Observer Error in Reporting on Liver Scans for Space-Occupying Lesions

Abstract

Ninety-nine 99mTc-sulfur colloid γ camera-imaged liver scans were examined by seven observers who decided whether a space-occupying lesion was present. The true diagnosis was determined by operation, autopsy, or biopsy. Clinical information was at first withheld from, then given to, the observers. Four observers reread the scans after being informed of their high false-positive rates. On best performance, the average over-all error rate was 20% (36% false-negative, 8% false-positive). The seven observers had closely similar over-all error rates, but widely differing distributions of false-negative and false-positive rates. Attempts to bias the observers toward reducing their false-positive error rates were successful in the case of untrained observers, unsuccessful in the case of nuclear physicians. It is concluded that, because of the high rate of observer error in interpreting liver scans, they should be used to create an index of suspicion of abnormality rather than being regarded as diagnostic.