A Geometric Model for Physiologic State Classification and Severity Classification of Critically Ill Patients
A Geometric Model for Physiologic State Classi cation and Severity Classi cation of Critically Ill Patients
Herman P. Friedman and John H. Siegel A m...
A Geometric Model for Physiologic State Classi cation and Severity Classi cation of Critically Ill Patients
Herman P. Friedman and John H. Siegel A model that represents physiologic states of patients as vectors in a 3-dimensional Euclidean space is presented. The model has been developed from observational data that were obtained from 3250 sets of 17 quantitative physiologic measurements from 919 patients at three centers. Concepts and tools of cluster analysis along with related methods of multivariate data analysis were used to de ne a reference state and six prototype physiologic patterns of adaptation, viewed as departures from the reference state, of the host defense response to events such as injury and sepsis. Within the model, the vectors from the origin (reference state) to each of the prototype states de ne directions that are indicative of the physiologic patterns associated with that state. These patterns provide a basis for a useful clinical interpretation for the nature and severity of a patient's response as function of location in the 3-dimensional space. The collaborative process used for the development of this model is described. This process had to blend prior clinical knowledge with statistical tools and concepts to provide a rational basis for de ning objectives for the classi cation problem. Key elements in this process are discussed and related to current issues in the application of classi cation theory and cluster analysis methods. The eective use of this model to organize the multifactor data necessary to cope with complex clinical decisions that are omnipresent in the care and treatment of patients with post-trauma critical illness is discussed. Herman Friedman will describe the model and discuss key elements of the collaborative process and Dr. Siegel will delineate the physiologic basis of the model and its clinical applications to critically ill and injured patients.
Herman P. Friedman Statistical Science and Technology Associates, Inc. 14C Spruce Lane Princeton, NJ 08540 email: hpfriedm@ix:netcom:com John H. Siegel Department of Surgery New Jersey Medical School:UMDNJ Newark, New Jersey email: siegeljh@umdnj:edu, 1