Medical University of Vienna Jeroen S. de Bruin Validation of Fuzzy Logic in Infection Surveillance Jeroen S. de Bruin, Alexander Blacky, Walter Koller, Klaus-Peter Adlassnig
Medical University of Vienna Jeroen S. de Bruin About me… University Assistant at the Medical University of Vienna Main research topic The electronic detection of healthcare-associated infections
Medical University of Vienna Jeroen S. de Bruin Healthcare-Associated Infections Definition according to the ECDC: An infection is considered as [healthcare]-associated if it occurs later than 48 hours after admittance to a [healthcare] facility. Commonly abbreviated as either HAI, or HCAI.
Medical University of Vienna Jeroen S. de Bruin HAI types Various (main) types of HAI, depending on infection site: Blood stream infection (BSI) Pneumonia (PN) Urinary tract infection (UTI) Central venous catheter-related infection (CRI) Surgical site infection (SSI)
Medical University of Vienna Jeroen S. de Bruin Electronic detection data PDMS Biochemistry Microbiology
Medical University of Vienna Jeroen S. de Bruin Electronic detection system
Medical University of Vienna Jeroen S. de Bruin Fuzzy set theory Perform a qualitative abstraction on quantifiable data. Calculating the compatibility between the patient’s measurable health status and an abstract linguistic clinical concept Fuzzy logic Inference mechanisms to reason about more abstract clinical concepts using fuzzy sets. Fuzzy set theory & Logic
Medical University of Vienna Jeroen S. de Bruin Why use Fuzzy? Fuzzy set theory and logic introduce graduality Infections and infection signs no longer simply appear, but the development process can be seen and tracked Potential clinical uses: Patterns & prediction Early intervention Correct classification of HAIs
Medical University of Vienna Jeroen S. de Bruin Practical example Make fixed (crisp) thresholds fuzzy! Fuzzy region of fever between 37.5 and 38 degrees
Medical University of Vienna Jeroen S. de Bruin Fever fuzzy set
Medical University of Vienna Jeroen S. de Bruin Blood stream infection
Medical University of Vienna Jeroen S. de Bruin Choice of fuzzy threshold How to determine if the threshold was adequately chosen? Can it be wider? Is it too wide? Hypothesis Patients with a fuzzy indication of HAI tend to have fuzzy values for infection indicators (e.g. fever, hypotension, leukopenia, etc) more often Fuzzy threshold valid?
Medical University of Vienna Jeroen S. de Bruin Validation experiment Period: January – December 2011 #Stations: 10 intensive care units #Patients: 2,429 #Patient days: 24,487 Infection subset: CRI
Medical University of Vienna Jeroen S. de Bruin Validation results Infection ParameterNo infection signsFuzzy CRI signsp #Fuzzy values (%) Increased body temperature Shock < Increased C-reactive protein < Leukopenia 2.77 < Leukocytosis Fever < Hypotension Clinical signs of BSI < 0.001
Medical University of Vienna Jeroen S. de Bruin Validation results Infection ParameterNo infection signsFuzzy CRI signsp #Fuzzy values (%) Increased body temperature Shock < Increased C-reactive protein < Leukopenia 2.77 < Leukocytosis Fever < Hypotension Clinical signs of BSI < 0.001
Medical University of Vienna Jeroen S. de Bruin Conclusions Fuzzy logic can be used effectively to detect patients with mild or partial signs of infection Potential clinical uses for this method include: –Prediction –Early intervention –Accurate classification of HAI
Medical University of Vienna Jeroen S. de Bruin Thank you! Many thanks go out to: –Dr. Harald Mandl –The Clinical Institute of Hospital Hygiene