Site-Specific Comparison of Summary Stage 1977 and Summary Stage 2000 Coding
Holly L. Howe
Vivien W. Chen
and members of the NAACCR Collaborative Research Work Group1
NAACCR has conducted two assessments of the summary stage coding changes that were introduced in 2000 to elucidate the impact that they would have on researcher use of the data in evaluation of time trends in stage of disease at diagnosis. The first study was a reabstract study that compared the differences in coding cancers of the breast and lung using both 1977 and 2000 rules. The results of this activity are provided in a report posted on the NAACCR web site (see below).
The second assessment uses data from the Surveillance, Epidemiology, and End Results (SEER) Program to examine selected site-specific impacts for the most common cancer types of all cases diagnosed in 1998-2001 where differences in indirect assignment to either a summary stage 1977 or a summary stage 2000 code would occur using the more detailed extent of disease coding system (EOD) as a basis for staging. Cases were indirectly coded to summary stage twice using EOD: once based on the 1977 rules and the second based on the 2000 rules. EOD codes are not precise enough to capture all of the staging differences between summary stage 1977 and 2000. For example, for breast cancer no differences in stage distribution are observed when summary stage 1977 and summary stage 2000 are derived from EOD. However, a comparison of staging rules indicates that infraclavicular lymph nodes were distant in SS77 and regional in SS2000. There isn't a separate code in EOD for infraclavicular lymph nodes; they are grouped with regional lymph nodes. Therefore, the infraclavicular nodes are converted to regional for both schemes even though they should be distant for SS77 and their impact can't be measured. There are some instances where EOD cannot be used to evaluate the complete impact of a change between SS77 and SS2000. The changes in stage distribution presented in this report represent minimum, measurable stage shifts. Cancer sites (breast and testis) for which no differences were observed between the two EOD-derived staging schemes are not presented.
The protocol of the assessment is described in greater detail on the next page. The results for each site are summarized in a table with stage distributions provided according to the 1977 and 2000 schemas; a graph of these results; and a summary kappa statistic, which compares the agreement beyond chance in the resulting proportional distributions. The NAACCR work group defined their interpretation based on the magnitude of the kappa statistic and the dispersion of the disagreement (e.g., whether it was across all stage groups or one or two groups and which stage groups disagreed).
Generally, a kappa statistic greater than 0.97 was interpreted to mean that the two methods produced similar results; a kappa statistic from 0.90 to 0.96 yielded a cautionary statement as a part of the interpretation of stage comparisons that span the two sets of coding rules; and a kappa value below 0.90 was interpreted that the two methods did not result in comparable data. Further, when the differences in the distributions involved more than one or two stage categories, the interpretation of comparability became more cautionary than the kappa statistic alone might suggest (see e.g., cancer of the sinus). This interpretation of the kappa statistic is more stringent than that recommended by Fleiss2 because of the impact that small differences, commonality of the site, and source of discrepancy will have on interpretation of data that are compared from both coding systems.
1Dee West, Brenda Edwards, Maria Schymura, Barry Miller, Joellyn Ellison
2Fleiss JL. Statistical Methods for Rates and Proportions. Second Edition. (NY):John Wiley and Sons, 1981.