Wearable textile antennas are basic components in body-centric communication systems. Flexible wearable patch antennas, when integrated into a body-worn garment are subjected to bending, causing variation in the resonance frequency when compared to the flat-antenna. Bending conditions vary statistically among different human subjects. Therefore, it is very important to be able to predict performance variations due to bending. We propose novel models which allow to predict the deterministic and statistical variation in resonance frequency of rectangular wearable patch antennas. They consist of an analytical model for cylindrical-rectangular patch antennas, expressing resonance frequency as a function of the bending radius, and a novel technique based on polynomial chaos, that quantifies statistically the variations of the resonance frequency under randomly varying bending conditions. The proposed models have been experimentally and numerically verified, and proven to be much faster and computationally less expensive than traditional techniques based on EM solvers and Monte Carlo simulations, making them very advantageous tools for the design and characterization of body-worn patch antennas.