Abstract
Objective:
To investigate the effects of discharge opioid supply following surgery for musculoskeletal injury on subsequent opioid usage
Study Design:
Instrumental variables analysis of retrospective administrative data
Methods:
Data was acquired on 1,039 patients treated operatively for a musculoskeletal injury between 2011 and 2015 at two Level I trauma centers. State registry data were used to track all post-operative opioid prescription fills. Discharge surgical resident was identified for each patient. We categorized residents in the top third of opioid prescribing as high-supply residents and others as low-supply residents, with adjustment for service attending and month. Primary outcome was subsequent opioid use, defined as new opioid prescriptions and cumulative prescribed opioid supply 7–8 months after injury.
Results:
On average, patients of high-supply residents received an additional 96 morphine milligram equivalents (MME) at discharge (95% CI: 29,163; p<.01), or 16% more than patients of low-supply residents and equivalent to an additional 2-day supply at a typical dosage. In the 7th or 8th month after surgery, patients of high-supply residents received a greater total MME volume than patients of low-supply residents (13.0 MME; 95% CI: 3.1,22.9; P<0.01) despite receiving a greater cumulative supply of opioid medications through the 6th month after surgery.
Conclusions:
After surgery for musculoskeletal injury, patients discharged by residents who prescribe greater supplies of opioid pain medications received higher supplies of opioids 7–8 months after surgery than patients discharged by residents who tend to prescribe less. Thus, limiting post-operative supplies of opioid pain medication may help reduce chronic opioid use.
Summary:
We use surgical resident assignment as an instrumental variable for discharge opioid prescribing, and estimate the impact of discharge opioid supply on subsequent use.
Introduction
Physicians struggle with the challenge of managing acute pain while guarding against the risk of long-term use when prescribing opioids.1 Recent policies have restricted the supply of opioids physicians can prescribe based on studies that suggest reducing opioid supplies could mitigate subsequent use and dependence.2–5 These correlational studies may be prone to significant bias if patient or injury characteristics influence both discharge supplies and subsequent usage. For example, if patients with more serious injuries receive higher supplies at discharge, their propensity to use opioids in the short- and long-term may reflect the severity of their injuries rather than the supplies of their initial prescriptions.
One study addressed this source of bias by examining the association between physician prescribing tendencies and patient outcomes in the emergency department (ED) setting, where patients may be naturally randomized to physicians.6 For a general population of ED patients with relatively low rates of opioid medication receipt, that study found a very small (0.35 percentage point) increase in subsequent use associated with higher-intensity ED physician prescribing patterns and did not account for the contribution of a given patient’s clinical needs to both the physician’s prescribing tendency and subsequent use. Moreover, the long-term consequences of high opioid supplies for acute pain in clinical settings where opioids are often necessary, as opposed to discretionary, remain unclear.
In this study, we investigate the effects of discharge opioid supply on subsequent opioid usage after surgery for musculoskeletal injury at two academic trauma centers. Traumatic injury is a clinical setting where opioid prescribing rates are high, opioids are often unavoidable, and the opioid supply is consequently the relevant variable.7 Since a patient’s assigned resident at a teaching hospital is determined by the idiosyncrasies of the monthly rotation schedule and daily on-call schedules, assignment to residents with variable prescribing tendencies creates a natural experiment in discharge opioid supplies. Thus, we use resident prescribing tendencies as a source of variation that should not be related to patient needs, and we employ methods that eliminate a given patient’s contribution to the physician’s prescribing profile. We also account for variation between patients of different attending physicians that may not be attributable to the resident. Our analysis identifies a group of “treated” patients of “high-supply” residents who are comparable to “control” patients of “low-supply” residents in all aspects other than the discharge opioid supply.
