Development of risk prediction model for new-onset atrial fibrillation in elderly patients undergoing video-assisted thoracoscopic surgery: a retrospective cohort study.

Background: Postoperative atrial fibrillation (POAF) is among the most frequent complications after video-assisted thoracoscopic surgery (VATS). Although there are numerous risk prediction models and advanced age is a notable factor, only a handful of these models specifically address older, high-risk groups. This study therefore aimed to construct a POAF risk model for patients aged 65 years and older undergoing VATS to support precise perioperative monitoring and preventive measures. Methods: We conducted a retrospective analysis of 4,449 eligible patients aged ≥65 years who underwent VATS at Peking Union Medical College Hospital from January 2013 to March 2024. The data were split into a training cohort (n=2,918) and a testing cohort (n=1,531) using a cutoff date of January 1, 2022. POAF was defined as follows: (I) any recorded atrial fibrillation episode lasting at least 6 minutes; (II) no history of atrial fibrillation (AF) prior to surgery; and (III) onset within 30 days postoperatively. Variables with a univariable P<0.20 and those deemed clinically important from prior literature were included in the multivariable logistic regression to identify predictors of POAF. Internal validation was performed within the training cohort. Discrimination was measured by the area under the receiver operating characteristic curve (AUROC), while calibration was assessed with the Hosmer-Lemeshow test and calibration plots. Overall predictive accuracy was summarized by the Brier score. Results: POAF occurred in 4.2% of patients (187/4,449). Five predictors were ultimately incorporated into the final model and depicted as a nomogram: male sex, elevated N-terminal pro-brain natriuretic peptide (NT-proBNP), enlarged left atrial anteroposterior diameter, lobectomy, and left-sided tumor location. In the training cohort, the model achieved an AUROC of 0.718 [95% confidence interval (CI): 0.675-0.764], with a sensitivity of 80.5% and a specificity of 48.5%. In the testing cohort, the AUROC was 0.753 (95% CI: 0.671-0.835), with a sensitivity of 86.1% and a specificity of 54.4%. Calibration was satisfactory, as indicated by Hosmer-Lemeshow P values above 0.05 in both groups. The Brier scores were 0.025 (training) and 0.021 (testing). Conclusions: We identified five readily obtainable clinical predictors and integrated them into a nomogram that demonstrated good discrimination and calibration for predicting POAF in patients aged65 years and older who underwent VATS. This practical tool supports perioperative risk stratification and could inform targeted prevention and monitoring strategies in high-risk surgical populations.