BACKGROUND: Much debate exists over the significance of the number of lymph nodes (LN) examined after colon resection. MATERIALS AND METHODS: The Surveillance, Epidemiology and End Results (SEER) database was queried for patients who presented with colonic adenocarcinoma. Multiple Cox proportional hazard regressions were run using successive LN cut-offs (6-26), first controlling for and then stratifying by T-stage. This was repeated in subsets of patients delineated by LN status. Additional variables controlled for in every regression were age, gender, ethnicity, marital status, number of positive LN, grade, metastases, and extent of surgery. After each regression, a Harrell's C statistic and an Akaike's information criterion (AIC) were performed to test the predictive capacity and fit of the model, respectively. RESULTS: 128,071 patients met selection criteria. The highest Harrell's C statistics among all patients were the cutoffs at 14 LN and 15 LN. Between those, the AIC shows that the cutoff at 15 LN fit the data more closely than the 14 LN cutoff. The models with the best predictive ability and best fit by T-stage were T1, 14 LN; T2, 10 LN; T3, 10 LN; T4, 12 LN. CONCLUSIONS: Using a population-based dataset, we show the optimal number of LN examined is dependent upon the patient's tumor stage. Across all T-stages, the highest optimal number of LN resected was 15. Since it is possible to estimate but not perfectly predict the stage of a patient's tumor preoperatively, we believe the recommendation should be based on the most conservative measure.
BACKGROUND: Much debate exists over the significance of the number of lymph nodes (LN) examined after colon resection. MATERIALS AND METHODS: The Surveillance, Epidemiology and End Results (SEER) database was queried for patients who presented with colonic adenocarcinoma. Multiple Cox proportional hazard regressions were run using successive LN cut-offs (6-26), first controlling for and then stratifying by T-stage. This was repeated in subsets of patients delineated by LN status. Additional variables controlled for in every regression were age, gender, ethnicity, marital status, number of positive LN, grade, metastases, and extent of surgery. After each regression, a Harrell's C statistic and an Akaike's information criterion (AIC) were performed to test the predictive capacity and fit of the model, respectively. RESULTS: 128,071 patients met selection criteria. The highest Harrell's C statistics among all patients were the cutoffs at 14 LN and 15 LN. Between those, the AIC shows that the cutoff at 15 LN fit the data more closely than the 14 LN cutoff. The models with the best predictive ability and best fit by T-stage were T1, 14 LN; T2, 10 LN; T3, 10 LN; T4, 12 LN. CONCLUSIONS: Using a population-based dataset, we show the optimal number of LN examined is dependent upon the patient's tumor stage. Across all T-stages, the highest optimal number of LN resected was 15. Since it is possible to estimate but not perfectly predict the stage of a patient's tumor preoperatively, we believe the recommendation should be based on the most conservative measure.