Association Rules of DCI Patient Clusters and Reliability of Clustering Analysis

Baris Aksoy (1) , Vincent Labatut (1) , Egi Murat (1) , Tamer Ozyigit (1) , Petar Denoble (2) , Costantino Balestra (3) , Richard Vann (2) , Alessandro Marroni (4)

1 Bit Lab
2 Duke Center for Hyperbaric Medicine and Environmental Physiology
3 Environmental & Occupational Physiology Laboratory
4 Divers Alert Network Europe - Research Division

Decompression illness (DCI) is an adverse outcome of decompression and has a wide spectrum of signs and symptoms. The treatment plan often depends on the classification of DCI, which makes the correct classification of DCI crucial; however, there is no consensus on the classification of DCI. [1-4]. We have previously attempted DCI clustering with statistical methods [5, 6] and suggested that data mining techniques can be used as a decision support tool to determine the type of DCI. Our recent study was based on a classification of DCI using multivariate statistics to assess naturally associated clusters of signs and symptoms based on 1929 cases reported by hyperbaric chambers to the Divers Alert Network from 1999 to 2003 [7]. The aim of this study is to validate the reliability of the previous work by applying three different alternative clustering methods, by comparing the results of two-step clustering analysis with the Perceived Severity Index (PSI) [4] and to validate the characteristics of patient clusters using association rules. Additionally, we will present the most interesting association rules detected by the A Priori algorithm on the same data.

Domaines: Bio-informatique [q-bio.QM]; Bio-Informatique, Biologie Systémique [q-bio.QM]

Baris Aksoy, Vincent Labatut, Egi Murat, Tamer Ozyigit, Petar Denoble, et al.. Association Rules of DCI Patient Clusters and Reliability of Clustering Analysis. International Conference on Diving Hyperbaric Medicine, 2009, Aberdeen, United Kingdom. pp.58. ⟨hal-00633616⟩