This study evaluated factors affecting the change in 5-year survival among Medicare beneficiaries diagnosed with CC and RC between 1992 and 2000. We found that patients diagnosed in 2000 were about a year older on average, with slightly lower Charlson scores, and were more likely to be female than those diagnosed in 1992. CC and RC patients diagnosed in 2000 were more likely to receive surgery only and surgery, chemotherapy and radiotherapy, respectively, than were patients diagnosed in 1992. We found that there have been considerable changes in stage distribution between 1992 and 2000, and that unadjusted 5-year survival improved by 7.7% (from 43.0% to 46.3%) for CC patients and by 7.1% (from 39.4% to 42.2%) for RC patients. As a comparison, unadjusted 5-year survival for patients in our larger study, who were matched to our CRC patients on age and gender, was 70.9% (unpublished authors' analyses).
Controlling for year of diagnosis and treatment mix, being diagnosed at later stages was shown to have a substantial negative impact on 5-year survival, as we would expect. Similarly, controlling for stage at diagnosis, patients diagnosed in later years (1996–2000) had significantly better survival odds than did patients diagnosed in 1992. In decomposing the effects of demographics, treatment mix, stage at diagnosis and technological change, we found that changes in demographics between 1992 and 2000 had a large negative effect on 5-year survival, while year of diagnosis, which we hypothesized as a marker for technological improvements, had the largest positive effect on 5-year survival. These results suggest that, despite the amount of effort that has gone into screening and colonoscopies, changes in stage distribution have contributed only a small amount to improvements in 5-year survival. This may be due, in part, to a possible bias caused by improved staging. Whereas screening and colonoscopies may have led to finding cancers earlier, improvements in staging accuracy may have contributed to a stage migration effect[11, 12].
Our study contributes to the existing literature on survival in colorectal cancer by focusing explicitly on older patients (the primary age group affected by this cancer), and by evaluating the latest available data, thus reflecting some of the recent changes in treatment patterns and detection for CRC. In addition, to our knowledge, ours is the first study that has attempted to disentangle the effects of earlier detection, demographics, treatment mix, and improvements in technology on 5-year survival in CRC patients using logistic regression techniques.
Comparisons with Other Literature
Despite some methodological differences, our findings are consistent with previous research on changes in stage distribution and survival in CRC. Recently, Gross et al used 11 years of SEER-Medicare data (1992–2002) to evaluate changes in the distribution of stage at diagnosis among colon cancer patients. They compared 3 time periods: before the first change in Medicare's reimbursement policy (1992–1997), between the first and second changes (1998–June 2001), and after the second change (July 2001–2002). They showed that the proportion of CC patients diagnosed at an early stage increased from 22.5% in period 1 to 25.5% in period 2 and 26.3% in period 3 (all significant increases), and our results were nearly identical.
Three previous studies have analyzed increases in US colorectal cancer survival rates over time, all of which examined a wider range of ages than our study. Jemal et al compared 5-year relative survival rates from the mid-1970s to 1995–2000 and found that colorectal cancer had the second-highest improvement in survival rates of all cancer types in that time period. They found that 5-year survival increased among men from 50% to 64% and among women from 52% to 63%. Brenner et al conducted a period analysis of SEER data on 5- and 10-year relative survival and found that both colon and rectal cancer patients had improved survival from 1998 to 2003. They found that for colon cancer patients, 5-year relative survival improved from 63% to 66% and 10-year survival improved from 57% to 62%, and for rectal cancer patients, 5-year relative survival improved from 64% to 67% and 10-year survival improved from 56% to 60%. Finally, Clegg and colleagues generated Kaplan-Meier estimates from SEER data and determined that relative survival had significantly improved from 1988 to 1997. Using similar data (1986–1997 SEER data), another study found that there were no significant improvements in 5-year relative survival. However, this study split the years of interest into 4 groups, concluding that there was no significant improvement in survival from 1986–1988 to 1995–1997.
Our results differ from previous studies for several reasons: 1) our population was limited to older patients; 2) we excluded patients who were enrolled in an HMO, who may have been healthier or had other characteristics that differed from the patients in our sample; and 3) we measured observed, rather than relative, survival. Observed survival takes all causes of death into account, whereas relative survival – the ratio of observed survival to expected survival in a comparable group of cancer-free individuals – is substantially higher than observed survival in older CRC patients. Using cause-specific survival to estimate relative survival requires that the information on cause of death be reliably coded. Even when cause of death is available to a cancer registry via the patient's death certificate, the information is often difficult to interpret. For example, Welch et al studied deaths among cancer patients that occurred within 1 month of diagnosis and found that 41% of deaths were not attributed to the coded cancer.
Few previous studies have attempted to isolate the impact of individual influences on survival in CRC, with most analyzing survival using joinpoint analysis and annual percentage changes instead[3, 8]. However, in a recent analysis, Sun and colleagues evaluated the impact of treatment and stage distribution on survival for multiple cancers using SEER data. They found that advances in treatment accounted for the majority of observed changes in survival (78% for all cancers examined) between 1980 and 2000. Since these researchers did not distinguish between treatment mix and technological advances in treatment and did not account for the influence of demographics, it is not possible to compare our results directly. However, their findings are consistent with ours in suggesting that advances in treatment (ie, technology) had the greatest impact on changes in survival over time.
This study is subject to certain limitations that are common to all studies that rely on retrospective claims data, such as potential coding errors and incomplete data. Our use of the SEER-Medicare database, which includes complete claims only for Medicare-eligible patients aged 65 years and older, introduces additional limitations[24, 25]. While older patients comprise the vast majority of patients with CRC, this sample is not representative of all US patients, particularly those with other forms of health insurance (eg, managed care, private pay). Despite its limitations, SEER-Medicare data have been used in numerous published studies of colon cancer, as well as cancers of the breast, prostate, and lung, among others. In addition, this study observed improvements in survival among CRC patients that may not be independent of improved survival for the general population. This study relied on year dummy variables to measure changes over time, which required the assumption that all determinants of survival were included and accurately measured in our specification.
Until recently, a combination of fluorouracil (5-FU) and leucovorin was the standard of care for CRC; numerous other drugs have been approved for use in CRC patients since 2004, including oxaliplatin, bevacizumab, cetuximab, and panitumumab. Because our study period ended in 2005, the influence of these new treatments on survival could not be measured. Future researchers will have the ability to use data collected since 2004, which will likely show greater changes in survival attributable to changes in treatment mix. Future researchers will also be able to use the methods we have described in this study to investigate whether changes in survival among cancer patients or other groups of patients may be attributable to changes in demographics, stage distribution, treatment mix, or technology.