Association of Surgical Treatment With Adverse Events and Mortality Among Medicare Beneficiaries With Proximal Humerus Fracture | Orthopedics | JAMA Network Open | JAMA Network
[Skip to Content]
Sign In
Individual Sign In
Create an Account
Institutional Sign In
OpenAthens Shibboleth
[Skip to Content Landing]
Table 1.  Assumptions Required and Treatment Effect Parameters Estimated for Risk-Adjusted Regression and Instrumental Variable Estimators
Assumptions Required and Treatment Effect Parameters Estimated for Risk-Adjusted Regression and Instrumental Variable Estimators
Table 2.  Medicare Proximal Humerus Fracture Patient Characteristics by Treatment Group
Medicare Proximal Humerus Fracture Patient Characteristics by Treatment Group
Table 3.  Characteristics of Medicare Patients With Proximal Humerus Fracture in 2011 by Local Area Surgery Ratios
Characteristics of Medicare Patients With Proximal Humerus Fracture in 2011 by Local Area Surgery Ratios
Table 4.  Risk-Adjusted Regression and Instrumental Variable Estimates of the Association of Surgery With Adverse Events and Mortality for Medicare Patients With Proximal Humerus Fracture in 2011 and Stratified Subgroups
Risk-Adjusted Regression and Instrumental Variable Estimates of the Association of Surgery With Adverse Events and Mortality for Medicare Patients With Proximal Humerus Fracture in 2011 and Stratified Subgroups
Table 5.  Risk-Adjusted Regression and Instrumental Variable Estimates for the Association Between Surgery Utilization and Cost Outcomes for Medicare Patients With Proximal Humerus Fracture in 2011 and Stratified Subgroups
Risk-Adjusted Regression and Instrumental Variable Estimates for the Association Between Surgery Utilization and Cost Outcomes for Medicare Patients With Proximal Humerus Fracture in 2011 and Stratified Subgroups
1.
Sabharwal  S, Patel  NK, Griffiths  D, Athanasiou  T, Gupte  CM, Reilly  P.  Trials based on specific fracture configuration and surgical procedures likely to be more relevant for decision making in the management of fractures of the proximal humerus: findings of a meta-analysis.  Bone Joint Res. 2016;5(10):470-480. doi:10.1302/2046-3758.510.2000638PubMedGoogle ScholarCrossref
2.
Bell  JE, Leung  BC, Spratt  KF,  et al.  Trends and variation in incidence, surgical treatment, and repeat surgery of proximal humeral fractures in the elderly.  J Bone Joint Surg Am. 2011;93(2):121-131. doi:10.2106/JBJS.I.01505PubMedGoogle ScholarCrossref
3.
Floyd  SB, Campbell  J, Chapman  CG, Thigpen  CA, Kissenberth  MJ, Brooks  JM.  Geographic variation in the treatment of proximal humerus fracture: an update on surgery rates and treatment consensus.  J Orthop Surg Res. 2019;14(1):22. doi:10.1186/s13018-018-1052-2PubMedGoogle ScholarCrossref
4.
Floyd  SB, Chapman  CG, Thigpen  CA, Brooks  JM, Hawkins  RJ, Tokish  JM.  Shoulder arthroplasty in the US Medicare population: a 1-year evaluation of surgical complications, hospital admissions, and revision surgery.  JSES Open Access. 2018;2(1):40-47. doi:10.1016/j.jses.2017.10.002PubMedGoogle ScholarCrossref
5.
Birkmeyer  JD, Reames  BN, McCulloch  P, Carr  AJ, Campbell  WB, Wennberg  JE.  Understanding of regional variation in the use of surgery.  Lancet. 2013;382(9898):1121-1129. doi:10.1016/S0140-6736(13)61215-5PubMedGoogle ScholarCrossref
6.
McCulloch  P, Nagendran  M, Campbell  WB,  et al.  Strategies to reduce variation in the use of surgery.  Lancet. 2013;382(9898):1130-1139. doi:10.1016/S0140-6736(13)61216-7PubMedGoogle ScholarCrossref
7.
Bernstein  J.  Not the last word: Choosing Wisely.  Clin Orthop Relat Res. 2015;473(10):3091-3097. doi:10.1007/s11999-015-4490-8PubMedGoogle ScholarCrossref
8.
Wennberg  J.  Which rate is right?  N Engl J Med. 1986;314(5):310-311. doi:10.1056/NEJM198601303140509PubMedGoogle ScholarCrossref
9.
Brooks  JM, Chapman  CG, Suneja  M,  et al.  Use of angiotensin-converting enzyme inhibitors and angiotensin receptor blockers for geriatric ischemic stroke patients: are the rates right?  J Am Heart Assoc. 2018;7(11):e009137. doi:10.1161/JAHA.118.009137PubMedGoogle Scholar
10.
Brooks  JM, Chrischilles  EA.  Heterogeneity and the interpretation of treatment effect estimates from risk adjustment and instrumental variable methods.  Med Care. 2007;45(10)(suppl):S123-S130. doi:10.1097/MLR.0b013e318070c069PubMedGoogle ScholarCrossref
11.
Brooks  JM, Cook  E, Chapman  CG,  et al.  Statin use after acute myocardial infarction by patient complexity: are the rates right?  Med Care. 2015;53(4):324-331. doi:10.1097/MLR.0000000000000322PubMedGoogle Scholar
12.
Phelps  CE.  Health Economics. 2nd ed. Reading, MA: Addison-Wesley; 1997.
13.
Park  T-R, Brooks  JM, Chrischilles  EA, Bergus  G.  Estimating the effect of treatment rate changes when treatment benefits are heterogeneous: antibiotics and otitis media.  Value Health. 2008;11(2):304-314. doi:10.1111/j.1524-4733.2007.00234.xPubMedGoogle ScholarCrossref
14.
Brooks  JM, McClellan  M, Wong  HS.  The marginal benefits of invasive treatments for acute myocardial infarction: does insurance coverage matter?  Inquiry. 2000;37(1):75-90.PubMedGoogle Scholar
15.
Harris  KM, Remler  DK.  Who is the marginal patient? understanding instrumental variables estimates of treatment effects.  Health Serv Res. 1998;33(5):1337-1360.PubMedGoogle Scholar
16.
McClellan  M, Newhouse  JP.  Overview of the special supplement issue.  Health Serv Res. 2000;35(5):1061-1069.PubMedGoogle Scholar
17.
Brooks  JM, Chapman  CG, Schroeder  MC.  Understanding treatment effect estimates when treatment effects are heterogeneous for more than one outcome.  Appl Health Econ Health Policy. 2018;16(3):381-393. doi:10.1007/s40258-018-0380-zPubMedGoogle ScholarCrossref
18.
Fang  G, Brooks  JM, Chrischilles  EA.  Apples and oranges? interpretations of risk adjustment and instrumental variable estimates of intended treatment effects using observational data.  Am J Epidemiol. 2012;175(1):60-65. doi:10.1093/aje/kwr283PubMedGoogle ScholarCrossref
19.
Brooks  JM, Chapman  CG, Cozad  MJ.  The identification process using choice theory is needed to match design with objectives in CER.  Med Care. 2017;55(2):91-93. doi:10.1097/MLR.0000000000000615PubMedGoogle ScholarCrossref
20.
Garrido  MM, Dowd  B, Hebert  PL, Maciejewski  ML.  Understanding treatment effect terminology in pain and symptom management research.  J Pain Symptom Manage. 2016;52(3):446-452. doi:10.1016/j.jpainsymman.2016.01.016PubMedGoogle ScholarCrossref
21.
McGuire  KJ, Bernstein  J, Polsky  D, Silber  JH.  The 2004 Marshall Urist award: delays until surgery after hip fracture increases mortality.  Clin Orthop Relat Res. 2004;(428):294-301. doi:10.1097/01.blo.0000146743.28925.1cPubMedGoogle Scholar
22.
