Development of a diagnostic computer decision support system for children presenting with fever

M Codarini*^1,2, JC Craig^#S,1,2, RM Hanson², L Irwig¹, D Isaacs^S,2, P Macaskill¹.
STEP Program, Department of Public Health and Community Medicine, University of Sydney¹; The Children's Hospital at Westmead²

Introduction Fever in children is a common diagnostic problem. Diagnosis can be difficult, especially in children with no clinical focus of infection, ten percent of whom may have serious bacterial infections with risk of disability and death. Clinicians usually make a diagnosis by evaluating the child's history, physical examination and preliminary investigation results. In theory, each item is like a diagnostic test. Combining test results to estimate explicit post-test probabilities of infection may assist clinical decision-making.

Current decision support systems for clinicians assessing febrile children without a clinical focus of infection have limitations. Estimates of the predictive value of history, examination and investigations are only available for young children. Recommendations from expert consensus guidelines for empiric laboratory testing and antibiotic treatment are controversial. The applicability of existing data since the introduction of new conjugate vaccines and the documentation of increasing antibiotic resistance is also unknown. Furthermore, there are no comparable data or diagnostic models in Australia.

The aims of this study are to estimate the combined accuracy of history, examination and preliminary investigations for the diagnosis of serious bacterial infections in children with fever of unknown source. A valid and accurate diagnostic algorithm will be developed and incorporated into a computer decision support system (CDSS).

Methods This prospective cohort study will be conducted in the Emergency Department of The Children's Hospital at Westmead over 24 months. Children younger than 15 years old with a history of fever or a recorded temperature of >38.00C at triage but no known congenital or acquired immunodeficiency will be eligible for participation. Clinicians will enter patient data into the Emergency Department database at presentation. Fields will be added to the existing database to standardise data collection. History and physical examination items and laboratory and radiological test results will be recorded. About 7,500 of the 16,000 expected eligible children will be followed by telephone contact to confirm primary diagnoses. Meningitis, occult bacteraemia, pneumonia and urinary tract infection will be diagnosed using standard microbiological and radiological criteria. Experts will be consulted if the child has had prior antibiotics or test results are equivocal. Additional tests will be arranged if fevers persist.

The accuracy of various combinations of history, examination and investigation items will be estimated for febrile children without a clinical focus of infection, for all bacterial infections combined, and for meningitis, occult bacteraemia, pneumonia and urinary tract infection separately. Logistic regression analysis will be used for both binary and multinomial outcomes. Expected pre-test and post-test probabilities of bacterial infection will be computed using the parameters of the diagnostic model. This new model will be added to Emergency Department software as a CDSS.

Discussion Prospective and comprehensive collection of data regarding history, examination and investigation results when children present with fever will contribute to the validity and accuracy of a diagnostic algorithm. This will be the first diagnostic model in Australia for children with fever. The combined accuracy of history, examination and investigations will be incorporated into a CDSS.

A valid and accurate computer decision support system may help clinicians improve their diagnostic accuracy in children with fever but no focus of infection. Benefits of prompt diagnosis should include less complication rates, hospitalisations, unnecessary investigations and inappropriate antibiotic use.

Presented at From Cell to Society 3, Research Conference 2002, College of Health Sciences, University of Sydney, 18 - 19 Sept 2002 Leura, NSW, Australia