METHODS
Study Data and Population
We identified 2,645 patients with traumatic musculoskeletal injuries who were treated operatively within 14 days of initial presentation and discharged by a resident physician at the Massachusetts General Hospital or Brigham and Women’s Hospital between 2011–2015. Detailed demographic and clinical data, including prescriber and discharge opioid prescription, were matched to each patient using the institutions’ Enterprise Data Warehouse (EDW). We excluded patients with multiple injuries and multiple traumatic events in a single episode. We also excluded patients who received a discharge opioid prescription exceeding 10,000 morphine milligram equivalents (MME, <1% of population). We used hospital discharge summaries to link patients to their care team of resident and attending physicians at discharge. Resident and attending physician names were identified using an electronic parser coded in Python, version 2.7 (Python Software Foundation).
As consistent with the acute nature of the injuries, 85% of patients (2,247) received an opioid at discharge. To track subsequent new prescriptions and re-prescriptions (including refills) of opioids throughout the state of Massachusetts, we linked hospital data to the Prescription Drug Monitoring Program data of the Massachusetts Department of Public Health’s Public Health Data Warehouse (PHD) using patient name, address, race, and gender (91% match rate). Patients from the hospital data who could not be matched to the PHD data were removed from the analysis. To limit our analysis to opioid-naïve patients, we excluded patients who were prescribed an opioid in the six months prior to presentation based on their prescription history in the PHD or EDW, excluding 48% of the original sample and leaving us with 1,051 patients. Finally, we excluded patients of residents and attendings who treated less than 5 patients in the sample (1% of the sample).
After applying these restrictions, our sample included 1,039 orthopaedic patients treated by 61 distinct residents and 34 distinct attending physicians. Attending physicians saw an average of 69 patients (range: 3–131), while residents saw an average of 22 patients (range: 5–42) meeting inclusion criteria over the four-year study period.
Study Variables
Discharge Opioid Prescription and Resident Dosing Intensity
We identified the discharge prescription as the first outpatient prescription within thirty days of initial injury. We used a 30-day window for initial prescription to ensure we included patients with longer hospitalizations; we used the earliest prescription to ensure we captured the prescription closest to discharge. Discharge prescriptions were converted into total MME by applying standard conversion tables.8 MME was summed in cases where more than one opioid prescription was filled on the same day.
Outcomes
We examined usage in each month following injury and the cumulative supply by the end of each month. We pre-specified our primary as subsequent opioid use 7–8 months after injury, defined as opioid prescription fills during those months, including new prescriptions and re-prescriptions. We considered both the volume of opioid supply in MME and whether or not an opioid was received during this interval.
Covariates
We recorded patient age, gender, race, and insurance type. We also considered injury year, location, and severity (AO-OTA fracture classification). Comorbidity was defined as the count of unique medication classes prescribed in the 180 days prior to the index visit (previously validated in trauma populations).2
Statistical Analysis
We ran ordinary least squares and two-staged least squares (2SLS) regressions to estimate the relationship between resident prescribing tendency and subsequent use. In our design, resident assignment acts as an instrument for discharge opioid prescription. Resident prescribing tendencies were adjusted for attending effects as described below. Our main analysis uses ordinary least squares regression to estimate the direct relationship between resident prescribing tendency and subsequent opioid use.
Subsequent analyses use 2SLS (reported in Tables A2 and A3). In our design, 2SLS estimates the relationship between discharge opioid prescription (directly estimated from resident prescribing tendency) and subsequent opioid use. The first stage of the estimator is the relationship between resident prescribing tendency and patient discharge prescription. The reduced form of the estimator is the direct relationship between resident prescribing tendency and subsequent use. The 2SLS estimate is produced by dividing the reduced form coefficient divided by the first stage, hence scaling the relationship between resident prescribing tendency and long-run use by the degree by which resident prescribing tendency predicts discharge prescription.
Resident Prescribing Tendencies
In our regressions, resident prescribing tendencies were identified using a “leave-one-out” method to adjust for heterogeneous features of attendings, attending panels, and patients.9 Our approach allowed us to compare the MME quantity prescribed by each resident for a typical patient to those of other residents in similar settings. To estimate an “attending effect” on each prescription, we measured the mean discharge prescription MME by the attending physician’s other residents, leaving out the prescriptions written by the patient’s own resident so the resident’s prescribing tendencies do not enter into the attending effect. We then adjusted the resident prescription for each patient by the attending effect for each patient’s attending physician. Therefore, the estimated resident prescribing tendency is demeaned by the attending effect. Adjusting resident prescribing practices for this attending effect was important to eliminate differences in residents’ exposure to different attending physicians, because while the resident assigned to a given patient should be random, assignment of surgical cases to different attending physicians may not be.
