Changing prognosis of metastatic colorectal adenocarcinoma: Differential improvement by age and tumor location
Talia Golan MD,
The Oncology Institute at the Chaim Sheba Medical Center, Tel-Hashomer, Israel
The Sackler School of Medicine, Tel Aviv University, Tel Aviv, Israel
Corresponding author: Talia Golan, MD, Gastrointestinal Oncology Unit, The Oncology Institute, Sheba Medical Center, Tel Hashomer, 52621 Israel; Fax: (011) 972-3-5352214; E-mail: firstname.lastname@example.org.
Over the past 2 decades, significant progress has been made in the field of metastatic colorectal cancer (mCRC) regarding new imaging techniques, surgical interventions, and systemic therapy. It is not known whether the benefit from these interventions has extended overall survival (OS) within the general mCRC population. A population-based survival analysis of newly diagnosed patients who presented with mCRC was therefore performed.
Survival statistics were extracted from the Surveillance, Epidemiology, and End Results (SEER) database for patients diagnosed with mCRC between 1988 and 2008. Demographic variables collected included age, race, and tumor grade. Survival was analyzed using the Kaplan-Meier method and extended Cox proportional hazard model as appropriate.
The study population consisted of 42,347 patients diagnosed with mCRC between 1988 and 2008 (52% women; mean age, 67 years). The 1- and 2-year estimated OS rates were 44% and 22%, respectively. Prognostic variables included race, sex, age, tumor location, and year of diagnosis. Median OS improved from 8 months to 14 months between 1988 and 2008. Significant improvements in OS were seen for all disease sites, but especially for descending colon cancers. Whereas the median OS increased by 13 months in patients ≤50 years of age and by 7 months in patients 51-70 years of age, the median OS of patients >70 years of age increased by only 1 month between 1988 and 2008.
During 2012, an estimated 143,460 new cases of colorectal cancer (CRC) will be diagnosed in the United States. Approximately 20% of these patients are diagnosed with metastatic colorectal cancer (mCRC), for whom systemic therapy has been the cornerstone of therapy.
Progress in the treatment of mCRC has been substantial. Between 1991 and 2008, the US Food and Drug Administration approved 7 new drugs for the treatment of mCRC (5-fluorouracil, irinotecan, oxaliplatin, capecitabine, bevacizumab, cetuximab, and panitumumab). Most of these agents have been shown to extend OS in patients with metastatic disease in randomized phase 3 trials, with extended survival ranging from 14 to 21 months.[2-9] Furthermore, surgical techniques have evolved over the past 20 years, enabling the resection of metastases in the liver and lung.[10, 11] In parallel, other ablative therapies have been developed, including radiofrequency ablation and extracranial radiosurgery. Based on retrospective studies, the resection or ablation of oligometastases appears to extend OS and cure a proportion of patients.[12, 13]
The overall net impact of these advances in the general population is unclear. The randomized phase 3 trials of new pharmaceutical agents are typically performed on selected populations (ie, younger patients, healthier patients, patients of better socioeconomic status), and it is not clear how generalizable these results are to other cohorts, such as patients with poorer health or the elderly. Likewise, the aggressive surgical/ablative management of metastases has not been tested in a phase 3 trial. In order to assess whether true progress against mCRC has been made in the wider community, we performed a population-based survival analysis of newly diagnosed patients with mCRC.
MATERIAL AND METHODS
This study utilized data from the 9 longest running registries within the Surveillance, Epidemiology, and End Results (SEER) database (Atlanta, Connecticut, Detroit, Hawaii, Iowa, New Mexico, San Francisco–Oakland, Seattle–Puget Sound, and Utah). The inclusion criterion was mCRC diagnosed between 1973 and 2008. Exclusion criteria were stage 1-3 disease, unknown histology, nonadenocarcinoma, or disease diagnosed only at autopsy. Data on patient demographics, tumor grade, primary site, year of diagnosis, and survival until death or follow-up as of December 31, 2008, were extracted. In addition, the median monthly family income for each patient's county was obtained from 1990 census data as provided through the SEER-Stat program version 7.1.0.
