Psychological Risk Assessment in Plastic Surgery via a DeepSeek Large Language Model: A Retrospective Cohort Study.

BACKGROUND: Preoperative psychological risk assessment is critical yet challenging in plastic surgery because of high rates of body dysmorphic disorder (BDD: 3.2-16.6%) and limited clinical screening tools. This study evaluated an artificial intelligence large language model (DeepSeek-LLM) for psychological risk stratification. METHODS: In this retrospective cohort study, preoperative questionnaires (14-item instrument assessing demographics, psychological status, and surgical expectations) and postoperative dispute data from 1826 patients were analyzed. Three blinded plastic surgeons scored patients' surgical risk via a 10-point visual analog scale (VAS). The concordance between physician scores and DeepSeek-R1 LLM-generated risk assessments was evaluated via Bland‒Altman analysis and Pearson correlation. RESULTS: Patients with BDD (12.0%, n = 24/200) and those experiencing postoperative disputes (2.4%, n = 40/1627) presented significantly elevated preoperative risk scores compared with controls (BDD: median 7.00 [6.0, 11.0] vs. 6.00 [3.0, 9.0], p = 0.027; disputes: 7.00 [5.3, 9.0] vs. 6.00 [3.0, 8.0], p = 0.015). "Satisfaction with previous surgical outcomes" and "surgical financial stress" emerged as the most significant risk factors for postoperative disputes. DeepSeek-generated surgical risk scores demonstrated strong concordance with physician evaluations (Pearson r = 0.883, p < 0.001), with 93.9% of the data points within the 95% limits of agreement (Bland‒Altman analysis). CONCLUSIONS: According to the results of this study, the DeepSeek can be a promising efficient tool for communication and assessment, generating profiles that detail patients' needs, psychology, and economic status. They can assist doctors in conducting more effective preoperative conversations and in assessing surgical risks to a certain extent. LEVEL OF EVIDENCE IV: This journal requires that authors assign a level of evidence to each article. For a full description of these evidence-based medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266.