Then, the “resident’s prescribing tendency” was defined as the mean discharge prescription (in MME) by the resident relative to all other residents, after adjusting for the attending effect and the clinical characteristics of the resident’s patients. To ensure the “resident’s prescribing tendency” was independent of the patient’s own characteristics, we excluded the patient’s own prescription. Leaving out the patient’s prescription was important to eliminate correlation between unobserved patient characteristics and the estimated prescribing tendencies of the patient’s resident. The resident prescribing tendency is therefore unique for each attending-patient combination. We defined “high-supply” residents as those in the top third of resident prescribing tendencies, with robustness to top fourth and top fifth.
Analysis
To test the assumption that patients were naturally randomized to residents, we checked whether patient characteristics were balanced across high and low-supply residents. We also assessed the strength of the instrumental variable by estimating the association between the prescribing tendency of a patient’s assigned resident and discharge supply.
For our main analysis, we estimated the direct (“reduced form”) relationship between resident prescribing tendency and long-run opioid use using ordinary least squares regression. In 2SLS analyses reported in the appendix, we further adjusted the reduced form estimate with the relationship between resident tendency and patient discharge MME prescription (“first stage”), producing an estimate of the relationship between MME prescription at discharge and subsequent opioid use. All regressions used heteroskedasticity-robust standard errors, and were performed using SAS Studio, Version 3.5 (SAS Corporation, Cary, NC).
As a falsification test, we checked whether resident prescribing tendencies were predictive of patients’ opioid usage in the 7– 8 months prior to surgery. Since treatment in the present should not affect opioid use in the past, uncovering such a relationship would suggest an omitted factor like chronic morbidity drove resident prescribing behavior and patient opioid use patterns. In contrast, a null result would support our study design, since many omitted factors that would create a misleading relationship between resident prescribing tendencies and opioid use would also affect prior use.
This study was approved by the Massachusetts General Hospital Institutional Review Board (IRB), which determined that individual patient and prescriber consent was not required. The work in the PHD was mandated by law, and conducted by a public health authority. The Massachusetts Department of Public Health was not engaged in human subjects research, and thus no IRB review was required. No patients were involved in determining the research question, outcome measures, or study design. There are no plans to involve patients in the dissemination of research findings.
RESULTS
Population Characteristics
Out of 1,039 opioid-naïve patients, 347 were treated by “high-supply” residents and 692 were treated by “low-supply” residents. Both groups had similar observable patient and injury characteristics (Table 1), suggesting that unobserved characteristics were also likely comparable.
Table 1:
Balance Check of Covariates Between Patients of High and Low-supply Residents
| Covariate | Patients of Low-supply Residents | Patients of High-supply Residents | Difference (Standard Error) |
p-value |
|---|---|---|---|---|
|
| ||||
| Age | 60.30 | 60.74 | 0.44 (1.17) |
0.71 |
| Comorbidities | 0.37 | 0.34 | −0.034 (0.051) |
0.50 |
| White | 83.7% | 81.56% | −2.11% (2.51) |
0.40 |
| Female | 52.75% | 53.88% | 1.15% (3.28) |
0.73 |
| White, female | 43.79% | 44.67% | 0.88% (3.27) |
0.79 |
| Medicaid | 8.24% | 8.25% | 0.12% (1.82) |
0.95 |
| Self-pay | 2.75% | 1.74% | −1.02% (0.94) |
0.28 |
| Leg | 32.80% | 32.57% | −0.23% (3.08) |
0.94 |
| Pelvis | 2.02% | 2.59% | 0.57% (1.01) |
0.57 |
| Upper Body | 8.09% | 7.21% | −0.89% (1.73) |
0.61 |
| AO/OTA A | 16.76% | 14.99% | −1.78% (2.39) |
0.46 |
| AO/OTA B | 16.91% | 16.72% | −0.19% (2.46) |
0.94 |
| AO/OTA C | 9.25% | 10.66% | 1.41% (1.99) |
0.48 |
|
| ||||
| 2011 | 4.48% | 1.16% | −3.33 (0.97) |
<0.001 |
|
| ||||
| 2012 | 16.33% | 22.48% | 6.14 (2.65) |
<0.001 |
|
| ||||
| 2013 | 22.40% | 31.70% | 9.30 (2.96) |
<0.001 |
|
| ||||
| 2014 | 41.04% | 38.04% | −3.00 (3.21) |
<0.001 |
|
| ||||
| 2015 | 15.75% | 6.63% | −9.12 (1.92) |
<0.001 |
AO/OTA is the AO Foundation/Orthopedic Trauma Association fracture classification.
