The autonomous vehicle industry faces significant challenges in validating safety performance, as traditional approaches require extensive testing to demonstrate reliability for rare safety-critical events. This paper addresses this limitation by introducing a framework that enables statistically rigorous safety assessment from limited testing data. We analyze statistical patterns of near-collision events using the Brake Threat Number (BTN) metric to predict the likelihood of potential collisions. Our methodology leverages Extreme Value Theory (EVT) with a multi-criteria optimization approach for threshold determination. Testing with real field data from Volvo Cars Corporation vehicles demonstrates the framework’s ability to establish quantitative Mean Time Between Failures (MTBF) estimates with defined confidence intervals. These results provide a foundation for evidence-based deployment decisions for Autonomous Driving/Advanced Driver Assistance Systems (AD/ADAS) while reducing the validation burden compared to conventional methods, offering a practical path toward balancing technological advancement with safety requirements.