To Register      SMDM Homepage

Tuesday, 19 October 2004

This presentation is part of: Poster Session - Clinical Strategies; Judgment and Decison Making

DETERMINING INDICATIONS FOR CARE COMMON TO COMPETING GUIDELINES BY USING THE CLASSIFICATION TREE METHOD. APPLICATION TO THE PREVENTION OF VENOUS THROMBOEMBOLISM IN MEDICAL INPATIENTS

Jose Labarere, MD, CHU BP 217, Unite d'Evaluation Medicale, GRENOBLE Cedex 9, France, Jean-Luc Bosson, MD, PhD, TIMC-IMAG, Service d'Information et Informatique Medicales, Grenoble Cedex 9, France, Loic Belle, MD, CH de la region annecienne, Service de cardiologie, Annecy, France, and Claudine Robert, PhD, Institut National de Recherche en Informatique et en Automatique, INRIA Rhone Alpes, Montbonnot, France.

Purpose: Substantial variations have been reported in the advice given by competing guidelines addressing prophylaxis of venous thromboembolism. The aim of this study was to determine positive and negative indications of prophylactic heparin treatment common to four competing guidelines disseminated in France from 1998 to 2000.

Method: We retrospectively applied the guidelines to data derived from a cross-sectional study of 818 patients hospitalized in the adult medical wards of a university hospital. For each patient, we determined the number of guidelines recommending the use of prophylactic heparin treatment, discretized into three categories: "0" corresponded to an agreement of the four guidelines to recommend no prophylactic heparin treatment (n=301 patients), "4" to an agreement of the four guidelines to recommend prophylactic heparin treatment (n=273), and "1–3" to a disagreement between the four guidelines (n=244). We displayed the level of agreement between the guidelines by using recursive partitioning analysis, with the number of guidelines recommending the use of prophylactic heparin treatment as the dependent variable and venous thromboembolism risk factors as covariates. We used the C4.5 tree-growing algorithm, which relies on Shannon entropy as an impurity measure of a node and on gain ratio as a splitting criterion. The appropriateness of each indication was illustrated with regard to the rate of deep vein thrombosis detected by systematic compression ultrasound examination.

Results: The resulting classification tree involved ten terminal nodes. Its accuracy estimated by performing tenfold cross-validation was 82% (standard deviation = 3). The covariates determining the structure of the tree included history of venous thromboembolism, acute stroke, recent myocardial infarction, congestive heart failure, current cancer, respiratory insufficiency, bedridden status, acute infectious disease, and varicose veins. Five consistent positive indications of prophylactic heparin treatment were identified. They involved 257 patients (31.4%) and were supported by robust scientific evidence. Deep vein thrombosis was detected in 10.5% (27/257) of these patients. Two consistent negative indications involved 347 patients (42.4%). Deep vein thrombosis was detected in 2.6% (9/347) of these patients. Three indications involving 214 patients (26.2%) were discordant over the four guidelines.

Conclusion: Classification tree analysis of real patient data is a useful strategy to identify indications common to competing guidelines. These indications should be considered for inclusion when updating guidelines. Further randomized trials are needed to test discordant indications.


See more of Poster Session - Clinical Strategies; Judgment and Decison Making
See more of The 26th Annual Meeting of the Society for Medical Decision Making (October 17-20, 2004)