Data Mining & Data Exploration

We have applied data mining and data exploration techniques to clinical studies and to post-authorisation observational studies. Currently, we do also perform some methodological research in the area of robust linear modelling.

Often, this helps to better understand and interpret the data, as compared to the planned analyses.

Examples:

A. As part of a Regulatory Support activity, we have re-analysed a previously reported pivotal study of a customer. We have identified issues in the original analysis, and suboptimal use of statistical methods in this study which had only partly achieved its objectives. Our re-analysis according to best current statistical practice has achieved statist-ical significance for most of the study objectives, which could thus help in discussions with regulators. Moreover, our results could be used to optimise the design of further pivotal studies as part of portfolio management activities.

B. In a large proof-of-concept study which used new (thus so far not very well understood) surrogate diagnostic methods to assess key objectives, our explorative analyses did help under-standing the probable nature of the observed effects. These results could be used to improve the original report, and to generate hypotheses enabling decisions and design for the Phase III program.

C. In a post-authorisation observational study, our exploratory analyses did confirm prognostic factors and their effects on events, as known from very large controlled clinical studies, in a practical non-controlled setting.

Recommendations:

Data mining should be used routinely for proof-of-concept and pivotal studies, in order to improve the statistical interpretation and conclusions, and thus to increase the chances of success.

Data mining can be combined cost-effectively with Verification of Statistical Results.