Abstract
Background:
Decision making amongst surgical treatment options for osteochondritis dissecans (OCD) of the knee depends on many factors, including lesion mobility/type. The ROCK Arthroscopy Classification of OCD lesions was created to help define lesion mobility. Different types of mobile lesions may be treated differently at the time of surgery.
Hypothesis:
We hypothesize that a multivariable model based on patient preoperative demographics and radiographic findings could predict arthroscopic mobile lesion type with high sensitivity/specificity.
Methods:
Demographic, preoperative physical examination, and radiographic data were collected from a multi-center national prospective cohort of patients with OCD of the knee. The inclusion criteria were patients <19 years of age and arthroscopically confirmed mobility status of the OCD lesion. 407 patients from the prospective cohort were included (mean age of 13.7 (SD±2.2 years); 62% male). Arthroscopic evaluation yielded 172 with mobile lesions including 58 trap door and 69 locked door lesions. Multivariate imputation by chained equations were used for missing demographic, physical examination, or radiographic data. A 75% partition of the data was used for model training, and 25% was held out for model validation testing.
Multivariable logistic regression analysis using stepwise model selection was used to determine factors associated with the likelihood of a specific trap door vs. locked door lesion. Quantitative model fit statistics were computed using the hold-out data, including sensitivity, specificity, and the area under the ROC curve (AUC), along with the corresponding 95% confidence interval (CI).
Results:
The most predictive model of a trap door vs locked door OCD lesion utilized imaging data including radiographic AP lesion width >15 mm (p=0.004), grade 2 or more effusion on MRI (p=0.004), and cartilage breach on proton density (PD) (p=0.02) while controlling for passive range of motion loss. In the 25% holdout validation sample, a sensitivity of 70%, specificity of 78%, and an AUC of 0.77 (95% CI, 0.82-0.95) was achieved with training dataset predictive factors.
Conclusion:
The resultant predictive model for knee OCD lesions indicates that a patient presenting with a radiographic AP lesion width >15 mm, grade 2+ effusion on MRI, and cartilage breach is 81%-96% likely to have a trap door lesion compared to a locked door lesion using the ROCK Arthroscopy Classification. The ability to predict the type of mobile lesion before surgical intervention can improve patient/family counseling, surgical planning, and outcomes.
