Truncation Due to Death. In public health research, when an outcome of interest is missing because a patient died before it could be measured, some researchers treat this observation as simply missing, and omit it from analyses, while others insert an outcome value corresponding to the lowest or worst possible outcome. Both methods have advantages and disadvantages. In the case of omitting the missing, this is an accurate reflection of lack of data, but produces an obvious bias, not to mention reduces sample size. Assigning death the lowest outcome value may be an accurate measure of the ‘value’ of death on the outcome scale, but it may not. Ideally, an analysis would take into account the information provided by the fact that an observation was truncated by death - although this results in a missing outcome value, these data are not missing in the same sense as an observation that is lost to follow-up. We know what happened to that patient, that patient died. It is this train of thought that proposes that placing death on the lowest end of an outcome scale is misleading, since death is not actually located on the same measurement scale as the outcome measure. As Xxxxx states in reference to a quality of life (QOL) study, “[t]o assign a particular value to QOL when dead is to assume we know how to trade off a particular QOL and being dead (and out of misery). Not only do we not know how to do this, but the trade-off could vary by individual, so we prefer simply to represent the actual truth at this point, and not bring in such extraneous value judgements [2006].” For example, in a study analyzing the effect of Progesterone on recovery from Traumatic Brain Injury (TBI), the outcome of interest was a measure of functional status, but this outcome was missing for several patients who died before functional status could be assessed [Xxxxxx et al., 2007]. Although one could argue that death corresponds to a functional status of zero, this does not necessarily represent the value judgements of individual patients. Indeed, if asked, individual patients may rank death above (as ‘preferred’) to certain lower levels of functional status, such as permanent vegetative state or severe disability. It is this philosophical argument, that individual patients may assign different values to death, and this should be reflected in estimates of treatment effect, that motivates Zhang and Xxxxx’x [2003] application of the principle stratification method to these types of problems. When applied to the tr...
