%0 Journal Article
%T Methoden zur Pr diktion von Hochnutzern: ein systematischer Literatur-Review [Methods to predict high users: a systematic literature review]
%A Hartmann
%A Justyna
%A Schauer
%A Svenja
%A Krauth
%A Christian
%A Amelung
%A Volker
%J GMS Medizinische Informatik, Biometrie und Epidemiologie
%D 2012
%I German Medical Science, D¨¹sseldorf
%R 10.3205/mibe000126
%X [english] Background: A small group of patients accounts for a high amount of health care expenditures in Germany as well as in other countries. A portion of these expenses could be prevented by early identification of potential high users. This is possible through predictive modelling which offers various of methodical approaches. Therefore, the aim of this study is to identify different methodological approaches of predictive modelling of potential high users and to aid the decision-making process for the selection of appropriate method.Method: A systematic literature search was done in the scientific database SciVerse Scopus in October 2011 and supplemented by a manual search. Two persons selected identified citations in a two-step procedure independently, according to predetermined inclusion and exclusion criteria.Results: From the 216 identified publications, 18 articles remained after the final selection process. Two different approaches for dealing with this topic can be identified. On the one hand, there is an approach that focuses on patient-characteristics. Therefore, studies using this approach define high cost patients based on the frequency of health care utilization. The methods used for this approach are logistic, linear and negative binomial regression, with logistic regression as the most common one. On the other hand, there is a cost-oriented approach. Papers with this focus are primarily interested in testing different methods and new ways of prediction. The common method of logistic regression is used as well as the very special method of extreme regression. Data-mining techniques and classification systems like diagnostic cost groups are utilized as well. These methods are suitable for preparation and information processing of a large amount of diagnostic data. Conclusion: Different methods to predict high users exist. The choice of the method depends on the research question, the aim, the data and the available resources. When research focuses on predictors of high usage, logistic regression is a suitable and commonly used method. [german] Hintergrund: Auf einen kleinen Anteil von Patienten entf llt ein gro er Anteil der Krankheitsausgaben. Dies zeigen sowohl deutsche als auch internationale Studien. Ein Teil dieser Ausgaben k nnte durch fr¨¹hzeitige Identifikation potentieller Hochnutzer vermieden werden. Dies ist unter anderem durch die Entwicklung eines Pr diktionsmodells m glich, wobei die methodische Umsetzung eines solchen Modells sehr unterschiedlich aussehen kann. Ziel dieser Arbeit ist deshalb herauszuarbeiten, welche method
%K frequent attenders
%K high-cost cases
%K heavy user
%K predictive modelling
%K methods
%K Hochnutzer
%K Hochkostenf lle
%K Pr diktion
%K Pr diktionsmodell
%K Methoden
%U http://www.egms.de/static/en/journals/mibe/2012-8/mibe000126.shtml