Preoperative Opioid Use Is Associated with Early Revision After Total Knee Arthroplasty: A Study of Male Patients Treated in the Veterans Affairs System

Ben-Ari, Alon MD; Chansky, Howard MD; Rozet, Irene MD

Journal of Bone & Joint Surgery - American Volume: 4 January 2017 - Volume 99 - Issue 1 - p 1–9
doi: 10.2106/JBJS.16.00167
Scientific Articles

Background: Opioid use is endemic in the U.S. and is associated with morbidity and mortality. The impact of long-term opioid use on joint-replacement outcomes remains unknown. We tested the hypothesis that use of opioids is associated with adverse outcomes after total knee arthroplasty (TKA).

Methods: We performed a retrospective analysis of patients who had had TKA within the U.S. Veterans Affairs (VA) system over a 6-year period and had been followed for 1 year postoperatively. The length of time for which an opioid had been prescribed and the morphine equivalent dose were calculated for each patient. Patients for whom opioids had been prescribed for >3 months in the year prior to the TKA were assigned to the long-term opioid group. A natural language processing-based machine-learning classifier was developed to classify revisions due to infectious and non-infectious causes on the basis of the postoperative note. Survival curves for the time to knee revision or manipulation were used to compare the long-term opioid group with the patients who did not take opioids long-term. Hazard and odds ratios for knee revision and manipulation were obtained as well.

Results: Of 32,636 patients (94.4% male; mean age [and standard deviation], 64.45 ± 9.41 years) who underwent TKA, 12,772 (39.1%) were in the long-term opioid group and 734 (2.2%) had a revision within a year after the TKA. Chronic kidney disease, diabetes, and long-term opioid use were associated with revision within 1 year—with odds ratios (95% confidence intervals [CIs]) of 1.76 (1.37 to 2.22), 1.11 (0.93 to 1.31), and 1.40 (1.19 to 1.64), respectively—and were also the leading factors associated with a revision at any time after the index TKA—with odds ratios (95% CIs) of 1.61 (1.34 to 1.92), 1.21 (1.08 to 1.36), and 1.28 (1.15 to 1.43), respectively. Long-term opioid use had a hazard ratio of 1.19 (95% CI = 1.10 to 0.24) in the analysis of its relationship with knee revision, but the hazard was not significant in the analysis of its association with knee manipulation. The accuracy of the text classifier was 0.94, with the area under the receiver operating characteristic curve being 0.99. There was no association between long-term use of opioids and the specific cause for knee revision.

Conclusions: Long-term opioid use prior to TKA was associated with an increased risk of knee revision during the first year after TKA among predominantly male patients treated in the VA system.

Level of Evidence: Therapeutic Level III. See Instructions for Authors for a complete description of levels of evidence.

1Department of Anesthesiology and Pain Medicine (A.B.-A. and I.R.) and Department of Orthopaedic Surgery and Sports Medicine (H.C.), VA Puget Sound Health Care System, University of Washington, Seattle, Washington

E-mail address for A. Ben-Ari: alon.benari@gmail.com

E-mail address for H. Chansky: chansky@uw.edu

E-mail address for I. Rozet: irozet@uw.edu

Article Outline

Liberal prescription of opioids is endemic in the U.S. and is on the rise1. Mortality associated with prescription of opioids has been widely covered in the national media2,3, National Institutes of Health (NIH) reports, and surveys1,4-6. Moreover, chronic prescription of opioids to relieve musculoskeletal pain in patients without cancer has been reviewed, and its efficacy has been questioned7-9. The high prevalence of opioid prescription and associated morbidity and mortality led for a call for the establishment of a new discipline of addiction medicine10. In light of this, the U.S. Centers for Disease Control and Prevention (CDC) revised its opioid prescription guidelines11. Given the high prevalence of opioid prescriptions, many patients are on long-term opioid therapy before they present for surgery. The issue of long-term opioid therapy before total knee arthroplasty (TKA) has been discussed12-14, and there is a wide body of literature with regard to risk factors associated with TKA failures15,16. Surprisingly, with the exception of a single report suggesting an association between preoperative chronic use of opioids and lower Knee Society Scores and a higher complication rate following TKA compared with controls17,18, to our knowledge not much is known about the epidemiology of chronic preoperative use of opioids and its possible impact on orthopaedic surgical outcomes. A substantial number of patients treated in the U.S. Veterans Affairs (VA) system are long-term opioid users, which led to an Inspector General report commissioned by the U.S. Senate19.