Charlson  ME, Pompei  P, Ales  KL, MacKenzie  CR.  A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.  J Chronic Dis. 1987;40(5):373-383. doi:10.1016/0021-9681(87)90171-8PubMedGoogle ScholarCrossref
23.
Chrischilles  E, Schneider  K, Wilwert  J,  et al.  Beyond comorbidity: expanding the definition and measurement of complexity among older adults using administrative claims data.  Med Care. 2014;52(suppl):S75-S84. doi:10.1097/MLR.0000000000000026PubMedGoogle ScholarCrossref
24.
Angrist  JD.  Treatment effect heterogeneity in theory and practice.  Econ J (Lond). 2004;114:C52-C83. doi:10.1111/j.0013-0133.2003.00195.xGoogle Scholar
25.
Heckman  JJ.  Econometric causality.  Int Stat Rev. 2008;76(1):1-27. doi:10.1111/j.1751-5823.2007.00024.xGoogle ScholarCrossref
26.
Heckman  JJ, Urzua  S, Vytlacil  E.  Understanding instrumental variables in models with essential heterogeneity.  Rev Econ Stat. 2006;88(3):389-432. doi:10.1162/rest.88.3.389Google ScholarCrossref
27.
Newhouse  JP, McClellan  M.  Econometrics in outcomes research: the use of instrumental variables.  Annu Rev Public Health. 1998;19:17-34. doi:10.1146/annurev.publhealth.19.1.17PubMedGoogle ScholarCrossref
28.
Imbens  GW, Angrist  JD.  Identification and estimation of local average treatment effects.  Econometrica. 1994;62(2):467-475. doi:10.2307/2951620Google ScholarCrossref
29.
Greenland  S, Morgenstern  H.  Confounding in health research.  Annu Rev Public Health. 2001;22:189-212. doi:10.1146/annurev.publhealth.22.1.189PubMedGoogle ScholarCrossref
30.
Heckman  JJ, Robb  R. Alternative methods for evaluating the impact of interventions. In: Heckman  JJ, Singer  B, eds.  Longitudinal Analysis of Labor Market Data. New York, NY: Cambridge University Press; 1985:156-245. doi:10.1017/CCOL0521304539.004
31.
Chapman  CG, Cozad  MJ, Brooks  JM.  Conditions for generating treatment effect estimates in line with objectives: beyond confounding.  Med Care. 2017;55(2):97-99. doi:10.1097/MLR.0000000000000614PubMedGoogle ScholarCrossref
32.
Brooks  JM, Fang  G.  Interpreting treatment-effect estimates with heterogeneity and choice: simulation model results.  Clin Ther. 2009;31(4):902-919. doi:10.1016/j.clinthera.2009.04.007PubMedGoogle ScholarCrossref
33.
Rassen  JA, Brookhart  MA, Glynn  RJ, Mittleman  MA, Schneeweiss  S.  Instrumental variables II: instrumental variable application—in 25 variations, the physician prescribing preference generally was strong and reduced covariate imbalance.  J Clin Epidemiol. 2009;62(12):1233-1241. doi:10.1016/j.jclinepi.2008.12.006PubMedGoogle ScholarCrossref
34.
Brooks  JM, Chrischilles  EA, Scott  SD, Chen-Hardee  SS.  Was breast conserving surgery underutilized for early stage breast cancer? instrumental variables evidence for stage II patients from Iowa.  Health Serv Res. 2003;38(6 Pt 1):1385-1402. doi:10.1111/j.1475-6773.2003.00184.xPubMedGoogle ScholarCrossref
35.
Tang  Y, Brooks  JM, Wetmore  JB, Shireman  TI.  Association between higher rates of cardioprotective drug use and survival in patients on dialysis.  Res Social Adm Pharm. 2015;11(6):824-843. doi:10.1016/j.sapharm.2014.12.007PubMedGoogle ScholarCrossref
36.
Polgreen  LA, Cook  EA, Brooks  JM, Tang  Y, Polgreen  PM.  Increased statin prescribing does not lower pneumonia risk.  Clin Infect Dis. 2015;60(12):1760-1766. doi:10.1093/cid/civ190PubMedGoogle ScholarCrossref
37.
Brooks  JM, Chrischilles  EA, Landrum  MB,  et al.  Survival implications associated with variation in mastectomy rates for early-staged breast cancer.  Int J Surg Oncol. 2012;2012:127854. doi:10.1155/2012/127854PubMedGoogle Scholar
38.
Brooks  JM, Tang  Y, Chapman  CG, Cook  EA, Chrischilles  EA.  What is the effect of area size when using local area practice style as an instrument?  J Clin Epidemiol. 2013;66(8)(suppl):S69-S83. doi:10.1016/j.jclinepi.2013.04.008PubMedGoogle ScholarCrossref
39.
Brooks  JM, Cook  EA, Chapman  CG,  et al.  Geographic variation in statin use for complex acute myocardial infarction patients: evidence of effective care?  Med Care. 2014;52(suppl):S37-S44. doi:10.1097/MLR.0b013e3182a7fc3dPubMedGoogle ScholarCrossref
40.
Brookhart  MA, Wang  PS, Solomon  DH, Schneeweiss  S.  Evaluating short-term drug effects using a physician-specific prescribing preference as an instrumental variable.  Epidemiology. 2006;17(3):268-275. doi:10.1097/01.ede.0000193606.58671.c5PubMedGoogle ScholarCrossref
41.
Stukel  TA, Fisher  ES, Wennberg  DE, Alter  DA, Gottlieb  DJ, Vermeulen  MJ.  Analysis of observational studies in the presence of treatment selection bias: effects of invasive cardiac management on AMI survival using propensity score and instrumental variable methods.  JAMA. 2007;297(3):278-285. doi:10.1001/jama.297.3.278PubMedGoogle ScholarCrossref
42.
Fang  G, Brooks  JM, Chrischilles  EA.  Comparison of instrumental variable analysis using a new instrument with risk adjustment methods to reduce confounding by indication.  Am J Epidemiol. 2012;175(11):1142-1151. doi:10.1093/aje/kwr448PubMedGoogle ScholarCrossref
43.
Schroeder  MC, Tien  YY, Wright  K, Halfdanarson  TR, Abu-Hejleh  T, Brooks  JM.  Geographic variation in the use of adjuvant therapy among elderly patients with resected non-small cell lung cancer.  Lung Cancer. 2016;95:28-34. doi:10.1016/j.lungcan.2016.02.010PubMedGoogle ScholarCrossref
44.
Fang  G, Brooks  JM, Chrischilles  EA.  A new method to isolate local-area practice styles in prescription use as the basis for instrumental variables in comparative effectiveness research.  Med Care. 2010;48:710-717. doi:10.1097/MLR.0b013e3181e41bb2PubMedGoogle ScholarCrossref
45.
Cochran  WG.  Some methods for strengthening the common χ2 tests.  Biometrics. 1954;10(4):417-451. doi:10.2307/3001616Google ScholarCrossref
46.
Armitage  P.  Tests for linear trends in proportions and frequencies.  Biometrics. 1955;11:375-386. doi:10.2307/3001775Google ScholarCrossref
47.
Wooldridge  J.  Introductory Econometrics: A Modern Approach. Mason, OH: Thomson South-Western; 2003.
48.
Staiger  D, Stock  J.  Instrumental variables regression with weak instruments.  Econometrica. 1997;65(3):557-586. doi:10.2307/2171753Google ScholarCrossref
49.
Schnell  S, Friedman  SM, Mendelson  DA, Bingham  KW, Kates  SL.  The 1-year mortality of patients treated in a hip fracture program for elders.  Geriatr Orthop Surg Rehabil. 2010;1(1):6-14. doi:10.1177/2151458510378105PubMedGoogle ScholarCrossref
50.
Brauer  CA, Coca-Perraillon  M, Cutler  DM, Rosen  AB.  Incidence and mortality of hip fractures in the United States.  JAMA. 2009;302(14):1573-1579. doi:10.1001/jama.2009.1462PubMedGoogle ScholarCrossref
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Views 1,673
    Citations 0
    Original Investigation
    Orthopedics
    January 10, 2020

