Development of Validated Computer Based Pre-operative Predictive Model for Proximal Junction Failure (PJF) or Clinically Significant PJK with 86% Accuracy Based on 510 ASD Patients with 2-year Follow-up.

STUDY DESIGN: Retrospective review of large, multicenter adult spinal deformity (ASD) database OBJECTIVE.: To build a model based on baseline demographic, radiographic, and surgical factors that can predict clinically significant proximal junctional kyphosis (PJK) and proximal junctional failure (PJF).

SUMMARY OF BACKGROUND DATA: PJF and PJK are significant complications and it remains unclear what are the specific drivers behind the development of either. There exist no predictive model that could potentially aid in the clinical decision making for adult patients undergoing deformity correction.

METHODS: Inclusion criteria: age ≥18, ASD, ≥4 levels fused. Variables included in the model were: demographics, primary/revision, use of 3-column osteotomy, UIV/LIV levels and UIV implant type (screw, hooks), number of levels fused, and baseline sagittal radiographs (PT, PI-LL, TK, and SVA). PJK was defined as an increase from baseline of proximal junctional angle ≥ 20° with concomitant deterioration of at least 1 SRS-Schwab sagittal modifier grade from 6wks postop. PJF was defined as requiring revision for PJK. An ensemble of decision trees were constructed using the C5.0 algorithm with 5 different bootstrapped models, and internally validated via a 70:30 data split for training and testing. Accuracy and the area under a receiver operator characteristic curve (AUC) were calculated.

RESULTS: 510 patients were included, with 357 for model training and 153 as testing targets (PJF: 37, PJK: 102). The overall model accuracy was 86.3% with an AUC of 0.89 indicating a good model fit. The 7 strongest (importance ≥0.95) predictors were: age, LIV, pre-operative SVA, UIV implant type, UIV, pre-operative PT, and pre-operative PI-LL.

CONCLUSIONS: A successful model (86% accuracy, 0.89 AUC) was built predicting either PJF or clinically significant PJK. This model can set the groundwork for preop point of care decision making, risk stratification, and need for prophylactic strategies for patients undergoing ASD surgery.