- Email: [email protected]
Proximal humeral fracture classification systems revisited
J Shoulder Elbow Surg (2011) 20, 1125-1132
www.elsevier.com/locate/ymse
Proximal humeral fracture classification systems revisited Addie Majed, MRCSa,*, Iain Macleod, FRCS(Orth)b, Anthony M.J. Bull, PhDc, Karol Zyto, MDd, Herbert Resch, MDe, Ralph Hertel, MDf, Peter Reilly, MD, FRCS(Orth)a, Roger J.H. Emery, MS, FRCSa a
Division of Surgery Oncology Reproductive Medicine and Anaesthetics, Imperial College London, UK ImperialCollege Healthcare NHS Trust, St. Mary’s Campus, London, UK c Department of Bioengineering, Imperial College London, UK d Queen Sophia Hospital, Stockholm, Sweden e UniversityHospital, Salzburg, Austria f Lindenhofspital, Berne, Switzerland b
Hypothesis: This study evaluated several classification systems and expert surgeons’ anatomic understanding of these complex injuries based on a consecutive series of patients. We hypothesized that current proximal humeral fracture classification systems, regardless of imaging methods, are not sufficiently reliable to aid clinical management of these injuries. Materials and methods: Complex fractures in 96 consecutive patients were investigated by generation of rapid sequence prototyping models from computed tomography Digital Imaging and Communications in Medicine (DICOM) imaging data. Four independent senior observers were asked to classify each model using 4 classification systems: Neer, AO, Codman-Hertel, and a prototype classification system by Resch. Interobserver and intraobserver k coefficient values were calculated for the overall classification system and for selected classification items. Results: The k coefficient values for the interobserver reliability were 0.33 for Neer, 0.11 for AO, 0.44 for Codman-Hertel, and 0.15 for Resch. Interobserver reliability k coefficient values were 0.32 for the number of fragments and 0.30 for the anatomic segment involved using the Neer system, 0.30 for the AO type (A, B, C), and 0.53, 0.48, and 0.08 for the Resch impaction/distraction, varus/valgus and flexion/extension subgroups, respectively. Three-part fractures showed low reliability for the Neer and AO systems. Discussion: Currently available evidence suggests facture classifications in use have poor intra- and interobserver reliability despite the modality of imaging used thus making treating these injuries difficult as weak as affecting scientific research as well. This study was undertaken to evaluate the reliability of several systems using rapid sequence prototype models. Conclusion: Overall interobserver k values represented slight to moderate agreement. The most reliable interobserver scores were found with the Codman-Hertel classification, followed by elements of Resch’s trial system. The AO system had the lowest values. The higher interobserver reliability values for the
Joint RNOH/IOMS National Research Ethics Committee approval was received to analyze deidentified medical images of proximal humeral fractures (REC reference number: 07/H0724/40). *Reprint requests: Addie Majed, MRCS, Clinical Research Fellow, Division of SORA (Surgery Oncology Reproductive Medicine and
Anaesthetics), Imperial College London, 10th Flr QEQM Building, St. Mary’s Hospital, Praed St, London W2 1NY, UK. E-mail address: [email protected] (A. Majed).
1058-2746/$ - see front matter Ó 2011 Journal of Shoulder and Elbow Surgery Board of Trustees. doi:10.1016/j.jse.2011.01.020
1126
A. Majed et al.
Codman-Hertel system showed that is the only comprehensive fracture description studied, whereas the novel classification by Resch showed clear definition in respect to varus/valgus and impaction/distraction angulation. Level of evidence: Level III, Diagnostic Study. Ó 2011 Journal of Shoulder and Elbow Surgery Board of Trustees. Keywords: Proximal humerus; fracture classifications; prototype-modelling; interobserver and intraobserver reliability
The incidence of proximal humeral fractures is reported to be 6% per 10,000 fractures, with approximately 15% of these being complex 3-part and 4-part fractures.27 The incidence is increasing, along with an increasing tendency to more severe fractures.8,12,18 The management of these often debilitating injuries is based on various fracture classification systems.16,18 A fracture classification should aim to improve the understanding and therefore the management of fractures based on the morphology of the fracture, the biologic and mechanical behavior, and to provide therapeutic guidelines based on clinical outcomes.16 It should be comprehensive with all fracture types classifiable. Its application should lead to reproducible results. The main criterion for a good classification system is that it addresses the clinically relevant question. Studies that have applied current proximal humeral fracture classification systems to plain radiographs show poor interobserver and intraobserver reliability,2,13,23,24 with 2-dimensional imaging being attributed to low levels of agreement. Some authors have advocated the need for formal training in using the classification systems.3,22 Despite the application of computed tomography (CT) scans and 3-dimensional (3D) computational CT reconstructions, the reliability of these classifications systems remains in doubt.1,21,25,26 Recently, the effect of stereo visualization of 3D CT datasets has suggested some improvement in overall reliability.4 The ability to represent proximal humeral fracture morphology physically using rapid sequence prototype modelling provides an experimental technique that may allow classification systems to be tested by ruling out imaging as a cause of low reliability. Through this technology, the observer could be provided with highly accurate physical models of the fracture that can be orientated freely. After a literature review, we selected 4 fracture classification systems that are in use or have the potential for use: The Neer system19 groups each fracture by the number of fracture segments and describes the fractured anatomic segment as a part. The groups are: group I, nondisplaced; group II, 2-part; group III, 3-part; and group IV, 4-part. There are 16 different potential types of fracture. The AO system describes each fracture as being 1 of 3 types, with 3 subdivisions each. Type A indicates extraarticular unifocal fractures; type B, extraarticular bifocal
fractures; and type C, articular fractures. The 3 subdivisions (groups 1, 2, and 3) are related to the pattern of the fracture.16 A further subdivision into 3 subgroups (1, 2 and 3) is achieved based on the degree of fragmentation, giving 27 different fracture patterns. The Codman-Hertel binary fracture description system10 was derived from the original drawings of Codman6 and was based on the analysis of fracture planes and not on the number of fragments. It comprises 12 different basic fracture patterns that are described with numbers from 1 to 12. There are 6 possible fracture combinations dividing the humerus into 2 parts, 5 possible fractures dividing the humerus into 3 parts, and 1 fracture type dividing the humerus into 4 parts. However, the system does not address fracture pathomechanics. The fourth system tested, devised by Resch, is a proposed addendum to the Codman-Hertel classification system and addresses fracture angulation and pathomechanics. It describes 3 biomechanical planes of injury that are classified by the observer: First, the fracture is described as an ‘‘impaction’’ or ‘‘distraction’’ injury in the coronal plane. An impaction injury occurs when the length position of the fractured greater tuberosity is unchanged and the total length of the humerus is reduced due to impaction of the head (Fig. 1, A). A distraction injury is defined as increased distance on the lateral side between shaft and head fragment (subcapital fracture) or between the shaft and the fractured greater tuberosity (subcapital fracture accompanied by a greater tuberosity fracture; Fig. 1, B). If there is little impaction or distraction, a ‘‘neutral’’ injury is assigned. Second, the fracture is classified as ‘‘varus,’’ ‘‘valgus,’’ or ‘‘neutral,’’ depending on the deformation of the head relative to the shaft (head inclination) in the coronal plane. Finally, the head-shaft angle in the transscapular plane is assessed and described as ‘‘flexion,’’ ‘‘extension,’’ or ‘‘neutral’’ (Fig. 2). Thus, addressing fracture angulation may also facilitate fracture reduction. We hypothesized that current proximal humeral fracture classification systems, regardless of imaging methods, are not sufficiently reliable to aid clinical management of these injuries. Therefore, the study used physical models to test the reliability of the Neer,19 AO,16 CodmanHertel10 proximal humeral fracture classification systems, as well as a primary version of a newly-devised
Proximal humeral fracture classification systems revisited
1127
Figure 1 (A) Impaction injury using the Resch description. The arrows depict the direction of the acting force. (B) Distraction injury using the Resch description. The arrows depict the direction of the acting force.