    Association of Surgical Treatment With Adverse Events and Mortality Among Medicare Beneficiaries With Proximal Humerus Fracture

    Author Affiliations
    • 1Center for Effectiveness Research in Orthopaedics, University of South Carolina, Greenville
    • 2Department of Health Services Policy and Management, Arnold School of Public Health, University of South Carolina, Columbia
    • 3ATI Physical Therapy, Greenville, South Carolina
    • 4Steadman Hawkins Clinic of the Carolinas, Prisma Health System, Greenville, South Carolina
    JAMA Netw Open. 2020;3(1):e1918663. doi:10.1001/jamanetworkopen.2019.18663
    Key Points español 中文 (chinese)

    Question  Are higher rates of surgery for proximal humerus fracture associated with adverse events, mortality, and cost?

    Findings  In this comparative effectiveness research study of 72 823 fee-for-service Medicare beneficiaries with a proximal humerus fracture in 2011, instrumental variable analysis showed that higher rates of surgery were significantly associated with increased costs, adverse event rates, and mortality risk at 1 year. These associations were even more striking for older patients, those with higher comorbidity burdens, and those with increased frailty.

    Meaning  Orthopedic surgeons should be aware of the harms of extending the use of surgery to more clinically vulnerable patient subgroups.

    Abstract

    Importance  Meta-analyses of randomized clinical trials suggest that the advantages and risks of surgery compared with conservative management as the initial treatment for proximal humerus fracture (PHF) vary, or are heterogeneous across patients. Substantial geographic variation in surgery rates for PHF suggests that the optimal rate of surgery across the population of patients with PHF is unknown.

    Objective  To use geographic variation in treatment rates to assess the outcomes associated with higher rates of surgery for patients with PHF.

    Design, Setting, and Participants  This comparative effectiveness research study analyzed all fee-for-service Medicare beneficiaries with proximal humerus fracture in 2011 who were continuously enrolled in Medicare Parts A and B for the 365-day period before and immediately after their index fracture. Data analysis was performed January through June 2019.

    Exposure  Undergoing 1 of the commonly used surgical procedures in the 60 days after an index fracture diagnosis.

    Main Outcomes and Measures  Risk-adjusted area surgery ratios were created for each hospital referral region as a measure of local area practice styles. Instrumental variable approaches were used to assess the association between higher surgery rates and adverse events, mortality risk, and cost at 1 year from Medicare’s perspective for patients with PHF in 2011. Instrumental variable models were stratified by age, comorbidities, and frailty. Instrumental variable estimates were compared with estimates from risk-adjusted regression models.

    Results  The final cohort included 72 823 patients (mean [SD] age, 80.0 [7.9] years; 13 958 [19.2%] men). The proportion of patients treated surgically ranged from 1.8% to 33.3% across hospital referral regions in the United States. Compared with conservatively managed patients, surgical patients were younger (mean [SD] age, 80.4 [8.1] years vs 78.0 [7.2] years; P < .001) and healthier (Charlson Comorbidity Index score of 0, 14 863 [24.4%] patients vs 3468 [29.1%] patients; Function-Related Indicator score of 0, 20 720 [34.0%] patients vs 4980 [41.8%] patients; P < .001 for both), and a larger proportion were women (49 030 [80.5%] patients vs 9835 [82.5%] patients; P < .001). Instrumental variable analysis showed that higher rates of surgery were associated with increased total costs ($8913) during the treatment period, increased adverse event rates (a 1–percentage point increase in the surgery rate was associated with a 0.19–percentage point increase in the 1-year adverse event rate; β = 0.19; 95% CI, 0.09-0.27; P < .001), and increased mortality risk (a 1–percentage point increase in the surgery rate was associated with a 0.09–percentage point increase in the 1-year mortality rate; β = 0.09; 95% CI, 0.04-0.15; P < .01). Instrumental variable mortality results were even more striking for older patients and those with higher comorbidity burdens and greater frailty. Risk-adjusted estimates suggested that surgical patients had higher costs (increase of $17 278) and more adverse events (a 1–percentage point increase in the surgery rate was associated with a 0.12–percentage point increase in the 1-year adverse event rate; β = 0.12; 95% CI, 0.11 to 0.13; P < .001) but lower risk of mortality after PHF (a 1–percentage point increase in the surgery rate was associated with a 0.01–percentage point decrease in the 1-year mortality rate; β = −0.01; 95% CI, −0.015 to −0.005; P < .001).

    Conclusions and Relevance  This study found that higher rates of surgery for treatment of patients with PHF were associated with increased costs, adverse event rates, and risk of mortality. Orthopedic surgeons should be aware of the harms of extending the use of surgery to more clinically vulnerable patient subgroups.

    Introduction

    A meta-analysis1 of existing randomized clinical trials suggests that the benefits and risks of surgery compared with conservative management vary, or are heterogeneous, across patients with proximal humerus fracture (PHF). The continued use of surgery over time suggests that surgeons have observed many patients improve after surgical treatment.2,3 In addition, lower surgery rates for more elderly patients with PHF suggest that surgeons recognize that the risks of surgery (eg, surgical complications, infections, and mortality) increase with patient complexity.4

    If the outcomes of surgery are heterogeneous across patients, the relevant research question is not whether surgery or conservative management is the single most effective treatment for patients with PHF, but rather, what is the effective surgery rate across patients with PHF?5-7 The effective surgery rate in a population of patients with PHF is defined as the rate that results when all patients receive the treatment that is best for them.8-13 Can additional patients benefit from higher surgery rates, or would higher surgery rates put additional patients with PHF at unnecessary risk with little benefit? Substantial and persistent geographic variation in surgery rates for patients with PHF suggests that there is no consensus as to what the effective rate of surgery is.2,3

