The diagnosis of hypovolemia using advanced statistical methods.

Link to article at PubMed

The diagnosis of hypovolemia using advanced statistical methods.

Comput Biol Med. 2011 Nov;41(11):1022-32

Authors: Bárdossy G, Halász G, Gondos T

Abstract
AIM: Diagnosing hypovolemia is not a trivial task. Hypovolemia itself has several physical signs, but their specificity and sensitivity is limited, even using sophisticated monitoring techniques. However, diagnosing hypovolemia is crucial in critically ill patients to avoid worse outcomes. The aim of this paper is to provide methods for better estimation of the degree of hypovolemia in ill patients.
METHODS: The so-called hypovolemic index (HVI) is introduced which classifies the degree of hypovolemia with a number in the interval [0, 1]. Four new methods are presented for the more precise diagnosis of hypovolemia. All methods rely on fuzzy logic. In the first method, clinical thresholds are used in the fuzzy rule system. The second method uses an iterative ROC analysis to determine the thresholds. The third one determines the thresholds using one single ROC analysis ("One step" method). The fourth method uses a genetic algorithm (GA) for the determination of the thresholds. The HVI is calculated using the data of patients from a previous investigation. Each method (except the first one) is tuned on a so called training database. Afterwards, they are carried out on a test database in order to determine the potential of the method.
RESULTS: All four methods are capable of differentiating between hypovolemic and normovolemic patients. However, using the first and the second methods, several patients get a HVI of around 0.5, therefore, their degree of hypovolemia is ambiguous. The third and fourth methods deliver a better classification, hypovolemic and normovolemic patients are clearly separated from each other.
CONCLUSION: All four novel methods deliver powerful tools for the diagnosis of hypovolemic patients. The degree of the hypovolemic state of each patient can be estimated with a hitherto unattained degree of reliability. Using ROC analysis and GA the estimation can be improved further.

PMID: 21945236 [PubMed - indexed for MEDLINE]

Leave a Reply

Your email address will not be published.