Materials and methods
Figure 2 Flexion of the head-shaft angle, assessed in the transscapular plane using the Resch description.
classification designed by Resch that aims to complement the Codman-Hertel classification system to include pathomechanics.
We analyzed deidentified images from a database of 100 consecutive patients who were treated at a single center for fractures of the proximal humerus and were investigated with computed tomography (CT) scanning between 2001 and 2007. Three patients were excluded because the fractures showed radiologic evidence of union and the data set for one was corrupted, leaving 96 patients available for this study. All CT scans were performed on a Philips MX8000 or Philips Brilliance 64 multiscanner (Koninklijke Philips N.V, Netherlands) with a slice thickness of 2.0 mm. The primary image plane was axial with a restricted field of view to the shoulder of 250 mm, and reformatting was performed using the Philips bone algorithm with a slice thickness of 1.0 mm. The DICOM data sets were transferred to a computer workstation using MIMICS 11.0 software (Materialise, Leuven, Belgium), and thresholding was conducted using bone window values (range, 226‑2799 Hounsfield units). For each image set, standard 3D volume‑rendered images were produced, and these files were then converted to Rapid Sequence Prototype models after formatting with ZPrint software, which allowed each model to be printed on a ZPrinter 310 printer (ZCorp, Burlington, MA, USA), with resolution of 300 450 dpi, and 0.0875 to 0.1‑mm layer thickness (Fig 3). The models were deidentified and arranged in a randomized order with no discernable markers. Each observer performed the assessment individually and was not given any feedback. Each observer was provided with diagrams of the Neer,19 AO,16 Codman-Hertel,10 and Resch classification systems and was given 1 minute to classify each model. One observer repeated the study 8 weeks later to allow for intraobserver reliability scores to be calculated.
1128
A. Majed et al.
Figure 3 (A) A photograph of a rapid sequence prototype model of a proximal humeral fracture viewed in the coronal plane. (B) A photograph of a rapid sequence prototype model show an inferomedial view of a proximal humeral fracture. Four senior observers (P.R., K.Z., H.R., and R.H.), who work in regional referral units and are perceived as experts in proximal humeral fracture management, were selected for this study. Two observers (R.H., H.R.) were the authors of the classifications used in this study.
Statistical analysis To assess interobserver and intraobserver reliability, k correlation coefficients were calculated using the method reported by Cohen et al7 using Stata 10.0 software (StataCorp LP College Station, Texas, USA). When all the observers agree in all cases, the k value is þ1. A k of 0.00 indicates that agreement is equal to that occurring by chance, with k values of less than 0.00 indicating poor agreement. The k coefficients were interpreted using the Landis and Koch criteria,14 where a k of more than 0.8 represents excellent agreement, between 0.6 and 0.8 is good agreement, between 0.4 and 0.6 is moderate agreement, between 0.2 and 0.4 is fair agreement, between 0 and 0.2 is slight agreement, and of less than 0 is poor agreement. Subgroup interobserver reliability k correlation coefficients were also calculated.
Results The k coefficient values for the interobserver reliability using the Neer classification was 0.33, with a respective intraobserver value of 0.57. The Neer classification was assessed using the ‘‘group’’ and the number of fracture fragments involved (eg, 3-part or 4-part). The k coefficient interobserver and intraobserver reliability values were 0.32 and 0.59, respectively. The Neer classification was simplified assessing only the fracture type, the anatomic segment involved (eg, greater tuberosity fracture), and the reliability of the system was also tested. The k coefficient interobserver and intraobserver reliability values were 0.30 and 0.54, respectively. The interobserver reliability based on
specific fracture types is reported in Table I, and the reliability based on a simplification of the classification to the number of parts and group (anatomic segments) involved are summarized in Tables II and III, respectively. Higher scores were attained for 2-part and 4-part fractures than with 3-part fractures. The k coefficient value for interobserver reliability using the AO classification was 0.11, with a respective intraobserver value of 0.42. Simplified interobserver and intraobserver reliability assessing whether the fracture was AO type (A, B, C) was 0.30 and 0.63, respectively. The interobserver reliability based on specific subgroups is reported in Table IV. The k coefficient value for interobserver reliability using the Codman-Hertel classification was 0.44, with a respective intraobserver value of 0.65. The interobserver reliability based on specific subgroups is reported in Table V. The k coefficient value assessing the interobserver reliability of the simplified system devised by Resch for whether the fracture is simply an impaction or distraction type was 0.52, valgus or varus type was 0.48, and flexion or extension type was 0.08. The k coefficient values assessing the intraobserver reliability of the system devised by Resch whether the fracture was impaction or distraction type was 0.68, valgus or varus type was 0.70, and flexion or extension type was 0.45. The overall k coefficient value for interobserver reliability using the system devised by Resch was 0.15, with a respective intraobserver value of 0.54. The agreement based on specific subgroups is reported in Tables VI, VII, and VIII.