    Estimates of the outcomes of surgery compared with conservative management for patients with PHF on the extensive margin are needed to address this question.10-14 Patients with PHF on the extensive margin are those who would be next to undergo surgery if surgery rates increased, or the first not to undergo surgery if surgery rates decreased. The objective of this study was to assess the outcome implications associated with higher surgery rates for Medicare patients with PHF. Instrumental variable estimators using local area surgery rates as instruments provide a natural empirical approach to address this question. Instrumental variable estimators yield estimates that are directly generalizable to patients with PHF on the extensive margin, or the set of patients with PHF whose surgery choice is sensitive to local area surgery practices.13,15-20 Instrumental variable estimators have been used to assess mortality after treatment for hip fracture.21 They are contrasted with risk-adjusted regression (RAR) estimates, which provide information on surgery outcomes for the patients who underwent surgery in our data.13,15-20

    Methods
    Data and Sample

    This was a retrospective comparative effectiveness research cohort study of all 130 959 Medicare beneficiaries in the United States with a PHF in 2011. This project was approved by the University of South Carolina institutional review board. A waiver of informed consent was granted because this study used deidentified Medicare data. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

    Individual patients with a radiography-confirmed diagnosis of PHF in 2011 were identified using Medicare Part B carrier, outpatient, and Medpar Part A inpatient claims. eTable 1 in the Supplement contains a full list of diagnosis codes used to identify patients with PHF. The index date of PHF was defined for each beneficiary as the first date of PHF in 2011. Full cohort inclusion criteria are described in more detail in our previous work3 and in eTable 2 in the Supplement. Patients were required to survive the treatment period, which was the 60-day period following the index date of PHF. The full cohort of patients with PHF was stratified into subgroups to examine the differential outcomes of surgery on each subgroup of patients: aged less than 80 years vs 80 years or older, Charlson Comorbidity Index22 score less than 2 vs 2 or higher, and Function-Related Indicators frailty score23 less than 2 vs 2 or higher.

    Treatment Measures

    Patients undergoing 1 of 4 surgical procedures (reverse shoulder arthroplasty, hemiarthroplasty, open reduction internal fixation, and closed reduction internal fixation) during the 60 days following the index PHF diagnosis date were classified as surgically managed patients. Surgery claims were identified using Part B carrier, outpatient, and Medpar inpatient claims files. Patients not undergoing surgery during the 60-day treatment window were classified as conservatively managed patients. Complete definitions of treatment variables are provided in eTable 3 in the Supplement.

    Outcome Measures

    The outcome period began on day 61 and ran through day 365 after the index fracture. Study outcomes included the occurrence of an adverse event, mortality, and shoulder-related, non–shoulder-related, and total health care costs from Medicare’s perspective. In the year following the index fracture, any claim with at least 1 diagnosis of a surgical or medical complication was identified and defined as an adverse event. Adverse events were defined as medical and surgical complications typically associated with elderly adults or surgical repair of PHF and included infection, nerve injury, prosthetic complication, hematoma, avascular necrosis, adhesive capsulitis, shoulder instability or dislocation, pneumonia, cardiac dysrhythmias, congestive heart failure, and deep vein thrombosis or pulmonary embolism.4 Mortality dates were obtained from the 2011 and 2012 Medicare Beneficiary summary files. The mortality variable was calculated during the 305-day period following the 60-day treatment window, which equaled 1 if the patient died during the period, and equaled 0 otherwise.

    Health care costs were summed across all Part A and Part B claims during 2 distinct outcome periods. The treatment window represented the 60-day period beginning on the day of the index fracture and ending 60 days thereafter (days 0-60). The outcome period began on day 61 and ran through day 365. Those claims with a diagnosis of 1 of 192 shoulder diagnoses were deemed a shoulder-related health care cost. All other claims without a shoulder diagnosis were categorized as non–shoulder-related costs. Total costs were a sum of shoulder-related and non–shoulder-related costs. All costs were from the perspective of the Medicare program and included payment amounts made by Medicare to health care professionals.

    Covariates

    The list of all patient demographic and clinical covariates specified in all estimation equations can be found in eTable 4 in the Supplement and in a previous publication.3 Demographic characteristics included age, sex, race, dual eligibility, Charlson Comorbidity Index score, and Frailty Risk Index score. Clinical covariates included previous-year shoulder diagnoses of osteoarthritis, rheumatoid arthritis, rotator cuff arthropathy, and avascular necrosis, as well as Medicare spending in the previous year.

    Analytical Approach

    It is well known that different estimation approaches for observational data yield average treatment effect estimates over distinct subgroups of patients and that these approaches rely on different assumption sets to avoid confounding bias.10,15,17,18,20,24-31 Consequently, when treatment outcomes are heterogeneous across patients, estimates from these approaches can differ and reveal different information about the study population.20,31 Here we used instrumental variable estimators to assess our research question. The RAR estimators were provided for comparison. Table 1 summarizes the properties of these estimators when assessing the outcomes of surgery compared with conservative management. Instrumental variable estimators yield the average surgery effect for the subgroup of the patients whose surgery choices were sensitive to the instrumental variable used in the analysis, which is known as the local average treatment effect in the instrumental variable literature.24,26 To be unbiased, instrumental variable estimators assume that the distribution of potential outcomes under conservative management are the same for patients with different instrumental variable values.24,26 The RAR estimators yield estimates of the average surgery effect for the subgroup of the patients who chose surgery (eg, the average treatment effect on the treated).10,17,30,32 For the estimate of average treatment effect on the treated to be unbiased, RAR estimators assume that the distribution of potential outcomes under conservative management are the same for patients ultimately treated with either surgery or conservative management.24,26

    The instrumental variables for our instrumental variable models were based on measures of local area practice styles that have been a practical and rich source for instrument development.10,11,14,33-44 The use of local area practice style measures as an instrument enabled a direct estimate of the outcomes of higher surgery rates on study outcomes. We used quintiles of risk-adjusted area surgery ratios (ASRs) across hospital referral regions (HRRs) as measures of local area practice styles. The ASRs were calculated for each HRR as the ratio of the number of patients with PHF in an HRR who underwent surgical treatment over the sum across these patients of their estimated probabilities of undergoing surgery. Estimated probabilities for each patient were produced from a logistic regression model of surgery choice as a function of measured covariates over the entire sample. Measured patient covariates in the model of surgery choice can be found in eTable 4 in the Supplement. The range of ASRs across all HRRs was divided into 5 quintiles, which became our instruments. More details on the construction of our instruments can be found in the eAppendix in the Supplement. Surgeons in HRRs with larger ASR values are theorized to have stronger beliefs as to the effectiveness of surgery compared with conservative management for patients with PHF. Patients in our sample were assigned the surgery rate and ASR value according to their residence HRR. Instrumental variable analysis was conducted using the procedure IVREG2 in Stata statistical software version 13.1 (StataCorp). All models controlled for patient demographic and clinical characteristics. We used RAR models to estimate the association between surgery and each study outcome after controlling for measured patient covariates associated with comorbidities and frailty. The RAR analysis was conducted using the Stata procedure REG.

