Abstract: Meta-regression is usually thought of in the context of a meta-analysis where heterogeneity is found. In these circumstances, meta-regression can help explain the source of the heterogeneity - perhaps due to differing patient characteristics, treatment details, or aspects of study design or execution. Here we highlight how meta-regression can be useful in different circumstances - trying to answer complex questions that are not directly addressed in the literature. We use as an example the important and interesting question "is a patient better off being in a trial?" - a question clearly difficult to address directly via a trial. Important questions are often composed, on closer inspection, of many 'contributory' sub-questions. In our example these include "do trial clinicians tend to be better clinicians?"; "if so, are trials a cause or an effect?"; "do trial treatments tend to be better?"; "do clinicians tend to give treatments better (better regimens) to patients in trials?" (protocol effect); "does the consent process benefit patients?". It may be possible to find studies that address, more or less directly, one or a combination of the contributory sub-questions. Studies may also be exposed to particular biases. The results of each study can be assumed to be due to an appropriate combination of sub-effects and biases, plus random variation. All together, the sub-studies can be modelled in a meta-regression which may provide estimates of individual 'sub-effects' (and biases), and of combinations of these as desired. Such a meta-regression would also highlight the sort of data most needed to shed more light on the main and sub-questions. In our example, we find that "clinician selection bias" (the potential bias arising from better clinicians being more likely to choose to recruit for trials) is confounded with 'protocol effect' in most of the studies we found. The question "is a patient better off being in a trial?" is now the topic of a Cochrane review - see http:// hiru.mcmaster.ca/ebm/trout/