Surgeon Variation in Perioperative Opioid Prescribing and Medium- or Long-term Opioid Utilization after Total Knee Arthroplasty: A Cross-sectional Analysis

Our data consisted of healthcare claims for Medicare fee-for-service patients submitted between January 1, 2010, and December 31, 2017. These data include demographic information (i.e., race and sex) and information about the utilization of medical services, such as diagnosis codes (International Classification of Diseases, Ninth or Tenth Revision) and procedure codes (Current Procedural Terminology). The data also contain information about drug utilization, including drug name, fill date, quantity dispensed, and number of days supplied.⁹  This study follows the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines and received approval from the Institutional Review Board at Stanford University, which also issued a waiver of consent. A data analysis and statistical plan was agreed to and recorded in the investigators’ files before the data were analyzed.

We identified Medicare patients ages 65 and older who underwent total knee arthroplasty between January 1, 2011, and December 31, 2016, by searching for claims submitted by an orthopedic surgeon with a Current Procedural Terminology code of 27447. Using this search, a study sample was created consisting of patients who were continuously enrolled for the time period spanning 1 yr before and 1 yr after the surgery date and for whom the surgeon’s claim could be matched to a corresponding hospital claim that (1) had an appropriate International Classification of Diseases, Ninth or Tenth Revision procedure code and (2) had an admission-to-discharge time range covering the surgery date.

From this initial sample of 732,919 patients, we then applied the following exclusion criteria. First, patients treated by orthopedic surgeons who performed fewer than 25 total knee arthroplasties during the study period were excluded (n = 58,467). Second, we excluded patients in the top 1% of opioid utilization, measured in oral morphine milligram equivalent (see the Outcomes section for further details of the construction of this measure) to avoid extreme outliers and potential data errors (n = 15,645). Third, we excluded patients with a cancer diagnosis (n = 46,429), resulting in a sample of 604,093 patients who were treated by 7,141 surgeons. This full sample was used to classify surgeons as high or low intensity based on their patients’ opioid utilization in the immediate postoperative period (7 days before to 7 days after surgery) using the methods described below. From this full sample, 8,285 patients were missing a zip code of residence and therefore were excluded from the remaining statistical analyses. A flowchart of sample construction is provided in the supplement (Supplemental Digital Content, fig. 1, http://links.lww.com/ALN/C856). Since our study included all observations meeting the inclusion and exclusion criteria, no statistical power calculation was conducted before the study.

Primary outcomes consisted of opioid utilization during the midterm postoperative period (postoperative days 8 to 90) and the long-term postoperative period (postoperative days 91 to 365) measured in oral morphine equivalent (morphine milligram equivalents). Prescriptions for fentanyl (patch or oral), hydrocodone, hydromorphone (oral), methadone, morphine, oxycodone, tramadol, and oxymorphone were identified during the study period. Prescriptions containing hydrocodone in cough/cold formulation and analgesic preparations containing codeine were excluded. For each prescription, morphine milligram equivalents were calculated by multiplying an opioid-specific conversion factor by the drug dose and quantity supplied.¹⁰  The total morphine milligram equivalents during the relevant periods were calculated and then divided by the number of days in each period (275 days for long-term postoperative period and 83 days for midterm postoperative period) to obtain the average daily morphine milligram equivalents during the period.

The primary exposure of interest was whether a total knee arthroplasty was performed by a high-intensity or non–high-intensity surgeon, based on opioid prescriptions filled in the immediate perioperative period (7 days before to 7 days after surgery). We chose this time window because most patients who fill their initial postoperative opioid prescription do so during this time, including some patients who fill postoperative prescriptions before the day of surgery.⁴ 

Since opioid utilization may be based on observable patient characteristics (e.g., age and comorbidities), we first used a regression analysis to identify the extent to which a given surgeon’s patients were more likely to utilize opioids in the immediate perioperative period after adjusting for patient factors. To do so, we performed a patient-level multivariable linear regression in which the dependent variable was average daily opioid utilization (in morphine milligram equivalents) during the immediate perioperative period and the independent variables were patient characteristics (age, sex, race, Medicare–Medicaid dual eligibility, and the medical comorbidities described below), year fixed effects, and a fixed effect for each surgeon. In this regression, the surgeon fixed effect can be interpreted as the extent to which a given surgeon’s patients deviate from the overall average opioid utilization in the immediate perioperative period, after adjusting for patient factors. For example, a fixed effect of 5 would mean that even after adjusting for observable patient factors, a given surgeon’s patients tended, on average, to utilize 5 morphine milligram equivalents per day more than average during the immediate perioperative period.