Discussion Managing complex proximal humeral fractures based on currently available literature is challenging. The most
Proximal humeral fracture classification systems revisited Table I
Neer classification fracture type agreement
Fracture type
k coefficient
One part (minimally displaced) Two partsdanatomic neck Two partsdsurgical neck Two partsdgreater tuberosity Two partsdlesser tuberosity Three partsdgreater tuberosity Three partdlesser tuberosity Four parts Four partsdfracture dislocation
0.34 0.00 0.48 0.43 0.00 0.10 0.00 0.35 0.00
1129 Table III group only Group
k coefficient
Minimally displaced (I) Anatomic neck (II) Surgical neck (III) Greater tuberosity (IV) Lesser tuberosity (V) Fracture dislocation (VI)
0.25 0.00 0.47 0.00 0.00 0.00
Table IV Table II Neer classification agreement considering the number of parts only Number of parts
k coefficient
Minimally displaced Two parts Three parts Four parts
0.34 0.40 0.12 0.35
commonly applied Neer and AO classification systems achieve poor interobserver agreement despite the imaging modality, and some have argued that this problem may jeopardize the interpretation of trial outcomes.5,11 Neer argued that special knowledge of the pathoanatomy is required to address a complicated anatomic problem, and each fragment and its location needs to be identified.17 He and others have reiterated that his classification system is based on a specific series of radiographs as well as intraoperative findings.17,20 More recent studies found the use of stereo-visualization of 3D CT data set reconstructions improved the interobserver reliability of both AO and Neer systems to ‘‘good’’ (k values between 0.6 and 0.8) and found significant improvement of the intraobserver reliability to ‘‘good’’ for the AO and to ‘‘excellent’’ (k > 0.8) for the Neer classifications when compared with plain radiographs and 2D CT.4 However, visualization of the fracture will not benefit in the description of a fracture if the classification system is not readily applicable. Imaging may be a confounding cause of low reliability applied to the classification systems, and thus, to accurately assess surgeons’ understanding of the relevant pathoanatomy, we performed this study using rapid sequence prototype models; to our knowledge, it is the first of its kind. Compared with CT studies1,25,26 our study showed comparably low levels of interobserver reliability using the Neer system. Simplification of the system depending on group or type did not significantly improve the reliability of the system, with moderate intraobserver values attained throughout. Assessing the AO system in its entirety, we found poor interobserver reliability (k ¼ 0.11) which was comparably lower than studies have reported using CT
Neer classification agreement considering the
AO classification fracture type agreement
Fracture type
k Coefficient
A1.1 A1.2 A1.3 A2.1 A2.2 A2.3 A3.1 A3.2 A3.3 B1.1 B1.2 B1.3 B2.1 B2.2 B2.3 B3.1 B3.2 B3.3 C1.1 C1.2 C1.3 C2.1 C2.2 C2.3 C3.1 C3.2 C3.3
0.27 0.75 0.00 0.00 0.13 0.10 0.06 0.21 0.07 0.01 0.03 0.01 0.00 0.00 0.01 0.00 0.16 0.02 0.21 0.00 0.00 0.06 0.04 0.01 0.00 0.06 0.11
(k ¼ 0.32).25,26 An improvement after simplification of the system to type only reached comparable levels with other studies. Neither the Neer or AO classifications systems, nor their simplification, reached the levels of reliability seen using stereo-visualization of rendered data sets.4 Sidor et al23 showed in their study the highest reproducibility was achieved by the shoulder surgical specialist; however, our study only used senior shoulder surgeons as observers. The 2-part and 4-part fractures were better understood and agreed upon using the Neer system compared with 3-part fractures. Indeed, when comparing the k values for 2-part greater tuberosity fractures using the Neer system (k ¼ 0.43) and the AO 1.12 (k ¼ 0.75), there appears to be
1130 Table V
A. Majed et al. Codman-Hertel fracture type agreement
Fracture type
k coefficient
1 2 3 4 5 6 7 8 9 10 11 12
0.69 0.49 0.53 0.43 0.39 0.00 0.35 0.15 0.14 0.00 0.16 0.46
Table VI Resch fracture type agreement considering impaction and distraction subgrouping Fracture mechanism type
k coefficient
Distraction Impaction Neutral Unclassifiable
0.50 0.53 0.00 0.51
considerable difference, suggesting the observer is able to recognize the anatomy involved but application of the rules of angulation and displacement remains problematic. The low values seen with 3-part fractures may be attributed to angulation and displacement, especially when complicated by fractures with articular involvement in a head-splitting fashion. Higher interobserver reliability scores were seen using the Codman-Hertel classification system. There was improved agreement of surgical neck, anatomic neck, isolated lesser/greater tuberosity fractures, and 3-part and 4part fractures compared with the Neer system (Tables I and V). Again, this suggests the observer is able to appreciate the anatomy involved but difficulties arise when applying rules of angulation and displacement. Simplified items (impaction/distraction; varus/valgus) from the system devised by Resch had moderate interobserver agreement. An interesting finding was that there was also higher agreement of those fractures that are not classifiable (Tables VI, VII, and VIII). These fractures are isolated greater and lesser tuberosities or head-splitting injuries whose injury mechanism is not encompassed by this system. Problems with classification systems are not limited to their interpretation or reliability. Sallay et al21 described a fracture not included in the Neer or AO systems involving the lesser and greater tuberosities held together by the bicipital groove. This finding was independently noted by Tamai et al28 during their comparative study of plain radiographic and surgical findings. This fracture pattern
Table VII Resch fracture type agreement considering varus and valgus subgrouping Fracture mechanism type
k coefficient
Valgus Varus Neutral Unclassifiable
0.46 0.59 0.05 0.51
Table VIII Resch fracture type agreement considering flexion and extension subgrouping Fracture mechanism type
k coefficient
Extension Flexion Neutral Unclassifiable
0.05 0.00 0.05 0.51
was subsequently described by Edelson et al9 as the ‘‘shield-type’’ injury, but also appears in Codman’s diagram depicting fracture types.6 Furthermore, Meyer et al15 have also noted a variant of this fracture configuration, with the humeral shaft attached to the head by the bicipital groove. Tamai et al28 also noted a 3-part fracture configuration consisting of the humeral head and greater tuberosity as 2 separate segments, with another fragment consisting of the lesser tuberosity attached to the humeral shaft. They argued that this fracture does not conform to the Neer classification description of a 3-part fracture and may resemble AO C2.2 fractures with respect to the radiologic scheme but is not a 4-part fracture as defined in the Jakob classification.28 We applied prototype engineering techniques to provide the surgeon/observer with the fracture in vivo to be able to manipulate, study, and interpret it in all planes. Because of perioperative visualization restrictions secondary to soft tissue coverage and accessibility issues, in some circumstances, the prototype models may give the observer potentially more information or from a novel perspective than during surgery. We believe that using these models addresses Neer’s concept first described in 197019 and then in 200217 that fracture classification is performed not only by imaging but also at the time of surgery. We believe the quality of the models is comparable to any achievable 3D reconstruction because rendering is the initial step of our process, and compared with other studies, whose CT slice cuts were between 1.5 mm4 and 3.0 mm,25,26 ours were 2.0 mm. The number of cases in our study was nearly double that in the study group assessed by Brunner et al.4 Finally, we have addressed the concepts of pathoanatomy by assessing the Codman-Hertel system10 and newly devised system by Resch. The Codman-Hertel system elegantly describes the number of parts and which anatomic segments are involved, and the system by Resch system applies the pathomechanical forces to the injury.
Proximal humeral fracture classification systems revisited Limitations in our study include the nature of our regional service, such that patients may not have undergone CT as part of their initial investigation and were thus excluded from the study. This may have lead to bias of our cohort with an artificially high proportion of complex fracture patterns compared with a more general unit. Further limitations included the number of patients in the study.