An initial analysis was performed on the years 1973-2008. A subsequent and more detailed analysis was performed on the years 1988-2008. The same inclusion and exclusion criteria were applied for the 2 analyses. For the purposes of analysis, continuous variables (eg, age and median regional income) were converted into categorical variables. Statistical analyses were performed using the Stata statistical package, version IC 11.1 (Stata, College Station, Texas). Chi-squared tests were used to assess associations between categorical variables. OS was defined from the time of initial diagnosis to the date of death and was calculated using the Kaplan-Meier method. The effects of demographic, pathologic tumor grade, primary tumor site, and age variables on survival were tested via Cox univariate analysis. Multivariate analysis was performed with a Cox proportional hazard model. Possible interactions were investigated by creating an interaction variable, which was included in the survival analysis together with the original variables; continuous variables were categorized. If the P value on multiple regression was <.05, the interaction was considered significant.
The assumption of the proportional hazards model was verified using the Grambsch and Therneau test as implemented in the Stata “estat phtest” command. When the assumption was not met, the extended Cox proportional hazards model with time-dependent variables was applied. Patients who had missing data were excluded from the multivariate analysis. All P values were 2-sided, and P<.05 was considered statistically significant.
An initial analysis of metastatic colorectal adenocarcinoma was performed examining the years 1973-2008. As demonstrated in Fig. 1 there was little change in survival among patients who were diagnosed from the early 1970s until the late 1980s, after which median survival began to increase. Subsequent in-depth analyses were performed for the period 1988-2008, during which the large changes in prognosis took place. This cutoff date was also chosen because the SEER database was modified considerably in 1988.
A total of 463,675 patients with colorectal cancer were identified through the SEER database as of its update on December 31, 2008. Patients were excluded if they had been diagnosed before 1988 (172,885 patients), had stage 1-3 disease (225,852 patients), had unknown histology or nonadenocarcinoma tumors (19,683 patients), or were diagnosed at autopsy (2,908), leaving 42,347 patients who presented with mCRC between 1988 and 2008. The characteristics of the patients are presented in Table 1. The proportion of patients diagnosed with metastatic disease during the period 1988-2008 increased from 16.18% in 1988 to 18.04% in 2008.
Table 1. Baseline Characteristics of the Study Population With Metastatic Colorectal Cancer
Diagnosed 1988-1997 (n=19,947)
Diagnosed 1998-2008 (n=22,400)
Abbreviation: NS, not significant.
Age, y, mean (range)
Age group, y, no. (%)
Sex, no. (% men)
Race, no. (%)
Location of primary tumor, no. (%)
Colon, side not stated
1990 family monthly income, median (range)
The mean age of our cohort was 67.7 years, with 47.7% being men. Over time, there has been a slight change in the age distribution of new cases, with the mean age of diagnosis decreasing from 70 to 67 years, and an increasing proportion diagnosed under the age of 60 years, from 22.4% in 1988 to 35.3% in 2008. As of the end of 2008, 89.7% of the subjects had died, with 77.5% of the deaths being attributed to colorectal cancer. Whereas 90.9% of the deaths in the ≤50 years age group were attributed to colon cancer, only 70.7% of the deaths in the >70 years age group were attributed to the disease.
The median OS of the entire study population was 10 months. Over time, the median life expectancy of patients with colorectal cancer has improved. For the entire population, median life expectancy improved from 8 months in 1988 to 14 months in 2008 (Fig. 1), with the sharp increase appearing from 1998 onward: the 18-month survival for patients diagnosed before 1997 was approximately 25% and rose to 41% for patients diagnosed in the year 2004. (Fig. 3).