    Statistical Analysis

    Descriptive statistics summarizing patient characteristics for surgery and conservative management groups were assessed by the 2-sample independent t test for continuous variables and Pearson χ2 test for categorical data. The Cochrane-Armitage test was used to assess trends of patients grouped by ASR quintiles.45,46 A 2-sided P < .05 was considered statistically significant. F statistics testing whether instruments described a significant amount of variation in surgery rates were assessed for the sample population and subgroups according to age, comorbidities, and frailty. An F statistic greater than 10 is considered indicative of a strong instrument.47 SAS statistical software version 9.4 (SAS Institute) was used for building our analytic database, and Stata was used for statistical analyses. Sensitivity analyses were conducted to assess the impact of cohort inclusion criteria and data extremes on the robustness of our findings and study conclusions. Data analysis was performed January through June 2019.

    Results
    Sample Characteristics

    The final cohort included 72 823 patients (mean [SD] age, 80.0 [7.9] years; 13 958 [19.2%] men). In the subgroups, 36 216 patients were younger than 80 years, 36 607 patients were aged 80 years or older, 33 389 patients had a Charlson Comorbidity Index score less than 2, 39 434 patients had a Charlson Comorbidity Index score of 2 or higher, 44 820 patients had a Function-Related Indicators score less than 2, and 28 003 patients had a Function-Related Indicators score of 2 or higher.

    Surgery within 60 days of diagnosis was used for 16.3% of the study cohort, and when compared with conservatively managed patients, surgically treated patients were younger (mean [SD] age, 80.4 [8.1] years vs 78.0 [7.2] years; P < .001), had fewer comorbidities (Charlson Comorbidity Index score of 0, 14 863 [24.4%] patients vs 3468 [29.1%] patients; P < .001), and had a lower Function-Related Indicators score (Function-Related Indicator score of 0, 20 720 [34.0%] patients vs 4980 [41.8%] patients; P < .001). Women (9835 [82.5%] patients vs 49 030 [80.5%] patients; P < .001) and white beneficiaries (11 388 [95.1%] patients vs 56 663 [93.0%] patients; P < .001) were more likely to be treated surgically than conservatively managed, and a lower percentage of surgical patients were dual eligible for Medicaid the month of their index fracture compared with conservatively managed patients (1086 [9.1%] patients vs 8728 [14.3%] patients; P < .001). A lower percentage of surgical patients, compared with conservatively managed patients, had a history of shoulder conditions (osteoarthritis, 2819 [23.6%] patients vs 15 556 [25.5%] patients [P < .001]; rheumatoid arthritis, 886 [7.4%] vs 4890 [8.0%] patients [P = .02]; rotator cuff arthropathy, 752 [6.3%] vs 4077 [6.7%] patients [P = .06]). Surgical patients had lower Medicare cost in the year preceding the index fracture compared with conservatively managed patients (mean [SD], $11 833 [$20 403] vs $15 472 [$26 365]; P < .001). Table 2 also shows average unadjusted 1-year outcomes for the full cohort by treatment group. Surgically managed patients experienced more adverse events (6979 [58.5%] patients vs 30 934 [50.8%] patients; P < .001) and had higher shoulder-related (mean [SD], $15 065 [$10 745] vs $3306 [$6330]; P < .001) and non–shoulder-related (mean [SD], $27 584 [$32 563] vs $23 690 [$32 308]; P < .001) health care costs in the year following their index diagnosis compared with conservatively managed patients. A larger proportion of conservatively managed patients died in the outcome period compared with surgically managed patients (10.8% vs 7.0%; P < .001).

    The proportion of patients treated surgically ranged from 1.8% to 33.3% across HRRs in the United States. Table 3 compares patient characteristics across ASR quintiles, which served as our instrumental variable. The mean percentage of patients who underwent surgery after PHF varied from 10.6% to 22.5% from the lowest to highest ASR quintile. Areas with higher surgery rates had higher rates of adverse events (53.6% vs 52.1%; P < .001) and higher mortality rates (10.5% vs 9.8%; P = .16) compared with areas with the lowest surgery rates. Measured patient factors associated with study outcomes, such as Charlson Comorbidity Index score, Function-Related Indicators score, and age, were similar across ASR quintiles. Although some trends in patient characteristics across ASR quintiles did reach statistical significance (quintile 1 vs quintile 5, 2.6% vs 3.3% for black race [P < .01]; 1.0% vs 0.9% for Hispanic ethnicity [P = .02]; 14.9% vs 11.0% for dual eligibility [P < .001]; and 23.9% vs 26.4% for osteoarthritis [P < .001]), the absolute differences across quintiles were small in comparison with the differences in the surgically and conservatively managed patients in Table 2.

    Instrumental Variable Results

    Table 4 shows instrumental variable estimates for the clinical outcomes. Table 4 also shows the F statistic assessing the effects of our instrumental variable on surgery choice. Instruments with an F statistic less than 10 are considered weak and more susceptible to bias in the instrumental variable literature.48 All F statistics in this study are much greater than 10 (total cohort, 205.9; age <80 years, 98.9; age ≥80 years, 111.3; Charlson Comorbidity Index score <2, 95.2; Charlson Comorbidity Index score ≥2, 110.9; Function-Related Indicators score <2, 136.3; Function-Related Indicators score ≥2, 69.9), signifying that local area treatment signatures, measured by risk-adjusted ASRs, were strong instruments. Table 4 shows the mean percentage of patients undergoing surgery in each patient subgroup and the corresponding range in rates across the ASRs. Our instrumental variable surgery effect estimates directly stem from the variation in surgery rates within these ranges. In addition, the surgery rates in Table 4 for stratified subsets of our sample show that patients with PHF who were older, had higher comorbidity burdens, and greater frailty were less likely to undergo surgery (mean [range], age <80 years, 19.9% [13.9%-26.4%] vs age ≥80 years, 12.8% [7.7%-18.4%]; Charlson Comorbidity Index score <2, 18.1% [12.2%-24.4%] vs Charlson Comorbidity Index score ≥2, 14.9% [9.4%-20.9%]; Function-Related Indicators score <2, 18.3% [12.1%-24.9%] vs Function-Related Indicators score ≥2, 13.2% [8.3%-18.6%]).

    Instrumental variable estimates show that higher surgery rates for PHF were associated with higher adverse event rates and higher mortality in the full cohort and select patient subsets. These instrumental variable estimates show the absolute average effect of a 1–percentage point increase in surgery rates on study outcome rates. For example, the instrumental variable estimate for 1-year adverse event risk for the full cohort implies that a 1–percentage point increase in the surgery rate was associated with a 0.19–percentage point increase in the 1-year adverse event rates (β = 0.19; 95% CI, 0.09-0.27; P < .001). The instrumental variable estimate for 1-year mortality for the full cohort implies that a 1–percentage point increase in the surgery rate was associated with a 0.09–percentage point increase in the 1-year mortality rates (β = 0.09; 95% CI, 0.04-0.15; P < .01). Furthermore, the adverse event and mortality risks associated with higher surgery rates were greatest in older patients and those with higher comorbidity burdens and frailty index scores. For example, the instrumental variable estimate for 1-year adverse event risk for the patients with higher comorbidity burdens (Charlson Comorbidity Index score ≥2) suggests that a 1–percentage point increase in surgery rates was associated with a 0.24–percentage point increase in the 1-year adverse event rates (β = 0.24; 95% CI, 0.11-0.36; P < .001) and a 0.15–percentage point increase in 1-year mortality rates (β = 0.15; 95% CI, 0.06-0.24; P < .01). Among the subset of patients with a Function-Related Indicator score of 2 or higher, a 1–percentage point increase in the surgery rate was associated with a 0.22–percentage point increase in the 1-year adverse event rate and a 0.11–percentage point increase in the 1-year mortality rate. Among the subset of patients aged 80 years or older, a 1–percentage point increase in the surgery rate was associated with a 0.20–percentage point increase in the 1-year adverse event rate and a 0.11–percentage point increase in the 1-year mortality rate.

