Improving phase II oncology trials using best observed RECIST response as an endpoint by modelling continuous tumour measurements