Abstract
Aim
The B.A.C.H. system is a new classification for long bone osteomyelitis. It uses the four key inter-disciplinary components of osteomyelitis, namely, bone involvement, anti-microbial options, soft tissue status and host status. This study aims to assess the inter-user reliability of using the B.A.C.H. classification system.
Method
We identified 20 patients who had a diagnosis of long bone osteomyelitis using a previously validated composite protocol. For each patient, osteomyelitis history, past-medical history, clinical imaging (including radiology report), photographs of the affected limb and microbiology were presented to clinical observers on an online form. Thirty observers, varying in clinical experience (training grades and consultants, with a variety of exposure to osteomyelitis) and specialty (orthopaedic surgery, infectious diseases and plastic surgery) were asked to rate the twenty cases of osteomyelitis. Before rating, an explanation of how to use the classification system was given to the observers, in a structured ‘user key’. The responses were assessed by accuracy against a reference value and Fleiss' kappa value (Fκ).
Results
The 30 users individually classified all 20 cases. The observers comprised 14 orthopaedic surgeons, 13 physicians (either microbiology, infectious diseases or anesthetists) and 3 plastic surgeons. The users had a variety of exposure to osteomyelitis ranging from less than one case per month to greater than one case per week.
The accuracy across all variables was 86.2% (95% CI 83.9% – 88.6%, SD 6.2%), with the ‘C’ variable scoring the highest at 92.5% (95% CI 88.5% – 95.6%, SD 8.2%) and the ‘B’ variable scoring the lowest at 77.0% (95% CI 71.2% – 82.8%, SD 15.5%). The variable with the highest agreement between users was the anti-microbial options with a Fκ of 0.815 (95% CI 0.811 – 0.819) which correlated to an ‘almost perfect agreement’. Despite this, the classification of the more complex isolates proved problematic. The most variability was seen in the bone involvement variable with an Fκ of 0.479 (95% CI 0.475 – 0.483) which correlated to a ‘fair agreement’.
Conclusions
The B.A.C.H. classification system for long bone osteomyelitis demonstrated a substantial agreement between observers according to the Fκ value. This was supported by a high level of accuracy of classification within each of the variables. The bone involvement category had a moderate agreement amongst users. This could be due to the nature of the 2-D presentation of cases within the online form. The Fκ was not influenced by clinical experience or clinical specialty, suggesting that B.A.C.H. is applicable by all levels. Further work is required to assess and optimise the descriptions of the bone involvement and anti-microbial options variables.
