3D Patellar instability Anatomical Severity Score (3D-PASS): A Novel Machine Learning Score Using 3D Bone Position From MRI to Predict Outcomes of Patellar Instability Treatment (Using a Subset of Data in the JUPITER Cohort).

Overview

abstract

BACKGROUND: Patellar instability treatment outcomes vary. Early prediction of patient responses to treatment is needed to personalize interventions, reduce recurrent instability, and improve outcomes. Existing scores that predict outcomes rely on 2-dimensional (2D) imaging measures that do not fully capture 3-dimensional (3D) anatomical data. PURPOSE/HYPOTHESIS: We aimed to develop a 3D medical imaging-based anatomical score of patellar instability severity. It was hypothesized that our 3D score would demonstrate stronger associations with instability history and posttreatment patient-reported outcomes (PROs) than a 2D imaging score. STUDY DESIGN: Cohort study (diagnosis); Level of evidence, 2. METHODS: Pretreatment magnetic resonance images from 26 control patients and 244 patients with first-time and recurrent patellar instability were retrospectively analyzed from 2 sites associated with the prospective JUPITER (Justifying Patellar Instability Treatment by Results) study group. Using statistically derived 3D features that reflect relative bone positions and bone shape, as well as 2D imaging measures, we developed several preliminary anatomical severity scores. We tested these scores based on the ability to distinguish first-time from recurrent instability patients. Then, we assessed correlations between these scores and baseline and 1-year posttreatment PROs, using the Kujala Anterior Knee Pain Scale (Kujala) and Banff Patellar Instability Instrument 2.0 (BPII 2.0). A final score-the 3D Patellar instability Anatomical Severity Score (3D-PASS)-was created to best correlate with PROs. RESULTS: 3D-PASS, based on patellar and tibial positions relative to the femur, distinguished first-time from recurrent instability patients (P = .002) and correlated with 1-year outcomes (r _{nonop Kujala} = -0.70; r _{nonop BPII 2.0} = -0.68, r _{op Kujala} = -0.23; r _{op BPII 2.0} = -0.25). 3D relative bone positions were more informative than 2D imaging measures and 3D bone shape, neither of which correlated with outcomes. CONCLUSION: A higher 3D-PASS is associated with instability history and worse outcomes across all patients (first-time and recurrent) treated nonoperatively or operatively. While 3D relative bone positions correlated with outcomes, 3D bone shapes did not.