investigator_user investigator user funding collaborators pending menu bell message arrow_up arrow_down filter layers globe marker add arrow close download edit facebook info linkedin minus plus save share search sort twitter remove user-plus user-minus
  • Project leads
  • Collaborators

Missing and imperfect data in clinical trials and observational studies

Ian White

0 Collaborator(s)

Funding source

Medical Research Council (MRC)
This programme focuses on practical difficulties commonly encountered in randomised trials and observational studies. Much of the work is motivated by our collaborative research but seeks wide generalisability. We aim to provide feasible yet theoretically well-grounded solutions in three main areas: Missing data in RCTs and observational studies: We will use rich data structures and plausible assumptions to explore whether data are informatively missing. This work will extend existing models to the complexities of the QUATRO trial, the ALSPAC study, the CFAS study and a programme of smoking cessation trials. By estimating the missing data mechanisms in these and other studies, we will begin to establish an evidence base for the selection of suitable methods for dealing with missing data. Alongside this work we will identify and disseminate suitable methods in single studies and meta-analyses. Departures from randomised treatment in RCTs: We will extend existing methodology to handle the full complexity of departures typically found in RCTs. To handle switches to non-trial treatments, or cessation of all treatment in comparative trials, we will use information external to the trial about non-trial treatment contrasts. To handle time-dependent treatment changes with survival outcomes, we will develop a new potential outcomes model. We will also explore the implications of departures for health economic evaluations. Measurement error in observational studies: We will explore issues arising in the Fibrinogen Studies Collaboration, the Emerging Risk Factors Collaboration and the Centre for Nutrition and Cancer. Specifically, we will develop methods for practical adjustment for measurement error in exposure and confounders using replicate measurements of biochemical risk factors, taking account of possible non-linear associations, and methods for allowing for bias in dietary reports by using replicate measurements of different types. We will extend these methods to the meta-analysis situation, allowing for patterns of replicate measurements which differ between studies. These three areas are linked both practically, because studies rarely suffer from a single problem in isolation; theoretically, by common methods such as instrumental variables; and by a common formulation as missing data problems. They also all raise important design issues. In addition to the above, we will also maintain collaborations on randomised trials, including the evaluation of an internet intervention to reduce heavy drinking, and on observational studies, including development of new methods for the antenatal diagnosis of Downs syndrome, and we will pursue new methodological issues that arise from these collaborations.

Related projects