Conclusion Interobserver k values were moderate to low for all classification systems assessed. The simplification of the Neer and AO systems did not improve outcome. Improved outcomes occurred with the Codman-Hertel system, followed by simplified elements of the Resch system, reflecting a better understanding of proximal humeral fractures according to anatomic fragments and pathomechanics. In general, surgeons appear to have difficulty applying fracture pattern into specific groups, and perhaps there is a need for a clear definition of the groups. In defining the validity of a classification system, it must address the pertinent questions asked to decide on the management of the fracture and thus the patient.
Acknowledgments The authors would like to thank Shirley Fetherston for her radiological expertise and Joseph Eliahoo for his statistical advice.
Disclaimer Addie Majed received a charitable grant from the Sir Siegmund Warburg Voluntary Settlement Fund, which was an outside source of funds involved in data collection, data analysis, and preparation of the manuscript. The authors, their immediate families, and any research foundations with which they are affiliated did not receive any financial payments or other benefits from any commercial entity related to the subject of this article. No benefits were received or will be received from any commercial party related directly or indirectly to the subject of this article.
References 1. Bernstein J, Adler LM, Blank JE, Dalsey RM, Williams GR, Iannotti JP. Evaluation of the Neer system of classification of proximal humeral fractures with computerized tomographic scans and plain radiographs. J Bone Joint Surg Am 1996;78:1371-5.
1131 2. Brien H, Noftall F, MacMaster S, Cummings T, Landells C, Rockwood P. Neer’s classification system: a critical appraisal. J Trauma 1995;38:257-60. 3. Brorson S, Bagger J, Sylvest A, Hrøbjartsson A. Improved interobserver variation after training of doctors in the Neer system. A randomised trial. J Bone Joint Surg Br 2002;84:950-4. doi:10.1302/ 0301-620X.84B7.13010 4. Brunner A, Honigmann P, Treumann T, Babst R. The impact of stereo-visualisation of three-dimensional CT datasets on the interand intraobserver reliability of the AO/OTA and Neer classifications in the assessment of fractures of the proximal humerus. J Bone Joint Surg Br 2009;91:766-71. doi:10.1302/0301-620X. 91B6.22109 5. Burstein AH. Fracture classifications: do they work and are they useful? J Bone Joint Surg Am 1993;75:1743-4. 6. Codman EA. Fractures in relation to the subacromial bursa. In: Codman EA, editor. The shoulder, rupture of the supraspinatus tendon and other lesions in or about the subacromial bursa. Malabar, FL: Krieger Publishing; 1934. p. 313-31. 7. Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas 1960;20:37-46. 8. Copeland SA. Fractures of the proximal humerus. Curr Orthop 1995;9: 241-8. 9. Edelson G, Kelly I, Vigder F, Reis ND. A three-dimensional classification for fractures of the proximal humerus. J Bone Joint Surg Br 2004;86:413-25. doi:10.1302/0301-620X.86B3.14428 10. Hertel R, Hempfing A, Stiehler M, Leunig M. Predictors of humeral head ischemia after intracapsular fracture of the proximal humerus. J Shoulder Elbow Surg 2004;13:427-33. doi:10.1016/j.jse.2004.01. 034 11. Jakob R, Kristiansen T, Mayo K, Ganz R, M€uller ME. Classification and aspects of treatment of fractures of the proximal humerus. In: Bateman JE, Welsh RP, editors. Surgery of the houlder. Philadelphia: BC Decker Inc; 1984. p. 330-43. 12. Kannus P, Palvanen M, Niemi S, Parkkari J, J€arvinen M, Vuori I. Osteoporotic fractures of the proximal humerus in elderly Finnish persons: sharp increase in 1970-1998 and alarming projections for the new millennium. Acta Orthop Scand 2000;71:465-70. 13. Kristiansen B, Andersen UL, Olsen CA, Varmarken JE. The Neer classification of fractures of the proximal humerus. An assessment of interobserver variation. Skeletal Radiol 1988;17:420-2. 14. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159-74. 15. Meyer DC, Espinosa N, Hertel R. Combined fracture of the greater and lesser tuberosities with intact connection of the humeral head to the shaft. J Trauma 2006;61:206-8. doi:10.1097/01.ta.0000224111. 21436.7f 16. M€uller ME, Nazarian S, Koch P, Schatzker J. The comprehensive classification of fractures of long bones. Berlin: Springer Verlag; 1990. p. 120-1. 17. Neer CS 2nd. Four-segment classification of proximal humeral fractures: purpose and reliable use. J Shoulder Elbow Surg 2002;11:389400. doi:10.1067/mse.2002.124346 18. Neer CS 2nd. Displaced proximal humeral fractures, part II. Treatment of three-part and four-part displacement. J Bone Joint Surg Am 1970; 52:1090-103. 19. Neer CS 2nd. Displaced proximal humeral fractures, part I. Classification and evaluation. J Bone Joint Surg Am 1970;52:1077-89. 20. Rockwood CA Jr. Fracture classification systems: do they work and are they useful? [correspondence]. J Bone Joint Surg Am 1994; 76:790. 21. Sallay PI, Pedowitz RA, Mallon WJ, Vandemark RM, Dalton JD, Speer KP. Reliability and reproducibility of radiographic interpretation of proximal humeral fracture pathoanatomy. J Shoulder Elbow Surg 1997;6:60-9. 22. Shrader MW, Sanchez-Sotelo J, Sperling JW, Rowland CM, Cofield RH. Understanding proximal humerus fractures: image
1132 analysis, classification, and treatment. J Shoulder Elbow Surg 2005; 14:497-505. doi:10.1016/j.jse.2005.02.014 23. Sidor ML, Zuckerman JD, Lyon T, Koval K, Cuomo F, Schoenberg N. The Neer classification system for proximal humeral fractures: an assessment of interobserver reliability and intraobserver reproducibility. J Bone Joint Surg Am 1993;75:1745-50. 24. Siebenrock KA, Gerber C. The reproducibility of classification of fractures of the proximal end of the humerus. J Bone Joint Surg Am 1993;75:1751-5. 25. Sj€ oden GO, Movin T, G€untner P, Aspelin P, Ahrengart L, Ersmark H, et al. Poor reproducibility of classification of proximal humeral
A. Majed et al. fractures: additional CT of minor value. Acta Orthop Scand 1997;68: 239-42. 26. Sj€oden GO, Movin T, Aspelin P, G€untner P, Shalabi A. 3Dradiographic analysis does not improve the Neer and AO classifications of proximal humeral fractures. Acta Orthop Scand 1999;70: 325-8. 27. Stimson B. A manual of fractures and dislocations. 2nd ed. Philadelphia: Lea & Febiger; 1947. 28. Tamai K, Hamada J, Ohno W, Saotome K. Surgical anatomy of multipart fractures of the proximal humerus. J Shoulder Elbow Surg 2002;11:421-7. doi:10.1067/mse.2002.126101
www.elsevier.com/locate/ymse
Proximal humeral fracture classification systems revisited Addie Majed, MRCSa,*, Iain Macleod, FRCS(Orth)b, Anthony M.J. Bull, PhDc, Karol Zyto, MDd, Herbert Resch, MDe, Ralph Hertel, MDf, Peter Reilly, MD, FRCS(Orth)a, Roger J.H. Emery, MS, FRCSa a
Division of Surgery Oncology Reproductive Medicine and Anaesthetics, Imperial College London, UK ImperialCollege Healthcare NHS Trust, St. Mary’s Campus, London, UK c Department of Bioengineering, Imperial College London, UK d Queen Sophia Hospital, Stockholm, Sweden e UniversityHospital, Salzburg, Austria f Lindenhofspital, Berne, Switzerland b
