Reliability for an imaging test: evaluation of scintigraphy to reveal liver state in gastrointestinal cancer.

: In 185 consecutive, surgical patients with suspected or proven gastrointestinal cancer a preoperative liver scintigraphy was performed; liver metastases were bioptically verified in 28 patients. A seven-class descriptive system was used for blind classification of the scintigraphies according to conferential consensus. Two years later (with the knowledge of the verified incidence of metastatic liver disease) the scintigraphies were reclassified. Considering the verified state of the patients, the latter classification was substantially improved. Probability for metastatic disease increased with more abnormal class label. At best, reliability for a twice interpreted scintigraphy as a binary test to denote the true state of the liver was 92 percent. The best cut-off level for diagnostic purposes yields a probability of 1.00 for an abnormal scintigraphy to denote liver metastases, accepting that 54 percent of patients with liver metastases were not found. For screening purposes, at best, a probability of 0.98 for a normal scintigraphy to denote no liver metastases was noted; thus, 83 percent of patients with normal liver state were missed. In conclusion, we find the liver scintigraphy classification system useful to find or exclude metastases in the test population and we think the procedure applied could be useful as a vehicle to report test results for any imaging modality.