National Cancer Data Base/Surveillance Epidemiology and End Results: Potential insensitive-measure bias

Objectives. Abstraction of data from National Cancer Data Base (NCDB)/Surve illance Epidemiology and End Results (SEER) for reasons other than incidenc e, mortality, and patterns of care has risen. A potential problem with thes e data is that insensitive-measure bias can exist because of possible inacc uracies in hospital tumor registry staging. The purpose of this study is to assess the accuracy of tumor registry staging from six community hospitals . Methods. Staging of 103 consecutive cancers operated on by a gynecologic on cologist (one of the authors) as a surgical consultant to a gynecologist or surgeon was reviewed. Hospital tumor registry staging forms were arbitrari ly assigned to be completed by the nongynecologic oncologist versus the gyn ecologic oncologist by the medical records department. The authors reassess ed cancer staging by medical chart review. The tumor registry staging was c ompared with the actual staging as determined by the authors. Major staging violations were defined as errors that would significantly change stage en ough to alter prognosis or change recommended adjuvant treatment. All other violations were defined as minor. Results. Twenty-eight (27%) cancers were staged by the gynecologic oncologi st and 75 (73%) by nongynecologic oncologists. Eighty (78%) cancers were en dometrial and 14 (13%) ovarian. Eighty-three (81%) tumors were stage I or I I. Major staging violations occurred in 0% of cancers staged by the gynecol ogic oncologist and 22% (16/75) by a nongynecologic oncologist (P = 0.002). Minor staging violations occurred in 14% (4/28) of cancers staged by the g ynecologic oncologist and 42% (32/75) by a nongynecologic oncologist (P = 0 .005). Minor violations were due to omission of histologic subtype and/or g rade. Conclusion. The 22% major staging violation rate represents significant ins ensitive-measure bias. If additional studies produce similar results, abstr action of data from NCDB/SEER for reasons other than incidence, mortality, and patterns of care cannot be accepted as evidence-based scientific medici ne. (C) 2000 Academic Press.