Comprehensive Databases: A Cautionary Note

Abstract

Over the last decade, there has been in an increase in the number and availability of registries of cancer patients. Two of the more commonly accessed sources of cancer patient data are the Surveillance Epidemiology and End Results (SEER) and the National Cancer Data Base (NCDB). These two databases differ fundamentally in how patient data are acquired; the NCDB is hospital based, while SEER is population based. Although this difference in method of data acquisition is extremely different, there are commonalities, both positive and negative, when large data sets such as these are queried to ascertain clinical outcomes on cancer patients. The large number of patients available for analysis allows for a broader perspective on treatment patterns and epidemiology. Additionally, it allows the opportunity to study rare cancers and to perform subset analysis of specific groups within a common cancer (i.e., postmastectomy radiotherapy in patients with locally advanced breast cancer; adjuvant chemotherapy in stage II colon cancer). Most recently, these types of data sets are being used to provide cancer-specific outcomes and minimum performance benchmarks, which are being made available for quality assessment and quality improvement. Clearly, the quality of the information obtained from these types of investigations depends on accuracy of the data collected. Questions regarding completeness, accuracy, and selection bias are common to both of these databases. It is important to note that the NCDB is a much larger data set than SEER. With increasing size of the data set, there is a greater potential opportunity for inaccuracies. Data are acquired from 1,500 American College of Surgeons Commission on Cancer (CoC)-approved hospitals, representing approximately 70 % of all newly diagnosed cancer cases in the United States. Another notable difference between the SEER and NCDB databases has been the issue of user access. The SEER database has had public access—that is, since 1975, it has been available to anyone who had the resources and who was willing to submit a signed usage agreement. In contrast, access to the NCDB has been limited. Recently, however, the CoC has undertaken a pilot program allowing access to the NCDB data. This access has been limited to investigators at CoCapproved programs who were selected to participate in an alpha test program to evaluate the potential for a more open access program to the NCDB data set. This pilot program is currently in a beta testing phase and is expected to expand to any CoC-approved program sometime in the next year. As access to large, comprehensive patient databases has increased, so too has the scientific investigation based on this information. There has been a relative explosion of outcomes research based on information obtainable from databases such as SEER and the NCDB. These two frequently used sources of oncologic patient data are maintained by the National Cancer Institute and the Commission on Cancer of the American College of Surgeons, respectively. The comprehensive nature of these databases, along with their large volume of data, almost dictates that they fall under the auspices of a large organization (such as those mentioned), both out of necessity born of the significant resources (financial, personnel, and technical) required, and to permit oversight to maintain standards and accuracy. Additionally, the use of these types of large databases has become so popular that they have spawned a wealth of new societies, degree opportunities, and meeting sections. These have been described by a number of terms including, but not limited to, ‘‘health services research,’’ ‘‘outcomes research,’’ and ‘‘public health research.’’ Along with all of these new meetings, degrees, and societies come new experts in these fields. It appears that as long as there remains access to these type of databases, the number of experts asking questions of these data appears limitless. Unsupervised access to these Society of Surgical Oncology 2013