Article Text

Download PDFPDF

Colorectal cancer survival in Europe: the Will Rogers phenomenon revisited
  2. D J AHNEN
  1. Gastroenterology Division, Department of Medicine
  2. University of Colorado School of Medicine, Colorado, USA
  3. and
  4. Department of Veterans Affairs Medical Center
  5. Denver, Colorado, USA
  6. dennis.ahnen{at}

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

See article on page 533

Colorectal cancer (CRC) is one of the leading causes of cancer death in Europe. In this issue of Gut, Gatta and colleagues1 report survival results from a population based study involving 2270 cases of CRC from 11 cancer registries in six European countries (see page 533). The authors report a surprisingly wide range of overall survival among the registries; three year survival varied from a low of 25% in Cracow, Poland to a high of 59% in Modena, Italy. An understanding of the causes for this wide variation could serve as the database for efforts to improve the overall outcome of CRC.

It is well established that the prognosis of CRC is dependent on factors related to the tumour, the patient, and the treatment of the disease. More advanced stage, location of the cancer in the rectum, poorly differentiated histology, vascular invasion, and older age are all associated with a poorer prognosis of CRC.2 The surgical approach (extent of resection, electivev emergency, skill of the surgeon) and appropriate use of adjuvant therapy also significantly affect overall survival.2 ,3 Much of the variation between registries in the report by Gatta and colleagues1 appears to be due to differences in known prognostic factors. Not surprisingly, the four registries that reported the poorest overall survival also had the four lowest reported percentages of early stage disease and had four of the five highest rates of rectal cancer. After adjustment for these factors, however, the relative risk of death among the registries still ranged from 0.76 to 1.81, with four of the sites being statistically different from the reference site.

Differences in surgical treatment appeared to account for much of the remainder of the differences in CRC survival among the registries. There appears to be a substantial difference in the surgical approach to CRC among the sites. The Cracow registry reported particularly low overall rates of surgical resection (53%), lower rates of resections for potentially curable disease (96% for Dukes stages A+B and 91% for stage C) and much lower rates of elective versus emergency resections (56%) than the other centres. In a model controlling for resection rates, but not elective versus emergency surgery, the variation among registries narrowed, and only the Cracow site, with a relative risk of death of 1.82, remained significantly different from the reference site.

What is the reason for the remaining substantial variance between Cracow and the other registries? A number of established prognostic factors such as the degree of differentiation of the cancers, presence of vascular invasion, or use of surgical adjuvant therapy were not measured in this study. If the registry data were controlled for these parameters, perhaps the differences in survival would narrow further.

It is tempting to speculate that the residual variation in prognosis could be because the aetiology and biology of CRC is different in Cracow than at the other sites. The lower age of patients and higher per cent of rectal cancers in Cracow suggests that there could be differences in aetiological factors or genetic predisposition to CRC in Cracow, and such discrepancies could be reflected in differences in the biological behaviour of CRC. Recently, several specific genetic alterations have been reported to have independent prognostic significance in CRC.4-7 The presence of microsatellite instability which occurs in almost all CRCs of the hereditary non-polyposis colorectal cancer syndrome and in about 15% of sporadic colorectal cancers has been reported to be associated with an improved prognosis.4 In contrast, loss of heterozygosity in the region of the DCC gene on chromosome 18q and Ki-ras mutations have been reported to be independently associated with a poorer prognosis in stage II CRC.5-7 None of these newer genetic prognostic markers was measured in the registry study.1 Although differences in the biological and genetic basis of CRC could account for the residual variation in reported survival across Europe, a more likely possibility is a variant of the Will Rogers effect.

While commenting on geographic migration during the economic depression of the 1930s, the American humourist Will Rogers is alleged to have said “When the Okies left Oklahoma and moved to California, they raised the average intelligence levels in both states”. An analogous phenomenon, stage migration, occurs with more careful staging of cancer.8 If a population of patients is more accurately staged it will improve the survival of all stages because patients with subtle advanced disease will be upstaged. For example, if liver imaging is routinely included in the staging workup, patients with subclinical liver metastases will be identified, removed from the stage III group and added to the stage IV group thus improving the expected survival of both groups. There is direct evidence in the report by Gatta and colleagues1 that there was variability in the intensity of the staging protocol among the registries. In 45% of cases, the staging of the CRC was not explicitly stated in the clinical record and had to be reconstructed from the medical records; often the information necessary for accurate staging was not available. In Cracow, for example, a lower percentage of subjects underwent tumour resection so that accurate pathological staging was not possible in 47% of subjects. Similarly, only 44% of subjects with CRC from Cracow had a liver imaging study performed compared with a mean of 67% for all of the centres. There was also a wide variation in the intensity of the lymph node evaluation among the registries, with as low as 2% and as high as 31% of the resected specimens having 12 or more lymph nodes examined. It seems likely that the reverse of the Will Rogers phenomenon—that is, less careful staging of patients leading to a lowering of the survival of all stages—accounts for much of the residual unexplained variation between CRC survival reported among the registries. Gatta and colleagues1 recognised this possibility and tried to adjust for it, using the frequency of 12 or more lymph nodes examined and the rate of liver imaging as surrogates for accuracy of staging. This adjustment decreased relative risk from 1.82 to 1.69 for the Cracow registry but the authors could not completely adjust for incomplete staging information.

What can we learn from these types of comparative data between cancer registries? The current study1 has identified at least two important differences in healthcare patterns that can be used to direct efforts to improve survival of CRC. It appears that late stage of presentation of disease is a major contributor to the poorer survival in some regions, and programmes to increase screening and early detection would probably have a major impact on CRC survival in those areas. Patterns of surgical treatment for CRC appear to account for much of the residual variation in CRC survival across Europe. Efforts to apply elective surgery more consistently and earlier in the course of CRC would pay dividends in improved survival for this disease. Finally, the report by Gatta and colleagues teaches us once again that survival comparisons are limited by the quality of the staging information available and the accuracy with which it is recorded. The Will Rogers phenomenon has been used to illustrate the risks of using historic control groups and to argue for the need for concurrent controls in intervention trials.8 The same reasoning is applicable to and limits the interpretation of comparisons of registry data. Despite this limitation, Gatta and colleagues1 have performed a service by identifying important differences in patterns of health care across Europe that are worthy of attention.


Linked Articles