Hypothesis: This study evaluated several classification systems and expert surgeons’ anatomic understanding of these complex injuries based on a consecutive series of patients. We hypothesized that current proximal humeral fracture classification systems, regardless of imaging methods, are not sufficiently reliable to aid clinical management of these injuries. Materials and methods: Complex fractures in 96 consecutive patients were investigated by generation of rapid sequence prototyping models from computed tomography Digital Imaging and Communications in Medicine (DICOM) imaging data. Four independent senior observers were asked to classify each model using 4 classification systems: Neer, AO, Codman-Hertel, and a prototype classification system by Resch. Interobserver and intraobserver k coefficient values were calculated for the overall classification system and for selected classification items. Results: The k coefficient values for the interobserver reliability were 0.33 for Neer, 0.11 for AO, 0.44 for Codman-Hertel, and 0.15 for Resch. Interobserver reliability k coefficient values were 0.32 for the number of fragments and 0.30 for the anatomic segment involved using the Neer system, 0.30 for the AO type (A, B, C), and 0.53, 0.48, and 0.08 for the Resch impaction/distraction, varus/valgus and flexion/extension subgroups, respectively. Three-part fractures showed low reliability for the Neer and AO systems. Discussion: Currently available evidence suggests facture classifications in use have poor intra- and interobserver reliability despite the modality of imaging used thus making treating these injuries difficult as weak as affecting scientific research as well. This study was undertaken to evaluate the reliability of several systems using rapid sequence prototype models. Conclusion: Overall interobserver k values represented slight to moderate agreement. The most reliable interobserver scores were found with the Codman-Hertel classification, followed by elements of Resch’s trial system. The AO system had the lowest values. The higher interobserver reliability values for the
Joint RNOH/IOMS National Research Ethics Committee approval was received to analyze deidentified medical images of proximal humeral fractures (REC reference number: 07/H0724/40). *Reprint requests: Addie Majed, MRCS, Clinical Research Fellow, Division of SORA (Surgery Oncology Reproductive Medicine and
Anaesthetics), Imperial College London, 10th Flr QEQM Building, St. Mary’s Hospital, Praed St, London W2 1NY, UK. E-mail address: [email protected] (A. Majed).
1058-2746/$ - see front matter Ó 2011 Journal of Shoulder and Elbow Surgery Board of Trustees. doi:10.1016/j.jse.2011.01.020
1126
A. Majed et al.
Codman-Hertel system showed that is the only comprehensive fracture description studied, whereas the novel classification by Resch showed clear definition in respect to varus/valgus and impaction/distraction angulation. Level of evidence: Level III, Diagnostic Study. Ó 2011 Journal of Shoulder and Elbow Surgery Board of Trustees. Keywords: Proximal humerus; fracture classifications; prototype-modelling; interobserver and intraobserver reliability
The incidence of proximal humeral fractures is reported to be 6% per 10,000 fractures, with approximately 15% of these being complex 3-part and 4-part fractures.27 The incidence is increasing, along with an increasing tendency to more severe fractures.8,12,18 The management of these often debilitating injuries is based on various fracture classification systems.16,18 A fracture classification should aim to improve the understanding and therefore the management of fractures based on the morphology of the fracture, the biologic and mechanical behavior, and to provide therapeutic guidelines based on clinical outcomes.16 It should be comprehensive with all fracture types classifiable. Its application should lead to reproducible results. The main criterion for a good classification system is that it addresses the clinically relevant question. Studies that have applied current proximal humeral fracture classification systems to plain radiographs show poor interobserver and intraobserver reliability,2,13,23,24 with 2-dimensional imaging being attributed to low levels of agreement. Some authors have advocated the need for formal training in using the classification systems.3,22 Despite the application of computed tomography (CT) scans and 3-dimensional (3D) computational CT reconstructions, the reliability of these classifications systems remains in doubt.1,21,25,26 Recently, the effect of stereo visualization of 3D CT datasets has suggested some improvement in overall reliability.4 The ability to represent proximal humeral fracture morphology physically using rapid sequence prototype modelling provides an experimental technique that may allow classification systems to be tested by ruling out imaging as a cause of low reliability. Through this technology, the observer could be provided with highly accurate physical models of the fracture that can be orientated freely. After a literature review, we selected 4 fracture classification systems that are in use or have the potential for use: The Neer system19 groups each fracture by the number of fracture segments and describes the fractured anatomic segment as a part. The groups are: group I, nondisplaced; group II, 2-part; group III, 3-part; and group IV, 4-part. There are 16 different potential types of fracture. The AO system describes each fracture as being 1 of 3 types, with 3 subdivisions each. Type A indicates extraarticular unifocal fractures; type B, extraarticular bifocal
fractures; and type C, articular fractures. The 3 subdivisions (groups 1, 2, and 3) are related to the pattern of the fracture.16 A further subdivision into 3 subgroups (1, 2 and 3) is achieved based on the degree of fragmentation, giving 27 different fracture patterns. The Codman-Hertel binary fracture description system10 was derived from the original drawings of Codman6 and was based on the analysis of fracture planes and not on the number of fragments. It comprises 12 different basic fracture patterns that are described with numbers from 1 to 12. There are 6 possible fracture combinations dividing the humerus into 2 parts, 5 possible fractures dividing the humerus into 3 parts, and 1 fracture type dividing the humerus into 4 parts. However, the system does not address fracture pathomechanics. The fourth system tested, devised by Resch, is a proposed addendum to the Codman-Hertel classification system and addresses fracture angulation and pathomechanics. It describes 3 biomechanical planes of injury that are classified by the observer: First, the fracture is described as an ‘‘impaction’’ or ‘‘distraction’’ injury in the coronal plane. An impaction injury occurs when the length position of the fractured greater tuberosity is unchanged and the total length of the humerus is reduced due to impaction of the head (Fig. 1, A). A distraction injury is defined as increased distance on the lateral side between shaft and head fragment (subcapital fracture) or between the shaft and the fractured greater tuberosity (subcapital fracture accompanied by a greater tuberosity fracture; Fig. 1, B). If there is little impaction or distraction, a ‘‘neutral’’ injury is assigned. Second, the fracture is classified as ‘‘varus,’’ ‘‘valgus,’’ or ‘‘neutral,’’ depending on the deformation of the head relative to the shaft (head inclination) in the coronal plane. Finally, the head-shaft angle in the transscapular plane is assessed and described as ‘‘flexion,’’ ‘‘extension,’’ or ‘‘neutral’’ (Fig. 2). Thus, addressing fracture angulation may also facilitate fracture reduction. We hypothesized that current proximal humeral fracture classification systems, regardless of imaging methods, are not sufficiently reliable to aid clinical management of these injuries. Therefore, the study used physical models to test the reliability of the Neer,19 AO,16 CodmanHertel10 proximal humeral fracture classification systems, as well as a primary version of a newly-devised
Proximal humeral fracture classification systems revisited
1127
Figure 1 (A) Impaction injury using the Resch description. The arrows depict the direction of the acting force. (B) Distraction injury using the Resch description. The arrows depict the direction of the acting force.
