Aim: The use of artificial intelligence is set to be a paradigm shift in the way vascular surgery is practiced. In this study, we aimed to investigate the use of generic large language models for the planning of endovascular aneurysm repair (EVAR). Method: 21 patients who had been selected by a surgeon for a Cook standard infrarenal EVAR at Royal Hobart Hospital were identified between September 2021 to September 2025. A single surgeon’s measurements of main body, contralateral and ipsilateral limb diameters and lengths were collected through Cook planning worksheets. A ChatGPT-5 large language model was then used to predict EVAR component selection. The model was trained to oversize by ~10-20%, leaving enough space to cannulate the contralateral gate, and maximising overlap between main body and limbs. Measurements were inputted using a standardised dialogue and a plan was generated. The Cook worksheet plan and ChatGPT plan were compared to intra-operative graft sizes used, and feasibility of the ChatGPT plan was assessed by two investigators. Results were analysed using correlation matrices and Spearman coefficients were calculated. Results: All ChatGPT plans were feasible for overall anatomy although there were variations in main body and limb lengths. The ChatGPT plans demonstrated statistically significant correlation to the intra-operative graft sizes used. Main body diameter/length, contralateral limb diameter, and ipsilateral limb diameter/length showed high degrees of correlation (p<0.001). Contralateral limb length also correlated (p = 0.013). Conclusion: It is feasible to use a trained generic large language model for EVAR planning.

Background: There is a major deficiency of a national comprehensive thoracic oncology surgery data especially in regional and remote areas such as Tasmania1. Unique to this region is its large proportion of elderly persons. The PUNT score utilises high risk features of pleural involvement, unassignable histology, neutrophil-to-lymphocyte ratio (NLR) >3.5 and tumour size to help identify patients with node-negative NSCLC at significantly discordant risk of recurrence to their TNM staging. Lymphovascular invasion, a risk factor for recurrence, was not incorporated in this score. Methods: This retrospective analysis examined all thoracic oncology cases undergoing surgical resection at Royal Hobart Hospital (2010-2022). A subgroup analysis was done on elderly node-negative lung cancer patients. We developed the PLUNT scoring system, incorporating lymphovascular invasion into the established PUNT framework, and compared prognostic performance using Cox proportional hazards regression and Kaplan-Meier survival analysis. Results: Eighty-six elderly patients with node-negative NSCLC were analysed using the PUNT and PLUNT scoring (mean age 75.2 years, 51.2% male). PLUNT > 2.0 identified 12 high-risk patients (14.0%) with 33.3% recurrence rate, while PUNT > 1.5 identified 10 patients (11.6%) with 30.0% recurrence rate. In multivariate analysis, PLUNT > 2.0 achieved statistical significance as an independent predictor of recurrence-free survival (HR = 3.700, 95% CI: 1.033-13.252, p = 0.0456), while PUNT > 1.5 failed to reach significance (p = 0.1609). Conclusions: PLUNT scoring demonstrates superior prognostic performance over PUNT thus providing enhanced risk stratification crucial for clinical decision-making. The incorporation of lymphovascular invasion significantly improves prognostic accuracy, supporting personalized treatment approaches such as the potential role in adjuvant therapy for high-risk patients.

Purpose: Lung cancer is the leading cause of cancer-related mortality worldwide and in Australasia, prompting the introduction of lung cancer screening programs in Australia and New Zealand. As screening identifies increasing numbers of pulmonary nodules, surgical pathways must ensure timely and effective intervention. This study evaluated the impact of pre-operative biopsy on time to surgery and rates of benign lung resection in patients with high clinical suspicion of malignancy. Methodology: A retrospective review was conducted of all patients undergoing lung resection between January 2023 and December 2024 at a tertiary referral centre responsible for approximately one third of New Zealand’s thoracic surgical workload. Cases were identified from hospital databases, with clinical data obtained from electronic medical records. Patients were grouped according to pre-operative biopsy versus direct resection based on high clinical suspicion confirmed at multidisciplinary meeting (MDM). Statistical analysis was performed using IBM SPSS version 31. Results: A total of 440 patients were included. Mean age was 66 years (SD 12), with 177 patients (40%) female. Mean Eastern Cooperative Oncology Group (ECOG) performance status was 0 (SD 1). Lobectomy was performed in 307 patients (70%). Pre-operative biopsy was undertaken in 130 patients (30%), while 295 patients (70%) proceeded directly to resection. Median time to surgery was longer in the biopsy group (42 days [IQR 30–55]) compared with the non-biopsy group (32 days [IQR 21–55]). Final histopathology demonstrated primary lung cancer in 378 patients (86%), metastatic disease in 22 patients (5%), and benign pathology in 59 patients (13%). Conclusion: Most patients undergoing lung resection based on high clinical suspicion had malignant pathology despite low biopsy utilisation. The benign resection rate of 13% aligns with international benchmarks. Pre-operative biopsy was associated with delayed time to surgery, supporting the importance of patient selection for pre-operative biopsy.