# Expected Value of Information and Decision Making in Health Technology Assessment

Health decision makers often have to decide whether to adopt a new health care intervention (e.g.: pharmaceuticals, new procedures, etc.) or keep the existing practice. If one assumes that the new intervention has positive but uncertain net benefits over the existing procedure, should the new technology be adopted?

A paper by Eckermann and Willan (2007) looks at this problem and create a theoretical framework to find which actions are best suitable for which situations. They claim that decision makers face three options:

1. adopt the new intervention without further research (A);
2. adopt the new intervention and undertake a trial (AT); or
3. delay the decision and undertake a trial (DT).

The authors adopt the notation of the expected net gain (ENG) where ENGA gives the expected gain from choose AT over A and ENGD gives the expected gain from choosing DT over A. ENGA represents the difference between the value of the trial (sample) information (assuming adoption) minus the cost of the trial. ENGD is the difference between the value of the trial (sample) information (assuming delay) minus the cost of the trial.

Delaying the decision allows time for more information to be collected, but creates direct costs (the cost of the trial) and opportunity costs (the cost of non-treatment of affected individuals during the delay). Deciding to adopt the technology and preform a trial has the benefit of providing more information to the decision maker but the additional direct cost of the trial as well as reversal costs (discussed later). The expected value of the of preforming the trial (expected value of sample information-EVSID) is calculated as follows.

• EVSI=N[∫ -b*{f0(b)-f1(b)} db] = N*[E0(b|b<0) - E1(b|b<0)]

The argument inside the integral is integrated between negative infinity and zero. The distributions f0 and f1 represent the predicted distribution of benefits at the present (0) and after the trial (1). The benefit level is given by the variable b, and N is the number of people affected by the disease. The information is only valuable if researchers find that there are more ‘bad’ outcomes than previously expected. If the new treatment proves safer than expected, choice A would have been optimal. Also, the trial is more valuable when the number of people affected with the disease, N, is larger since the decision will be a more important one to society.

Eckermann and Willian also add to the model the concept of the cost of reversal. After a new treatment is adopted, a subsequent reversal has costs. These costs include reversing information flows (e.g.: public health messages, changing med school training curriculum, etc.) and sunk cost investments in specific equipment or training. Taking into account these reversal costs makes option A seem (relatively) more attractive to the DT and AT cases since a reversal is impossible if a trial is not conducted.
Using a cost benefit analysis, the following decision rules are established.

• choose A when ENGA and ENGD<0;
• choose AT when ENGA>0 and ENGD<0;
• choose DT when ENGA<0 and ENGD>0;

The authors give a more intuitive explanation as well.

• AT is preferred where expected costs of reversal per patient are small relative to the expected distribution of net benefit below 0, E(b|b<0).
• A is preferred where there is little uncertainty of the positive benefits and costs of reversal are large.
• DT is preferred when there is significant uncertainty, the opportunity costs of delay are small and data collection and analysis proceeds quickly.

The authors use the framework constructed above to analyze the prospect of adopting external cephalic version (ECV) treatment for pregnant women presenting in the breech position. ECV attempts to manipulate the fetus into a cephalic presentation and avoid a caesarian delivery.

This paper is interesting and possibly useful. The theoretical model is enlightening and warns decision makers to evaluate opportunity costs and reversal costs in addition to simply the direct cost of conducting a trial. The researchers also advise decision-makers when to choose which course of action in a comparative sense. Further, in the example using ECV, the authors actually mathematically calculate which course of action is preferred in this real-life situation. Much of the value of the framework, empirically, depends and having accurate information. Do decision-makers know how many people are affected by the disease? Are the cost of referrals known? Can we accurately estimate the value of information gained from a future trial? If the answer to these questions is ‘yes’, then this paper is very useful; if not, it is still a clever model.