We a priori categorized surgeons into quartiles based on the surgeon fixed effects and defined those in the upper quartile as high-intensity opioid surgeons and the remaining group as non–high-intensity. It is important to highlight that our exposure is not based on the actual morphine milligram equivalents a patient utilized during the immediate perioperative period; rather, the exposure is essentially based on the average morphine milligram equivalents utilized across a given surgeon’s patient population, after adjusting for potential variables that could influence opioid utilization (i.e., medical comorbidities).

Our analysis adjusted for several potential confounders including the following demographic factors: age, sex, race or ethnic group, and Medicare–Medicaid dual eligibility. We also defined a dichotomous opioid-naïve group based on whether a patient filled an opioid prescription in the 12 months before surgery. In addition, we identified the presence of Elixhauser comorbidities.¹¹ 

We first compared patient demographics, Elixhauser comorbidities, and primary outcomes between patients treated by surgeons who were high-intensity versus non–high-intensity surgeons. We calculated means and standard deviations for both groups, and two-sample, unpaired tailed t tests for continuous variables (e.g., age) and chi-square tests for discrete variables (e.g., comorbidities) were used to assess statistical differences between the groups.

A simple comparison of outcomes between patients who received care from high-intensity versus non–high-intensity opioid surgeons may be confounded if high-intensity surgeons are more likely to treat patients who are at higher risk for higher opioid utilization during the mid- and long-term postoperative periods. To address this issue, we adjusted for a robust set of potential confounders, including patient demographics (age, race, sex, Medicare–Medicaid dual enrollment), comorbidities (set of indicator variables for comorbidities comprising the Elixhauser index; shown in table 1), opioid naïve (yes or no), and year and state fixed effects. Standard errors were clustered at the hospital level.

To address the possibility of residual confounding—namely that unobservable patient factors may be correlated with both the likelihood of receiving surgery from a high-intensity versus non–high-intensity surgeon and higher opioid utilization during the mid- and long-term postoperative periods—we also conducted an instrumental variable analysis. This approach identifies the causal association between a treatment and an outcome through the use of an instrument, which is any variable that (1) influences the independent variable of interest (in this case, whether the patient received care from a high-intensity surgeon) but (2) is otherwise independent of the outcomes of interest. In effect, the instrument is used to quasirandomize patients to treatment and control arms of a study (in this case, whether a patient received care from a high-intensity vs. non–high-intensity surgeon). Following previous studies,^12–14  we used distance to the nearest high-intensity surgeon as an instrument. This approach mirrors several other studies that have used distance as an instrumental variable.^15,16  For example, one study¹⁷  used distance to the nearest hospital with regional anesthesia capabilities to examine the association between regional anesthesia and outcomes after hip fracture. The underlying assumption of this approach is that distance to a high-intensity surgeon is unlikely to be correlated with a patient’s risk for higher opioid utilization during the mid- and long-term postoperative periods.

To implement our instrumental variable approach, we first calculated the distance between the patient and the nearest high-intensity surgeon based on (1) the zip code of the patient’s residence and (2) the zip codes for all the hospitals utilized by high-intensity surgeons. We then implemented our instrumental variable approach using a multivariable two-stage least-squares linear probability model. The regression model included adjustments for potential confounders (e.g., patient sex, clinical comorbidities) previously described and used the distance to the nearest high-intensity surgeon as an instrument for whether the patient received care from a high-intensity surgeon. Further details of our instrumental variable approach, such as the methods used to verify the validity of the approach, are provided in the technical appendix found in the Supplemental Digital Content (http://links.lww.com/ALN/C856). All statistical analyses were performed using Stata 14.0 (Stata Corporation, USA).

We conducted a preplanned subgroup analysis on patients who were predicted to be at high-risk for higher opioid utilization during the long-term postoperative period. To define this population, we performed a multivariable linear regression in which the dependent variable was opioid utilization in the long-term postoperative period, and the independent variables were the patient characteristics (e.g., demographics and comorbidities) shown in table 1. The results of this model were used to obtain the predicted opioid utilization in the long-term postoperative period for each patient, and high-risk patients were defined as those in the upper quartile of predicted utilization during this period. The instrumental variable approach used for the main analysis was then repeated on this subgroup.

