Rationale for modelling protocol change

BackgroundOur protocol specified that the models of Chapters 5 and 7 would be based on parameters estimated from data from St Luke’s Hospital, Tokyo, and recalibrated to the UK using CPRD data. During early analyses of both St Luke’s data and CPRD data it became apparent that this variability was substantially lower when estimated from St Luke’s data than from CPRD data by a factor close to two. Hypotheses to explain this included that the CPRD estimate was so inflated by the nature of routine clinical data (missed appointments, varying laboratory methods, etc.) that the St Luke’s estimate was indeed an improved estimate but to a much greater extent than anticipated, or, that it is unsafe to make the modelling assumption that this short-term variability would be transportable between ethnic groups and settings. Under the former hypothesis it would be highly advisable to combine parameter estimates from St …