Reliability of scintigraphic diagnosis of liver metastases.
Three experienced specialists in nuclear medicine independently read 52 liver scans for the presence or absence of metastases. The accuracy of the assessments was, for each of the observers and for the observers' agreed-upon assessments, determined against findings at laparotomy or autopsy. Overall accuracy, predictive values of a positive and negative test result, sensitivity, and specificity were 0.71-0.79, 0.80-0.90, 0.63-0.72, 0.63-0.77, and 0.77-0.91. The accuracy was not higher for the agreed-upon assessment than for each of the observers' assessments. The interobserver variation was calculated as overall agreement between pairs of observers and ranged from 0.85 to 0.94. After adjustment of overall agreement for chance agreement, kappa values from 0.70 to 0.88 were obtained. In the determination of intraobserver variation overall agreement rates and kappa values ranged from 0.90 to 0.94 and 0.84 to 0.88. The scintigraphic diagnosis of hepatic metastases is reliable but, to increase the reliability further, it should be examined whether observer error reflects error in the primary detection of findings in the scans or error in the overall interpretation.