Prescribing Behavior
As expected, the majority of patients (85%) received an opioid prescription soon after injury. The average patient received a discharge prescription of 745 MME, which would be a 15-day supply at the standard of 50 MME per day.10 In unadjusted analysis, patients of “high-supply” residents received an average of 820 MME compared to 707 MME among patients of “low-supply” residents, a difference of 113 MME or ~2 days of a typical supply (95% CI: 47.3,177.8; P=0.001, Table 2). After adjustment for the set of covariates listed in Table 1, the estimated effect of resident prescribing tendencies on discharge supply remained stable (Appendix Table A1). Specifically, in our first-stage specification, patients of high-supply residents received 96.1 more MME than patients of low-supply residents (95% CI: 28.4,163.7; P=0.006).
Table 2:
Differences in Discharge MME and Short- Term and Subsequent Usage Between Patients of High and Low-supply Residents
| Variable | Patients of Low-supply Residents | Patients of High-supply Residents | Difference (Standard Error) |
p-value | Difference After Controls (Standard Error) |
p-value after controls |
|---|---|---|---|---|---|---|
| Discharge MME | 707.1 | 819.7 | 112.6*** (33.15) |
<0.001 | 96.09** (34.50) |
0.01 |
| MME 1–2 Months After Discharge | 18.29 | 5.21 | −13.1* (6.67) |
0.06 | −13.6** (5.47) |
0.01 |
| MME 7–8 Months After Discharge | 4.59 | 19.22 | 14.6** (5.57) |
0.01 | 13.0** (5.08) |
0.01 |
| % with Opioid 1–2 Months After Discharge | 8.09% | 3.46% | −4.63*** (1.43) |
0.002 | −6.01*** (1.51) |
<.0001 |
| % with Opioid 7–8 Months After Discharge | 6.06% | 9.86% | 3.80** (1.85) |
0.05 | 2.66 (1.79) |
0.14 |
MME indicates total effective volume of opioids filled during specified time period. Controls include patient demographics—comorbidities, insurance status (Medicaid and self-pay versus the default of private insurance or Medicare), age and age squared, and indicators for being white, female, and a white female—as well as descriptions of the injury—the location (leg and upper body versus elsewhere on the body), the type of fracture (based on AO/OTA classifications), and the year of the injury. Standard errors that are robust to heteroscedasticity are in parentheses.
Evolution of cumulative supply of prescribed opioids after discharge
Due to the higher opioid supply at discharge, patients of high-supply residents started with a higher volume of MME (Figure 1A). This gap initially closed modestly as patients of high-supply residents received lower total volumes of opioids within the 5 months after injury, but the gap remained substantial over the entire period and began to widen starting in the 6th month after injury as patients of high-supply residents received higher volumes of newly prescribed opioids (Figure 1A and 1B). We found the same pattern after adjusting for patient characteristics and the year of surgery. By 7 months after injury, patients of high-supply residents received significantly greater new supplies of opioids (Figure 2).
Figure 1: (A) Cumulative Difference in Opioid Volume (B) Opioid Volumes (in MME) in Months after Injury.
