conducted all statistical analyses using SAS software, version 9.4 (SAS Institute Inc.).
Ethics approval
The study was approved by the Ethics Committee of Université de Montréal (no. 18-102-CERES-D).
Results
The study cohort consisted of 124 664 eligible adult patients without cancer who initiated a prescription opioid in the outpatient setting, of whom 4172 (3.3%) went on to long-term use of prescription opioids within 1 year. Patients who progressed to long-term use accounted for 51.1% of the MMEs dispensed to the entire cohort during follow-up.
Patient characteristics
Patients who progressed to long-term use were older, more likely to be female and more likely to have a lower income than those with short-term use (Table 2). They also had a higher frequency of risk factors for opioid use disorder, including previous diagnosis of psychiatric disorders and dementia, and had received more psychotropic medications in the year before the index date. There was, however, no statistically significant difference in history of substance use disorder. The frequency of previous long-term use of pain medications was also higher among those with long-term use of prescription opioids. The most frequent claims preceding the initiation of a prescribed opioid among patients who progressed to long-term use were related to chronic pain (47.4%), whereas among those with short-term use the most frequent claims were related to accident, fracture or surgery (31.2%). Overall, 24.1% and 15.0% of those with long-term and short-term use, respectively, had no claims or claims that were not associated with acute or chronic pain.
Prescribing characteristics
Among those for whom opioids were prescribed over the long term, the mean number of days’ supply over 12 months was 242.7 (standard deviation [SD] 86.6), and the mean dosage intensity was 41.5 (SD 57.4) MME/day. Among patients with short-term use, the mean number of days’ supply was 10.9 (SD 15.3), and the mean dosage intensity was 36.3 (SD 23.2) MME/day.
For both groups, hydromorphone was the most frequently dispensed opioid at the index date, followed by morphine (Table 3). There were, however, important differences in mode of action. At initiation, a long-acting opioid was dispensed to 24.0% of patients who progressed to long-term use compared with 1.3% of those who had short-term use. Of patients who went on to long-term use of prescription opioids, 27.1% had obtained an initial supply of 30 days or more, compared with 2.8% of those who had short-term use. A mean initial daily dose of at least 90 MME was also more frequent among patients who progressed to long-term use than among those without such progression (5.7% v. 3.6%).
Family physicians accounted for the majority of prescribers, regardless of patient group. Surgeons and urologists prescribed the highest mean dose at initiation (47.6 and 42.9 MME/d, respectively) whereas family physicians prescribed the lowest mean dose (33.3 MME/d). The number of days of initial supply was highest among family physicians (mean 6.6, SD 6.7) and lowest among gynecologists (mean 3.8, SD 3.2) and dentists (mean 3.8, SD 2.5). Family physicians accounted for the majority of prescriptions with initial supply of 30 days or more (82.5%).
As shown in Table 4, the prescription characteristics highly associated with progression to long-term use were long-acting opioid at treatment initiation, concomitant opioid dispensing during the index period, initial dose of at least 90 MME/day and initial supply of at least 30 days. Compared with morphine, there were no differences among opioid products (except codeine, which decreased the likelihood of progression to long-term prescription of opioids). Patient characteristics significantly associated with long-term opioid use were previous claims for diagnosis associated with chronic pain and women aged 85 years or older. Previous use of pain medications was also a risk factor, especially long-term use of prescribed acetaminophen–codeine. Psychiatric diagnosis, a known risk factor for opioid use disorder, was not associated with long-term opioid use after adjustment for all other variables. Compared with recent claims for accident or fracture (associated with acute pain), all other pain diagnosis categories increased the probability of long-term use, especially chronic pain.
Predictors of long-term opioid use
The predictive model including all variables was associated with area under the curve of 0.8736 (Figure 1). The parsimonious model, which included only initial dispensing of at least 30 days’ supply and a previous chronic pain diagnosis, had area under the curve of 0.7618. The association between each variable included in the parsimonious model and long-term use is shown in Table 5. For a patient with a chronic pain diagnosis and an initial supply of 30 days or more, the predicted probability of progression to long-term use was 31.7%, compared with an observed probability of 28.9%. When the initial supply was less than 15 days, the predicted and observed probabilities decreased to 3.9% and 4.1%, respectively. Predictive performance of all other combinations of variables is presented in Appendix 3 (available at www.cmajopen.ca/content/9/1/E96/suppl/DC1).
Interpretation
In this population of community-dwelling adults without cancer, several factors were associated with long-term use of prescription opioids, 4 of which are immediately modifiable at the point of care: long-acting opioids at initiation, a first filled opioid prescription of 15 days or more, initial dose of 90 MME/day or more, and concomitant administration of multiple opioids. Health care providers should be aware of the profiles of patients at greatest risk for progressing to long-term use, including women 75 years of age or older, those using acetaminophen–codeine for 90 days or more in the preceding 12 months and those with a diagnosis associated with chronic pain. Prescriptions written by family physicians provided, on average, a week’s duration of opioids and most frequently led to long-term use, which suggests that, at best, initial opioid prescriptions should not exceed 7 days’ supply. Although the US Centers for Disease Control and Prevention has advised against initiating therapy with a long-acting opioid,38 24.0% of those for whom opioids were prescribed over the long term had received a long-acting opioid at the index date. These results are likely not explained by under-ascertainment of previous opioid usage, given that our analysis accounted for recent hospitalizations. Topical analgesics can be proposed for such patients. Unfortunately, these formulations are not covered by the Quebec public drug plan.
