In this review, we assess the real-world effectiveness of Advanced Driver Assistance Systems (ADAS) in preventing vehicle crashes. We propose a new, data-driven framework of safety performance based on dimensions urgency and level of control as an alternative to existing taxonomies. We identified 28 ADAS and collected data on (real-world) safety performance of from grey (technical reports) and white (scientific) literature. ADAS were categorized by functional class (longitudinal/lateral control, monitoring, information systems) and by interaction type (informing, warning, intervening, comfort-enhancing). The data analysis showed that Lane Keeping Assist (LKA) (-19.1%) and Driver Monitoring Systems (DMS) (-14%) had the strongest crash rate reduction effects, followed by Automatic Emergency Braking (AEB) (-10.7%). However, systems like Adaptive Cruise Control (ACC) and Cruise Control (CC) were associated with increased crash rates (+8%, +12%). Categorizing systems by either functional class or interaction type revealed central tendencies favoring safety of longitudinal control and intervening systems, while comfort-enhancing systems showed detrimental effects. From the categorizations, we derived dimensions urgency and level of control, scoring individual ADAS accordingly. A linear model based on these dimensions (pseudo-R2 = 0.103) explained a similar amount of variance as the categorizations (functional class: 0.140, interaction type: 0.087). The analysis indicated that low urgency and high level of control, typical of comfort-enhancing systems, did not improve safety. Our findings support the positive safety effects of ADAS, but also point to risks, particularly for comfort- enhancing technologies. The proposed framework offers an explanation for the observations. It is simple and generalizable, and avoids disadvantages inherent to categorical classifications, making it a potentially valuable tool for designers and policymakers.