A univariate extended Cox proportional hazard survival analysis was performed to assess the impact of the following variables on OS: age group, location of primary tumor, grade, sex, race, income quartile and year of diagnosis. Younger age, left-sided tumors/rectum, well-differentiated histology, sex, higher median regional income, and later year of diagnosis were all associated with a more favorable prognosis (Table 2). Interactions on survival analysis were demonstrated between year of diagnosis and age (P<.001), sex and year of diagnosis (P = .02), and primary location and year of diagnosis (P<.001). An extended Cox multivariate analysis was performed stratified by year-group: age and tumor grade were found to be significant (Table 2).
Table 2. Univariate and Multivariate Analysis of Variables Associated With Overall Survival in Metastatic Colorectal Cancer Patients, 1988-2008: Extended Cox Model
Stratified by Year of Diagnosis
For the purposes of description, the hazard ratios and P values for the data for two periods 1988-1997 and 1998-2008 are provided.
The multivariate analysis was stratified by year of diagnosis because this variable interacted with multiple covariates.
Age had a major influence on outcome; overall, the median survival for patients aged <50 years was 16 months compared with 7 months for patients aged >70 years. Net improvement in OS was highly dependent upon age; for patients aged <50 years, OS improved from 9 to 21 months; for patients 50-70 years from 10 to 17 months; for patients >70 years from 7 to 8 months (Fig. 2). Age was a significant factor in both the univariate and multivariate analysis, and showed a strong interaction with year of diagnosis.
The Influence of Grade
Tumor grade influenced outcome on both univariate and multivariate analysis (Table 2 and Fig. 4); the 18-month survival for patients with well- to moderately differentiated tumors was 38% versus 24% for patients with poorly differentiated tumors. There was, however, no demonstrable interaction between grade and year of diagnosis, and the prognosis of both low- and high-grade tumors has improved over time.
The Influence of Tumor Location
Tumor site also influences outcome; patients with rectal cancer or left-sided tumors had superior OS compared with patients who had right-sided tumors. Furthermore, the prognosis of left-sided tumors has improved more than the prognosis of right-sided tumors over the 20-year period (Fig. 5). In 1988, the median OS for right-sided, rectal, and left-sided tumors was 7, 10, and 10 months, respectively, whereas in 2008, the median OS for these 3 groups was 11, 15, and 18 months, respectively. On statistical analysis, there was a notable interaction between tumor location and year of diagnosis on survival analysis. Once stratified by year, tumor location was no longer an independent predictor of survival.
Additional prognostic factors investigated included sex and regional income (Table 2); however, the hazard ratios for these covariates approached unity. Women survived longer than men; however, the difference was small and not significant on multivariate analysis.
We note that in 1988, the percentage of patients diagnosed with metastatic disease was only 16.18%, whereas it was 18.04% in 2008. A modeling experiment was performed to assess whether the improved OS was merely the influence of the Will Rogers stage migration phenomenon (ie, that about 2% of the patients previously thought to be nonmetastatic based on 1988 standards were actually metastatic had 2008 standards been applied (eg, frequent use of positron emission tomography–computed tomography), and that this accounted for the apparent improvement in OS. We therefore performed a hypothetical analysis in which 255 patients with stage 3 disease from the 1988 dataset were randomly selected and reassigned as stage 4 disease, resulting in 18.04% of the 1988 subjects having metastatic disease (Fig. 6). Even with a diluted dataset, there was a highly significant difference between the 1988 and 2008 survival curves (P<.001), although there was an increase in the hazard ratio from 0.67 (confidence interval, 0.60-0.74) to 0.75 (confidence interval, 0.68-0.84).
Our population-based analysis demonstrates an improvement in the prognosis of mCRC in the general population from 8 to 14 months from 1988 to 2008. This benefit was seen in all tumor subgroups but was most prominent among patients aged <70 years and patients with left-sided tumors. The sharp increase in OS was seen from 1998 onward.
