This work investigates the incorporation of a risk metric based on time to collision for Adaptive Cruise Control applications on multi-lane motorways. To improve the fuel-efficiency and safety, a stochastic model predictive control approach is suggested that limits the violation probability of the imposed risk metric. For this reason, a Bayesian network is used to predict the probability distributions of the surrounding vehicles' future motion based on actual measurements. Subsequently, these distribution functions are incorporated into a stochastic model predictive control algorithm. The novel strategy is evaluated in simulation using an artificial evaluation cycle and shows significant improvements in terms of fuel efficiency and safety in comparison to conventional Adaptive Cruise Control when sharp cut-off maneuvers occur.