Prediction of socioeconomic disadvantage using AI analysis of CT imaging in kidney cancer patients.

Abstract

428 Background: Socioeconomic status (SES) has been linked to higher rates of mortality after major urologic surgery; however, SES is difficult to assess at the individual level with current measures assessed at the neighborhood level. Artificial intelligence (AI) provides the ability to appreciate additional clinical information from vast amounts of data to make clinical inferences and predictions. For example, AI powered image analysis has shown that predicted “biologic age” is a stronger predictor of survival and chemotherapy tolerance than chronological age in kidney cancer patients. This finding suggests that AI-derived analysis can indicate frailty or advanced biologic age that is not reflected in actual age. In this study, our objective was to apply AI image analysis to predict Area Deprivation Index (ADI) score. Methods: Retrospectively, patients who underwent nephrectomy and had preoperative contrast-enhanced abdominal CT imaging available from 10/2009 to 7/2024. ADI, a validated metric for SES, was obtained from the Neighborhood Atlas. A ResNet-50 neural network was fine-tuned to predict ADI from CT image inputs. 5-fold cross validation was used to obtain predictions for all patients. Predicted ADI was compared to actual ADI with linear regression. Results: A total of 1349 nephrectomy patients had available imaging and ADI. Predicted ADI was correlated with actual ADI (r = 0.18, p = 1.79x10-11, y = 0.0552 * x + 0.4661). Predicted ADI was not associated with oncologic or survival outcomes following nephrectomy. Conclusions: In a preliminary analysis of a computer-vision based AI model, we demonstrated the ability to predict Area Deprivation Index from CT images in kidney cancer patients, suggesting socioeconomic status may be embedded within medical imaging data. Further refinement of the model is needed. A robust model for SES prediction may provide a better and more personalized estimation of SES exposures and may offer critical insights into how these exposures impact clinical outcomes in patients afflicted by oncologic diseases.

Ähnliche Arbeiten