Non-homogeneous semi-Markov reward process for the management of health insurance models