Abstract
Objectives Automation of test follow-up offers potential reductions in workload for clinicians.
The primary objective of the study was to evaluate the performance of MicrobEx, a regular expression-based algorithm in classifying urine culture reports in primary
care.
Methods A retrospective validation of MicrobEx was performed using urine culture reports abstracted from a single academic family
health team. MicrobEx classifications were compared with labels assigned manually by a human reviewer.
Measures of diagnostic performance were calculated.
Results MicrobEx achieved 95.3% accuracy, 88.6% sensitivity, and 100% specificity in classifying 1,999
urine culture reports.
Conclusion The accuracy of MicrobEx was comparable to its performance in the original development and validation study
by Eickelberg. Additional work is required to explore and improve the accuracy of
MicrobEx and assess its performance across primary care settings and with more complex urine
culture reports.
Keywords
urinary tract infections - natural language processing - electronic medical records
- family medicine
