A multilevel extension of a Heckman-type model for longitudinal quality of life data subject to drop out in a multi-centre clinical trial