50 likes | 191 Views
University Medical Center Utrecht. Radboud University Nijmegen. U sing a B ayesian-network M odel for the A nalysis of C linical T ime-series D ata. Peter Lucas Karin Schurink Marc Bonten Stefan Visscher. AIME 2005. Introduction.
E N D
University Medical Center Utrecht Radboud University Nijmegen Using a Bayesian-network Model for the Analysis of Clinical Time-series Data Peter Lucas Karin Schurink Marc Bonten Stefan Visscher AIME 2005
Introduction • Time is an essential element in the clinical management of patients as disease processes develop in time. • A static Bayesian network was developed previously to support clinicians in the diagnosis and treatment of Ventilator-Associated Pneumonia (VAP) as treating and diagnosing critically ill patients is a challenging task. Aim • We have investigated whether this static Bayesian network can also be used to analyse the temporal data collected in the ICU to suggest appropriate antimicrobial treatment. AIME 2005 Using a Bayesian-network Model for the Analysis of Clinical Time-series Data
pathogen1 coverage1 pathogen2 coverage2 pathogen3 coverage3 Antibiotic therapy pathogen4 coverage4 Spectrum very narrow narrow intermediate broad pathogen5 coverage5 pathogen6 coverage6 pathogen7 coverage7 AIME 2005 Using a Bayesian-network Model for the Analysis of Clinical Time-series Data
Methods • A temporal database of 17710 records was used. Each record represents data of a mechanically ventilated patient in the ICU during a period of 24 hours. For 157 of these episodes, VAP was diagnosed by two infectious-disease specialists. • We consider the period from admission to discharge of the patient as a time series Xt , t = 0,…, np, where t = np is the time of discharge of patient p. For each record, we collected the output of the Bayesian network, i.e., the best possible antimicrobial treatment, taken into account colonisation data from 3 days before. H. Influenzae Acinetobacter Enterobacter1 P. aeruginosa t = np t = 0 VAP + treatment AIME 2005 Using a Bayesian-network Model for the Analysis of Clinical Time-series Data
Conclusions and Future Research • For 38% of the cases, the antibiotic spectrum was too broad. Further research is needed to improve the therapeutic performance of the Bayesian network Poster ~ ~ ~ ~ ~ ~ VAP ~ ~ ~ ~ ~ ~ • S.Visscher@umcutrecht.nl AIME 2005 Using a Bayesian-network Model for the Analysis of Clinical Time-series Data