We performed several preplanned sensitivity analyses to examine the robustness of our results. First, we replicated the main analyses by focusing on the comparison between patients treated by high-intensity surgeons or low-intensity surgeons (upper quartile vs. lower quartile of surgeon fixed effects) by excluding patients in the middle two quartiles. Second, we considered an alternative outcome—incidence of chronic opioid utilization, defined as (1) having filled 10 or more prescriptions or (2) having more than 120 days’ supply during the postoperative days 91 to 365 within the first postsurgical year.

Several post hoc sensitivity analyses were performed as a result of the manuscript review process. First, in our baseline analysis, high-intensity surgeons were defined as those in the upper quartile based on the surgeon fixed effects previously described; we performed two sensitivity analyses in which high-intensity surgeons were defined as the top 5th and the top 10th percentile of surgeons based on these surgeon fixed effects. Second, the immediate perioperative period was defined as preoperative day 7 to postoperative day 7 in our baseline analysis; we performed a sensitivity analysis in which this period was redefined as postoperative days 0 through 7. Third, we performed a sensitivity analysis in which we adjusted for whether the patient lived in a rural or urban area, using rural–urban commuting area codes¹⁸  to define rural or urban areas, based on the patient’s zip code. Fourth, we performed a sensitivity analysis in which we excluded both (1) patients who received a bilateral total knee arthroplasty and (2) patients who received a second total knee arthroplasty within 365 days of their initial surgery. Fifth, we performed a weighted regression in which the weights were the number of patients treated by each surgeon. Sixth, we performed an analysis where we adjusted for additional measures of preoperative opioid utilization (number of prescriptions, number of days supplied, and average daily morphine milligram equivalents during preoperative days 8 to 365). Seventh, we performed an analysis where we adjusted for the use of nerve blockade. The use of a nerve blockade was identified by searching for claims that were: (1) submitted on the same date as the total knee arthroplasty itself and (2) corresponded to the use of a central or peripheral nerve block.¹⁹ 

The study sample included 604,093 Medicare fee-for-service beneficiaries, with an average age of 74 yr (SD 6); there were 413,121 (68.4%) females. A total of 180,926 patients (30%) were treated by high-intensity surgeons, and 423,167 patients (70%) were treated by non–high-intensity surgeons. The number of patients treated by high-intensity surgeons is slightly larger than 23% of the total sample because these surgeons tended to see more patients on average (230 patients per surgeon) compared to surgeons who were not high-intensity (167 patients per surgeon, difference 63 patients per surgeon; 95% CI, 62 to 64; P < 0.001 for the difference). Compared to their counterparts, patients treated by a high-intensity surgeon were slightly less likely to be female (67.5% vs. 68.8%; P < 0.001) and to be dually eligible for Medicare and Medicaid (9.3% vs. 11.5%; P < 0.001; table 1). Patients treated by high-intensity surgeons were also slightly younger (74.0 vs. 74.3 yr; P < 0.001) and more likely to be non-Hispanic white (88.4% vs. 88.1%; P < 0.001). They also had slightly fewer comorbidities (Elixhauser index 3.3 vs. 3.4; P < 0.001) and slightly higher opioid utilization in the year before surgery (4.2 vs. 3.7 morphine milligram equivalents per day; P < 0.001). However, while statistically significant, the standardized mean difference between the two groups for most variables was less than 0.1, which generally indicates a difference of small magnitude.²⁰  Exceptions included opioid utilization in the perioperative period (standardized difference, 0.63) and during postoperative days 8 to 90 (standardized difference, 0.12).

After using multivariable regression to adjust for patient characteristics and time trends, there was significant variation in opioid utilization during the immediate perioperative period (7 days before surgery to 7 days after surgery). Figure 1 shows, for each surgeon in our sample, the extent to which their patients on average utilized more (or less) opioid in the immediate perioperative period compared to the median surgeon. For example, the upper quartile of figure 1 was 7 morphine milligram equivalents per day, indicating that after multivariable adjustment, patients for the upper quartile of utilized at least 7 more morphine milligram equivalents per day compared to the median surgeon during the immediate perioperative period. Similarly, the bottom quartile was –5.7, meaning that the patients of the bottom 25% of surgeons utilized 5.7 or less morphine milligram equivalents per day compared to the median surgeon’s patients.