This study’s 3.3% progression rate to long-term opioid use was similar to estimates from the US (3.3%),18 Australia (2.6%),39 Germany (1.3%)40 and Norway (3.0%).41 Higher progression rates were found across Canadian provinces, ranging from 9.9 to 17.7%,42 likely because of differences in the definition of long-term use. Those who progressed to long-term use accounted for 51.1% of total MMEs dispensed to our cohort, which is lower than the 87% found in British Columbia26 but consistent with data from other studies.43–46
Predictors of long-term opioid use were also found in previous studies, such as chronic pain,43 history of substance use disorder or dementia,15 prior use of psychotropic drugs or analgesics,39,45,47 use of hydromorphone or oxycodone at initiation, concomitant opioids and initial daily dose of 90 MME or more.46 In the current study, patients receiving at least 90 MME/day at the index date had mostly initiated opioid therapy with short-acting hydromorphone or oxycodone and had not been hospitalized before the index date. Overall, 24.1% of those with long-term prescribing of opioids and 15.0% of those with short-term prescribing did not have any previous pain diagnoses, surgery or recent hospital admission. An Ontario study, also using claims data, found a similar proportion of patients without these factors (12%).34 To our knowledge, the development of a predictive model for long-term use was previously attempted in only 1 study, in which long-acting opioid and use of tramadol were leading predictors.45
In this study, the large number of patients classified as having long-term use of prescription opioids allowed for the inclusion of all available variables in the regression model, which is a strength for assessing potential risk factors.
Limitations
The Quebec drug claims database does not record inpatient medication use. We mitigated this limitation by excluding patients admitted for 70% or longer of the follow-up period and, for the other patients, by taking into account opioid dispensing before and after admission. The alternative (i.e., excluding hospitalized patients) would have introduced selection bias.
We did not examine use of tramadol, a predictor of long-term opioid use,43,45 because it was not covered during the study period.
Absence of indication is a limitation of studies conducted using claims databases. The categories developed a priori were partially based on clinical expert opinion. Incomplete recording of diagnoses in claims may explain the substantial frequency of patients with no potential indications for opioids, but likely does not explain the difference between those with long-term and short-term prescribing of opioids.
The public drug plan in Quebec excludes residents with private coverage plans, those in hospital and those in long-term care. This limits the generalizability of our results, although the proportion of patients with long-term use was consistent with that of previous studies.
The area under the curve assesses only model discrimination. One area for future study would be model calibration. By selecting a parsimonious model that met the area under the curve threshold of 0.70, we excluded models with greater area under the curve and hence, better predictive performance. However, we were aiming to develop a simple tool that could be used to alert clinicians to patient profiles with a higher likelihood of transition to long-term use and to underscore risky prescribing behaviours, such as initial prescriptions with duration of 1 month or longer. Furthermore, ORs provide complementary insights to the area under the curve, which is why elements of both types of analysis were displayed.
Given the timelines for data acquisition and study conduct, these findings may not reflect contemporary practices.
Conclusion
This study identified factors associated with long-term use of prescription opioids. Health care providers can help to limit the progression to long-term opioid use by ensuring that initial prescriptions do not exceed 7 days or the equivalent of 90 MME/day without reassessment and by avoiding long-acting formulations and concomitant opioids. Regular re-evaluation of opioid prescriptions should be performed, with simple questions about type and intensity of pain, to prevent unnecessary dose titration. Thorough, periodic re-evaluations of opioid prescriptions by health care providers (perhaps with introduction of non-opioid alternatives, according to existing guidelines) might help limit long-term opioid use.
Footnotes
Competing interests: For activities unrelated to the current study, Philippe Vincent has received an educational grant from Janssen and personal fees from Janssen, Otsuka and Proxim. As president of YolaRx Consultants, Yola Moride has served as a consultant for various opioid manufacturers, including Teva Canada Innovation and Paladin Labs. No other competing interests were declared.
This article has been peer reviewed.
Contributors: Yola Moride and Audrey Béliveau were the main contributors to conception and design of the study and to acquisition, analysis and interpretation of the data. Cara Tannenbaum and Philippe Vincent contributed to defining the scope of the study, developing the study protocol and interpreting the study findings, especially as they relate to the clinical implications. Cristiano Soares de Mouras and Sasha Bernatsky contributed to developing the study protocol, performing the analysis and interpreting the findings. Anne-Marie Castilloux planned and conducted all data management activities, as well as the cohort programming. Audrey Béliveau developed the first draft of the manuscript, and Yola Moride developed all subsequent drafts. All of the authors revised the manuscript for important intellectual content, approved the final version for publication and agreed to be accountable for the work. Yola Moride was the senior investigator and supervised all components of the study. At the time of this study, Audrey Béliveau was a student at the Université de Montréal, and this work was part of her MSc thesis.
Funding: This project was funded by the Drug Safety and Effectiveness Network through Canadian Institutes for Health Research grant DES-149322.
Data sharing: The study data are not available to others.
Supplemental information: For reviewer comments and the original submission of this manuscript, please see www.cmajopen.ca/content/9/1/E96/suppl/DC1.
This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY-NC-ND 4.0) licence, which permits use, distribution and reproduction in any medium, provided that the original publication is properly cited, the use is noncommercial (i.e., research or educational use), and no modifications or adaptations are made. See: https://creativecommons.org/licenses/by-nc-nd/4.0/
References
- © 2021 Joule Inc. or its licensors