Merging claims databases with a tumor registry to evaluate variations in cancer mortality: results from a pilot study of 698 colorectal cancer patients treated at one hospital in the 1990s.

BACKGROUND: Prognostic models are essential for evaluating variations in cancer mortality statistics. While cancer stage is the most widely accepted and commonly used predictor of survival for cancer, electronic claims databases contain large amounts of information on cancer patients. Previous studies have used Medicare databases and tumor registry information from the Surveillance Epidemiology and End Results data sets to evaluate variations in outcomes for older cancer patients. We evaluated if similar analytic efforts could be carried out with readily available data sets for colorectal cancer patients of all ages who received care at a single hospital during the 1990s.

METHODS: Hospital tumor registry and discharge claims data for patients at one mid-western hospital with surgically treated stage I-III colorectal cancer from 1990-1998 were used to model survival. Kaplan-Meier logrank tests and Cox proportional hazards models tested the statistical significance of demographic, operative, and clinicopathological factors as predictors of survival. Survival probabilities also were compared to U.S. population life table data to determine if survival deficits were larger for younger cancer patients.

RESULTS: Of the 698 colorectal cancer patients, overall five-year survival probability was 65%, with a median follow-up of 44.7 months. Factors associated with higher relative risks of death included sociodemographic characteristics [female gender (1.5, 95% CI: 1.1-1.9), ages 70-79 years (1.7, 95% CI: 1.2-2.3), and > or = 80 years (3.3, 95% CI: 2.4-4.7) as compared to younger patients], clinical characteristics [moderate (1.5, 95% CI: 1.1-2.1) or severe (2.1, 95%: 1.4-3.2) comorbid illness, as compared to mild or no comorbid illnesses and emergency admission (2.1, 95% CI: 1.5-2.9)], pathological characteristics [positive surgical margins (3.5, 95% CI: 2.3-5.3): and higher cancer stage (stage II RR = 1.5, 95% CI = 1.1-2.2; stage III RR = 2.2, 95% CI = 1.5-3.2), as compared to stage I]. A comparison to the age- and gender-matched survival probabilities of the general population demonstrated similar deficits in survival for older patients (> or = 70 years) and younger patients (< 70 years).

CONCLUSIONS: While cancer stage is a reliable predictor of survival, other sociodemographic and clinical data elements can improve the evaluation of expected survival rates for patients with surgically resectable colorectal cancers. To facilitate comparative interpretations of mortality data, consideration should be given to merging hospital discharge claims data sets with tumor registry information in a manner analogous to that which has been done for older cancer patients who are covered by the Medicare program.