Evaluation of the Maruyama computer program accuracy for preoperative estimation of lymph node metastases from gastric cancer.

Controversy still exists about the optimal lymph node (LN) dissection for potentially curable gastric cancer. For rational LN dissection it is important to know the incidence of metastasis at each LN station. For this purpose a computer program was developed using data from 4302 primary gastric cancers treated at the National Cancer Center Hospital in Tokyo between 1969 and 1989. To evaluate the accuracy of the computer program, the differences between the individual reports generated by the computer and the stored data were investigated in 282 Italian patients submitted to curative gastrectomy and D2 or more extended LN dissections for gastric cancer. Receiver operating characteristic (ROC) analysis was used to assess the sensitivity and specificity of the program for predicting LN metastases in each of the 16 regional LN stations. The computer program showed good predictive ability for LN metastases in most of the 16 LN stations, as the areas under the curve ranged from 0.741 (station 15) to 0.944 (station 8), with a mean of 0.856. A critical cutoff point of 18% of the program's expected percentage was the value maximizing the validity of the prediction. Using an "absolute" cutoff point of 0%, the overall rate of false-negative (FN) predictions in 176 N+ patients was 11.9%; of these, 11 (6.2%) were absolute FNs, in which the program totally failed to estimate LN metastases; the remaining 10 cases (5.7%) were relative FNs because the specific prediction was positive for a different depth of stomach invasion. The low number of D3/D4 lymph-adenectomies in the historical database may affect the low estimate of metastases to N3/N4 nodes generated by the program. Based on these data, the program predicts with good accuracy the extent of LN metastases from gastric cancer, but it is not recommended for directing the surgeon to perform more extensive lymphadenectomy.