Proximal humerus fractures have a great variability and complexity. In general only two systems are used for classification: the Neer classification and the classification of the Arbeitsgemeinschaft für Osteosynthesefragen/Association for the Study of Internal Fixation (AO/ASIF) [1–4]. During the last years, a few studies presented fracture types, which were not accurately described with either of the two classification systems [5–7]. Several studies have also shown high rates of inter- and intraobserver variability in both classifications [8–16]. Only the analysis of fractured parts according to Neer and the AO-main-types led to a moderate to substantial reliability. Reliability studies that specifically analysed a very detailed classification (Neer parts and groups, AO – types and groups and subgroups) showed only fair to slight reliability. A low reliability reported in these studies could be a possible reason for discrepancies when comparing outcome studies especially after treatment of complex three-and four-part-fractures [17].

Therefore, new classification systems were introduced during the last years [18, 19]. A new system based on the analysis of more than 250 humeral fractures-was developed, which allows a detailed modular classification of proximal humeral fractures condensing Neer and AO with special regard to defined topographic and morphological criteria (MTM-Classification). The current study assessed the reliability of the MTM-classification and compared the results with fractured parts of the proximal humerus according to Neer-classification using plain x-rays. We also wanted to answer the question whether classification of proximal humeral fractures in this very detailed way, exceeding the analysis of fractured parts (according to Neer), may be limited by its reliability.

In a prospective study during March 2005 and August 2005 a consecutive series of 22 patients with 22 proximal humeral fractures presented at the BG Trauma Centre of the University of Tuebingen were included into the study. All patients were diagnosed with standardized plain x-rays in a.p. and axillary views with the patient supine using a shoulder splint with at least 60° abduction of the arm[21]. In addition, suspected instability in minimally displaced fractures was assessed using fluoroscopy (n = 4). All x-rays were reviewed by an experienced trauma surgeon, who was not participating in the evaluation to ensure that the x-ray quality of all 22 patients was sufficient for later classification. All x-rays were collected and numerated; patient identification data was rendered anonymously. Then, the x-rays were submitted to three observers. Observer 1 was an experienced orthopaedic surgeon with fellowship training in shoulder surgery. Observer 2 was an experienced trauma surgeon with a special interest in bone and joint orthopaedic surgery. Observer 3 was a fellow for trauma surgery with a special interest in bone and joint surgery. None of the examiners were employed by the hospital, in which the fractures were seen. During routine work of the three observers only the Neer and AO classification were used. None of the observers had any experience with the MTM classification before study onset to exclude the influence of training on reliability.

At the beginning of the study, the MTM classification was provided to the examiners in English and German language. In addition, the first author gave a 15-minute presentation of the MTM-Classification. A goniometer and a pen were given to the examiners. The data acquisition of the three examiners took place independently.

All fractures were classified by the observers according to the following guidelines:

a) Topographic analysis by individual fracture types (M to CTM) with consideration of individual specification (S1 to S4, Figures 6, Table 1).

b) Topographical analysis by individual fracture types (M to CTM) (see Table 1).

c) Analysis by a combination of both main fracture types (A, B, C, D) and number of parts (2–4) (see Table 1).

a) Analysis by the 4 main fracture types (A, B, C or D) (Table 1).

d) Analysis by number of main parts (2-part, 3-part, and 4-part according to Neer-classification) (see Table 1).

e) Morphological distinction between minimal-displaced fracture and displaced fracture (S1/S2 or S3/S4) (see Figures 6).

After initial analysis, a second evaluation of the x-rays took place after 4 months. Prior to this part of the study, none of the examiners had knowledge about this second evaluation. The classification was presented again in English and German languages. A routine application of the MTM classification during routine work did not take place. The second evaluation was performed as described above except that radiographs were presented in reverse order. All except one fracture, that was treated conservatively, were stabilized by open reduction and internal fixation with an angular-stable plate osteosynthesis.

The MTM-Classification was developed to advance understanding of proximal humeral fractures. Especially by introduction of incomplete articular fractures (B-fractures) a topographic classification could by possible. By adding morphological aspects and the possibility of a modular application of the system a precise evaluation of almost all fractures may be possible as well. However, the classification of proximal humeral fractures with the MTM-systems results in various differences in reliability depending on the short or extensive version of application. Analyzing the short version of the classification, the interobserver analysis resulted in moderate kappa values in the category 'main parts' (according to Neer-parts) and substantial kappa values for analyzing 'fracture displacement'. Those results were in accordance with the literature[8, 23]. Analyzing the extensive version of the MTM-Classification like evaluation of the main or individual fracture type with inclusion of the specification, only fair results can be expected.

