An approach to formal decision making in exploratory development
Mick Looby and Amy Racine
M&S Novartis
Decision making is an everyday activity in pharmaceutical research and development. Sometimes decision making is formalized according to predefined mathematical criteria such as performed in hypothesis tests for pivotal trials. The exact criteria for such trials are often drawn up in agreement with regulatory authorities or standards. However such formal decision procedures are applied less often in exploratory development. Moreover, applying the "pivotal" paradigm to the exploratory setting is often not appropriate in any case.
From a business perspective, good decision making in exploratory development is essential as it can act as a "gate-keeper" which prevents poor candidates from unnecessarily entering large expensive full development trials. This presentation will provide an example of the steps required to apply a formal decision making at Proof of Concept (PoC).
PoC is the exploratory development decision point at which a candidate is promoted to full development or not. Arriving at this decision requires the following:
- definition of a treatment difference that is relevant for the development decision
- definition of a probabilistic threshold which balances the risk of false positives and negatives against current developmental costs
- definition of the data needs to support the decision and definition of the analytical methods that will be used to turn this information into decision relevant information
The definition of a relevant treatment response, although an essential clinical task, is often mistaken for a statistical question of statistically relevant difference. It is surprising how often this mistake is made and how difficult it can be to pin down the difference that would be considered relevant for further development.
The definition of the probabilistic threshold is perhaps one of the most difficult tasks as it requires balancing current investment (i.e. expense of PoC programme) with the consequences of making a bad decision for the next stage of development. If the next stage of development is very expensive, then a high threshold has to be chosen to reduce the chance of false positives; however, if the threshold is set too high we may also kill potentially good compounds as false negatives. It should be remembered the expense of the next step is sometimes itself a design issue. Setting this threshold is a team activity, driven by the statistician, which requires much discussion and iteration to weigh up the various development scenarios and knowledge of the next steps beyond PoC.
Given an out line of these decision properties, it is now possible to focus on efficient generation of information to support the decision. Data generation, i.e. PoC program design, is intimately linked with the analytical methods that will be used to transform the data into decision relevant information. Model based approaches which take into account the underlying pharmacology are an efficient means of generating quantitative information which not only guides the strategy in terms of study designs, but also provides the information necessary to underwrite the decisions themselves. Defining and employing these methods requires both pharmacological and statistical input.
Thus, the aim of a PoC strategy becomes one of designing studies to provide data, which is in turn transformed into decision relevant information and fed into the pre-defined decision making process. The overall strategy will be defined by the clinical team who need to take into account both the developmental constraints and the ultimate deliverables of the overall strategy.
This approach to the PoC strategy will be illustrated by some generic examples.