In this study, we aimed to determine whether, after we controlled for relevant confounding variables, opioid use altered the risk of knee revision and knee manipulation in the first year following a primary TKA.

Back to Top | Article Outline

Materials and Methods

Following approval from the local institutional review board, the national VA databases20 were queried using Current Procedural Terminology (CPT) codes 27446 (unicompartmental knee arthroplasty), 27438 (patellar arthroplasty), and 27447 (TKA) for procedures performed within the VA system nationwide from January 1, 2006, to January 1, 2012. In addition, the databases were queried for knee manipulations (CPT code 27570) and knee revisions (CPT codes 27487 and 27486) done from January 1, 2006, to January 1, 2013, thus allowing at least 1 year of follow-up after arthroplasties done by January 1, 2012. Patients were matched by patient identification (ID) numbers and the operatively treated side, with the latter determined by text-mining of the procedure description text field using regular expressions21. Patients were excluded if they (1) had had both knees replaced at the same time, (2) had had a knee revision in the VA system but the primary knee arthroplasty on the revised side could not be identified in the VA system, (3) had dates of procedures that could not be reconciled (for example, a revision before a primary arthroplasty on the same side), (4) were on long-term opioid therapy for which a morphine equivalent dose (MED) could not be calculated, (5) had procedures for which the operative duration had been documented as 0 minutes, or (6) had a CPT code other than 27447. When a patient had had both knees replaced at different times within the study period, only the first procedure was included in the cohort.

Abstracted data included demographic information and weight within 6 months prior to the TKA. Relevant comorbidities (diabetes, congestive heart failure, hypertension, chronic kidney disease, obstructive sleep apnea, depression within 1 year before the surgery, and posttraumatic stress disorder [PTSD]) were abstracted using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes. We did not include smoking because the ICD coding does not capture smoking status accurately. The outpatient pharmacy records of opioid prescriptions during the year before the TKA were queried for all patients in the cohort. Dosage, duration of the prescription, number of times that the patient filled the prescription, time between fills, and brand names of the drugs were abstracted. Brand names were then converted to generic names using text mining for each patient. The MED was calculated for each patient by using a conversion table22. Only VA prescriptions were included. For each patient, we counted the number of unique opioids prescribed and the number of months between consecutive refills in the year before the TKA. Patients were considered to be on long-term opioid therapy if opioids had been prescribed consecutively for longer than 3 months in the year before the TKA. The underlying cause of revision—i.e., whether or not it was due to infection—was abstracted from the full text of the postoperative surgical note using natural language processing methods and a machine-learning classifier (see Appendix).