Figure 2: Differences in MME Month-by-Month and Cumulatively for Patients of High and Low-supply Residents in the Months After Injury.
We show the cumulative differences in MME and the month-by-month differences between patients of high and low-supply residents, after adjusting for patient demographics—comorbidities, insurance status, age, race, and gender—as well as injury type—location, type of fracture, and the year of the injury. Confidence intervals reflect standard errors that are robust to heteroscedasticity.
Short-term and subsequent opioid use
In the first two months after surgery, 3.5% of patients of high-supply residents received an opioid prescription compared to 8.1% of patients of low-supply residents (adjusted difference: 6.0pp; 95% CI: −9.0,−3.1; P < 0.001; Table 2). The opposite pattern emerged 7–8 months after discharge: 9.9% of patients of high-supply residents received an opioid prescription compared to 6.1% of patients of low-supply residents (adjusted difference: 2.7pp; 95% CI: −0.7,6.2; P= 0.13; Table 2). Unadjusted estimates were similar (Appendix Table A3).
In the first two months after injury, patients of high-supply residents received 5.2 MME and patients of low-supply residents received 18.3 MME (adjusted difference: −13.6 MME; 95% CI: −24.4, −2.9; P=0.013) of opioid pain medications from new prescriptions. In the 7–8 months after surgery, patients of high-supply residents received 19.2 MME of newly prescribed opioid medication on average compared to 4.6 MME among patients of low-supply residents (adjusted difference: 13.0 MME; 95% CI: 3.1,22.9; P=0.011). Unadjusted estimates were similar (Appendix Table A2). Residents in the top percentile of prescribing tendencies tended to prescribe 190 MME more than low supply residents, suggesting their patients would receive 26.9 less MME 1–2 months after discharge and 25.7 MME more 7–8 months after discharge.
Sensitivity Analysis
In the six months prior to injury, opioid use was zero for all patients by construction. In the 7–8 months prior to injury, there was no significant gap in opioid volumes between patients of high-supply and low-supply residents (difference =1.8; 95% CI: −3.4,15.4; P=0.26; Table 3). In addition, results were similar when employing alternative definitions of “high-supply” residents (i.e. top fourth or top fifth of prescribing tendencies). Adjusted and unadjusted results did not differ substantively.
Table 3:
Placebo Checks: Opioid Usage 7 to 8 Months Before Injury
| % with Opioid 7–8 Months Before Injury |
MME 7–8 Months Before Injury |
|
|---|---|---|
|
High-supply
(SE) p-value |
1.79 (1.63) (0.29) |
6.03 (4.80) (0.26) |
|
Mean
(Std. Dev) |
6.45 (24.6) |
10.6 (72.3) |
| N | 1,039 | 1,039 |
The past opioid use of patients of high supply residents is compared to the past use of patients of low supply residents. The difference between these groups is presented. The same controls used in Table 2 are used here; standard errors that are robust to heteroscedasticity are in parentheses.
DISCUSSION
We found that patients who received a higher initial supply of opioids were more likely to fill and receive higher volumes of opioids 7 or 8 months after surgery than patients discharged by residents with more judicious prescribing tendencies, despite receiving a substantially greater cumulative supply through the first 6 months after surgery. While a higher discharge supply reduced re-prescription shortly after surgery, the increased subsequent opioid volumes 7–8 months later are suggestive of the development of misuse as a consequence of higher discharge quantities.
Our findings build on prior literature which has attempted to characterize the link between initial opioid supplies and subsequent use.1–3,5,6 While those studies may have been prone to significant bias if patient or injury characteristics influenced both discharge supplies and subsequent usage, we succeeded in identifying a natural source of prescribing variation that was unrelated to patients’ observable clinical needs based on balance checks and the similarity of results with and without adjustment. Building on a prior study,6 we found that physician prescribing in terms of opioid supply was an important factor in subsequent use in clinical settings where opioids are often necessary and the standard of care for treatment of acute pain.
