PhD in Medical Statistics - Developing innovative approaches for adaptive design and analysis of clinical trials in type 1 diabetes based on “platform trials”.

The Innovative Medicines Initiative funded a proposal (INNODIA) with many European partners. At the heart of this project is an observational cohort of newly diagnosed type I diabetics. The platform trial is well suited to the design of multiple sub-trials to increase efficiency in protocol development, recruitment and laboratory measurements. Currently there are four different trials being planned within this platform. The plans are this platform will continue for as long as possible so we keep accruing information from the numerous trials and the observational cohort. One of the challenges in type I diabetes is the current “best” measure of disease is the stimulated C-peptide response and this is often summarised over 2 or 4 hours. The test requires a clinical visit and places a burden on the participants, and leads to higher drop out rates during a trial/follow-up. Additionally, there are many other outcomes that are being measured more frequently at home such as dried blood spot C-peptide and continuous glucose monitors. From the clinical visits, blood samples are used to measure many immune markers and other biomarkers that could be used as outcomes or predictors of treatment response. This project will use these data to investigate the quality of endpoints, and how to use all this information to improve the design and analysis of future sub-trials.

Often when trialists are faced with multiple outcomes create a combined composite score and simplify analysis by using a single outcome. Combination of outcomes could use equal weights or other weighting schemes to improve the power of a trial/analysis. This approach may lead to inefficient trials, diluting treatment effects and with results that are hard to interpret. This project will investigate methods that can handle multiple outcomes with the appropriate multiplicity correction whilst still retaining efficiency, such as closed testing procedures or methods of k-Family-wise error rate. As the data are measured over time there are issues of missing data, either sporadic missingness or loss to follow-up. Currently, we design trials assuming a certain amount of dropout and thereby increasing the sample size to compensate. However, this project will investigate different modelling strategies and try to use historical missingness patterns to determine better sample size calculations

Work will include obtaining data from the INNODIA observational cohort and model the outcome data using longitudinal data techniques. This will give an idea of how the data are distributed and an idea of the estimated correlations, variances and potential effect sizes. The analysis will then lead on to how the trials are currently designed but how to design a new study incorporating multiple outcomes. This will include the use of the k-familywise error rate, composite endpoints, various multiplicity corrections such as sequential closed testing procedures.

The main methodological focus should be on the joint modelling MMTT and DBS C-peptide values and how to design a trial that encorporates longitudinal outcome data. In particular given historical missing data mechanisms how to calculate sample sizes that are more realistic. Then the project will investigate how to do interim analyses in a two-arm parallel groups trial to either stop the trial or continue. This will require the adaptation of methods in Jennison and Turnbull’s book on group sequential trials and also the development of methodology to work out the full distribution of the test statistics and how to calculate the most efficient boundary conditions to either minimise expected sample sizes or a loss function that encorporates expected sample sizes under the null, alternative and the maximum sample size. This work will either be extended by including more outcome variables and or treatment arms.