Back to Top | Article Outline

Analysis

Survival curves were plotted to describe the time to knee manipulation or knee revision within a year after TKA for patients on long-term opioid therapy compared with those who were not. We used the Kaplan-Meier estimator and the log-rank test to assess the statistical difference between the 2 patient populations. As knee revision may necessitate >1 surgical intervention (debridement, placement of an antibiotic-loaded spacer, etc.), some patients had undergone several surgical interventions to complete the procedure. For the primary end point, we narrowed the scope of our analysis to 1 year following TKA and analyzed the time to knee revision or manipulation only in that period. The Cox proportional hazard model was used to calculate the hazard of knee revision or knee manipulation being needed within a year following TKA. The Cox proportional model was fit using long-term opioid use and known risk and protective factors for knee revision, such as diabetes and age. A logistic regression model was fit to the data to calculate the odds ratio for knee revision within 1 year, and any time following TKA in patients treated with long-term opioid therapy. A full model, including all major comorbidities (congestive heart failure, chronic kidney disease, obstructive sleep apnea, depression, PTSD, and diabetes), age, obesity (defined as a body mass index [BMI] of >30 kg/m2), and a history of long-term opioid therapy, was created. Backward model selection was done to select the best model. From the selected model, the odds ratio and 95% confidence interval (CI) for knee revision in the year following TKA were obtained. Estimates are reported as the mean and standard deviation or median and interquartile range as appropriate. A p value of <0.05 was considered significant. The queries and code for analysis were written in SQL (Structured Query Language), R23, and Python24.

Back to Top | Article Outline

Results

In the VA system, 39,636 TKAs were performed from January 1, 2006, to January 1, 2012, and 7,097 knee revisions and 1,634 knee manipulations were done from January 1, 2006, to January 1, 2013. Surgical laterality could not be determined for 407 TKAs, 218 revisions, and 76 knee manipulations, which were excluded. The generic name and opioid dose were abstracted from 165,816 prescriptions; only 54 prescriptions could not be assigned to any generic opioid drug. After exclusion criteria were applied, 32,636 patients were included in the study cohort (Fig. 1). The mean age of the cohort was 64.45 ± 9.41 years; 94.4% of the patients were male. A total of 1,645 (5.04%) of the patients had undergone at least 1 revision, and 734 revisions were done within the first year following TKA. The total number of manipulations was 1,012 (3.1%). The height of 4,494 patients and the weight of 758 patients were not recorded. The BMI was known for 27,850 patients, for whom it averaged 32.04 ± 8.02 kg/m2. A comprehensive search for the postoperative notes on all first knee revisions was attempted; the notes were not found for 223 (13.6%). A machine learning text classifier applied to the entire set of surgical notes to detect the etiology of the revision identified 597 revisions due to infection (36.3%). The classifier demonstrated an accuracy of 0.94.

Opioids were used long-term by 12,772 patients (39.1%) in the year before the TKA. The mean age of those patients was 62.38 ± 9.02 years compared with 65.8 ± 9.4 years for the patients not treated with long-term opioid therapy, with no significant difference between the 2 groups (Fig. 2). Figure 3 demonstrates that more knee revisions were done in the patients on long-term opioids during the study period. We found an association between the proportion of patients who had undergone at least 1 knee revision and the long-term use of opioids (chi square = 34.44, p < 0.05).

The monthly MED was calculated for each patient on long-term opioid therapy and presented on a semi-logarithmic scale with projection histograms showing the distribution of MEDs and the cumulative time for which the patients were on opioids in the year before TKA (Fig. 4-A). The median MED was 832.5 mg/mo, with an interquartile range of 400 to 1,785 mg/mo. The median cumulative duration of opioid consumption in the long-term opioid therapy group was 10 months (interquartile range, 7 to 11 months). Figure 4-B depicts the distribution of the number of unique opioid formulations prescribed in the long-term opioid group; 67.2% (8,587) of the patients were prescribed >1 opioid formulation in that group. Following model selection and controlling for age, BMI, and clinically relevant comorbidities such as diabetes, depression, PTSD, congestive heart failure, and chronic kidney disease, a logistic regression model was fit with revision within 1 year and with revision at any time as dependent variables. We found that patients on long-term opioids before TKA were more likely to undergo revision within 12 months, with an odds ratio of 1.40 (95% CI, 1.19 to 1.64; p < 0.005). There was no difference in the likelihood of undergoing a knee manipulation between patients on long-term opioids and those not on long-term opioids. Table I shows the odds ratios for diabetes, long-term opioid use, and chronic kidney disease as markers for knee revision in the first year after TKA. Patients on long-term opioid therapy were more likely to undergo any knee revision during the study period, with an odds ratio of 1.28 (95% CI, 1.15 to 1.43). Table II shows the odds ratios for patients with diabetes, using opioids long-term, or with chronic kidney disease undergoing revision at any time. Older age was a protective factor in both models.

