Abstract
OBJECTIVE: To identify the clinical manifestations, metabolic factors, and comorbidities independently associated with interpersonal dysfunction in Chinese acromegaly patients. METHODS: We analyzed clinical, cognitive, and comorbidity data from 585 acromegaly patients across 112 tertiary hospitals in China (July 1995-December 2018). Interpersonal difficulties were quantified using the Inventory of Interpersonal Problems-Distress (IIP-D) and dichotomized into low (< 17) and high (≥ 17) groups. Group differences were tested with nonparametric tests. Supervised machine-learning models were developed to predict features associated with the high IIP-D group, with performance evaluated via five-fold cross-validation. The top-performing model was further validated on the held-out data set and the feature importance analysis identified the key predictors. Exploratory hierarchical clustering (Ward's method) was used to explore symptom groupings, though sampling adequacy was limited. RESULTS: Patients with high interpersonal distress exhibited significantly higher preoperative growth hormone (GH), frontal bossing, palpitations, and cognitive impairment (all p < 0.05). Among machine-learning models, extreme gradient boosting (XGBoost) demonstrated the highest performance, area under the curve (AUC = 0.868 in across-validation), and maintained strong accuracy in final testing (AUC = 0.873). Key independent predictors included frontal bossing, palpitations, cardiomyopathy, disease duration, preoperative GH, acral enlargement, arrhythmia, and atrial fibrillation. CONCLUSION: Physical disfigurement, palpitations, cardiac comorbidities, and elevated GH levels independently predict high IIP-D in acromegaly. Integrating systematic psychosocial screening into neuroendocrine care, alongside referral to psychoneuroendocrine teams, may help mitigate social disability and improve quality of life.