    Table 5 shows instrumental variable estimates for cost outcomes. During the 60-day treatment period, shoulder-related (β = $5358), non–shoulder-related (β = $3555), and total (β = $8913) costs were higher in local areas with higher surgery rates in the full cohort and most subgroups. During the outcome period, higher surgery rates were associated with lower shoulder-related, non–shoulder-related, and total costs for the full cohort and most patient subgroups, although most estimates did not reach significance. These results were especially pronounced in the more clinically vulnerable patient subgroups, which showed the greatest reductions in cost, although the estimates did not reach statistical significance.

    Risk-Adjusted Results

    Table 4 shows RAR estimates on clinical outcomes for the full sample and patient subgroups. In the full cohort and across all subgroups of patients with PHF, the probability of adverse events was higher for the surgically managed patients (a 1–percentage point increase in the surgery rate was associated with a 0.12–percentage point increase in the 1-year adverse event rate; β = 0.12; 95% CI, 0.11 to 0.13; P < .001). Across all subgroups, the risk-adjusted estimates suggest that surgery was associated with a decrease in the 1-year mortality rate compared with conservative management for the patients who underwent surgery. In the full cohort, a 1–percentage point increase in the surgery rate was associated with a 0.01–percentage point decrease in the 1-year mortality rate (β = −0.01; 95% CI, −0.015 to −0.005; P < .001).

    In Table 5, shoulder-related (β = $621), non–shoulder-related (β = $919), and total (β = $1541) costs were all higher among patients undergoing surgery compared with conservatively managed patients across all periods and subgroups. In the treatment period, surgical patients had total Medicare reimbursements that were a mean of $17 278 higher than those for conservatively managed patients, the majority of which was likely due to the surgical procedure. Shoulder-related, non–shoulder-related, and total costs were also higher for surgery recipients in the outcome period. Sensitivity analyses were conducted to assess the impact of cohort inclusion criteria and data extremes on the robustness of our findings but did not yield materially different conclusions.

    Discussion

    Our analysis showed that higher rates of surgery were associated with increased 1-year mortality and adverse event rates. This finding suggests that surgery rates in 2011 for patients with PHF were higher than the effective rate and that a reduction in surgery rates could reduce mortality and adverse event rates for patients with PHF. This result is especially pronounced for older cohorts and those with higher comorbidity burdens and increased frailty. Similar to more well-studied fracture populations, such as patients with hip fracture,49,50 patients with PHF are an especially frail population, and the surgery decision is associated with a series of events that can trigger poor outcomes, including death. Our instrumental variable estimates are unbiased if unmeasured patient differences between surgically and conservatively managed patients are unrelated to the instrument we selected. Consistent distributions in measured patient characteristics across patients grouped by our instrument in Table 3 support this assumption.

    For comparison, we used risk-adjusted estimators to estimate the association between surgery and each study outcome by controlling for a series of measured baseline covariates. Risk-adjusted estimators provide estimates of the outcomes of surgery for the patients who undergo surgery. For risk-adjusted estimates to be unbiased, outcomes for surgical patients if they had not undergone surgery would have to be the same as those observed for the conservatively treated patients.24,26 Risk-adjusted estimates suggested that surgical patients had lower risks of mortality after PHF. However, we found meaningful differences in patient age, comorbidity burden, and frailty index between surgically and conservatively managed patients, which suggests that there are likely additional unmeasured differences between treatment groups that are partly responsible for the estimated treatment outcomes we observed. Given that patients with PHF undergoing surgery are younger, less frail, and have lower comorbidity burdens, it is likely that the estimates of reduced mortality from surgery are biased high as the result of unmeasured differences between the patients selected for surgery and conservative care.

    Limitations

    Our results should be viewed through the limitations of the study design. First, our adverse event and mortality instrumental variable estimates should be interpreted only within the surgery rate ranges we provided in Table 4. Because surgery outcomes are likely heterogeneous across patients with PHF, it is risky to generalize our estimates to surgery rates below or above these surgery rate ranges. Our study sample was from 2011, and surgery rates for PHF continue to change. The survival implications associated with surgery rate changes from different baseline rates should be evaluated. Second, the International Classification of Diseases, Ninth Revision diagnosis codes available for this sample of Medicare data likely does not include information on many factors associated with patient survival, which are also associated with the choice of surgery, such as fracture complexity. Because, patients with complex fractures are more likely to undergo surgery, a portion of the mortality associated with surgery in our instrumental variable models will stem from differences in the distributions of fracture complexity across local areas. However, previous research38 suggests that this bias risk is attenuated when large geographic areas such as the HRRs are used, as we did here. However, future research using Medicare claims after 2015, which use International Statistical Classification of Diseases and Related Health Problems, Tenth Revision diagnosis codes that reflect fracture complexity, is needed to validate this assumption. Third, and related to our second limitation, is the assumption that patients are similar across quintiles of our instrument. Our instrumental variable estimates are unbiased if unmeasured patient differences between surgically and conservatively managed patients are not associated with the instrument we selected. Although some minor differences in shoulder health were noted across instrument quintiles, mostly consistent distributions in measured patient characteristics across patients grouped by our instrument in Table 3 support our assumption that patients were similar across high and low surgery areas. However, it must still be acknowledged that patient differences across areas, if they exist, could confound our findings. Fourth, we chose to exclude patients who did not survive the treatment period. Therefore, our results do not generalize to those patients who were very frail and not survive the 60 days following their fracture.

    Conclusions

    This study included a diverse, national sample of patients with PHF. We found wide treatment variation and sought to understand the implications associated with higher rates of surgery for patients with PHF on patient outcomes. Instrumental variable methods offer an approach to directly assess this question using populations of Medicare patients with PHF from across local areas.13,15-20 Our estimates provide evidence to suggest that current PHF surgery rates are higher than the effective surgery rate and that lowering surgery rates may be associated with lower 1-year mortality rates and adverse event risks. These results are especially pronounced for patients with PHF in high-risk patient subgroups. Because the choice of treatment can affect a number of measured and unmeasured variables, some of which could not be measured in this analysis, further analysis using patient-reported outcomes is warranted.

    Back to top
    Article Information

    Accepted for Publication: November 10, 2019.

    Published: January 10, 2020. doi:10.1001/jamanetworkopen.2019.18663

    Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2020 Floyd SB et al. JAMA Network Open.

    Corresponding Author: Sarah B. Floyd, PhD, Department of Health Services Policy and Management, Arnold School of Public Health, University of South Carolina, 915 Greene St, Ste 303C, Columbia, SC 29208 (floydsb@mailbox.sc.edu).

    Author Contributions: Dr Floyd had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: All authors.

    Acquisition, analysis, or interpretation of data: Floyd, Kissenberth, Brooks.

    Drafting of the manuscript: Floyd, Kissenberth, Brooks.

    Critical revision of the manuscript for important intellectual content: All authors.

    Statistical analysis: Floyd, Thigpen, Brooks.