On average, patients receiving treatment from a high-intensity surgeon utilized 36.1 morphine milligram equivalents per day during the immediate perioperative period (95% CI, 36 to 36.3) compared to 17.3 morphine milligram equivalents per day for all other patients (95% CI, 17.2 to 17.4; P < 0.001 for the difference; fig. 2). However, the difference between these two groups were largely diminished by the midterm postoperative period (postoperative days 8 to 90) and the long-term postoperative period (postoperative days 91 to 365). Unadjusted opioid utilization during the midterm postoperative period was 10.8 morphine milligram equivalents (95% CI, 10.6 to 11.0; P < 0.001) for patients treated by high-intensity surgeons and 9.2 morphine milligram equivalents (95% CI, 9.1 to 9.4; P < 0.001) for those treated by non–high-intensity surgeons, and unadjusted opioid utilization during the long-term postoperative period was 3.8 morphine milligram equivalents (95% CI, 3.6 to 3.9) for patients treated by high-intensity surgeons compared to 3.4 morphine milligram equivalents (95% CI, 3.3 to 3.5) for patients who were not treated by high-intensity surgeons. In both cases, these differences were statistically significant but small in magnitude.

However, these unadjusted differences may be confounded by underlying differences between high-intensity and non–high-intensity surgeons in the propensity of their patients for postoperative opioid utilization. After using an instrumental variable approach to adjust for potential confounders, patients receiving treatment from a high-intensity surgeon had higher opioid utilization in the midterm postoperative period than patients who did not (fig. 2; adjusted daily morphine milligram equivalents, 11.4 [95% CI, 10.9 to 12.0] versus 9.0 [95% CI, 8.7 to 9.2]; difference in adjusted daily morphine milligram equivalents, 2.4; P < 0.001), a difference that was statistically significant but clinically small. Patients who received treatment from a high-intensity surgeon had lower opioid utilization during the long-term postoperative period (adjusted daily morphine milligram equivalents, 2.8 [95% CI, 2.5 to 3.1]) compared to patients treated by non–high-intensity surgeons (adjusted daily morphine milligram equivalents, 3.8 [95% CI, 3.6 to 3.9]). Again, while statistically significant (P < 0.001), this difference was clinically small.

In subgroup analysis of patients at high predicted risk for higher opioid utilization during the long-term postoperative period (n = 150,819), there was a statistically significant association receiving treatment from a high-intensity surgeon and opioid utilization during the midterm postoperative period (table 2; adjusted daily morphine milligram equivalents, 17.6 for high-intensity surgeons [95% CI 16.6 to 18.6] versus 13.9 for non–high-intensity surgeons [95% CI 13.5 to 14.1]; difference, 3.7 morphine milligram equivalents; P < 0.001). In a test of interactions, the difference between the effect for this subgroup and the effect among non–high-risk patients was statistically significant (P = 0.008). However, among high-risk patients, there was no association between receiving treatment from a high-intensity surgeon and opioid utilization in the long-term postoperative period (adjusted daily morphine milligram equivalents, 7.3 [95% CI 6.5 to 8.1] vs. 8.3 [95% CI 7.9 to 8.6]; difference in adjusted daily morphine milligram equivalents, –1; P = 0.08). There was no statistically significant difference between this estimated effect and the estimated effect for low-risk patients (P = 0.81 in a test of interactions).

We performed a sensitivity analysis in which we considered the incidence of chronic opioid utilization as an alternative outcome. Compared with low-intensity surgeons (i.e., the bottom quartile of surgeons), patients treated by surgeons in the second, third, and fourth quartiles had higher odds of filling 10 or more prescriptions or having more than 120 days’ opioid supply during the postoperative days 91 to 365 (Supplemental Digital Content, fig. 2, http://links.lww.com/ALN/C856). We observed stepwise higher incidence of chronic opioid utilization with exposure to physicians in each quartile of perioperative opioid utilization. The results of our other sensitivity analyses were qualitatively similar to our main results (table 3). For example, sensitivity analyses that examined alternate thresholds for defining high-intensity surgeons produced results that were higher in magnitude than our baseline analyses, but the effects remained clinically small.

