When the user has decided on the diagnosis, the next step is to determine appropriate therapies. The program looks for candidate therapies by searching along the causal chains in the diagnosis that lead to undesirable outcomes. Therapies are included in the model as having corrective effects on the nodes (as well as possible detrimental effects). The therapies with the potential to break some of the causal paths are collected as a list of potential therapies. This approach allows the program to find therapies that are appropriate even though the findings that usually trigger their consideration are absent. For example, hydralazine decreases the systemic vascular resistance, so it is commonly used to decrease blood pressure. However, in a patient with primary cardiac muscle dysfunction, the systemic vascular resistance can be too high even though the blood pressure is normal when the cardiac output is low. In such a patient, the use of hydralazine will be suggested because of the high systemic vascular resistance and is quite appropriate. Since the therapies typically have multiple effects or the primary effect has multiple potential consequences, it is necessary to determine what the effects of the therapy actually will be. That is done by the therapy prediction algorithm. There are also other considerations when selecting therapies for the patient, such as side-effects outside the domain of this model, requirements for monitoring drug effects, and known patient sensitivities, that are not covered by the program.