Methods: The net benefit regression framework for cost-effectiveness analysis is used to discuss various strategies for analyzing observational data. The strategies include propensity scores, instrumental variables and an alternative method involving pseudo-randomization (non-binding randomization). The third method is illustrated using data from a hypothetical clinical trial of treatment for pneumonoultramicroscopicsilicovolcanokoniosis.
Results: When patients are able to decide whether to comply with new treatment or continue to receive usual care, it is unclear why an estimate is needed for those who will not actually use the new treatment. In this case, the policy relevant group is those whom we labeled compliers. The results for pseudo-randomization, which works like an instrumental variable, show an estimate of cost-effectiveness for compliers. However, this technique requires person-level data and the ability to pseudo-randomize.
Conclusions: We suggested an alternative estimation strategy to be used in cases where propensity scores could be misleading. Researchers may wish to apply other methods if propensity scores are felt not to work or if the researchers would like to avoid controversy. Because one can conduct the economic evaluation exercise for observational data in a regression framework, it is possible to bring regression methods to bear using the net benefit regression framework.