Introduction

The perfect classification system needs to be reproducible [1] in order to aid accurate communication between surgeons and standardise research. In turn, good reproducibility will influence the overall reliability of that classification. In addition, the perfect classification system should guide the surgeon in his treatment options and predict outcome.

There have been numerous classification systems for subtrochanteric fractures introduced over time as treatment options have been modified (Seinsheimer (1978) [2]; Russell-Taylor classification (1984) [3]; and AO classification [4]). See Figs. 1, 2, 3.

Fig. 1
figure 1

Seinsheimer classification

Full size image
Fig. 2
figure 2

AO classification

Full size image
Fig. 3
figure 3

Russell-Taylor (RT) classification

Full size image

Studies have already suggested poor reproducibility of the Seinsheimer [2] and AO [4] classification systems concluding that they are inaccurate for classifying subtrochanteric fractures.

One study investigating the reproducibility of the Seinsheimer classification showed that only 13 out of 50 fractures (26%) were identically classified by all 4 observers, increasing to 62% when solely identifying the fractures classified as 3a [5]. This is important as the 3a subgroup was shown to have the highest rate of failure of fixation and persistent non-union in the original paper [2]. The complex AO classification system has been shown to be significantly more reliable when classifying by group as opposed to the more detailed subgroup [6].

A cohort of patients was followed up prospectively during the use of a new reconstruction nail—the Synthes© Proximal Femoral Nail Antirotation (PFNA). The poor reproducibility of subtrochanteric fractures classification systems was observed whilst classifying the fractures suitable for this nail use. We, therefore, developed a new classification system, ‘MCG’, as an attempt to improve the reproducibility by simplifying the classification of subtrochanteric fractures with the clinical aid of alerting the surgeon to potential hazards: type 1—subtrochanteric fracture with intact greater and lesser trochanters; type 2—subtrochanteric fractures involving greater trochanter, thus increasing the difficulty of obtaining the correct entry point for nailing; type 3—subtrochanteric fractures involving the lesser trochanter, the most unstable configuration and fractures in which reduction could be difficult to achieve. See Fig. 4.

Fig. 4
figure 4

MCG classification

Full size image

The aim of this study was to assess the intra-observer and inter-observer reproducibility of the Seinsheimer [2], AO [4] and Russell-Taylor (RT) [3] classification systems along with the new system (MCG). Each classification system was noted to have varying numbers of subgroups Seinsheimer = 8, AO = 12, RT = 4 and MCG = 3. We did not use the classification systems during this study to guide treatment options as each fracture was treated with the same device (Synthes© PFNA).

Materials and methods

All patients treated for subtrochanteric fractures in our hospital between April 2006 and April 2007 were included in our study. Pathological fractures and those with inadequate quality radiographs were excluded leaving a total of 32 patients. All 32 anteroposterior (AP) and lateral radiographs were classified independently by 4 observers (2 Consultants and 2 Registrars) on two separate occasions. Each consultant had more than 10 years and each registrar more than 6 years experience in orthopaedics.

The radiographs were presented via a PowerPoint presentation (Microsoft Corp.) containing one patient’s radiographs per slide (one AP, one Lateral). Each observer was provided with printed definitions and diagrams demonstrating the four classification systems of Seinsheimer, AO, RT and MCG. In the form of a test, the observers individually classified and recorded their selections for each fracture according to these classification systems. The observers were permitted to refer to their printed sheets throughout their classification of the fractures.

In order to prevent bias, the observers were not informed and there would be a second test. Six weeks later, the test was repeated but with the sequence of the slides altered. The observers were not informed that it was the same cohort of X-rays as they had observed previously.

Kappa statistics were used to analyse the intra-observer and inter-observer reproducibility initially on each classification system.

In addition, Kappa statistics were also used to assess the reproducibility for certain individual classification grades of each system: Seinsheimer 3a, AO Group 3—subgroup 1 (31-A3.1), RT 1 or 2, RT a or b and MCG 3. The purpose of this element of the study was to simplify the classification systems and attempt to remove the inherent statistical bias that is observed when comparing classification systems that contain varying numbers of subgroups. The possibility is that the classification systems with fewer subgroups may only produce higher reproducibility results due to there being fewer options to choose from when reclassifying.

Kappa is a coefficient of agreement corrected for the probability of agreement by chance, which ranges from +1, representing perfect agreement, through 0, representing chance agreement, to −1, representing absolute disagreement. As defined by Landis and Koch, values of >0.80 are considered to be having almost perfect agreement; 0.61–0.80, substantial agreement; 0.41–0.60, moderate agreement; 0.21–0.40, fair agreement; 0–0.20, slight agreement; and 0, poor agreement [7]. SPSS® 11.0 software was used for the statistical analysis [8].

