Composite outcome measurement in clinical research: the triumph of illusion over reality?

Composite measures that combine different types of indicators are widely used in medical research; to evaluate health systems, as outcomes in clinical trials and patient-reported outcome measurement. The potential advantages of such indices are clear. They are used to summarise complex data and to overcome the problem of evaluating new interventions when the most important outcome is rare or likely to occur far in the future. However, many scientists question the value of composite measures, primarily due to inadequate development methodology, lack of transparency or the likelihood of producing misleading results. It is argued that the real problems with composite measurement are related to their failure to take account of measurement theory and the absence of coherent theoretical models that justify the addition of the individual indicators that are combined into the composite index. All outcome measures must be unidimensional if they are to provide meaningful data. They should also have dimensional homogeneity. Ideally, a specification equation should be developed that can predict accurately how organisations or individuals will score on an index, based on their scores on the individual indicators that make up the measure. The article concludes that composite measures should not be used as they fail to apply measurement theory and, consequently, produce invalid and misleading scores.