This is lesson 4.2.3, Decision Models. Decision modelling is a means of incorporating probabilities and future costs into a medical technology assessment. Both base rule and discounting are tools that maybe useful for calculating a decision model. It is also possibly more useful to consider decision modelling in the case of an example. So consider a health care worker who has a needle-stick injury and whether or not that health care worker should be given an HIV preventive treatment. HIV is an incurable chronic illness, and there is a risk of infection from a needle-stick injury. Preventive treatment or prophylaxis treatment can be given to prevent HIV infection, but there are many potential side effects and is not 100% effective. Decision modeling is one means of determining whether or not treatment is cost effective. This slide exemplifies the decision model graphically. The question being asked, should treatment be given? Has two possible outcomes. Yes, treatment should be given or no, treatment should not be given. If treatment is given, there's a possibility of side effects but there is also a possibility of no side effects. In either case, because the treatment is not 100% effective, there's also a probability of the healthcare worker still getting HIV, or potentially not getting HIV. If no treatment is given, there is also the probability of a healthcare worker not getting HIV, but there is a possibility that they will get HIV. All of these possibilities are depicted in this model. In this case, the objective of our example technology assessment is to determine if there are sufficient benefits to treatment for HIV for the healthcare worker who has a needle-stick injury. The relevant questions we need to address before beginning. Our calculations are, what is the risk of infection after the needle-stick? What drugs are available in order to prevent HIV? How effective are those drugs? What are their side effects? And I would like to point out in this case, we're not considering what any of the costs are. But in a more robust model, we'd want to answer all these questions and try to identify the cost associated with all of them. Although we're not directly considering costs, we are considering benefits and harms. So if there's no treatment, there's no side effects, and no unnecessary treatment. However, there's still a risk of developing HIV. If treatment is given, there's a reduced risk of HIV, but there is also potential unnecessary treatment. Because it's possible that the person would not have gotten HIV, even without having received the treatment. There's also the side effects from the treatment, and after the treatment there's still a risk of developing HIV. Note, this isn't no risk, it's a reduced risk. Now that we've reviewed the types of information that's necessary to evaluate the model, we can discuss how to evaluate the model. To do so, we apply probabilities to each possibility in the decision model, and the outcome measures to each end point. At each split in the decision tree or each branch off of one of the previous events, the probabilities have to add up to 100%. Also note that the probabilities now come measures should come good quality research evidence. And if assumptions are made about any particular probability or outcome it needs to be clear why and how those assumptions were made. Finally, to calculate the outcomes or the final measures used to evaluate the technology decision model. The probabilities are multiplied by the outcome measures which results in an expected value of the outcome measure. For this example the average risk of HIV transmission after exposure is approximately 0.3%. The effectiveness of the treatment is difficult to estimate, but there are some studies that have tried to do so. One study found that it reduced the odds of HIV infection by 81% which is equivalent to about a 5% chance of HIV. Now, there was some calculation that was done to get from 81% reduced odds ratio to a 5% chance. It's important to know how and what types of information are reported in studies, so that you don't blindly apply probabilities to the decision model that aren't appropriate. Finally, there's numerous side effects that include nausea and vomiting, fatigue, headache, abdominal pain, diarrhea, and possibly others. The probability of getting a side effect has a wide range of estimates, anywhere between 50 and 75%. So in order to estimate the model we'll assume it's around 63%. Just as it we have to identify the probabilities for the model. We also had to identify values of the outcomes. We can do this also from the literature. The existing preference scores associated with HIV infection and some side effects. Or the quality of life measures include preference scores for HIV which are between 0.5 and.0.75, and preference scores for some of the side effects. The values that we can apply to the model are shown below. Note, whichever values we apply to the model we still need to demonstrate where and how they came from. So that anyone using the results of the model are able to assess the validity and reliability of the model itself. If we again consider the graphic representation of the model for each possible outcome. No treatment or yes treatment, we would assign a probability, To each event occurring. And note, Probabilities in each case have to add the one. So the probability in first stag of 1 plus 2 has to equal 100%. The same here and the same here, and the same here. The outcome measures are then applied to the end points where outcomes are assessed. I've omitted the full calculations from this presentation. However, the expected utility value of no treatment is 0.99715, and the expected value of treatment is 0.9369. Therefore, based on this model and the assumptions made a person is better off with no treatment. However, this could change if the assumptions, either the probabilities or the outcome values, were changed. It's important to remember that the assumptions made in order to calculate the model impact the final results.