RISK FACTORS FOR PATELLAR INSTABILITY USING A QUANTITATIVE ANALYSIS OF TROCHLEAR DYSPLASIA

Abstract

Objectives:Multiple studies have described several anatomic and demographic risk factors of patellar instability (PI). Trochlear dysplasia (TD) has been shown to be a dominant risk factor for patellar instability but most prediction models have used the qualitative Dejour system to evaluate the influence of TD on PI. The lateral trochlear inclincation (LTI) angle is a described quantitative method to evaluate TD and a recent measurement technique has near perfect inter and intra rater reliability. Our hypothesis is that, in combination with other known radiographic and demographic risk factors of PI, that using a quantitative and numeric evaluation for TD, a highly reliable prediction model for PI can be created.Methods:98 patients in a pediatric and adolescent sports medicine practice were identified with documented PI that had magnetic resonance imaging (MRI) studies available for review. A matched cohort of 100 patients with no history of PI but with MRI’s were identified as a control group. Anatomic risk factors evaluated included the LTI, sulcus angle, lateral condyle index (LCI), lateral patellar inclination angle (LPI), proximal and distal tibial tubercle-trochlear groove distance (pTTTG and dTTTG), Caton-Deschamps ratio (CD ratio), and patellotrochlear index (PTI). Demographic data included age and sex. Receiver operator characteristic (ROC) curves were constructed for each variable to identify which variables were the best predictors of PI (ROC value >0.7). Using the ROC curves with a Youden’s J statistic and setting specificity at 0.9, cutoff values for each variable were created. Each radiographic and demographic variable was analyzed for significance and those that were found to be significant were analyzed. Area under the curve (AUC) was determined for each variable. Two predictive models were created. One was developed from the ROC curve results while the other evaluated all measured variables. The models were designed to produce the best possible fit while trying to limit the total number of predictors. These models were tested on a second cohort of 45 patients with PI and 42 control patients.Results:ROC curve data is in Table 1. Of the two models, the superior model was the model that evaluated all variables, regardless of ROC cutoff value. The model takes on the form of a general logistic regression (Eq 1, Eq 2). Model accuracy on the validation set showed 84% accuracy with 78% sensitivity and 88% specificity. These values are based on a probability of >90%. Patellar height measures had significant AUC’s but were not prime drivers of the final model. Age was not significant in the ROC analysis.Conclusion:This study establishes a highly reliable and predictive model for PI that is driven by various direct (LTI, sulcus angle, LCI) or indirect (dTT-TG, LPI) quantitative measurements of TD. Patellar height did correlate with PI but was not a prime driver of the model which suggests that patella alta is a less common risk factor for PI.