Proximal junctional kyphosis (PJK) is a prevalent and detrimental complication associated with corrective surgery for adult spinal deformity (ASD). While existing predictive models have been able to predict PJK, they have lacked individualized risk factor assessment, and the inclusion of non-modifiable predictive variables has limited their clinical usefulness. Thus, this study aimed to identify novel risk factors and establish a personalized risk calculator for PJK.We conducted a retrospective analysis of consecutive patients who underwent realignment surgery for ASD between January 2016 and October 2022. The primary outcome was the occurrence of PJK within a minimum two-year follow-up. The machine-learning (ML) algorithm LASSO was used to identify key predictors of PJK. Eight ML models were created to predict individualized PJK risk based on these identified predictors. The Shapley additive explanations (SHAP) method was used for model interpretation and to feature importance ranking. The optimal model and its patient-level SHAP values were deployed as a personalized risk calculator to facilitate individualized risk factor assessment and clinical decision-making.The analysis included 213 patients, of whom 78 (36.6%) developed PJK during follow-up. Seven features were selected as key predictors of PJK, all of which were clinically modifiable. Pelvic compensation and preoperative lower limb function scores were identified for the first time as predictors of PJK. The random forest model had the best predictive performance, with an area under the receiver operating characteristic curve of 0.847. The personalized risk calculator could identify individual risk factors for PJK and guide perioperative management. This calculator was made available online for clinical use.We developed a calculator to assess the risk of PJK after ASD surgery based on novel and modifiable clinical features. This tool facilitated the individualized risk factor assessment for PJK and served as a guide for the perioperative management of the identified modifiable risk factors. Due to the limitations of the study sample size and single-centre investigation, prospective multicentre validation is warranted to assess the generalizability of this risk calculator.© 2025 Wang et al.