RESUMO
For the purpose of predicting postoperative local recurrence and/or distant metastasis in differentiated thyroid cancer, a discriminant analysis was done. To assess the reliability of this method and the induced function, simulation was done by using random samples under various conditions. Simulation revealed that the results obtained by this method were relatively stable. The discriminant functions Z1 and Z2 were induced, which are required at many hospitals. If the ratio of recurrence is less than 1:8 in a hospital, function Z1 should be applied. If the ratio is 1:10, function Z2 should be applied. Z1 = -(sex x 2.104) + (age x 0.032) - (diameter of tumor x 0.033)-(site of tumor x 0.871) - (histology x 0.393) + 0.249 (discriminant boundary 0.451) Z2 = -(sex x 1.967) + (age x 0.018) + (diameter of tumor x 0.025) - (site of tumor x 0.646) + (histology x 0.014) - (local invasion x 0.840) - (lymph node metastasis x 0.009) + (operation x 0.364) - (lymph node dissection x 0.235) + 1.058 (discriminant boundary 0.400). In an internal check, sensitivity was 78%, specificity 64%; and false-negative rate 5%. In an external check, sensitivity was 68%, specificity 93%, and false-negative rate 9%. In the actual application to hospitals, sensitivity was 58% to 100%, specificity 47% to 63%, and false-negative rate 0% to 16%. In clinical usage, a high sensitivity and low false-negative rate are required. These results fulfilled this condition.
