Standardized mortality in eating disorders--a quantitative summary of previously published and new evidence

Ten eating disorder (ED) populations were reviewed using the standardized mortality ratio (SMR) presenting new evidence for several studies. In eight of the ten samples, strong evidence (in one sample weak evidence and in one sample no evidence) supports an hypothesis of elevated SMR. We found strong evidence for an increase in SMR for anorexia nervosa (AN), whereas no firm conclusions could be drawn for bulimia nervosa (BN). Bias caused by loss to follow-up was quantified and found non-negligable in some samples (possible increase in SMR from 25% to 240%). We did not find a significant effect of gender or time period on SMR. Survival analysis showed a significant difference among the life-tables for males and females; female risk of death averaged 0.59% per year, whereas all male deaths occurred within the first 2 years after presentation. Weight at presentation had a highly significant effect on SMR, and lower weight at presentation was associated with higher SMR. Age at presentation exerted a significant unimodal effect on SMR; aggregate overall SMR was 3.6 for the youngest age group (< 20 years), 9.9 for those aged 20-29 years, and 5.7 for those aged > or = 30 years at presentation. Length of follow-up had a highly significant inverse effect on SMR; maximal SMR was 30 for female AN patients in the first year after presentation. A statistically significant increase in SMR was documented for at least up to 15 years after presentation. One study indicated a treatment effect on SMR. New evidence on causes of death suggests there are more deaths from suicide and other and unknown causes and fewer deaths related to ED than previously reported. Our findings have both research and clinical implications, with the most important clinical implication being the need for vigorous and well-directed treatment efforts from the initial presentation for treatment. An important research implication is that no single measure of mortality is sufficient; that is, only a combination of different statistics will maximize the available information.