|
BACKGROUND/RATIONALE:
Colorectal cancer (CRC) is the third most common cancer in veterans, with approximately 3,200 new cases diagnosed in VA facilities each year. While both the IOM and VA leadership have recognized that cancer health services research (HSR) is a priority, the IOM has also recognized that the cancer data system infrastructure is a barrier to cancer HSR and should be expanded, for example by linking cancer registry data to electronic medical record (EMR) data. The VAs Central Cancer Registry (VACCR) has not been released to investigators (preventing linkage to other VA databases) and existing VA databases (e.g., Austin data) do not contain cancer-related indicators (e.g., cancer type) that are important covariates in HSR. National VA databases have been used previously for cancer HSR, but these studies have used simple methods to identify cancer cases without testing for validity and reliability. Thus, more reliable methods are required to identify incident cancer cases in national VA databases. Consequently, the development of methods to identify incident cancer cases in VA databases would be a major advance in VA cancer HSR
OBJECTIVE(S):
To develop and validate a clinically informed algorithm that uses solely VA existing, EMR data to identify incident colorectal cancer cases.
METHODS:
This is a pilot project designed to test the feasibility of using VA data to identify incident CRC cases. The overall goal of algorithm development is to determine the set of identifying factors that maximizes discrimination between CRC cases and controls. The study links VISN 11 cancer registry (the "gold standard" for case identification) and VISN 11 facility-level VISTA data to determine the optimal criterion for CRC case ascertainment. The VISN 11 VISTA databases will be used to develop and test various algorithms to identify incident CRC cases, using a combination of diagnoses, procedures, treatments, laboratory tests, and/or provider codes. Algorithm development will be based on VISN 11 CRC cases diagnosed from 2001-2006 (expected sample size, N=1,050 CRC cases) and a 5% random sample of controls without any type of cancer. Cases and controls will be split into a training cohort (used to create and test various algorithms) and a validation cohort (used to confirm the validity of a chosen algorithm). Algorithm development will be an iterative process with interaction between clinical insight and statistical analysis. In constructing potential algorithms, we will give consideration to diagnoses, procedures, treatments, laboratory codes, and provider specialty codes, as well as information on the coding position (e.g., primary, secondary), when and how frequently codes occurred, and data type (e.g., inpatient, outpatient, pharmacy) that might improve case identification. Logistic regression and likelihood ratio statistics will be used to evaluate candidate identifying factors that will be considered in model development. Receiver operating curves (ROC) and relative tradeoffs in sensitivity, specificity, and positive predictive value (PVP) will guide the final model selection.
FINDINGS/RESULTS:
None at this time.
IMPACT:
The proposed project to develop more reliable and valid methods for CRC case ascertainment will provide a foundation for investigators to more reliably use VA databases to study cancer care quality. The case identification strategy we develop will enable the critical distinction to be drawn between incident and prevalent cases. Algorithm development is a necessary initial step before research projects can begin using VA databases to conduct quality of cancer care studies. In future research, we will apply our methods to conduct cancer HSR and quality of care studies using national VA databases.
PUBLICATIONS:
None at this time.
DRA:
Chronic Diseases
DRE:
Quality of Care, Technology Development and Assessment
Keywords:
Cancer
MeSH Terms:
Databases
|