    Obtained funding: Brooks.

    Administrative, technical, or material support: Floyd, Brooks.

    Supervision: Brooks.

    Conflict of Interest Disclosures: Dr Kissenberth reported being a paid consultant for Arthrex. No other disclosures were reported.

    References
    1.
    Sabharwal  S, Patel  NK, Griffiths  D, Athanasiou  T, Gupte  CM, Reilly  P.  Trials based on specific fracture configuration and surgical procedures likely to be more relevant for decision making in the management of fractures of the proximal humerus: findings of a meta-analysis.  Bone Joint Res. 2016;5(10):470-480. doi:10.1302/2046-3758.510.2000638PubMedGoogle ScholarCrossref
    2.
    Bell  JE, Leung  BC, Spratt  KF,  et al.  Trends and variation in incidence, surgical treatment, and repeat surgery of proximal humeral fractures in the elderly.  J Bone Joint Surg Am. 2011;93(2):121-131. doi:10.2106/JBJS.I.01505PubMedGoogle ScholarCrossref
    3.
    Floyd  SB, Campbell  J, Chapman  CG, Thigpen  CA, Kissenberth  MJ, Brooks  JM.  Geographic variation in the treatment of proximal humerus fracture: an update on surgery rates and treatment consensus.  J Orthop Surg Res. 2019;14(1):22. doi:10.1186/s13018-018-1052-2PubMedGoogle ScholarCrossref
    4.
    Floyd  SB, Chapman  CG, Thigpen  CA, Brooks  JM, Hawkins  RJ, Tokish  JM.  Shoulder arthroplasty in the US Medicare population: a 1-year evaluation of surgical complications, hospital admissions, and revision surgery.  JSES Open Access. 2018;2(1):40-47. doi:10.1016/j.jses.2017.10.002PubMedGoogle ScholarCrossref
    5.
    Birkmeyer  JD, Reames  BN, McCulloch  P, Carr  AJ, Campbell  WB, Wennberg  JE.  Understanding of regional variation in the use of surgery.  Lancet. 2013;382(9898):1121-1129. doi:10.1016/S0140-6736(13)61215-5PubMedGoogle ScholarCrossref
    6.
    McCulloch  P, Nagendran  M, Campbell  WB,  et al.  Strategies to reduce variation in the use of surgery.  Lancet. 2013;382(9898):1130-1139. doi:10.1016/S0140-6736(13)61216-7PubMedGoogle ScholarCrossref
    7.
    Bernstein  J.  Not the last word: Choosing Wisely.  Clin Orthop Relat Res. 2015;473(10):3091-3097. doi:10.1007/s11999-015-4490-8PubMedGoogle ScholarCrossref
    8.
    Wennberg  J.  Which rate is right?  N Engl J Med. 1986;314(5):310-311. doi:10.1056/NEJM198601303140509PubMedGoogle ScholarCrossref
    9.
    Brooks  JM, Chapman  CG, Suneja  M,  et al.  Use of angiotensin-converting enzyme inhibitors and angiotensin receptor blockers for geriatric ischemic stroke patients: are the rates right?  J Am Heart Assoc. 2018;7(11):e009137. doi:10.1161/JAHA.118.009137PubMedGoogle Scholar
    10.
    Brooks  JM, Chrischilles  EA.  Heterogeneity and the interpretation of treatment effect estimates from risk adjustment and instrumental variable methods.  Med Care. 2007;45(10)(suppl):S123-S130. doi:10.1097/MLR.0b013e318070c069PubMedGoogle ScholarCrossref
    11.
    Brooks  JM, Cook  E, Chapman  CG,  et al.  Statin use after acute myocardial infarction by patient complexity: are the rates right?  Med Care. 2015;53(4):324-331. doi:10.1097/MLR.0000000000000322PubMedGoogle Scholar
    12.
    Phelps  CE.  Health Economics. 2nd ed. Reading, MA: Addison-Wesley; 1997.
    13.
    Park  T-R, Brooks  JM, Chrischilles  EA, Bergus  G.  Estimating the effect of treatment rate changes when treatment benefits are heterogeneous: antibiotics and otitis media.  Value Health. 2008;11(2):304-314. doi:10.1111/j.1524-4733.2007.00234.xPubMedGoogle ScholarCrossref
    14.
    Brooks  JM, McClellan  M, Wong  HS.  The marginal benefits of invasive treatments for acute myocardial infarction: does insurance coverage matter?  Inquiry. 2000;37(1):75-90.PubMedGoogle Scholar
    15.
    Harris  KM, Remler  DK.  Who is the marginal patient? understanding instrumental variables estimates of treatment effects.  Health Serv Res. 1998;33(5):1337-1360.PubMedGoogle Scholar
    16.
    McClellan  M, Newhouse  JP.  Overview of the special supplement issue.  Health Serv Res. 2000;35(5):1061-1069.PubMedGoogle Scholar
    17.
    Brooks  JM, Chapman  CG, Schroeder  MC.  Understanding treatment effect estimates when treatment effects are heterogeneous for more than one outcome.  Appl Health Econ Health Policy. 2018;16(3):381-393. doi:10.1007/s40258-018-0380-zPubMedGoogle ScholarCrossref
    18.
    Fang  G, Brooks  JM, Chrischilles  EA.  Apples and oranges? interpretations of risk adjustment and instrumental variable estimates of intended treatment effects using observational data.  Am J Epidemiol. 2012;175(1):60-65. doi:10.1093/aje/kwr283PubMedGoogle ScholarCrossref
    19.
    Brooks  JM, Chapman  CG, Cozad  MJ.  The identification process using choice theory is needed to match design with objectives in CER.  Med Care. 2017;55(2):91-93. doi:10.1097/MLR.0000000000000615PubMedGoogle ScholarCrossref
    20.
    Garrido  MM, Dowd  B, Hebert  PL, Maciejewski  ML.  Understanding treatment effect terminology in pain and symptom management research.  J Pain Symptom Manage. 2016;52(3):446-452. doi:10.1016/j.jpainsymman.2016.01.016PubMedGoogle ScholarCrossref
    21.
    McGuire  KJ, Bernstein  J, Polsky  D, Silber  JH.  The 2004 Marshall Urist award: delays until surgery after hip fracture increases mortality.  Clin Orthop Relat Res. 2004;(428):294-301. doi:10.1097/01.blo.0000146743.28925.1cPubMedGoogle Scholar
    22.
    Charlson  ME, Pompei  P, Ales  KL, MacKenzie  CR.  A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.  J Chronic Dis. 1987;40(5):373-383. doi:10.1016/0021-9681(87)90171-8PubMedGoogle ScholarCrossref
    23.
    Chrischilles  E, Schneider  K, Wilwert  J,  et al.  Beyond comorbidity: expanding the definition and measurement of complexity among older adults using administrative claims data.  Med Care. 2014;52(suppl):S75-S84. doi:10.1097/MLR.0000000000000026PubMedGoogle ScholarCrossref
    24.
    Angrist  JD.  Treatment effect heterogeneity in theory and practice.  Econ J (Lond). 2004;114:C52-C83. doi:10.1111/j.0013-0133.2003.00195.xGoogle Scholar
    25.
    Heckman  JJ.  Econometric causality.  Int Stat Rev. 2008;76(1):1-27. doi:10.1111/j.1751-5823.2007.00024.xGoogle ScholarCrossref
