Using “Big” NONMEM Dataset Generated from Standard SDTM to Review Data and Accelerate Model Building Process
Shafi Chowdhury (1), Rajwanur Rahman (2), Kantish Chowdhury (2)
1) Shafi Consultancy Limited, UK, 2) Shafi Consultancy Bangladesh
Objectives: The use of standard SDTM structured data has given us the opportunity to generate one “BIG” NONMEM dataset by merging required SDTM datasets together. This can merge all related data together by using the key identifying variables within those datasets which are always the same. The advantage of doing this is that we are then in a position to easily analyse how all the data relate to each other and see if there are any issues or unexpected trends in the data before finalizing the NONMEM dataset specification.
Methods: Once the "big" NONMEM dataset is created using the standard SDTM datasets, the data is ready for statistical analysis. Statistical analysis can be performed automatically to check associations of all types of data based on age, sex, race, country and all other covariates specified within a project. Associations can be highlighted and potential data quality issues or trends in the data can be explored on an ongoing basis prior to database lock. Graphical tools can be used to see how values are changing over time, and if there are unexpected results then actions can be taken to raise the quality of the data. This will not only lead to cleaner data, but also to an accelerated model determination process.
Results: The statistical methods display association between the different combinations of covariates both in tabular and graphical form. Those with significant results are then summarised and grouped so that they are easy to see by the pharmacometrician. All data outside the reference ranges are also highlighted.
Conclusions: Looking at all associations and summary data of individual covariates makes it easy to see if there are any data issues, and which covariates are more logical for inclusion in the NONMEM dataset specification. This allows a more focussed specification to be created, and as the "BIG" NONMEM dataset was already generated, the reduced dataset is produced with minimum additional time. As associations between covariates are also known, this can therefore aid the modelling process, helping to accelerate the process.
This helps to bring some of the exploration of the data prior to database lock, thus ensuring the Pharacometrician is more familiar with the data by the time they start the modelling process. This can speed up the delivery of the analysis and mean the results will be available in time to influence future decisions.