Materials and methods
Figure 2 Flexion of the head-shaft angle, assessed in the transscapular plane using the Resch description.
classification designed by Resch that aims to complement the Codman-Hertel classification system to include pathomechanics.
We analyzed deidentified images from a database of 100 consecutive patients who were treated at a single center for fractures of the proximal humerus and were investigated with computed tomography (CT) scanning between 2001 and 2007. Three patients were excluded because the fractures showed radiologic evidence of union and the data set for one was corrupted, leaving 96 patients available for this study. All CT scans were performed on a Philips MX8000 or Philips Brilliance 64 multiscanner (Koninklijke Philips N.V, Netherlands) with a slice thickness of 2.0 mm. The primary image plane was axial with a restricted field of view to the shoulder of 250 mm, and reformatting was performed using the Philips bone algorithm with a slice thickness of 1.0 mm. The DICOM data sets were transferred to a computer workstation using MIMICS 11.0 software (Materialise, Leuven, Belgium), and thresholding was conducted using bone window values (range, 226‑2799 Hounsfield units). For each image set, standard 3D volume‑rendered images were produced, and these files were then converted to Rapid Sequence Prototype models after formatting with ZPrint software, which allowed each model to be printed on a ZPrinter 310 printer (ZCorp, Burlington, MA, USA), with resolution of 300 450 dpi, and 0.0875 to 0.1‑mm layer thickness (Fig 3). The models were deidentified and arranged in a randomized order with no discernable markers. Each observer performed the assessment individually and was not given any feedback. Each observer was provided with diagrams of the Neer,19 AO,16 Codman-Hertel,10 and Resch classification systems and was given 1 minute to classify each model. One observer repeated the study 8 weeks later to allow for intraobserver reliability scores to be calculated.
1128
A. Majed et al.
Figure 3 (A) A photograph of a rapid sequence prototype model of a proximal humeral fracture viewed in the coronal plane. (B) A photograph of a rapid sequence prototype model show an inferomedial view of a proximal humeral fracture. Four senior observers (P.R., K.Z., H.R., and R.H.), who work in regional referral units and are perceived as experts in proximal humeral fracture management, were selected for this study. Two observers (R.H., H.R.) were the authors of the classifications used in this study.
Statistical analysis To assess interobserver and intraobserver reliability, k correlation coefficients were calculated using the method reported by Cohen et al7 using Stata 10.0 software (StataCorp LP College Station, Texas, USA). When all the observers agree in all cases, the k value is þ1. A k of 0.00 indicates that agreement is equal to that occurring by chance, with k values of less than 0.00 indicating poor agreement. The k coefficients were interpreted using the Landis and Koch criteria,14 where a k of more than 0.8 represents excellent agreement, between 0.6 and 0.8 is good agreement, between 0.4 and 0.6 is moderate agreement, between 0.2 and 0.4 is fair agreement, between 0 and 0.2 is slight agreement, and of less than 0 is poor agreement. Subgroup interobserver reliability k correlation coefficients were also calculated.
Results The k coefficient values for the interobserver reliability using the Neer classification was 0.33, with a respective intraobserver value of 0.57. The Neer classification was assessed using the ‘‘group’’ and the number of fracture fragments involved (eg, 3-part or 4-part). The k coefficient interobserver and intraobserver reliability values were 0.32 and 0.59, respectively. The Neer classification was simplified assessing only the fracture type, the anatomic segment involved (eg, greater tuberosity fracture), and the reliability of the system was also tested. The k coefficient interobserver and intraobserver reliability values were 0.30 and 0.54, respectively. The interobserver reliability based on
specific fracture types is reported in Table I, and the reliability based on a simplification of the classification to the number of parts and group (anatomic segments) involved are summarized in Tables II and III, respectively. Higher scores were attained for 2-part and 4-part fractures than with 3-part fractures. The k coefficient value for interobserver reliability using the AO classification was 0.11, with a respective intraobserver value of 0.42. Simplified interobserver and intraobserver reliability assessing whether the fracture was AO type (A, B, C) was 0.30 and 0.63, respectively. The interobserver reliability based on specific subgroups is reported in Table IV. The k coefficient value for interobserver reliability using the Codman-Hertel classification was 0.44, with a respective intraobserver value of 0.65. The interobserver reliability based on specific subgroups is reported in Table V. The k coefficient value assessing the interobserver reliability of the simplified system devised by Resch for whether the fracture is simply an impaction or distraction type was 0.52, valgus or varus type was 0.48, and flexion or extension type was 0.08. The k coefficient values assessing the intraobserver reliability of the system devised by Resch whether the fracture was impaction or distraction type was 0.68, valgus or varus type was 0.70, and flexion or extension type was 0.45. The overall k coefficient value for interobserver reliability using the system devised by Resch was 0.15, with a respective intraobserver value of 0.54. The agreement based on specific subgroups is reported in Tables VI, VII, and VIII.
