Performance analysis of a medical decision algorithm to mitigate spread of SARS due to interfacility patient...

To determine performance of a medical decision algorithm to mitigate spread of severe acute respiratory syndrome (SARS) from interfacility patient transfers during the Toronto SARS outbreak.

Records from the Provincial Transfer Authorization Centre and Toronto Public Health from April 1 to July 31, 2003, were linked using probabilistic methods. Authorization decision (transfer authorized or denied) and SARS status (probable case, suspect case, or patient under investigation for SARS; or non-SARS case) were obtained for linked records. Primary outcome was the number of patients where correct authorization decisions were made based on SARS status at the time of request. Secondary outcome was the number for whom, in retrospect, authorization decision was correct knowing final SARS status. Algorithm sensitivity, specificity, and predictive values were determined.

There were 14,571 requests for transfer and 2,132 patients investigated for SARS during the study period. The algorithm authorized 14,551 and did not authorize 20 requests. Sensitivity and specificity to make appropriate authorization decisions at the time of request were 100% (95% confidence interval [CI], 77.2%-100%) and 99.95% (95% CI, 99.9-100%), respectively. Positive and negative predictive values were 65% (95% CI, 44.1%-85.9%) and 100% (95% CI, 98.4%-100%), respectively. Sensitivity and specificity, in retrospect, within ten days of the transfer request were 100% (95% CI, 80.6%-100%) and 99.97% (95% CI, 99.9%-100%), respectively. Positive and negative predictive values were 80% (95% CI, 62.5%-97.5%) and 100% (95% CI, 98.4%-100%), respectively. Seven of the 20 patients with nonauthorized requests were not known to have SARS at the time of request. Within ten days, three of seven were under investigation for, a suspect case of, or a probable case of SARS.

The medical decision algorithm was highly sensitive and specific in correctly authorizing transfers. Despite its highly sensitive and specific algorithm, it did incorrectly deny authorization to a very small number of patients without SARS.