Rockwood, Green, and Wilkins' Fractures, 10e Package

116

SECTION ONE • General Principles

and thinking to the learner. When asked to review a fracture radiograph, junior residents often demonstrate their knowledge by citing a well-known classification of that fracture: how to classify the fracture becomes part of the educational discussion.

Compendium, many portions of the alphanumeric code are included for completeness yet have not been demonstrated to be valid with little likelihood that such validity would be iden tified if further investigated. For instance, diaphyseal fractures of the femur are categorized into three types, each of which is categorized into three groups without compelling evidence that these resulting nine categories are important to treatment or prognosis. Validity is linked to reliability. The categories may be import ant, but if they cannot be reliably assessed, the classification cannot be valid. As an example, there would be wide agreement that the severity of soft tissue injury associated with closed frac tures is important; but if a classification that assesses soft tissue injury cannot be reliably used since observers cannot agree, it is not valid or useful. MEASUREMENT Measurement may be embedded in some categorical classifica tions. For instance, a type I versus a type II radial head fracture (as described by Mason) is often defined by millimeters of dis placement and percentage of the articular surface involved. 36 Alternatively, a classification may be solely based on a measure ment, as when assessing calcaneal fractures through depression of Böhler angle. The use of measurement is laudable, and cate gories precisely defined by objective measurements are always preferable to imprecise subjective definitions. However, all mea surements contain some degree of error, either large or small. If the measurement is important, the degree of error inherent in that measurement should be assessed critically. First, consider the difficulty in precisely measuring centime ters or angles, that are hand-drawn with imperfect technique on imaging software. Some degree of error is inherent as part of this measurement technique. Second, many classification systems are designed to be measured on certain radiographic projections. For example, when measuring Pauwels angle in a femoral neck, this measurement is meant to be examined on an anteroposterior (AP) radiograph of a hip. 56 If neither of the orthogonal views presented are exactly a perfect AP view, the amount of error is by definition increased. While this is less important in the Pauwel measurements, as the ranges that need to be distinguished between are quite large (0–30 degrees, 30–50 degrees, and >50 degrees), an imperfect view can have significant ramifications as discussed with Böhler’s angle. In measuring scoliosis radiographs, the error of Cobb angles has been found to be small, but in two commonly used frac ture measurements (Böhler’s angle for calcaneus and degree of articular depression for tibial plateau) the error was found to be very high. 46 These measurements are commonly used and such a high margin of error calls into question their value for classifi cation, treatment, or assessment of prognosis. CATEGORIZING CONTINUOUS VARIABLES Swiontkowski et al. reported a study of pilon fractures using an older version of the OTA/AO fracture classification and high lighted the issue of individual judgment in taking a continuous variable such as fracture pattern and compartmentalizing it into

LIMITATIONS AND CHALLENGES OF FRACTURE CLASSIFICATION SYSTEMS

Improving fracture classification requires understanding and optimization of issues related to reliability, validity, measure ment error, and host/patient factors. A fracture classification must have reasonable reliability and validity to be important and worthwhile. RELIABILITY Reliability of classification has been assessed in many categor ical classifications in various medical disciplines. Given the central role of classification in the treatment of fractures, under standing how reliability is assessed, why it is important, and the limitations reliability imposes on the language we speak is a part of fracture care. For a classification to be reliable, there must be confidence that two observers separately shown the same set of radiographs will agree on the appropriate classification category a reasonable proportion of times. Determining the reliability of a fracture classification requires a scientific exercise where more than one observer classifies a number of fracture radiographs separately from each other and their degree of agreement is measured. Kappa is a statistic where 1.0 indicates perfect agree ment and 0.0 indicates no agreement. The Kappa statistic cor rects for agreement by chance alone. The guidelines of Landis and Koch are used to provide verbal descriptors of the degree of agreement measured by the Kappa statistic. 37 They character ized values less than 0 as indicating no agreement and 0 to 0.20 as slight, 0.21 to 0.40 as fair, 0.41 to 0.60 as moderate, 0.61 to 0.80 as substantial, and 0.81 to 1 as nearly perfect agreement. These guidelines have been applied to fracture classification, and Kappa statistics are widely quoted. Many fracture classifica tions in current use have K values in the fair to moderate range. There are many reasons for fair or even poor reliability of a fracture classification. Poor agreement can result from imprecise or inadequately defined categories (displaced or nondisplaced), nondiscrete categories (small open wound with a high-grade fracture pattern), complex fracture patterns that are hard to interpret (acetabular fractures), poor radiographs, catego ries that are defined by difficult measurements (45 degrees of angulation or 1 cm of displacement in the proximal humerus), nonclassifiable fracture patterns, and assignment of continuous variables to categories such as degree of comminution. VALIDITY For a fracture classification to be valid, the classification must be shown to relate to something that is particularly relevant. Most but not all fracture classifications meet this metric, at least at face value. For instance, in the OTA/AO 2018 Fracture