    26.
    Heckman  JJ, Urzua  S, Vytlacil  E.  Understanding instrumental variables in models with essential heterogeneity.  Rev Econ Stat. 2006;88(3):389-432. doi:10.1162/rest.88.3.389Google ScholarCrossref
    27.
    Newhouse  JP, McClellan  M.  Econometrics in outcomes research: the use of instrumental variables.  Annu Rev Public Health. 1998;19:17-34. doi:10.1146/annurev.publhealth.19.1.17PubMedGoogle ScholarCrossref
    28.
    Imbens  GW, Angrist  JD.  Identification and estimation of local average treatment effects.  Econometrica. 1994;62(2):467-475. doi:10.2307/2951620Google ScholarCrossref
    29.
    Greenland  S, Morgenstern  H.  Confounding in health research.  Annu Rev Public Health. 2001;22:189-212. doi:10.1146/annurev.publhealth.22.1.189PubMedGoogle ScholarCrossref
    30.
    Heckman  JJ, Robb  R. Alternative methods for evaluating the impact of interventions. In: Heckman  JJ, Singer  B, eds.  Longitudinal Analysis of Labor Market Data. New York, NY: Cambridge University Press; 1985:156-245. doi:10.1017/CCOL0521304539.004
    31.
    Chapman  CG, Cozad  MJ, Brooks  JM.  Conditions for generating treatment effect estimates in line with objectives: beyond confounding.  Med Care. 2017;55(2):97-99. doi:10.1097/MLR.0000000000000614PubMedGoogle ScholarCrossref
    32.
    Brooks  JM, Fang  G.  Interpreting treatment-effect estimates with heterogeneity and choice: simulation model results.  Clin Ther. 2009;31(4):902-919. doi:10.1016/j.clinthera.2009.04.007PubMedGoogle ScholarCrossref
    33.
    Rassen  JA, Brookhart  MA, Glynn  RJ, Mittleman  MA, Schneeweiss  S.  Instrumental variables II: instrumental variable application—in 25 variations, the physician prescribing preference generally was strong and reduced covariate imbalance.  J Clin Epidemiol. 2009;62(12):1233-1241. doi:10.1016/j.jclinepi.2008.12.006PubMedGoogle ScholarCrossref
    34.
    Brooks  JM, Chrischilles  EA, Scott  SD, Chen-Hardee  SS.  Was breast conserving surgery underutilized for early stage breast cancer? instrumental variables evidence for stage II patients from Iowa.  Health Serv Res. 2003;38(6 Pt 1):1385-1402. doi:10.1111/j.1475-6773.2003.00184.xPubMedGoogle ScholarCrossref
    35.
    Tang  Y, Brooks  JM, Wetmore  JB, Shireman  TI.  Association between higher rates of cardioprotective drug use and survival in patients on dialysis.  Res Social Adm Pharm. 2015;11(6):824-843. doi:10.1016/j.sapharm.2014.12.007PubMedGoogle ScholarCrossref
    36.
    Polgreen  LA, Cook  EA, Brooks  JM, Tang  Y, Polgreen  PM.  Increased statin prescribing does not lower pneumonia risk.  Clin Infect Dis. 2015;60(12):1760-1766. doi:10.1093/cid/civ190PubMedGoogle ScholarCrossref
    37.
    Brooks  JM, Chrischilles  EA, Landrum  MB,  et al.  Survival implications associated with variation in mastectomy rates for early-staged breast cancer.  Int J Surg Oncol. 2012;2012:127854. doi:10.1155/2012/127854PubMedGoogle Scholar
    38.
    Brooks  JM, Tang  Y, Chapman  CG, Cook  EA, Chrischilles  EA.  What is the effect of area size when using local area practice style as an instrument?  J Clin Epidemiol. 2013;66(8)(suppl):S69-S83. doi:10.1016/j.jclinepi.2013.04.008PubMedGoogle ScholarCrossref
    39.
    Brooks  JM, Cook  EA, Chapman  CG,  et al.  Geographic variation in statin use for complex acute myocardial infarction patients: evidence of effective care?  Med Care. 2014;52(suppl):S37-S44. doi:10.1097/MLR.0b013e3182a7fc3dPubMedGoogle ScholarCrossref
    40.
    Brookhart  MA, Wang  PS, Solomon  DH, Schneeweiss  S.  Evaluating short-term drug effects using a physician-specific prescribing preference as an instrumental variable.  Epidemiology. 2006;17(3):268-275. doi:10.1097/01.ede.0000193606.58671.c5PubMedGoogle ScholarCrossref
    41.
    Stukel  TA, Fisher  ES, Wennberg  DE, Alter  DA, Gottlieb  DJ, Vermeulen  MJ.  Analysis of observational studies in the presence of treatment selection bias: effects of invasive cardiac management on AMI survival using propensity score and instrumental variable methods.  JAMA. 2007;297(3):278-285. doi:10.1001/jama.297.3.278PubMedGoogle ScholarCrossref
    42.
    Fang  G, Brooks  JM, Chrischilles  EA.  Comparison of instrumental variable analysis using a new instrument with risk adjustment methods to reduce confounding by indication.  Am J Epidemiol. 2012;175(11):1142-1151. doi:10.1093/aje/kwr448PubMedGoogle ScholarCrossref
    43.
    Schroeder  MC, Tien  YY, Wright  K, Halfdanarson  TR, Abu-Hejleh  T, Brooks  JM.  Geographic variation in the use of adjuvant therapy among elderly patients with resected non-small cell lung cancer.  Lung Cancer. 2016;95:28-34. doi:10.1016/j.lungcan.2016.02.010PubMedGoogle ScholarCrossref
    44.
    Fang  G, Brooks  JM, Chrischilles  EA.  A new method to isolate local-area practice styles in prescription use as the basis for instrumental variables in comparative effectiveness research.  Med Care. 2010;48:710-717. doi:10.1097/MLR.0b013e3181e41bb2PubMedGoogle ScholarCrossref
    45.
    Cochran  WG.  Some methods for strengthening the common χ2 tests.  Biometrics. 1954;10(4):417-451. doi:10.2307/3001616Google ScholarCrossref
    46.
    Armitage  P.  Tests for linear trends in proportions and frequencies.  Biometrics. 1955;11:375-386. doi:10.2307/3001775Google ScholarCrossref
    47.
    Wooldridge  J.  Introductory Econometrics: A Modern Approach. Mason, OH: Thomson South-Western; 2003.
    48.
    Staiger  D, Stock  J.  Instrumental variables regression with weak instruments.  Econometrica. 1997;65(3):557-586. doi:10.2307/2171753Google ScholarCrossref
    49.
    Schnell  S, Friedman  SM, Mendelson  DA, Bingham  KW, Kates  SL.  The 1-year mortality of patients treated in a hip fracture program for elders.  Geriatr Orthop Surg Rehabil. 2010;1(1):6-14. doi:10.1177/2151458510378105PubMedGoogle ScholarCrossref
    50.
    Brauer  CA, Coca-Perraillon  M, Cutler  DM, Rosen  AB.  Incidence and mortality of hip fractures in the United States.  JAMA. 2009;302(14):1573-1579. doi:10.1001/jama.2009.1462PubMedGoogle ScholarCrossref
    ×