Determining the reliability of the Graf classification for hip dysplasia

We sought to establish the levels of interrater reliability and intrarater reliability of the Graf classification among orthopaedic surgeons in their final training year and who learned the method by instructed teaching or self study. Using standard teaching material developed by Graf, two groups of senior orthopaedic residents at the same training level received structured teaching sessions (Group A, n = 2) or performed self study (Group B, n = 2). Interrater reliability and intrarater reliability were determined (Cohen's weighted kappa). Proportions of correctly rated sonograms were compared between groups, implications of misclassifications were analyzed, and sensitivity analyses were performed. Interrater reliability was 0.59 (95% CI = 0.32-0.85) for Group A, and 0.47 (95% CI = 0.14-0.79) for Group B. Intrarater reliability showed an overall kappa of 0.57 (95% CI = 0.35-0.78) in Group A, and 0.47 (95% CI = 0.19-0.75) in Group B. The proportion of correctly rated sonograms between groups was similar in the original dataset and in the sensitivity analysis. Misclassifications influencing treatment were infrequent; one patient would have received unwarranted treatment and three patients would not have received warranted treatment. The Graf classification showed moderate reliability. Using self study, it can be learned almost, but not quite as effectively as by a structured program.