INTRODUCTION: The term decompression illness (DCI) describes maladies resulting from inadequate decompression, but there is little consensus concerning clinically useful DCI subclasses. Our aim was to explore an objective DCI classification using multivariate statistics to assess naturally associated clusters of DCI manifestations. We also evaluated their mapping onto other DCI classifications and investigated the association with therapeutic outcome. METHODS: We defined the optimal number of clusters using "two-step" cluster analysis and Bayesian information criterion with confirmation by hierarchical clustering with squared Euclidian distances and Ward's method. The data were 1929 DCI cases reported by hyperbaric chambers to the Divers Alert Network (DAN America) from 1999-2003. RESULTS: Four robust and highly significant clusters of DCI manifestations were demonstrated containing 300, 741, 333, and 555 patients. Each cluster had characteristic manifestations. Cluster 1 was effectively pain only. For Cluster 2, characteristic manifestations included numbness, paresthesia, and decreased skin sensitivity; for Cluster 3, malaise, paralysis, muscular weakness, and bladder-bowel dysfunction; and for Cluster 4, hearing loss, localized skin swelling, tinnitus, skin rash and mottling, confusion, dyspnea/chokes, muscular problems, vision problems, altered consciousness, headache, vertigo, nausea, fatigue, dizziness, and abnormal sensations. DISCUSSION: Internal reliability was confirmed by arbitrarily dividing the dataset into two parts and repeating the analysis. The clusters mapped poorly onto traditional DCI categories (AGE, Type I DCS, Type II DCS), but more specifically onto the Perceived Severity Index (PSI). All three classification methods (DCI, Cluster, PSI) predicted complete relief of manifestations equally well. We conclude that cluster analysis is an objective method for classifying DCI manifestations independent of clinical judgment.
INTRODUCTION: The term decompression illness (DCI) describes maladies resulting from inadequate decompression, but there is little consensus concerning clinically useful DCI subclasses. Our aim was to explore an objective DCI classification using multivariate statistics to assess naturally associated clusters of DCI manifestations. We also evaluated their mapping onto other DCI classifications and investigated the association with therapeutic outcome. METHODS: We defined the optimal number of clusters using "two-step" cluster analysis and Bayesian information criterion with confirmation by hierarchical clustering with squared Euclidian distances and Ward's method. The data were 1929 DCI cases reported by hyperbaric chambers to the Divers Alert Network (DAN America) from 1999-2003. RESULTS: Four robust and highly significant clusters of DCI manifestations were demonstrated containing 300, 741, 333, and 555 patients. Each cluster had characteristic manifestations. Cluster 1 was effectively pain only. For Cluster 2, characteristic manifestations included numbness, paresthesia, and decreased skin sensitivity; for Cluster 3, malaise, paralysis, muscular weakness, and bladder-bowel dysfunction; and for Cluster 4, hearing loss, localized skin swelling, tinnitus, skin rash and mottling, confusion, dyspnea/chokes, muscular problems, vision problems, altered consciousness, headache, vertigo, nausea, fatigue, dizziness, and abnormal sensations. DISCUSSION: Internal reliability was confirmed by arbitrarily dividing the dataset into two parts and repeating the analysis. The clusters mapped poorly onto traditional DCI categories (AGE, Type I DCS, Type II DCS), but more specifically onto the Perceived Severity Index (PSI). All three classification methods (DCI, Cluster, PSI) predicted complete relief of manifestations equally well. We conclude that cluster analysis is an objective method for classifying DCI manifestations independent of clinical judgment.
Authors: David Monnot; Thierry Michot; Emmanuel Dugrenot; François Guerrero; Pierre Lafère Journal: Diving Hyperb Med Date: 2019-06-30 Impact factor: 0.887