Results

We were unable to calculate Kappa values for both AO and Seinsheimer classification systems over the two sessions. This was due to the large number of subgroups and relatively small sample size. For the RT classification, the mean kappa values were 0.25 for intra-observer variation and 0.35 for inter-observer variation. The MCG classification results showed a mean kappa value of 0.3 for intra-observer variation and 0.31 for inter-observer variation. See Table 1.

Table 1 Assessment of all four classification systems reproducibility using mean kappa score
Full size table

When analysing the individual classifications to see whether reproducibility improved, the MCG3 showed the best mean kappa value, 0.308, and the narrowest range (0.188–0.412) for intra-observer variation and 0.461 (range 0.446–0.629) for inter-observer variation. The RT 1 or 2 mean kappa value was 0.367 (range 0.271–0.675) for intra-observer variation and 0.387 (range 0.25–0.538) for inter-observer variation. RT a or b mean kappa was 0.296 (range 0.019–0.538) for intra-observer variation and 0.304 (range 0.043–0.592) for inter-observer variation. Seinsheimer 3a group showed a mean kappa value 0.295 (range −0.185–0.634) for intra-observer variation and 0.304 (range 0.004–0.796) for inter-observer variation. The AO31-A3.1 had the lowest mean kappa values, 0.075 (range 0.027–0.188) for intra-observer variation and 0.261 (range 0.125–0.428) for inter-observer variation. See Table 2.

Table 2 Assessment of reproducibility for certain individual classification grades of each system using mean kappa score
Full size table

All of these fractures were treated with the Synthes© Proximal Femoral Nail Antirotation (PFNA), and all but 2 achieved union. The 2 patients who had not united at 4 months were unfortunately lost to follow-up. We defined union based on clinical (no pain, tenderness or abnormal movement at the fracture site) and radiological (3 out of 4 bridging cortices) assessment [9]. Thus, the classification systems did not guide treatment or predict outcome or complications in this small study.

Discussion

The aim of this study was to assess the intra-observer and inter-observer reproducibility of the Seinsheimer [2], AO [4] and Russell-Taylor (RT) [3] classification systems along with a new system (MCG) specifically designed to improve upon the failings of the existing classification systems. Each system has its individual strengths and weaknesses.

The Seinsheimer system is widely used but has already been shown in the literature to have poor reproducibility. It is a descriptive system which offers many possible classification groups, although this may have contributed to its poor reproducibility.

The AO system is even more comprehensive with a greater number of subgroups, although again the benefit to its descriptive value potentially results in a poor reproducibility. Alternatively, the Russell-Taylor system has only 4 subgroups therefore making it less descriptive but possibly more reproducible. The new MCG system was designed specifically in order to be easy to use, be reproducible and be descriptive, primarily for clinical relevance rather than to purely describe the intricacies of fracture configuration. The objective being to create a classification system that would highlight to the surgeon the potential hazards involved in treating such fractures.

Our study agreed with the conclusions of those published by Gehrchen et al. [5] and Pervez et al. [6] who found improvements in the reproducibility of classification systems when observers only had to identify certain classification grades. However, at best, the mean agreement of all four classification systems was fair and when narrowed to certain classification grades, it only improved to moderate agreement. Interestingly, the observers still attained poor reproducibility despite having the classifications available to refer to throughout both tests.

Damany et al. [10] performed a Literature review of subtrochanteric classification publications from 1966 to 2003 and of 110 studies involving 2,725 fractures and 16 classification systems, and none were shown to be of value in determining treatment or for predicting outcome of subtrochanteric fractures. Specifically, we made a detailed comparison of existing classification systems with the addition of a new system that was designed with the explicit intention of addressing the criteria of a classification system. In spite of this, only a moderate inter-observer variation was demonstrated at best with the new classification (MCG).

We were able to test these classifications under standardised conditions with the use of digital radiographs. The small sample size resulted in statistical difficulty when analysing the classification systems with large numbers of subgroups (Seinsheimer and AO). However, given the uncommon nature of this injury and the relatively large size of our prospective cohort, this reflects poorly on the construct of those classification systems, i.e. systems used to predict outcome in uncommon conditions should be simple with minimal subgroups to permit adequate reproducibility.

All our fractures were treated with the same implant. Although this permitted a standardised assessment of outcome due to fracture type, it did not permit comparison of outcomes with different treatment modalities. However, we demonstrated a union rate of greater than 93% at 4 months. This may suggest that the vast majority of subtrochanteric femoral fractures can be treated with modern cephalomedullary nailing devices, thereby eliminating the need for the classification to guide treatment in these fractures.

In conclusion, the four subtrochanteric classification systems which we assessed were not found to be sufficiently reproducible to be of significant value in clinical practice.