Survival curves for the time to revision within a year after TKA suggested that long-term opioid therapy in the year before surgery is associated with early knee revision following TKA (log-rank test, p < 0.05) (Fig. 5). There was no difference in the time to knee manipulation between the patients who were treated with long-term opioid therapy and those who were not (Fig. 6). The Cox proportional hazard model showed the relative risk of knee revision to be 19% higher in the long-term opioid group compared with the other group (Table III). Application of the machine-learning text classifier to the surgical notes on the revisions showed diabetes to be associated with revision for infection (chi square = 48.75, p < 0.05) but demonstrated no association between long-term use of opioids and the etiology of the revision.

Back to Top | Article Outline

Discussion

This study demonstrated that long-term use of prescription opioids is highly prevalent among the (mostly) male population undergoing knee replacement surgery in the VA system. After controlling for relevant clinical features, we found that male VA patients on long-term opioid therapy had an increased hazard of, and odds ratio for, undergoing knee revision within a year after TKA. Long-term opioid use was not associated with any specific cause of the revision. In agreement with others, we found diabetes to be a risk factor for revision due to surgical site infection25,26, although diabetes did not play a significant role in the short term after TKA. Regarding obesity, our cohort had a substantial lack of weight and height data, which may have impacted the results; since we did not distinguish between revisions due to infection and those due to mechanical failure in our analysis, we cannot comment on a specific association between failure and obesity. Of note, a study of 70,070 patients did not show obesity to significantly increase the relative risk for revision27. In agreement with our finding, opioid use was not associated with knee manipulation following TKA in a large cohort of 27 million patients28.

To our knowledge, this is the first study to describe the prevalence of opioid prescriptions in the year prior to TKA in a large cohort of patients. In our group of >30,000 mostly male VA patients, we found an association between opioid prescription before TKA and the prevalence of early knee revision surgery. Of equal importance is the fact that about 40% of the patients were prescribed non-trivial amounts of opioids for about a year prior to TKA. Knee arthroplasty is one of the most common procedures performed in the U.S., and a report29 predicted an increased demand for both primary and revision TKA. In that report29, the authors used demographic data, historical surgical data, and life tables to predict that there would be 3.48 million knee arthroplasties by 2030 and that the number of knee revisions would double in the period of 2005 to 2030. Our study suggests that, in light of the current trend of liberal opioid prescription, including to individuals who will require knee arthroplasty, demographic projections of future trends in knee arthroplasty may need to include the effects of prescription opioid use. It is possible that a conservative policy for opioid prescription may reduce the rate of knee revisions, with an ensuing decrease in associated morbidity and costs.

Regarding the generalizability of this study, we used VA data on a national scale and our findings probably reflect VA practice. However, the majority of TKAs are done in females, and we cannot comment on differences in prescription patterns between the sexes given that our data were derived mostly from men. Replication of this study in more balanced populations will determine its generalizability.

Our results agree with the previous (although not universal30) finding that diabetes is a risk factor for revision25,26 and the observation that older age is a protective factor for knee revision31,32. In addition, in agreement with others33, we found chronic kidney disease to be the leading risk factor for knee revision. The fact that 8.4% (2,741) of our patients had the diagnosis of chronic kidney disease makes a strong case that the presence of such disease needs to be a serious consideration in decision-making before TKA.

