Prevalence and Risk Factors Associated With Long-term Opioid Use After Injury Among Previously Opioid-Free Workers

Key Points Question What is the prevalence of long-term opioid use after injury among previously opioid-free workers in Tennessee, and what risk factors are associated with developing long-term opioid use after injury? Findings In this cohort study of 46 399 injured workers who were opioid free at the time of injury, 4.0% had long-term opioid use after injury. Long-term use was associated with receiving 20 or more days’ supply in the initial opioid prescription and visiting 3 or more prescribers. Meaning Prescribing characteristics were the strongest risk factors for long-term opioid use after injury.


Introduction
Long-term opioid use is about more than the number of days someone takes an opioid; it is also about the threshold at which the costs of opioid use begin to outweigh the advantages. Research on long-term opioid use for chronic pain was summarized in a systematic review of clinical trials and observational studies, 1 which reported no evidence of effectiveness for opioid use of greater than 3 months but increased risk of harms, including overdose, substance use disorder, fractures, and myocardial infarction. An epidemiological study 2 showed that long-term opioid use is associated with dependence, addiction, poor self-rated health, inactivity, unemployment, higher health care utilization, and poor self-rated quality of life. In injured workers specifically, long-term opioid therapy is associated with higher health care costs 3 and lower productivity. 4 A 2018 study of workers' compensation (WC) claimants in Maryland 5 examined factors associated with opioid use more than 90 days after injury, focusing on patient characteristics.
Associations with opioid use after 90 days were found for older age (Ն60 years vs 15-29 years: odds ratio [OR], 1.92; 95% CI, 1.56-2.36), higher annual income (Ն$60 000 vs <$20 000: OR, 1.31; 95% CI, 1.07-1.61), crush injuries vs soft-tissue or contusion injuries (OR, 1.55; 95% CI, 1.28-1.89), strains and sprains vs soft-tissue or contusion injuries (OR, 1.54; 95% CI, 1.36-1.75), and chronic joint pain (OR, 1.98, 95% CI 1.79-2.20). 5 A 2017 study of a commercially insured population 6 pointed to the number of days' supply of the initial prescription as the single largest factor in continued opioid therapy for more than 1 year and more than 2 years after age, sex, mental comorbidities, and dose were controlled for. A retrospective cohort study conducted in Oregon using prescription drug monitoring program data 7 similarly reported strong associations of dose (400-799 morphine milligram equivalents [MME] vs <120 MME in a month: OR, 2.96; 95% CI, 2.81-3.11) and number of prescriptions filled (2 fills/mo vs 1 fill/mo: OR, 2.25; 95% CI, 2.17-2.33) with risk of long-term use of opioids (defined as Ն6 fills/y after initiation) among previously opioid-free patients. Another study in Utah 8 found increased odds of long-term opioid use (defined as Ն120 prescription days or Ն90 prescription days with Ն10 fills in a year) in patients who received benzodiazepines in the first 14 days of care (OR, 1.87; 95% CI, 1.01-3.48).
Among injured workers with back injuries, higher dose in the first 3 months after injury has been shown to be associated with long-term opioid use even after baseline pain and injury severity are controlled for (900-1799 MME vs Յ899 MME in first 3 months after injury: OR, 4.01; 95% CI, 2.23-7.20). 9 Following the US Centers for Disease Control and Prevention guidelines, providing the lowest dose of a short-acting opioid analgesic for the fewest days possible (preferably 3 and no more than 7 days) when opioids are initiated may mitigate the risk of developing long-term use 10 ; however, data are limited on the association of prescription-based factors, such as days' supply and dose, with longterm use among injured workers. This cohort study was conducted to develop a predictive model for the development of long-term opioid use in previously opioid-free injured workers. Demographic and prescribing characteristics hypothesized to be factors associated with long-term use were examined in previously opioid-free injured workers using Tennessee's Controlled Substances Monitoring Database (CSMD).

