Acoustic neurinoma diagnostic model evaluation using decision support systems

Three acoustic neurinoma (hereafter called acoustic neuroma) diagnostic models (Jenkins, Le Liever, Kaseff) were implemented as rule-based decision support systems and evaluated from the perspective of sensitivity, specificity, and US dollar cost, using a data base of 95 case histories suggestive of acoustic neuroma. The specificities of the models were equivalent (.97). The Jenkins model had the highest sensitivity (.96) and the highest average cost ($1470.99). The sensitivities and average costs of the Le Liever and Kaseff models were comparable (.84 vs. .82, and $1092.38 vs $1114.17, respectively). We observed that omitting brain-stem evoked response and electronystagmography testing from the Le Liever model subjected four (4.2%) more patients without acoustic neuroma to air contrast computed tomography, increased sensitivity to .89, and decreased the average cost to $774.75, without affecting specificity. We discuss the reasons for the slightly improved sensitivity and the impact of decision support systems on the clinician.