Wayne S Kendal1. 1. Division of Radiation Oncology, The Ottawa Hospital Regional Cancer Center, and The Ottawa Health Research Institute, Ottawa, Ontario, Canada. wkendal@ottawahospital.on.ca
Abstract
OBJECTIVES: Clinicians will commonly individualize adjuvant cancer therapy, on the basis of the number of involved lymph nodes and other clinicopathological factors, under the assumption that despite the expected statistical variability of such data one can nonetheless garner useful information for the individual case. Here the scientific basis of this assumption will be examined. METHODS: Survival data from the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program for 19,107 breast, 4,234 gastric, and 4,058 rectal cancers were studied with Kaplan-Meier estimates and Cox proportionate hazard models. The minimal sample size required to discriminate between high and low-risk groups was determined from the hazard ratios between various comparative groups, and their respective frequencies. RESULTS: The number of involved nodes was the strongest prognostic factor for all 3 cancers, followed by tumor diameter and grade. Discrimination between high and low-risk nodal prognostic groups required samples of 30 to 200 cases, depending on the prognostics used and the specific tumor, to attain a two-sided alpha of 0.05% with 90% power. At the individual level such prognostications therefore were uninformative. CONCLUSIONS: Clinicopathological prognostics based upon the number of involved lymph nodes are subject to population heterogeneity that limits their application to large samples. At the individual level, these prognostics appear more spurious than useful. The use of such prognostics to tailor cancer treatment to individuals should be considered a specious practice; instead a more categorical approach, based on the results of randomized trials, should be used.
OBJECTIVES: Clinicians will commonly individualize adjuvant cancer therapy, on the basis of the number of involved lymph nodes and other clinicopathological factors, under the assumption that despite the expected statistical variability of such data one can nonetheless garner useful information for the individual case. Here the scientific basis of this assumption will be examined. METHODS: Survival data from the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program for 19,107 breast, 4,234 gastric, and 4,058 rectal cancers were studied with Kaplan-Meier estimates and Cox proportionate hazard models. The minimal sample size required to discriminate between high and low-risk groups was determined from the hazard ratios between various comparative groups, and their respective frequencies. RESULTS: The number of involved nodes was the strongest prognostic factor for all 3 cancers, followed by tumor diameter and grade. Discrimination between high and low-risk nodal prognostic groups required samples of 30 to 200 cases, depending on the prognostics used and the specific tumor, to attain a two-sided alpha of 0.05% with 90% power. At the individual level such prognostications therefore were uninformative. CONCLUSIONS: Clinicopathological prognostics based upon the number of involved lymph nodes are subject to population heterogeneity that limits their application to large samples. At the individual level, these prognostics appear more spurious than useful. The use of such prognostics to tailor cancer treatment to individuals should be considered a specious practice; instead a more categorical approach, based on the results of randomized trials, should be used.