Study Design and Population
A detailed description of cohort creation was previously published. 12 Briefly, data on workplace injuries from Tennessee WC were linked to prescription records in the CSMD. Injured worker data were obtained from WC records (First Report of Injury, shared by the Tennessee Department of Labor), which is required regardless of the employee's intention to pursue medical care or a WC claim. 13 The CSMD contains data on prescriptions for controlled substances that are filled, and Tennessee dispensers are generally required to report filled prescriptions within 1 business day, with the exception of veterinarians, who were excluded from this study. 14 Data cleaning (eg, correcting errors and standardizing fields to enable linkage and analysis) and linkage between data sources on name and date of birth has been previously described. 12 To allow for prescriptions to be measured from 60 days prior to injury through 180 days after injury, prescription records were accessed from January 1, 2013, to June 30, 2016, and WC records were The main study population was opioid free at the time of injury, defined as having no record of receiving an opioid prescription for 60 days before the injury. 15 Opioid dispensing in Tennessee is limited to 30 days' supply, 16 and a 60-day restriction allows for only 2 consecutive periods of 30-day prescriptions. No injured workers receiving opioids for medication-assisted treatment were included in the study. Eligibility required complete name and date of birth data in the WC record to enable matching to the CSMD, having sex data, having a physical injury, and being aged 15 to 99 years. To avoid the confounding effect of multiple injuries, eligibility was restricted to injured workers who reported only 1 injury during the study period. Opioid prescriptions were included if they were opioid class and schedule 2 to 4 (schedule 5 was excluded owing to very low dose, eg, cough syrups) and excluded if they were known to be dispensed by a veterinarian or Veterans Affairs pharmacy.
Veterans Affairs pharmacies had incomplete reporting, composed a low amount (1%) of all opioid prescriptions, and were excluded owing to concerns regarding missing data. Prescription criteria resulted in only prescriptions being excluded, not patients. Opioid prescriptions were measured from 60 days before each person's date of injury (earliest January 1, 2013) to 180 days after each person's injury (latest June 30, 2016).

Demographic Characteristics and Clinical Information
Age at the time of injury, sex, marital status, type of injury, part of body injured, and residence type were selected from WC records on the basis of availability, completeness, and previous literature. 7,17,18 Type of injury was categorized into strains, sprains, and tears; fractures; and other on the basis of frequency and hypothesized association with long-term opioid use. Part of body injured was categorized into lower back, finger(s), knee, and other on the basis of frequency and association with long-term use. Additional clinical data were not available. Residence type was identified from zip codes and classified as urban (residing in a county with one of Tennessee's 6 largest cities) or rural (residing in one of the other 89 counties).
Postinjury opioids were classified by type, formulation (short-acting or long-acting), payment (cash or other), and overlapping days with a benzodiazepine prescription. Maximum daily dose in MME and maximum days' supply were identified. Each prescription measure, including dose and days' supply, was completed for 7 days, 30 days, and 90 days after injury. Additional prescription characteristics included dose and days' supply of the first prescription, receiving benzodiazepine in the 60 days before the injury, and the number of prescribers and pharmacies visited during the 90 days after the injury. For workers who received multiple opioid prescriptions on the same day (1695 [2.9%]), MME and days' supply were summed for all eligible opioid prescriptions received on that day.

Primary Outcome
Long-term opioid analgesic use, the primary study outcome, was measured with prescription days, ie, the days' supply of an opioid prescription added to the date on which the opioid was received. For overlapping prescriptions, prescription days were only counted once. Corresponding to the Centers for Disease Control and Prevention definition of chronic pain as lasting longer than 3 months, 19 longterm use was defined as receiving an opioid on most days for a 90-day period, measured as 45 or more prescription days in 90 days after injury. Nonoverlapping prescription days were summed for all prescriptions that were received and did not need to be continuous.

Statistical Analysis
A split data set approach 20 (Figure 1 and

Long-term Opioid Use
The number of injured workers who received an opioid on most days decreased sharply from 1834 to 966 between the first and second month after injury and decreased less steeply between 31 and 120 days after injury. The number plateaued after 120 days (Figure 2).

