Abstract
We thank Dr Lines (1) for his attention to our commentary on outcome measures for trials of headache treatments. We noted that single outcome measures such as pain ratings may not accurately portray a patient’s overall experience with treatment (2). Dr Lines suggests that composite outcome measures are a promising alternative, not only because they incorporate multiple treatment attributes but also because they are statistically more efficient. Dr Lines gives two examples of composite outcome measures that might be used in evaluating headache treatments: ‘total migraine freedom’ (pain freedom and absence of photophobia, phonophobia and nausea) and ‘sustained pain freedom over 24 hours with no adverse events’.
While such outcomes may have statistical advantages, we do not fully share Dr Lines’ enthusiasm for these or other composite outcomes as a way to capture ‘key elements of treatment effects relevant to patients’. As other authors have noted, such outcomes are difficult to interpret; for one thing, some may assume that the results for the composite outcome apply to its individual components (3). In the case of ‘total migraine freedom’, measuring the ‘absence’ of photophobia, phonophobia and nausea only makes sense in patients who had all of these symptoms to begin with. We doubt this is true of most patients in clinical trials or in clinical practice, especially if headaches are treated at an early or mild stage.
In the case of ‘sustained pain freedom over 24 hours with no adverse events’, the implicit assumption is that the two events—pain freedom and absence of adverse events—are equally valued by patients. This is unlikely to be the case, which complicates interpretation (4). Rather, the importance of avoiding adverse events will vary depending upon their severity and the values and circumstances of individual patients. Furthermore, the two parts of this measure do not usually move in tandem. Powerful medications that eliminate pain in a large proportion of patients will, in most cases, also produce more adverse events; less effective medications will render fewer patients free of pain but may cause few adverse events. Judged using a composite of these outcomes, a modestly effective but well-tolerated drug might seem as good as or better than a highly effective drug that commonly produces mild adverse events. (Indeed, this seems to be the case in the post hoc analysis cited by Dr Lines.)
In short, there is good reason to be wary of composite outcomes. No single outcome measure, considered in isolation, can fully capture important elements of the treatment experience, but combining important events within a single outcome measure is not always an especially good solution to that problem. When composite outcomes are used in clinical trials of headache treatments, the individual outcomes should still be reported as separate measures. This will facilitate meta-analysis and avoid the possibility, described by other authors, that composite outcomes will be used to mislead or to ‘game’ trials (5).