The smart use of real-world evidence (RWE) is known to enable more flexible forms of access to novel medicines. This flexibility may be especially promising for targeted cancer medicines, which often do not align with the traditional approach to medicinal development, and for which therapeutic innovation (i.e. favourable and clinically significant benefits at an affordable price) in routine clinical practice is becoming ever more difficult to achieve due to its highly complex nature. Amongst others, RWE should include information on the performance of those medicines for every individual patient. However, the current estimation of this patientlevel performance using conventional methods from pharmaceutical and medical sciences is challenging. This is because these methods are unable to derive causal conclusions on medicinal performance from the complex real-world environment, as opposed to controlled and randomised clinical trial settings. Simultaneously, the increasing emergence of novel medicines, and their promising combined effects are now creating a new combinatorial complexity level on the captured data. At the same time, new and advanced analytical methods within the field of data science are continuously being developed and have recently been applied to pharmaceutical and medical research, including the domain of pharmacoepidemiology. These powerful methods include techniques such as machine learning and Bayesian approaches, both being recognised as having a transformative potential in clinical research and practice. Specifically, they may be used to gain new insights into the patient-level performance of novel medicines in the messy real world, thereby providing a better understanding of RWE. In doing so, studies may generate new hypotheses through the exploration of data sets, or test existing hypotheses prespecified during prior clinical research. In this dissertation, I present the specific methods of advanced analytics to unravel the complexity of RWE , therefore, increasing our understanding of the individual performance of cancer treatments. These methods are investigated for their use in both hypothesis generation (part 1) and hypothesis testing (part 2) studies. A general ntroduction into the field is provided in Chapter 1, followed by the research objectives. In Chapter 2, I validate the use of an advanced modelling technique, i.e. machine learning, as a personal performance prediction model for glioblastoma. Optimisations of this model to be used on novel medicines in more complex situations are proposed in Chapter 3. In Chapter 4, the importance of multi-product RWE assessments is explored, for which hypotheses. Lastly, for these investigated analytics to become useful in healthcare, the need for an insight-providing federated network is introduced in Chapter 6. Chapter 7, the last chapter of this dissertation, presents a general conclusion with discussion of the research contributions.

Novelties of September 2009

This paper proposes three main paradigms underlying theories on executive