Derivation Model Analyses
In unadjusted analyses of the derivation group (Table 3), long-term opioid use was associated with receiving 20 or more days' supply in the initial prescription compared with less than 5 days' supply In the multivariable predictive model (see Table 3 for covariates), long-term opioid use continued to be associated with receiving 20 or more days' supply in the initial prescription compared with less than 5 days' supply (OR, 28

Derivation and Validation Model Fit
The  and pharmacies, except that the OR for an initial supply of 20 or fewer days decreased and the OR for amount of MME increased (eTable 4 in the Supplement).
Risk factors shown to be associated with opioid receipt in a previous study of this cohort 12 were also evaluated for their association with long-term use. When the model was restricted to people with fractures, the model fit was lowered to a C statistic of 0.87, and the association of receiving 20 or fewer days' initial supply decreased (adjusted OR, 6.30; 95% CI 3.84-10.18) but was still significant, while lower back injuries and overlapping opioid and benzodiazepine prescriptions were no longer significant (eTable 2 in the Supplement). Restricting the model to the most common injury (ie, strains, sprains, and tears) lowered the model fit to a C statistic of 0.91, while associations were unchanged except that finger injuries and MME of 160 or more were no longer significantly associated with long-term opioid use (eTable 3 in the Supplement). Hydrocodone and oxycodone were the most commonly prescribed opioids in both injury types and long-term use statuses increased by more than 10 daily MME over 3 months, and 79 (13.6%) surpassed the Tennessee chronic pain guidelines threshold of 120 daily MMEs during this time. Among injured workers who qualified as having long-term use in the first 90 days after injury, 557 patients (31.5%) were still using opioids on most days 6 months after injury.

Discussion
In this study, 6 opioid-use characteristics (days' supply of first prescription, receipt of a long-acting opioid within 30 days of injury, overlapping opioid and benzodiazepine prescription, number of prescribers and pharmacies visited within 90 days of injury, and maximum MME received within 30 days of injury) were associated with long-term opioid use in previously opioid-free injured workers.
The strongest association was receiving 20 or more days' supply in the initial prescription, followed by visiting 3 or more prescribers for opioids, associations which have been found in other populations. 22 Characteristics of opioid prescriptions, especially the initial prescription, had a greater association with long-term use than patient demographic or injury characteristics.
A previous study of this cohort 12 estimated that the prevalence of receiving an opioid after injury was 22.8% within 1 week, 29.7% within 1 month, and 33.3% within 6 months. In this study, most workers (79.6%) who received opioids after injury were opioid free before injury, and 4.0% of these became long-term users. Notably, one-third of patients who qualified as long-term opioid users in the 90 days after injury continued to use opioids 180 days after injury. Although dose escalation was noted in only one-quarter of long-term users, it can contribute to opioid-related harms, and surpassing 50 MME per day can increase the risk of overdose by 30%. 10  Lack of data on comorbidities may also account for some of the differences seen in the model when the number of prescribers and dispensers visited were excluded. The number of contacts with medical care could indicate multiple health conditions, and medical comorbidities have been associated with long-term opioid use in other studies. 18 Multiple contacts may also be an indication of drug-seeking behavior among people with an opioid use problem. 30 Patient reports or flags in the CSMD about visiting multiple prescribers and pharmacies can still serve as a warning sign of potential long-term use.
The effect of race and ethnicity may be a larger concern; studies in other states show that white patients tend to be prescribed opioids more frequently 31 and are more likely to overdose on opioids than black patients, 32 but nonwhite workers are more likely to receive WC. 33 Stratifying by race/ ethnicity may provide more granular detail toward understanding the trends in the epidemic and directing resources toward groups that would benefit from increased attention. It may also decrease the association of residence type with opioid use, as white residents are more concentrated in rural areas of Tennessee. 25 Although not all potential factors associated with long-term opioid use could be assessed, the variables included in this study resulted in a predictive model with extraordinarily good fit.
This study provides a good picture of the development of long-term opioid use in previously opioid-free injured workers in Tennessee but may not be generalizable to injured workers who do not report injuries to WC, report more than 1 injury, or live and work in other areas. Additionally, this study measures only prescriptions that were filled legally in Tennessee and may have missed prescriptions that were not consumed or were obtained from other sources. Misclassification of opioid-free patients owing to opioid supply from nonlegal sources or sources not measured by the CSMD is possible.

Conclusions
To our knowledge, this was the first study to examine long-term opioid use in Tennessee and among the first nationally to use a prescription drug monitoring program to measure long-term opioid use in injured workers with prescription records instead of clinical or insurance records. Developing longterm use appears to be more associated with prescribing practices, especially days' supply of the initial prescription, than patient or injury characteristics. These practices can be modified to reduce patient risk of overdose and associated morbidities.