Abstract The field of oncology, as compared to other medical therapeutic areas, is at a disadvantage in terms of having new products (or ideas) to reach a state of concept proven. Reasons for these include:Lack of predictivity of preclinical modelsHeterogeneity of the disease(s)Difficulty of access to repeated sampling of material (as compared to repeated testing on blood samples for example)- A failure to produce convincing correlative data from pre-clinical to clinical workThe relative coarseness of the clinical outcomes (endpoints: typically a categorization into responders/non-responders at early stage) and the error margins on those outcomes This talk intends to expand the last two points. In the transition from preclinical work to Phase I/II, there is a point (or gap?) where a less rigid exploratory phase is followed by the machinery of clinical trials in humans, for example trying to prove that a drug affects a target. In oncology, Phase I is typically dedicated to dose finding and pharmacokinetic work, Phase II to establishing threshold efficacy, and Phase III to definitive comparative proof. Still, although dose finding studies focus on safety (as they should), we should try to use them for a read-out of potential efficacy (in this case tumor measurements). The existing gap could be bridged (depending on the case) by pairing tumor outcomes with marker data that pertain to the envisioned mode of action (pharmacodynamic markers). Typical “shortcomings” of Phase I studies when considering them for this purpose are:Few patients, who are late stage, resulting in restricted exposureA mix of tumor typesA mix of doses (however this may also be a benefit, allowing for experimental dose response investigations) There are however some points that could work in favor. Patients proposed for Phase I typically have a serious tumor burden, so that % changes in tumor burden can be interpreted as being on a common scale (as compared to % changes of smaller tumor burden, which may be argued to be oversensitive to clinically irrelevant changes). The doses or regimens that are proposed to be measured with the intent of showing an acceptable toxicity profile are the ones that will be used for Phase II and likely also Phase III. In other words, we can look for the therapeutic window as early as Phase I. Some data will be shown on the possibility - at least at an experimental level — to explore percent changes in tumor measurements as a quantitative variable, which should better enable correlative attempts as described above. Another point to consider is the potential comparative nature of early data. What do we mean by this? In late Phase trials, randomization is a necessity for valid comparison by minimizing allocation bias. However, in early Phase trials, one can obtain paired data (i.e. two different quantities within the same patient, or longitudinal observations within the same patient), and such data are “perfectly randomized”, because each individual serves as their own comparator. Other therapeutic areas typically use cross-over studies at the early stage, even with (different sequences of) multiple dose intensities within the same patient (including placebo). In oncology this is hard to do, but we can take (or rather make) benefit from paired data. Some examples:If a reading can be obtained of the (rate of) tumor growth prior to treatment, this can be compared to the rate of tumor burden evolution during and post treatment.If we can pair outcomes of a marker (pharmacodynamic effect, as assessed by an assay which may be in parallel development) with tumor burden evolution, this constitutes paired data. We will argue that it is essential to conserve the paired nature of such data in the analysis. It is not enough to show an effect on a marker, and — next to that — an effect on the tumor. The marker and the tumor should be shown to be linked. If they cannot be linked, that may mean a number of things (e.g. the assay is inconsistent, the drug does affect the marker, but not the tumor volume, etc.). We therefore encourage basic scientists, clinicians and statisticians to go far enough in the exploration of early data. Such exploration must be interdisciplinary. The complex thinking behind the science (which can be very case dependent) needs to be sufficiently clarified to enable the statistician to deliver the best contribution, and conversely the statistician needs to seek such understanding. Citation Information: Mol Cancer Ther 2009;8(12 Suppl):CN04-03.