Statistics from Altmetric.com
Editor,—The study by Karlen et al (Gut1998;42:711–14) makes the reader wonder how the authors could draw a conclusion that is suggestive of a positive outcome when the data analysis clearly shows no statistical significance.
It is well established that mortality from colorectal carcinoma is an important factor in patients with ulcerative colitis. However whether surveillance programmes have any effect on mortality is still matter of debate.1-4 Furthermore for the practical and ethical considerations mentioned by Karlen et al in their introduction, it is unlikely we would ever have a randomised controlled trial to look at the value of colonoscopic surveillance in this respect. It is likely that we will come to rely on studies such as Karlen et al’s.
It is for this reason that we should scrutinise the methods and results of such studies closely so that we can draw appropriate conclusions. Karlen et al studied 4664 subjects with a definite diagnosis of ulcerative colitis from an approximate background population of 3 million people in two regions of Sweden. The patient group was identified as “all patients in the study population that had died from colorectal cancer after 1975”, which amounted to 40 cases. The aim of the control population was to have three controls for each patient matched individually by age at diagnosis (± 5 years), duration of disease, extent of disease at diagnosis, sex, as well as needing to be alive at the time of death of the patient and to have some part of the colon intact five years prior to the diagnosis of the cancer of the patient. The criteria are extensive and valid and the authors’ aim was to obtain 120 individuals. However they give the impression that after the 120 individuals were chosen, tight scrutiny of the criteria confined them to 102. Surely you would select your controls based on defining your criteria initially rather than reassessing the number of individuals chosen and excluding some that do not meet the “criteria”.
Finally the results give a relative risk for which the reference is not clearly identifiable. One is left to assume that it is the relative risk of survival advantage having undergone a surveillance colonoscopy. Regardless of this, the confidence interval includes 1. Karlen et al go on to state that of those undergoing two or more surveillance colonoscopies the relative risk was 0.22 with a 95% confidence interval that is wider than the overall confidence interval for the relative risk of ever having a surveillance colonoscopy. This confidence interval also includes 1, and they state that somehow this indicates “a protective dose response relation”.
The final comment in the results alludes to 10% of controls having undergone colectomy within five years prior to diagnosis of the cancer of the patient but we are not informed of the reasons why these patients underwent colectomy. It is important to know whether their colectomy was for colorectal cancer or failure of medical treatment to control their ulcerative colitis.
One is left disconcerted that, given the criticisms outlined earlier, the authors could still conclude that, “the main finding, although not statistically significant, indicates that colonoscopic surveillance may have a protective effect against death from colorectal cancer”. Furthermore that this, “protective effect is even more pronounced if the patient underwent two or more surveillance colonoscopies, indicating a protective dose response relation”.
This is clearly a negative study, and the statement that there is a “protective dose response relation” must be considered to be unfounded and biased.
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.