Reply to Hoff et al.

Disease-Specific and All-Cause Mortality in Screening Protocols

In their letter to the editors, Geir Hoff and colleagues stress that the cumulative all-cause mortality occurring after a screening procedure should be considered as an important factor in the evaluation of the associated benefit. Errors in reporting the causes of death after a screening intervention may occur in opposite directions - either by falsely attributing death from other causes to the specific disease, or by falsely attributing to other causes deaths that are related to screening. With regard to colorectal cancer, death in direct relation to the screening should be classified as a complication of screening; death after endoscopic perforation or surgical treatment would be examples of this. On the other hand, persons in the screened group who are prevented from dying from colorectal cancer may die from other, much more common causes, such as ischemic diseases.

The role of screening in the prevention of colorectal cancer was recently reviewed in this journal [1]. The section entitled “Severe Complications of Screening” in the review was not clear enough regarding the distinction between the mortality related to screening and the cumulative global mortality in the group. Geir Hoff and colleagues must be thanked for emphasizing the distinction. They initiated the Telemark Polyp Study in 1983, and one group of population-sorted individuals (n = 400) was offered a screening intervention (flexible sigmoidoscopy). The causes of death in this group were checked against the cancer registry up to 1996. A second or control group (n = 399) of age-matched and sex-matched individuals selected by the same method were not offered screening in 1983. The basic results of this study, presented in 1999, confirm a significant decrease in the incidence of colorectal cancer in the group receiving screening. However, a significant increase also occurred in the cumulative all-cause mortality in the screened group. The excess death toll was not apparently related to the screening procedure, but to an excess of ischemic (mainly cardiac) deaths. In April 2001, Geir Hoff and colleagues [2] presented a further analysis of the difference in mortality between the two groups, including data on diet, weight, and smoking habits. It was shown that screenees with positive endoscopic findings tend to have a healthy style of life, while screenees with negative findings do not (persisting smoking and increased body mass index). Any survival benefit for the first group of screenees might therefore be neutralized by a loss in the other group. In addition, a high rate of mortality was observed in the small group (19 % of the sample) of noncompliant individuals, who were included in the screened group at the beginning of the trial but did not attend for follow-up. This observation is well in line with other data in the literature concerning screening trials, but there is no doubt that the figure increases the global burden of deaths in the screened group.

In general, the all-cause mortality is not affected by bias in classifying the causes of death, while the disease-specific mortality is. A recent publication by Black et al. [3] analyzes 12 published randomized cancer screening trials, and reports inconsistencies in the direction of variation of the disease-specific and all-cause mortality. Three of the analyzed trials are concerned with fecal occult blood testing. The decline in both mortality indexes is consistent only in the Funen study, while it is inconsistent in the Minnesota and Nottingham trials, where the decline in disease-specific mortality is in contrast with the variation in the all-cause mortality. The authors conclude that the all-cause mortality (a more reliable index) should always be examined for consistency with the disease-specific mortality when evaluating a randomized cancer screening trial.

The size of the difference in cumulative mortality (16 % vs. 10 %) between the screened and unscreened groups is surprisingly high in the Telemark study. This is most probably due to the relatively small size of the samples and unavoidable selection bias in a pseudo-randomized study. In any case, the Norwegian group has shown that the expected survival benefit following screening may be neutralized by multiple interfering factors associated with the lifestyle of the population concerned. The Norwegian group is right in stressing that screening should not be considered as an isolated intervention, and that it should be complemented by an educational program that emphasizes primary prevention.

References

• 1 Lambert R, Provenzale D, Ectors N. et al . Early diagnosis and prevention of sporadic colorectal cancer.  Endoscopy. 2001;  33 1042-1064
• 2 Hoff G, Thiis-Evensen E, Grotmol T. et al . Do undesirable effects of screening affect all-cause mortality in flexible sigmoidoscopy programmes? Experience from the Telemark Polyp Study, 1983 - 1996.  Eur J Cancer Prev. 2001;  10 131-137
• 3 Black W C, Haggstrom D A, Welch H G. All-cause mortality in randomized trials of cancer screening.  J Natl Cancer Inst. 2002;  94 167-173
• 4 Muller A D, Sonnenberg A. Prevention of colorectal cancer by flexible endoscopy and polypectomy: a case-control study of 32 702 veterans.  Ann Intern Med. 1995;  123 904-910
• 5 Thiis-Evensen E, Hoff G S, Sauar J. et al . Population-based surveillance by colonoscopy: effect on the incidence of colorectal cancer. Telemark Polyp Study I.  Scand J Gastroenterol. 1999;  34 414-420