Because the etiologies of the revisions had not been captured in a structured way in the records on our large cohort, we developed a machine-learning classifier to sift through the 1,422 postoperative notes of patients who had had at least 1 revision. This is an interesting demonstration of abstracting clinically meaningful data from unstructured text, and the fact that we found a significant association between diabetes and revision due to infection lends support to our methodology. Long-term opioid use was not associated with any specific cause of revision. This suggests that the mechanisms behind the association between revision and long-term opioid use might be related to behavioral differences and/or the sociopsychological environment of the long-term opioid users. The notion that a patient’s psychological state may affect surgical outcome has been studied34,35, although the effect of anxiety and depression needs to be quantified in a prospective investigation. Use of administrative databases to address a clinical question was a limitation of our study because the databases lacked clinically pertinent data (for example, preoperative and postoperative range of motion and knee severity scores). Because these parameters are important in the clinical assessment of patients preoperatively, looking at them in aggregate may allow us to better characterize patient populations. Nonetheless, large databases are a credible source of robust estimates and have aided both researchers and industry analysts in finding relevant information36.

Despite the limitations of our study, we believe that our observation that long-term use of opioids before TKA places patients at higher risk for an early revision holds true. We believe that future work should focus on (1) development of a decision support system that enables a physician discussing TKA with a patient to quantify the risk of an early revision, (2) replicating our findings in balanced populations and with data sets that include risk factors associated with success and failure of TKAs along with medication history, and (3) comparing the effects of liberal and conservative opioid prescription regimens on TKA outcomes in a prospective study.

Back to Top | Article Outline

Appendix Cited Here...

A description of our method of classifying the causes of the revisions on the basis of the postoperative surgical notes is available with the online version of this article as a data supplement at jbjs.org.

NOTE: The authors thank Seth S. Leopold, MD, Department of Orthopaedics and Sports Medicine, University of Washington, for his insightful remarks, thorough reading of the manuscript, guidance, and substantial contribution in bringing this manuscript to its final form. They also thank the VA for lending the technical support in the execution of this study, especially Mr. William Carson from the Linux support team. Special thanks are extended to Ms. Rebecca Felkey for her assistance in editing the manuscript.

Investigation performed at the Department of Anesthesiology and Pain Medicine and Department of Orthopaedic Surgery and Sports Medicine, VA Puget Sound Health Care System, University of Washington, Seattle, Washington

A commentary by Michael S. Reich, MD, and Richard H. Walker, MD, is linked to the online version of this article at jbjs.org.

Disclosure: No external funding was received for this study. The Disclosure of Potential Conflicts of Interest forms are provided with the online version of this article.