Discussion Managing complex proximal humeral fractures based on currently available literature is challenging. The most
Proximal humeral fracture classification systems revisited Table I
Neer classification fracture type agreement
Fracture type
k coefficient
One part (minimally displaced) Two partsdanatomic neck Two partsdsurgical neck Two partsdgreater tuberosity Two partsdlesser tuberosity Three partsdgreater tuberosity Three partdlesser tuberosity Four parts Four partsdfracture dislocation
0.34 0.00 0.48 0.43 0.00 0.10 0.00 0.35 0.00
1129 Table III group only Group
k coefficient
Minimally displaced (I) Anatomic neck (II) Surgical neck (III) Greater tuberosity (IV) Lesser tuberosity (V) Fracture dislocation (VI)
0.25 0.00 0.47 0.00 0.00 0.00
Table IV Table II Neer classification agreement considering the number of parts only Number of parts
k coefficient
Minimally displaced Two parts Three parts Four parts
0.34 0.40 0.12 0.35
commonly applied Neer and AO classification systems achieve poor interobserver agreement despite the imaging modality, and some have argued that this problem may jeopardize the interpretation of trial outcomes.5,11 Neer argued that special knowledge of the pathoanatomy is required to address a complicated anatomic problem, and each fragment and its location needs to be identified.17 He and others have reiterated that his classification system is based on a specific series of radiographs as well as intraoperative findings.17,20 More recent studies found the use of stereo-visualization of 3D CT data set reconstructions improved the interobserver reliability of both AO and Neer systems to ‘‘good’’ (k values between 0.6 and 0.8) and found significant improvement of the intraobserver reliability to ‘‘good’’ for the AO and to ‘‘excellent’’ (k > 0.8) for the Neer classifications when compared with plain radiographs and 2D CT.4 However, visualization of the fracture will not benefit in the description of a fracture if the classification system is not readily applicable. Imaging may be a confounding cause of low reliability applied to the classification systems, and thus, to accurately assess surgeons’ understanding of the relevant pathoanatomy, we performed this study using rapid sequence prototype models; to our knowledge, it is the first of its kind. Compared with CT studies1,25,26 our study showed comparably low levels of interobserver reliability using the Neer system. Simplification of the system depending on group or type did not significantly improve the reliability of the system, with moderate intraobserver values attained throughout. Assessing the AO system in its entirety, we found poor interobserver reliability (k ¼ 0.11) which was comparably lower than studies have reported using CT
Neer classification agreement considering the
AO classification fracture type agreement
Fracture type
k Coefficient
A1.1 A1.2 A1.3 A2.1 A2.2 A2.3 A3.1 A3.2 A3.3 B1.1 B1.2 B1.3 B2.1 B2.2 B2.3 B3.1 B3.2 B3.3 C1.1 C1.2 C1.3 C2.1 C2.2 C2.3 C3.1 C3.2 C3.3
0.27 0.75 0.00 0.00 0.13 0.10 0.06 0.21 0.07 0.01 0.03 0.01 0.00 0.00 0.01 0.00 0.16 0.02 0.21 0.00 0.00 0.06 0.04 0.01 0.00 0.06 0.11
(k ¼ 0.32).25,26 An improvement after simplification of the system to type only reached comparable levels with other studies. Neither the Neer or AO classifications systems, nor their simplification, reached the levels of reliability seen using stereo-visualization of rendered data sets.4 Sidor et al23 showed in their study the highest reproducibility was achieved by the shoulder surgical specialist; however, our study only used senior shoulder surgeons as observers. The 2-part and 4-part fractures were better understood and agreed upon using the Neer system compared with 3-part fractures. Indeed, when comparing the k values for 2-part greater tuberosity fractures using the Neer system (k ¼ 0.43) and the AO 1.12 (k ¼ 0.75), there appears to be
1130 Table V
A. Majed et al. Codman-Hertel fracture type agreement
Fracture type
k coefficient
1 2 3 4 5 6 7 8 9 10 11 12
0.69 0.49 0.53 0.43 0.39 0.00 0.35 0.15 0.14 0.00 0.16 0.46
Table VI Resch fracture type agreement considering impaction and distraction subgrouping Fracture mechanism type
k coefficient
Distraction Impaction Neutral Unclassifiable
0.50 0.53 0.00 0.51
considerable difference, suggesting the observer is able to recognize the anatomy involved but application of the rules of angulation and displacement remains problematic. The low values seen with 3-part fractures may be attributed to angulation and displacement, especially when complicated by fractures with articular involvement in a head-splitting fashion. Higher interobserver reliability scores were seen using the Codman-Hertel classification system. There was improved agreement of surgical neck, anatomic neck, isolated lesser/greater tuberosity fractures, and 3-part and 4part fractures compared with the Neer system (Tables I and V). Again, this suggests the observer is able to appreciate the anatomy involved but difficulties arise when applying rules of angulation and displacement. Simplified items (impaction/distraction; varus/valgus) from the system devised by Resch had moderate interobserver agreement. An interesting finding was that there was also higher agreement of those fractures that are not classifiable (Tables VI, VII, and VIII). These fractures are isolated greater and lesser tuberosities or head-splitting injuries whose injury mechanism is not encompassed by this system. Problems with classification systems are not limited to their interpretation or reliability. Sallay et al21 described a fracture not included in the Neer or AO systems involving the lesser and greater tuberosities held together by the bicipital groove. This finding was independently noted by Tamai et al28 during their comparative study of plain radiographic and surgical findings. This fracture pattern
Table VII Resch fracture type agreement considering varus and valgus subgrouping Fracture mechanism type
k coefficient
Valgus Varus Neutral Unclassifiable
0.46 0.59 0.05 0.51
Table VIII Resch fracture type agreement considering flexion and extension subgrouping Fracture mechanism type
k coefficient
Extension Flexion Neutral Unclassifiable
0.05 0.00 0.05 0.51
was subsequently described by Edelson et al9 as the ‘‘shield-type’’ injury, but also appears in Codman’s diagram depicting fracture types.6 Furthermore, Meyer et al15 have also noted a variant of this fracture configuration, with the humeral shaft attached to the head by the bicipital groove. Tamai et al28 also noted a 3-part fracture configuration consisting of the humeral head and greater tuberosity as 2 separate segments, with another fragment consisting of the lesser tuberosity attached to the humeral shaft. They argued that this fracture does not conform to the Neer classification description of a 3-part fracture and may resemble AO C2.2 fractures with respect to the radiologic scheme but is not a 4-part fracture as defined in the Jakob classification.28 We applied prototype engineering techniques to provide the surgeon/observer with the fracture in vivo to be able to manipulate, study, and interpret it in all planes. Because of perioperative visualization restrictions secondary to soft tissue coverage and accessibility issues, in some circumstances, the prototype models may give the observer potentially more information or from a novel perspective than during surgery. We believe that using these models addresses Neer’s concept first described in 197019 and then in 200217 that fracture classification is performed not only by imaging but also at the time of surgery. We believe the quality of the models is comparable to any achievable 3D reconstruction because rendering is the initial step of our process, and compared with other studies, whose CT slice cuts were between 1.5 mm4 and 3.0 mm,25,26 ours were 2.0 mm. The number of cases in our study was nearly double that in the study group assessed by Brunner et al.4 Finally, we have addressed the concepts of pathoanatomy by assessing the Codman-Hertel system10 and newly devised system by Resch. The Codman-Hertel system elegantly describes the number of parts and which anatomic segments are involved, and the system by Resch system applies the pathomechanical forces to the injury.
Proximal humeral fracture classification systems revisited Limitations in our study include the nature of our regional service, such that patients may not have undergone CT as part of their initial investigation and were thus excluded from the study. This may have lead to bias of our cohort with an artificially high proportion of complex fracture patterns compared with a more general unit. Further limitations included the number of patients in the study.