We examined the association between receiving treatment from high-intensity surgeons (those whose patients were in the upper quartile in terms of opioid utilization in the immediate perioperative period) and subsequent opioid utilization during the midterm and long-term postoperative periods for Medicare fee-for-service patients undergoing total knee arthroplasty. Because patients treated by high-intensity surgeons may be systematically different from those treated by lower-intensity surgeons, we relied on an instrumental variable approach to minimize confounding and selection bias. Across several analyses, we found significantly higher midterm (postoperative 8 to 90 days) prescription opioid utilization among patients who received surgery from high-intensity surgeons compared to patients treated by non–high-intensity surgeons. We also found significantly lower long-term (91 to 365 postoperative days) opioid utilization among patients who received surgery from high-intensity surgeons. However, any large study such as ours is likely to find statistically significant findings that are small in magnitude, and the differences found in this study (less than 3 morphine milligram equivalents per day), which correspond to less than 1 hydrocodone tablet per day, are unlikely to be clinically significant. Indeed, studies examining opioid utilization in chronic pain patients generally define clinically significant reductions in opioid utilization as being at least 8 to 30 morphine milligram equivalents per day.^21–23  As such, the primary finding of our study is that differences in surgeon prescribing behaviors are unlikely to meaningfully impact opioid utilization in the mid- and long-term postoperative periods among patients undergoing total knee arthroplasty.

Several factors could explain the lack of correlation between surgeon prescribing behaviors and long-term postoperative opioid utilization. First, the difference in perioperative opioid utilization for patients seeing high-intensity surgeons—roughly 19 morphine milligram equivalents per day during a 15-day period—may not have been sufficiently large to place patients at risk of higher opioid utilization in the mid- and long-term postoperative periods. Second, to the degree higher opioid utilization improves pain control in the immediate perioperative period, it may reduce the risk of chronic postoperative pain and therefore the transition to chronic opioid utilization.²⁴  Finally, high-intensity surgeons may also provide more patient education regarding the risks of longer-term opioid utilization.

Our study adds to evidence that there is a wide variation in opioid utilization during the immediate perioperative period for total knee arthroplasty, one of the most common surgical procedures among older adults. We additionally quantified the extent of this variation across surgeons and studied its relationship with opioid utilization in the mid- and long-term postoperative periods. Previous studies have also not employed quasiexperimental statistical approaches such as instrumental variable analysis to overcome confounding that may occur because patients who utilize opioids, or larger amounts of opioids, in the immediate perioperative period may systematically differ from patients who do not in ways that are also correlated with longer-term opioid utilization.

Our results have potential policy implications given a growing interest in limiting opioid prescribing after surgery though the use of clinical guidelines and legislation.^25,26  Since variation in opioid utilization during the immediate perioperative period appears to be associated with clinically insignificant effects on opioid utilization in the mid- and long-term postoperative periods, our results suggest that further efforts to implement these policies should consider take into account the possibility that these initiatives may not reduce mid- and long-term opioid utilization and weigh this fact against the potential costs of these policies, such as less well controlled pain.

Our study had several limitations. First, our study was limited to patients undergoing total knee arthroplasty, a procedure that is often indicated for the relief of pain, and the results may be different for other procedures (i.e., appendectomy) for which pain is not an indication. Second, our study was limited to Medicare fee-for-service patients and may not apply to other populations. However, Medicare patients comprise the majority of total knee arthroplasty patients.²⁷  Moreover, older adults may be particularly susceptible to risks associated with prolonged opioid utilization.^28,29  Third, while surgeons play an important role in opioid prescribing during the immediate perioperative period, we cannot exclude the possibility that some opioid prescribing during the immediate perioperative period is outside of the surgeon’s control, which could result in misclassification of surgeons into the high-intensity group. Along these lines, we were only able to observe opioid utilization (i.e., filled prescriptions) as opposed to actual prescribing behaviors or opioid consumption. Fourth, since this is an observational study, our findings should be interpreted as associations rather than causal relationships; however, our approach did utilize instrumental variable approach to minimize confounding from unobservable patient factors. Some of these potential factors include opioids administered during the hospital stay and the use of prescription adjuncts such as nonsteroidal anti-inflammatory drug. Fifth, our sample consisted of patients undergoing surgery from 2011 to 2016, and subsequent to this period, many laws and policies were enacted aimed at limiting opioid prescribing.^30,31 

In conclusion, patients undergoing total knee arthroplasty by high-intensity surgeons—those whose patients were more likely to utilize opioids in the immediate perioperative period—had slightly higher opioid utilization in the subsequent 8 to 90 days after surgery and slightly lower opioid utilization in the subsequent 91 to 365 days after surgery compared to patients who underwent the procedure by surgeons with a lower propensity to prescribe opioids in the perioperative setting. However, while statistically significant, both of these differences were of small clinical significance. Our findings, along with previous work on opioid prescribing after surgery, have implications for the development of clinical practice guidelines in reducing inappropriate opioid prescribing under the framework proposed by a report from the National Academies of Sciences, Engineering, and Medicine.²⁶  Future research may explore surgeon-level prescribing behaviors across a range of surgeries and quantify thresholds of overuse. Such studies could provide insights into clinical guidelines for pain management strategies associated with surgery.