Back to Top | Article Outline

References

1. Manchikanti L, Helm S 2nd, Fellows B, Janata JW, Pampati V, Grider JS, Boswell MV. Opioid epidemic in the United States. Pain Physician. 2012 ;15(3)(Suppl):ES9–38.
2. Catan T, Barrett D, Martin TW. Prescription for addiction. Wall Street J. 2012 . http://online.wsj.com/news/articles/SB10000872396390444223104578036933277566700. Accessed 2014 Sep 30.
3. The Seattle Times. Methadone and the politics of pain. 2012 . http://seattletimes.com/flatpages/specialreports/methadone/methadoneandthepoliticsofpain.html. Accessed 2016 Sep 12.
4. Centers for Disease Control and Prevention (CDC). Vital signs: overdoses of prescription opioid pain relievers—United States, 1999–2008. MMWR Morb Mortal Wkly Rep. 2011 ;60(43):1487–92.
5. Centers for Disease Control and Prevention (CDC). Drug overdose deaths—Florida, 2003-2009. MMWR Morb Mortal Wkly Rep. 2011 ;60(26):869–72.
6. Centers for Disease Control and Prevention (CDC). Overdose deaths involving prescription opioids among Medicaid enrollees - Washington, 2004-2007. MMWR Morb Mortal Wkly Rep. 2009 ;58(42):1171–5.
7. Krashin D, Sullivan M, Ballantyne J. What are we treating with chronic opioid therapy? Curr Rheumatol Rep. 2013 ;15(3):311.
8. Ballantyne JC, Sullivan MD. Why doctors prescribe opioids to known opioid abusers. N Engl J Med. 2013 ;368(5):484–5.
9. Krashin D, Murinova N, Ballantyne J. Management of pain with comorbid substance abuse. Curr Psychiatry Rep. 2012 ;14(5):462–8.
10. O’Connor PG, Sokol RJ, D’Onofrio G. Addiction medicine: the birth of a new discipline. JAMA Intern Med. 2014 ;174(11):1717–8.
11. Centers for Disease Control and Prevention (CDC). CDC Morbidity and Mortality Weekly Report. CDC guidelines for prescribing opioids for chronic pain— United States, 2016. 2016 Mar 15. http://www.cdc.gov/media/modules/dpk/2016/dpk-pod/rr6501e1er-ebook.pdf. Accessed 2016 Jun 21.
12. Devin CJ, Lee DS, Armaghani SJ, Bible J, Shau DN, Martin PR, Ehrenfeld JM. Approach to pain management in chronic opioid users undergoing orthopaedic surgery. J Am Acad Orthop Surg. 2014 ;22(10):614–22.
13. Buvanendran A, Kroin JS, Della Valle CJ, Kari M, Moric M, Tuman KJ. Perioperative oral pregabalin reduces chronic pain after total knee arthroplasty: a prospective, randomized, controlled trial. Anesth Analg. 2010 ;110(1):199–207. Epub 2009 Nov 12.
14. Buvanendran A, Kroin JS, Tuman KJ, Lubenow TR, Elmofty D, Moric M, Rosenberg AG. Effects of perioperative administration of a selective cyclooxygenase 2 inhibitor on pain management and recovery of function after knee replacement: a randomized controlled trial. JAMA. 2003 ;290(18):2411–8.
15. Chang MJ, Lim H, Lee NR, Moon YW. Diagnosis, causes and treatments of instability following total knee arthroplasty. Knee Surg Relat Res. 2014 ;26(2):61–7. Epub 2014 May 30.
16. Hofmann S, Seitlinger G, Djahani O, Pietsch M. The painful knee after TKA: a diagnostic algorithm for failure analysis. Knee Surg Sports Traumatol Arthrosc. 2011 ;19(9):1442–52. Epub 2011 Aug 6.
17. Zywiel MG, Stroh DA, Lee SY, Bonutti PM, Mont MA. Chronic opioid use prior to total knee arthroplasty. J Bone Joint Surg Am. 2011 ;93(21):1988–93.
18. Parker Vail T. Preoperative pain management decisions impact outcome after total knee arthroplasty-implications for opiate use: commentary on an article by Michael G. Zywiel, MD, et al.: “Chronic opioid use prior to total knee arthroplasty”. J Bone Joint Surg Am. 2011 ;93(21):e1291–1.
19. Department of Veterans Affairs Office of Inspector General. Healthcare Inspection-VA patterns of dispensing take-home opioids and monitoring patients on opioid therapy. Office of Healthcare Inspections Report No. 14-00895-163. http://www.va.gov/oig/pubs/VAOIG-14-00895-163.pdf. Accessed 2016 Sep 12.