The strengths of the study include the large size of the database, with its broad coverage of all age ranges and performance status, unlike clinical trials or studies based on single-institution experience. Furthermore, this is a well-validated and reliable dataset with long-term follow-up.[18-20] Our study is retrospective in nature. We have no data regarding systemic treatments, no details of surgery, and no data regarding other ablative treatments, including radiofrequency or radiotherapy were applied for oligometastatic disease. Therefore, the major weakness of our study is that we cannot determine the causative factors contributing to the improvement in OS in the younger subjects, and the comparative lack of improvement in survival among patients aged >70 years.
The net improvement in OS over the 20-year period is most likely the result of the new pharmacological agents and surgical advances for oligometastatic disease, although we were unable to confirm this due to the limited data contained within the SEER database. We cannot exclude the possibility that the Will Rogers stage migration phenomenon is partly responsible for the perceived improvement (ie, that due to advances in imaging, patients with minimal metastatic disease who have a comparatively favorable prognosis are now being labeled as metastatic, whereas they were previously considered to be nonmetastatic). The fact that the proportion of patients diagnosed with metastatic disease increased from 16.2% in 1988 to 18.0% in 2008 despite the increased use of screening during this period supports this possibility. Nonetheless, as described in the Results section, modeling experiments performed to account for this stage migration suggested that although the Will Rogers phenomenon may have led to some overestimation of survival gains, the effect is comparatively minor.
Previous population-based studies of colorectal cancer have shown an improvement in OS over time; however, they included all stages of CRC in their analyses.[22, 23] We found that patients with right-sided tumors had a worse prognosis compared with patients who had left-sided colon and rectal tumors, consistent with previous studies. The influence of regional income on OS is in line with previous reports and highlights well-described disparities in cancer care. The complex reasons behind this are beyond the scope of this study.
Elderly patients are less commonly offered surgery and chemotherapy, at least within certain communities worldwide.[26, 27] This is despite the fact that there is considerable evidence that chemotherapy alone or in combination with biological agents has a similar therapeutic ratio in the elderly with good functional status compared with younger patients.[28-32] Recent data from an Australian cancer registry have shown that elderly patients (>70 years of age) were less likely to receive chemotherapy; only 40% received treatment in comparison with 74% in younger patients (<70 years of age). However, when patients receiving chemotherapy were compared by age cohorts, no significant difference was seen in median OS (21.3 months for <70 years and 21.1 months for >70 years). Such inequalities have prompted the International Society of Geriatric Oncology to set up a task force to address these disparities in the care of elderly cancer patients.
We have shown that elderly patients with mCRC have poorer outcomes overall, and this subpopulation has benefitted little from the advances of the previous 2 decades regarding OS. We could not determine why elderly patients had a worse prognosis, however.
Age was a significant prognostic factor, even when stratifying by year of diagnosis. However, there was also a strong interaction between age and year of diagnosis. These results illustrate a complex relationship between a patient's age and the prognosis of mCRC. It appears that (1) older patients have (and always have had) a worse prognosis compared with younger patients and (2) the strong interaction between age and year of diagnosis implies that older patients have benefited less over time than younger patients. Older patients were more likely to have well-differentiated tumors but less likely to have left-sided tumors. This is somewhat in contrast to other reports that have assigned much of the discrepancies due to more advanced CRC at diagnosis and more right-sided tumors.[35, 36]
We hypothesize that the supposed mechanisms for the discrepancies seen between elderly patients and their lack of improvement in OS may include: (1) competitive mortality, because a higher percentage of elderly patients died of non–cancer-related death compared with younger patients (30% versus 10%); (2) undertreatment, for which the reasons are unknown but may include concomitant medical problems, worsening performance status, or patient–physician preference; and (3) distinct biology, with elderly patients presenting with more right-sided tumors.
In conclusion, we have demonstrated that there has been a very notable improvement in survival for patients with mCRC over the past 20 years; however, it does not appear that patients over the age of 70 years have benefitted substantially. The reasons behind the increasing disparity between younger and older patients will be the subject of further studies.