OBJECTIVES. The purpose of this study was to assess the adequacy of th
ree widely used models-Lognormal, Weibull, and Gamma-for describing th
e distribution of length of stay. This is a fundamental step in the de
velopment of outliers resistant (robust) methods for the statistical a
nalysis of this kind of data, where the main objective is to determine
measures of average and total resource consumption of groups of patie
nts. Current practice uses several types of trimming rules, many of wh
ich are based on the Lognormal model, although theoretical and experim
ental bases are still insufficient. METHODS. The three models were adj
usted using robust procedures based on M-estimators to approximately 5
million stays grouped by Diagnosis-Related Groups (DRGs). The resulti
ng 3,279 samples were collected in five European countries during 3 ye
ars. RESULTS. Most Of the distributions observed could be fitted with
one of these models. The descriptions provided by the Gamma and the We
ibull models were similar, and the Gamma model could be omitted. The c
asemix description provided by the Lognormal-Weibull family was, for c
ertain countries, significantly better than the one provided by the si
ngle Lognormal model. Often, for a given DRG and a given country, leng
th of stay distributions could be described with the same model during
several years. A given DRG, however, usually had to be described by m
eans of different models for different countries. CONCLUSIONS. Practic
al and conceptual consequences of the results are discussed. They can
be extended to the analyses of other consumption variables used in hea
lth services. Statistical procedures for casemix description, includin
g current rules of trimming, should be improved by means of more flexi
ble families of models.