20. US Department of Veterans Affairs. VA Informatics and Computing Infrastructure (VINCI). 2014. http://www.hsrd.research.va.gov/for_researchers/vinci/. Accessed 2016 Sep 12.
21. Wikipedia. Regular expression. http://en.wikipedia.org/wiki/Regular_expression. Accessed 2014 November 8.
22. Wikipedia. Equianalgesic. https://en.wikipedia.org/wiki/Equianalgesic. Accessed 2016 Sep 12.
23. R Core Team, R Foundation for Statistical Computing. R: a language and environment for statistical computing. 2014. http://www.gbif.org/resource/81287. Accessed 2016 Sep 12.
24. Python Software Foundation. Python language reference, version 2.7. https://www.python.org/downloads/release/python-2712/. Accessed 2016 Sep 12.
25. Marchant MH Jr, Viens NA, Cook C, Vail TP, Bolognesi MP. The impact of glycemic control and diabetes mellitus on perioperative outcomes after total joint arthroplasty. J Bone Joint Surg Am. 2009 ;91(7):1621–9.
26. Iorio R, Williams KM, Marcantonio AJ, Specht LM, Tilzey JF, Healy WL. Diabetes mellitus, hemoglobin A1C, and the incidence of total joint arthroplasty infection. J Arthroplasty. 2012 ;27(5):726–9.e1. Epub 2011 Nov 4.
27. Pfefferle KJ, Gil KM, Fening SD, Dilisio MF. Validation study of a pooled electronic healthcare database: the effect of obesity on the revision rate of total knee arthroplasty. Eur J Orthop Surg Traumatol. 2014 ;24(8):1625–8. Epub 2014 Feb 16.
28. Pfefferle KJ, Shemory ST, Dilisio MF, Fening SD, Gradisar IM. Risk factors for manipulation after total knee arthroplasty: a pooled electronic health record database study. J Arthroplasty. 2014 ;29(10):2036–8. Epub 2014 May 9.
29. Kurtz S, Ong K, Lau E, Mowat F, Halpern M. Projections of primary and revision hip and knee arthroplasty in the United States from 2005 to 2030. J Bone Joint Surg Am. 2007 ;89(4):780–5.
30. Adams AL, Paxton EW, Wang JQ, Johnson ES, Bayliss EA, Ferrara A, Nakasato C, Bini SA, Namba RS. Surgical outcomes of total knee replacement according to diabetes status and glycemic control, 2001 to 2009. J Bone Joint Surg Am. 2013 ;95(6):481–7.
31. Paxton EW, Inacio MC, Khatod M, Yue E, Funahashi T, Barber T. Risk calculators predict failures of knee and hip arthroplasties: findings from a large health maintenance organization. Clin Orthop Relat Res. 2015 ;473(12):3965–73. Epub 2015 Sep 1.
32. Dy CJ, Marx RG, Bozic KJ, Pan TJ, Padgett DE, Lyman S. Risk factors for revision within 10 years of total knee arthroplasty. Clin Orthop Relat Res. 2014 ;472(4):1198–207. Epub 2013 Dec 18.
33. Deegan BF, Richard RD, Bowen TR, Perkins RM, Graham JH, Foltzer MA. Impact of chronic kidney disease stage on lower-extremity arthroplasty. Orthopedics. 2014 ;37(7):e613–8.
34. Lavernia CJ, Villa JM, Iacobelli DA. What is the role of mental health in primary total knee arthroplasty? Clin Orthop Relat Res. 2015 ;473(1):159–63.
35. Blackburn J, Qureshi A, Amirfeyz R, Bannister G. Does preoperative anxiety and depression predict satisfaction after total knee replacement? Knee. 2012 ;19(5):522–4. Epub 2011 Aug 16.
36. Raghupathi W, Raghupathi V. Big data analytics in healthcare: promise and potential. Health Inf Sci Syst. 2014;2:3. Epub 2014 Feb 7.
37. Porter MF. An algorithm for suffix stripping. Program. 1980;14:130–7.
38. Cortes C, Vapnik V. Support-vector networks. Mach Learn. 1995;20:273–297.
39. Bird S, Klein E, Loper E. Natural language processing with Python. Sebastopol, CA: O’Reilly Media; 2009.
40. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Prettenhofer P, Weiss R, Dubourg V, Vanderplas J, Passos A, Cournapeau D, Brucher M, Perrot M, Duchesnay E. Scikit-learn: machine learning in Python. J Mach Learn Res. 2011;12:2825–30.
Copyright 2017 by The Journal of Bone and Joint Surgery, Incorporated