ASO Author Reflections: A Statistical Caution Regarding Missing Clinical Stage in the National Cancer Database

Excerpt

Increasingly, scientific advances are supported with research from large databases such as the National Cancer Database (NCDB).1^,2 However, one of the major flaws in retrospective reviews of large de-identified databases is missing data. Although this cannot be avoided, how missing data are handled in study design and statistical analyses can significantly alter the conclusions drawn. Extensive publications on missing data remedies, especially multiple imputation, exist, but practical advice specific to the data source often is not available. Other authors have tried to bridge this gap with approachable methods papers specific to using the NCDB3 and handling missing data.4 This article5 explains a specific missing data problem regarding clinical stage in the NCDB, illustrated in breast cancer, wherein missing clinical stage was much more frequent in 2004–2007 than in the more recent years due to a change in abstraction rules. The problem also was more common for patients treated with adjuvant chemotherapy than for those treated with neoadjuvant chemotherapy. …