We may have underestimated the effect of opioid prescribing at discharge on subsequent use if patients sought subsequent opioid prescriptions outside of the state health care system (i.e., via diversion from friends or family members) as a result of exposure to a higher initial supply, but we would not expect patient’s procurement of opioids through other means to vary by resident prescribing tendencies. We could not assess the clinical consequences of a low initial opioid supply as we could not assess pain control, but this is an important area for future research. The time period of the analysis (2011–2015) may not reflect more recent changes in opioid prescribing norms; however, the available setting allowed for significant heterogeneity in prescribing behavior. Our results showed that patients of high-supply residents filled higher quantities of opioids 7–8 months after discharge, however our analysis of the percentage of those patients filling prescriptions failed to reject the null hypothesis, with large standard errors. Average volume is a combination of prescription intensity and number of patients; further work with larger sample could estimate this decomposition with precision. (Table A3) Finally, we used an electronic parser to identify resident and attending physicians and we may have failed to link some care teams due to data quality. We do not believe this would bias our results since this failure was unlikely to occur in systematic fashion for specific attendings, residents, or patients.
Reducing exposure to opioids, while emphasized in prescribing guidelines, is often challenging following musculoskeletal injuries because of patients’ need for acute pain control.11 Thus, the typical choice facing trauma care teams is not whether to prescribe an opioid but how much to prescribe. Our results suggest that some physicians may have a natural tendency to give higher initial opioid supplies, perhaps to reduce the short-term demand for opioid prescription renewals that discharging physicians may have to address before patients are able to see their surgeon or primary outpatient physician in follow up. Indeed, we observed higher rates of new prescriptions initially after surgery for patients prescribed lower supplies at discharge. Taken together, our findings suggest that policies to lower discharge doses of opioid pain medication would reduce subsequent use but that limitations on discharge supply may need to be coupled with strategies to ensure adequate treatment of short-term pain. Strategies might include improving post-discharge access to appropriate renewals and non-opioid pain medication via electronic or telephonic pain assessments and e-prescribing.12,13 Other areas for improvement include better training on the balance between pain management and the risk of elevated opioid use14 and techniques to identify high-risk individuals.15
Conclusion
After surgery for musculoskeletal injury, patients discharged by residents who prescribe greater supplies of opioid pain medications received higher volumes of new prescriptions for opioids 7 or 8 months after surgery than patients discharged by residents with more judicious prescribing tendencies, despite receiving a greater cumulative supply through the first 6 months after surgery. Based on this finding, to minimize the risk of subsequent use while managing acute pain, our findings would support a strategy of low opioid supplies at discharge with a coordinated pain management plan in place to ensure adequate treatment of short-term pain.
Supplementary Material
Take Away Points:
We use surgical resident assignment as an instrumental variable for discharge opioid prescribing, and estimate the impact of discharge opioid supply on subsequent use.
Surgical trauma patients who receive higher opioid prescriptions at discharge have higher rates of subsequent use (supply of new opioids at 7–8 months)
Limiting initial opioid prescription at discharge may reduce subsequent opioid use
Resident assignment in orthopedic trauma surgery can be used as an effective instrumental variable, or “natural experiment,” for discharge opioid prescribing
Funding:
Pre-doctoral Fellowship on Aging NIA T32 AG 51108 from the National Institute on Aging
Contributor Information
Matthew Basilico, Department of Economics, Harvard University and Harvard Medical School, 98 Hancock Street, Unit 2, Cambridge, MA 02139
Abhiram R. Bhashyam, Harvard Combined Orthopaedics Residency Program
Emma K. Harrington, Department of Economics, Harvard University
Monica Bharel, Massachusetts Department of Public Health
J. Michael McWilliams, Department of Health Care Policy, Harvard Medical School, Division of General Internal Medicine and Primary Care, Brigham and Women’s Hospital
Marilyn Heng, Department of Orthopaedic Surgery, Massachusetts General Hospital, Harvard Orthopaedic Trauma Initiative, Harvard Medical School.
