TOWARDS AN AUTOMATED NOSOCOMIAL INFECTION CASE REPORTING - Framework to Build a Computer-aided Detection of Nosocomial Infection
Jimison Iavindrasana, Gilles Cohen, Adrien Depeursinge, Henning Müeller, Rodolphe Meyer, Antoine Geissbühler · 2009
The prevalence survey is a valid and realistic surveillance strategy for nosocomial infection surveillance but it is resource and labor-consuming. Querying the hospital data warehouse with a set of relevant features and applying a classification algorithm on the results can reduce the amount of cases to be evaluated by the infection control practitioners. The objective of this work is to provide a framework to build a nosocomial infection model with a set of pre-selected features with Fisher’s linear discriminant algorithm. Application of the methodology to two datasets provides promising results. It permits to predict respectively an average of 41.5% and 43.54% positive cases including respectively 65.37% and 82.56% true positive cases. The proposed framework can be applied to other classification algorithms, which are planned as future work.