427 Background: Our team previously introduced AI Age Discrepancy, a novel parameter for quantifying frailty and postoperative risk in kidney cancer patients undergoing nephrectomy. AI Age discrepancy was found to be a significant predictor of shorter overall survival and longer length of hospital stays post nephrectomy, independent of established factors. This work presents a replication of the previous study using a larger cohort from a different health system, along with an external validation. Methods: This retrospective study included 1800 patients treated at a single large health system for internal validation and 590 patients from an external validation dataset, all of whom had contrast-enhanced CT imaging and underwent partial or radical nephrectomy for suspected renal malignancy. Only patients over the age of 18 were included. A ResNet-50 neural network was fine-tuned on the primary institution patients' CT images to predict age as a continuous variable. 5-fold cross-validation was performed to obtain predictions for all patients. The ensemble of models was then used to predict age for the external validation cohort, with the final prediction being the average of the 5 models. The AI Age Discrepancy was calculated as the difference between the predicted age and the true age at nephrectomy. Multivariate Cox proportional-hazards regression was used to evaluate AI Age Discrepancy as a predictor of Length of Hospital Stay (LOS) and Overall Survival (OS) with established factors as model covariates. Results: The model age predictions showed a significant and high correlation to the true age for both the primary institution and external validation cohorts, with Pearson correlation coefficients of 0.75 (p = 1.86e-321) and 0.71 (p = 2.81e-91), respectively. For the primary institution, the AI Age discrepancy was a significant predictor of OS with a hazard ratio (HR) of 1.021 (p = 0.013). AI Age discrepancy was not a significant predictor of LOS for this cohort. For the external validation cohort, AI Age Discrepancy was a significant predictor of both OS and LOS, with HR of 1.073 (p = 0.023) and 0.985 (p = 0.015), respectively. Conclusions: This study confirms that the AI Age Discrepancy parameter is a significant predictor of overall survival for kidney tumor patients undergoing nephrectomy. The model demonstrated significant predictive power on both the primary institution and the external validation cohorts, demonstrating its generalizability. However, the association with length of hospital stay was not significant in the primary institution, suggesting that institutional factors may influence this relationship. These findings support the potential integration of AI Age Discrepancy into preoperative risk stratification to improve individualized patient management.
Seshadri et al. (Sun,) studied this question.