A structured approach to assessing the suitability of clinical gold standards for use in computational decision support algorithm development

Abstract

Advanced decision support diagnostic algorithm use in medicine is being hampered by the problem that high quality evidence is not available to support adoption. The strongest evidence for a diagnostic tool is obtained from accuracy comparison trials to a true gold standard. Clinical Gold standards are the best tests available for a condition and frequently have high false positive rates. This thesis presents and tests a novel methodology for reviewing clinical gold standards. Two test cases were used one for neonatal spells comparing it to an unmodified clinical gold standard, the second for late onset neonatal sepsis using a modified clinical gold standard. Before embarking on any algorithm testing a detailed review of the clinical gold standard is needed with an expert panel of clinicians. Most clinical gold standards will need modification before being used as a comparator for advanced clinical decision support systems.