At present the classifications by Neer and the AO for proximal humeral fractures are widely accepted and commonly used although both classifications have received some criticism during the last years [1, 3]. Neither of the two systems achieved the general requirement of classifications to adequately cover the complete spectrum of all possible fracture patterns. The Neer-system does not include 4-part valgus impacted fractures (covered by the AO-Classification as C2.1 or C2.2)[24]. Furthermore, the Neer-classification describes a large and rather inhomogeneous group (one-part fractures/minimal-displacement) that consists of all fractures that are angulated less than 45° and displaced less than 1 cm, independently of the number of fractured parts. More so, 3 and 4-part fractures are divided by Neer in types of lesser tuberosity and greater tuberosity fractures. In these complex fractures, it remains unclear whether a fracture runs through the anatomical or the surgical neck. Furthermore, the AO classification includes three types (A, B and C) each divided into 3 groups and 9 subgroups. For further subgroup description, 41 qualifications were used. Despite this detailed approach, complex injuries such as fractures of the anatomical neck with both tuberosities (4-part fractures according to Neer) or isolated lesser tuberosity fractures are inadequately defined.

New classifications systems such as the MTM classification were developed to address these critical points[18, 19]. As a main prerequisite for correct topographic and morphologic fracture assessment, standardized and good-quality anteroposterior and axillary radiographs are required. Tamai published a series of 22 3 and 4-part fractures. He compared plain x-rays in anteroposterior and scapula lateral views with the intraoperative gross anatomy. Eight cases (36.4%) could neither be classified by Neer nor by the AO classification system. The authors concluded that diagnostics in complex fracture situations are difficult and that the classifications used were not as accurate as one would hope. Furthermore, Tamai attributed present difficulties in the choice of correct therapeutic methods to an inadequate fracture classification system[7].

The MTM classification advanced available classifications by including the fracture model of Codman, the Neer analysis of fractured parts and the AO differentiation of fracture height, allowing for the description of almost all possible fracture types [1, 3, 25].

Also, comparisons of kappa values from different studies may be problematic, as kappa values change with the prevalence of the diagnosis[26]. Despite these problematic aspects, the author's selected a few studies with comparable methods for discussion.

Several authors assess the intraobserver reliability and interobserver reproducibility. Using conventional radiography and in some studies additional CT-scans, they found a fair to substantial mean intraobserver agreement as well as a fair to substantial mean interobserver agreement for the Neer and for the AO classification. (see Table 4).

The present study used standardized plain x-rays, enlisted experienced trauma surgeons and used the whole range of the classification spectrum. This approach led to an average percentage agreement of the three observers of 49.2% with a fair kappa value of 0.39 for fracture-analysis under consideration of individual fracture type and specification.

Siebenrock and Gerber also stated that the AO classification with three main types (A-C) resulted in better kappa values than the modified Neer classification with four choices of fracture types [11]. These findings were confirmed by our study, in which we found acceptable to good agreement in 2 to 3 groups for determination of fracture displacement or fractured main parts. On the other hand, Sidor found that a simplification of the Neer classification from 16 into 6 categories did not result in significant changes of interobserver reliability[10]. However, a 2-choice question, for example about therapeutic management or (as in the present study) about fracture displacement led to perfect and good reliability, indicating a dependency of interobserver reliability and the number of choices the observer has to select[8, 23].

Sidor also discussed that a differentiation of single fragments due to multiple fractures lines is difficult. He also stresses the importance of a high-quality radiological diagnostics, which makes an overlapping free presentation for the fractured region possible to avoid any classification restriction. For an excellent classification of proximal humeral fractures, a perfect radiological visualization of the fractured region is mandatory[10]. In accordance with other experts, it is possible to increase the information with an axillary view in terms of direction and dimension of fracture dislocation, humeral head involvement, and the degree of displacement of tuberosity fractures, especially considering fractures of the lesser tuberosity [27–29].

Brorson and Shrader analyzed the importance of training in a specific classification system. They could show that kappa values for interobserver reliability were significantly improved from fair to substantial after observers had received a classification training[23, 30, 31].

Since the training in a specific classification system improves reliability, one weakness of the current study was, that the MTM-classification was not used in daily clinical practice. With sufficient training, the reliability of this classification could be higher.

In summary, some complex fracture types are inadequately defined by classification systems such as the present Neer or AO classification. To allow a precise topographic and morphological description, the MTM classification was developed for a better understanding of individual fractures and to address the question whether a very detailed classification of proximal humeral fractures may be limited by its reliability. Unfortunately, the very detailed classification approach led only to fair or unacceptable results and is not helpful to improve reliability.

We concluded that a detailed classification exceeding the part analysis of Neer is not a practical approach to address current problems in classification systems regarding proximal humerus fractures.

For future projects an evaluation of ongoing developments of diagnostic imaging-technology like the CT and here specifically multiplanar visualisation of the fracture in thin-cut technique and 3 D visualisation should be undertaken.