We present a new framework for supporting decisions in sequential clinical risk assessment examinations. In this framework, the decision whether to perform a test depends on its expected contribution to risk assessment, given results of previous tests, and the contribution is quantified using information theory. In many cases adding an additional examination clearly improves the predictive model. However, there are cases in which the improvement is not constant for all values of previous tests, and quantification of possible improvement can support decision on further examinations. Using this approach can prevent many expensive, unpleasant or risky examinations. We demonstrate the use of this method on an example of type 2 diabetes onset study. The results show that reducing a considerable percent of the blood tests does not decrease the model's prediction power.