REFERENCES
- 1.Lembke A, Mackey S and Costello P, “The Problem of Prescription Opioids,” in Health Policy Forum, Stanford, CA. [Google Scholar]
- 2.Deyo RA, Hallvik SE, Hildebran C, Marino M. Association Between Initial Opioid Prescribing Patterns and Subsequent Long-Term Use Among Opioid-Naïve Patients: A Statewide Retrospective Cohort Study. Journal of General Internal Medicine. 2017;32(1):21–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Waljee JF, Li L, Brummett CM, Englesbe MJ. Iatrogenic Opioid Dependence in the United States: Are Surgeons the Gatekeepers? Ann Surg. 2016;265(4):728–730. doi: 10.1097/SLA.0000000000001904. [DOI] [PubMed] [Google Scholar]
- 4.Blackman K Prescribing Policies: States Confront Opioid Overdose Epidemic. [Google Scholar]
- 5.Sekhri S, Arora NS, Cottrell H, et al. Probability of Opioid Prescription Refilling After Surgery: Does Initial Prescription Dose Matter? Ann Surg. 2017;268(2):271–276. doi: 10.1097/SLA.0000000000002308. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Barnett ML, Olenski AR, Jena AB. Opioid-Prescribing Patterns of Emergency Physicians and Risk of Long-Term Use. N Engl J Med. 2017;376(7):663–673. doi: 10.1056/NEJMsa1610524. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Morris BJ, Mir HR. The opioid epidemic: impact on orthopaedic surgery. J Am Acad Orthop Surg. 2015;23(5):267–271. doi: 10.5435/JAAOS-D-14-00163. [DOI] [PubMed] [Google Scholar]
- 8.Dowell D, Haegerich TM, Chou R. CDC Guideline for Prescribing Opioids for Chronic Pain--United States, 2016. JAMA. 2016;315(15):1624–1645. doi: 10.1001/jama.2016.1464. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Aizer A and Doyle JJ Jr. Juvenile Incarceration, Human Capital and Future Crime: Evidence from Randomly-Assigned Judges. Quarterly Journal of Economics. 2015;130(2): 759–803. [Google Scholar]
- 10.Guy GP, Zhang K, Bohm MK, et al. Vital Signs: Changes in Opioid Prescribing in the United States, 2006–2015. MMWR Morb Mortal Wkly Rep. 2017;66(26):697–704. doi: 10.15585/mmwr.mm6626a4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Giannoudis PV, Hak D, Sanders D, Donohoe E, Tosounidis T, Bahney C. Inflammation, Bone Healing, and Anti-Inflammatory Drugs: An Update. Journal of Orthopaedic Trauma. 2015;29 Suppl 12:S6–9. doi: 10.1097/BOT.0000000000000465. [DOI] [PubMed] [Google Scholar]
- 12.Kaafarani HMA, Weil E, Wakeman S, Ring D. The Opioid Epidemic and New Legislation in Massachusetts: Time For a Culture Change in Surgery? Ann Surg. 2016;265(4):731–733. doi: 10.1097/SLA.0000000000002083. [DOI] [PubMed] [Google Scholar]
- 13.Meara E, Horwitz JR, Powell W, et al. State Legal Restrictions and Prescription-Opioid Use among Disabled Adults. N Engl J Med. 2016;375(1):44–53. doi: 10.1056/NEJMsa1514387. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Antman KH, Berman HA, Flotte TR, Flier J, Dimitri DM, Bharel M. Developing Core Competencies for the Prevention and Management of Prescription Drug Misuse: A Medical Education Collaboration in Massachusetts. Acad Med. 2016;91(10):1348–1351. doi: 10.1097/ACM.0000000000001347. [DOI] [PubMed] [Google Scholar]
- 15.Bohnert ASB, Valenstein M, Bair MJ, et al. Association between opioid prescribing patterns and opioid overdose-related deaths. JAMA. 2011;305(13):1315–1321. doi: 10.1001/jama.2011.370. [DOI] [PubMed] [Google Scholar]
- 16.Chen JH, Humphreys K, Shah NH. Distribution of opioids by different types of medicare prescribers. JAMA Intern Med. 2016;176(2):1315–1321. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.