Comparative effectiveness and cost-effectiveness of an artificial intelligence workflow for small renal mass diagnosis on computed tomography

INTRODUCTION: Small renal masses are increasingly detected on imaging, but their accurate classification as benign, indolent, or aggressive remains challenging. Such classification can aid further diagnostic work up, improving patient outcomes and saving costs. AI models show promise in this area. Here we evaluate cost effectiveness of 3 such models compared to standard of care (SOC). METHODS: We developed a Markov microsimulation model to simulate clinical and economic outcomes in 10,000 patients aged 40-75 with small renal masses over a 10-year horizon. We compared four strategies: SOC, MRI + AI, and two CT + AI models (with or without image embeddings). The model reported total costs, quality-adjusted life years gained (QALYGs), and incremental cost-effectiveness ratios (ICERs), with 100 simulations per scenario. RESULTS: CT + AI Model 1 (without embeddings) had the lowest cost ($9,079.65) and highest QALYGs (8.8207), outperforming SOC ($10,601.23, 8.8000), MRI + AI ($10,178.30, 8.8108), and CT + AI Model 2 ($9,177.93, 8.8205). CT + AI Model 2 (with embeddings) was dominant in 28% of simulations. The costs of CT + AI Model imaging could increase from $265 to $2,358.58 while remaining cost-effective (ICER <= WTP) compared to the MRI + AI Model. CONCLUSIONS: CT-based AI models are cost-effective alternatives to SOC and MRI-based AI, offering better outcomes at lower cost. Their value stems from accurate risk stratification, enabling timely intervention and reducing overtreatment. These findings support the clinical and economic utility of AI in renal mass evaluation.