Routinely collected healthcare data is becoming more commonly used for healthcare research. The increasing availability of such data promises advantages in the shape of largescale, representative data, but also brings many challenges which require statistical innovation. I will highlight some of these promises and challenges using four examples illustrating the use of routinely collected data, including modelling lung function trajectories of cystic fibrosis patients, dynamic prediction of cardiovascular disease, multi-state modelling of multimorbidity and predicting outcomes for intensive care patients.