Conclusion Interobserver k values were moderate to low for all classification systems assessed. The simplification of the Neer and AO systems did not improve outcome. Improved outcomes occurred with the Codman-Hertel system, followed by simplified elements of the Resch system, reflecting a better understanding of proximal humeral fractures according to anatomic fragments and pathomechanics. In general, surgeons appear to have difficulty applying fracture pattern into specific groups, and perhaps there is a need for a clear definition of the groups. In defining the validity of a classification system, it must address the pertinent questions asked to decide on the management of the fracture and thus the patient.
Acknowledgments The authors would like to thank Shirley Fetherston for her radiological expertise and Joseph Eliahoo for his statistical advice.
Disclaimer Addie Majed received a charitable grant from the Sir Siegmund Warburg Voluntary Settlement Fund, which was an outside source of funds involved in data collection, data analysis, and preparation of the manuscript. The authors, their immediate families, and any research foundations with which they are affiliated did not receive any financial payments or other benefits from any commercial entity related to the subject of this article. No benefits were received or will be received from any commercial party related directly or indirectly to the subject of this article.
References 1. Bernstein J, Adler LM, Blank JE, Dalsey RM, Williams GR, Iannotti JP. Evaluation of the Neer system of classification of proximal humeral fractures with computerized tomographic scans and plain radiographs. J Bone Joint Surg Am 1996;78:1371-5.
1131 2. Brien H, Noftall F, MacMaster S, Cummings T, Landells C, Rockwood P. Neer’s classification system: a critical appraisal. J Trauma 1995;38:257-60. 3. Brorson S, Bagger J, Sylvest A, Hrøbjartsson A. Improved interobserver variation after training of doctors in the Neer system. A randomised trial. J Bone Joint Surg Br 2002;84:950-4. doi:10.1302/ 0301-620X.84B7.13010 4. Brunner A, Honigmann P, Treumann T, Babst R. The impact of stereo-visualisation of three-dimensional CT datasets on the interand intraobserver reliability of the AO/OTA and Neer classifications in the assessment of fractures of the proximal humerus. J Bone Joint Surg Br 2009;91:766-71. doi:10.1302/0301-620X. 91B6.22109 5. Burstein AH. Fracture classifications: do they work and are they useful? J Bone Joint Surg Am 1993;75:1743-4. 6. Codman EA. Fractures in relation to the subacromial bursa. In: Codman EA, editor. The shoulder, rupture of the supraspinatus tendon and other lesions in or about the subacromial bursa. Malabar, FL: Krieger Publishing; 1934. p. 313-31. 7. Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas 1960;20:37-46. 8. Copeland SA. Fractures of the proximal humerus. Curr Orthop 1995;9: 241-8. 9. Edelson G, Kelly I, Vigder F, Reis ND. A three-dimensional classification for fractures of the proximal humerus. J Bone Joint Surg Br 2004;86:413-25. doi:10.1302/0301-620X.86B3.14428 10. Hertel R, Hempfing A, Stiehler M, Leunig M. Predictors of humeral head ischemia after intracapsular fracture of the proximal humerus. J Shoulder Elbow Surg 2004;13:427-33. doi:10.1016/j.jse.2004.01. 034 11. Jakob R, Kristiansen T, Mayo K, Ganz R, M€uller ME. Classification and aspects of treatment of fractures of the proximal humerus. In: Bateman JE, Welsh RP, editors. Surgery of the houlder. Philadelphia: BC Decker Inc; 1984. p. 330-43. 12. Kannus P, Palvanen M, Niemi S, Parkkari J, J€arvinen M, Vuori I. Osteoporotic fractures of the proximal humerus in elderly Finnish persons: sharp increase in 1970-1998 and alarming projections for the new millennium. Acta Orthop Scand 2000;71:465-70. 13. Kristiansen B, Andersen UL, Olsen CA, Varmarken JE. The Neer classification of fractures of the proximal humerus. An assessment of interobserver variation. Skeletal Radiol 1988;17:420-2. 14. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159-74. 15. Meyer DC, Espinosa N, Hertel R. Combined fracture of the greater and lesser tuberosities with intact connection of the humeral head to the shaft. J Trauma 2006;61:206-8. doi:10.1097/01.ta.0000224111. 21436.7f 16. M€uller ME, Nazarian S, Koch P, Schatzker J. The comprehensive classification of fractures of long bones. Berlin: Springer Verlag; 1990. p. 120-1. 17. Neer CS 2nd. Four-segment classification of proximal humeral fractures: purpose and reliable use. J Shoulder Elbow Surg 2002;11:389400. doi:10.1067/mse.2002.124346 18. Neer CS 2nd. Displaced proximal humeral fractures, part II. Treatment of three-part and four-part displacement. J Bone Joint Surg Am 1970; 52:1090-103. 19. Neer CS 2nd. Displaced proximal humeral fractures, part I. Classification and evaluation. J Bone Joint Surg Am 1970;52:1077-89. 20. Rockwood CA Jr. Fracture classification systems: do they work and are they useful? [correspondence]. J Bone Joint Surg Am 1994; 76:790. 21. Sallay PI, Pedowitz RA, Mallon WJ, Vandemark RM, Dalton JD, Speer KP. Reliability and reproducibility of radiographic interpretation of proximal humeral fracture pathoanatomy. J Shoulder Elbow Surg 1997;6:60-9. 22. Shrader MW, Sanchez-Sotelo J, Sperling JW, Rowland CM, Cofield RH. Understanding proximal humerus fractures: image
1132 analysis, classification, and treatment. J Shoulder Elbow Surg 2005; 14:497-505. doi:10.1016/j.jse.2005.02.014 23. Sidor ML, Zuckerman JD, Lyon T, Koval K, Cuomo F, Schoenberg N. The Neer classification system for proximal humeral fractures: an assessment of interobserver reliability and intraobserver reproducibility. J Bone Joint Surg Am 1993;75:1745-50. 24. Siebenrock KA, Gerber C. The reproducibility of classification of fractures of the proximal end of the humerus. J Bone Joint Surg Am 1993;75:1751-5. 25. Sj€ oden GO, Movin T, G€untner P, Aspelin P, Ahrengart L, Ersmark H, et al. Poor reproducibility of classification of proximal humeral
A. Majed et al. fractures: additional CT of minor value. Acta Orthop Scand 1997;68: 239-42. 26. Sj€oden GO, Movin T, Aspelin P, G€untner P, Shalabi A. 3Dradiographic analysis does not improve the Neer and AO classifications of proximal humeral fractures. Acta Orthop Scand 1999;70: 325-8. 27. Stimson B. A manual of fractures and dislocations. 2nd ed. Philadelphia: Lea & Febiger; 1947. 28. Tamai K, Hamada J, Ohno W, Saotome K. Surgical anatomy of multipart fractures of the proximal humerus. J Shoulder Elbow Surg 2002;11:421-7. doi:10.1067/mse.2002.126101