Supported by National Institutes of Aging (Bethesda‚ Maryland) grant No. R56 AG059620 and Agency for Healthcare Research and Quality (Rockville, Maryland) grant No. R01 HS26753-01A1 (to Dr. Jena); National Institute on Drug Abuse (Bethesda‚ Maryland) grant No. K08 DA042314 (to Dr. Sun); and National Institute on Drug Abuse grant No. K24 DA029262 (to Dr. Mackey). Data for this project were accessed using the Stanford Center for Population Health Sciences Data Core (Palo Alto‚ California), which is supported by National Institutes of Health National Center for Advancing Translational Science (Bethesda‚ Maryland) Clinical and Translational Science award No. UL1 TR001085 and by internal Stanford (Palo Alto‚ California) funding. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the paper.

Dr. Sun is on the advisory board of Lucid Lane, LLC (Los Altos‚ California) and reports receiving consulting fees unrelated to this work from Analysis Group (Boston‚ Massachusetts). Dr. Jena reports receiving (in the last 36 months) consulting fees unrelated to this work from Bioverativ (Waltham‚ Massachusetts), Merck/Sharp/Dohme (Kenilworth, New Jersey), Janssen (Titusville, New Jersey), Edwards Life Sciences (Irvine, California), Novartis (East Hanover‚ New Jersey), Amgen (Thousand Oaks, California), Eisai (Nutley‚ New Jersey), Otsuka Pharmaceuticals (Princeton‚ New Jersey), Vertex Pharmaceuticals (Boston, Massachusetts), Celgene (Summit‚ New Jersey), Sanofi Aventis (Bridgewater‚ New Jersey), Precision Health Economics and Analysis Group (Boston‚ Massachusetts). Dr. Jena also reports receiving (in the last 36 months) income unrelated to this work from hosting the podcast Freakonomics, M.D. (Boston, Massachusetts), and from book rights from Doubleday Books (New York‚ New York). Dr. Jena also reports being retained as an expert witness in lawsuits against opioid manufacturers and distributors. Dr. Mackey reports receiving research funding unrelated to this work from the Redlich Professorship and Rosenkrans Pain Research Endowment Fund at the Stanford University School of Medicine (Palo Alto, California), the Stanford Wu Tsai Neurosciences Institute (Palo Alto, California), the UCSF-Stanford Center of Excellence in Regulatory Science and Innovation (Palo Alto, California), the National Institutes of Health (R61NS118651, R03HD094577, R01DA045027, R01NS109450, R01AT008561, R01DA035484, P01AT00665105; Bethesda, Maryland), and the Patient Centered Outcomes Research Institute (Washington, DC). Dr. Mackey reports travel expenses and/or honoraria unrelated to this work from the Walter Reed National Military Medical Center (Bethesda, Maryland), The Georgetown University School of Medicine (Washington, DC), Harvard University (Cambridge, Massachusetts), the American Academy for Pain Medicine (Orlando, Florida), Washington University (St. Louis, Missouri), the US Food and Drug Administration (Silver Spring, Maryland), the National Institutes of Health (Bethesda, Maryland), the University of Washington (Seattle, Washington), George Washington University (Washington, DC), New York University (New York, New York), Weill Cornell Medical College (New York, New York), and the Canadian Pain Society (Markham, Ontario). Dr. Mackey reports consulting fees unrelated to this work from the American Society of Anesthesiologists (Schaumburg, Illinois), Fain, Anderson, VanDerhoef, Rosendahl, O’Halloran, and Spillane, PLLC (Seattle, Washington), Cox, Wootton, Lerner, Griffin, and Hansen (San Francisco, California), Lewis Brisbois Bisgaard and Smith (Los Angeles, California), Muro and Lampe (Folsom, California), and the Oklahoma Health Sciences Center (Norman, Oklahoma). Dr. Cen declares no competing interests.

Statistical Model Description and Supplemental Results, http://links.lww.com/ALN/C856