• anorexia nervosa;
  • length of stay;
  • practice-based research.

Objective:  We describe the establishment of an Australasian multisite research database for inpatient treatment of anorexia nervosa (AN). Using this database, the second aim of this study is to investigate the extent to which length of stay (LOS) in participating facilities could be predicted at admission from patient, clinical, and site variables.

Method:  Standardized demographic and clinical data were collated for 213 admission episodes involving 154 participants over a 20 month period from five Australian and one New Zealand specialist treatment centres.

Results:  While nine variables significantly predicted LOS on univariate analysis, linear regression determined that only body mass index, and having had 2−3 previous admissions made significant independent contributions to LOS.

Discussion:  Multisite databases offer a viable means by which to conduct clinical research, particularly in regard to low prevalence disorders such as AN. Their additional advantage is that of involving front-line practitioners recruiting participants more likely to be representative of cases seen across treatment centres. At just under a fifth of the total variance predicted by the best-fit model, LOS in hospital remains an aspect of AN treatment difficult to predict, and future studies need to explore variables other than the obvious demographic or clinical issues at admission. The clinical and planning implications are discussed.