Potential of joint modelling of longitudinal observations and time-to-event data to improve prognosis in chronic heart failure studies