Some study outcomes may only be applicable to a proportion of participants. For example, in subfertility trials the proportion of clinical pregnancies that miscarry following treatment is often reported. By definition this outcome excludes participants who do not achieve an interim state (clinical pregnancy), so the comparison is not of all participants randomized. As a general rule it is better to re-define such outcomes so that the analysis includes all randomized participants. In this example, the outcome could be whether the woman has a ‘successful pregnancy’ (becoming pregnant and reaching, say, 24 weeks or term). Another example is provided by a morbidity outcome measured in the medium or long term (e.g. development of chronic lung disease), when there is a distinct possibility of a death preventing assessment of the morbidity. A convenient way to deal with such situations is to combine the outcomes, for example as ‘death or chronic lung disease’.
Some intractable problems arise when a continuous outcome (say a measure of functional ability or quality of life following stroke) is measured only on those who survive to the end of follow-up. Two unsatisfactory alternatives exist: (a) imputing zero functional ability scores for those who die (which may not appropriately represent the death state and will make the outcome severely skewed), and (b) analysing the available data (which must be interpreted as a non-randomized comparison applicable only to survivors). The results of the analysis must be interpreted taking into account any disparity in the proportion of deaths between the two intervention groups.