How should informative censoring be analyzed

Dead people do not contribute data to studies of quality of life. When death competes with a non-mortality endpoint such as quality of life, a phenomenon called ‘informative censoring’ can occur. A treatment that increases mortality may preferentially lead to the death of debilitated patients who would have had very poor quality of life if they had survived. Similarly, a treatment that saves the lives of very ill debilitated patients may rescue these patients so that they are healthy enough to contribute very poor quality-of-life data to the study. Life-saving therapy in the ICU can seem to worsen quality of life, and harmful treatments can seem to improve quality of life. It is therefore problematic to adopt a therapy in the ICU solely on the basis of its apparent effect on quality of life unless we know that this effect is not purchased by increasing mortality in the most debilitated. There are statistical techniques for studying a non-mortality endpoint in the face of competing mortality and these should be used by intensivists to evaluate the effects of interventions on quality of life after critical illness [10,11]. Which outcome measures should be used?