 |
|
Analgesics and Symptomatic Diverticular Disease
Arch Fam Med. 1998;7:262-263.
NONEXPERIMENTAL (ie, observational) epidemiological methods are frequently employed to investigate the relationship between an exposure and a disease. The 3 most common observational research designs are case-control studies in which the cases and controls are chosen based on disease status, cohort studies in which the comparison groups are chosen based on exposure status, and cross-sectional studies in which disease and exposure status are determined after sample selection. As compared with randomized trials in which subjects are assigned to the comparison groups before exposure to eliminate selection bias, the comparability of the groups in observational studies is always of concern. A variety of design-phase restrictions (exclusion criteria, matching, and others) and statistical methods are used to reduce or adjust for the differences that have been anticipated and measured, but it is impossible to correct for unrecognized or unmeasured differences between the groups.
Generally, cohort designs such as the one used by Aldoori et al1 are subject to fewer sources of bias than case-control and cross-sectional designs, and are considered to be more robust and trustworthy. However, interpretation of the results of any single observational study is exceedingly difficult even when the study is done well. For this reason, criteria to be used to establish causality from epidemiological data have been suggested (Table 1).2
|
|
|
|
Criteria for Evaluating a Suspected Causal Association*
|
|
|
Having provided reason for caution, let us look more carefully at the study by Aldoori et al. The authors have ostensibly attempted to investigate whether nonsteroidal anti-inflammatory drugs (NSAIDs) and/or acetaminophen cause asymptomatic diverticular disease to become symptomatic. In fact, because they have given very little justification for suspecting acetaminophen to be a culprit a priori, I assume that they examined various potential causal factors (eg, corticosteroids, aspirin, physical activity, dietary fiber, and others) and, in the course of multiple analyses, identified NSAIDs and acetaminophen as significant. When multiple comparisons are made, the probability that an apparent association is due to chance increases. This is not reflected in the 95% confidence intervals or P values quoted, unless certain adjustments are made. In this case, we should demand that the lower confidence limit for relative risk be more than a little greater than 1.0 (eg,  2.0).
The population chosen was a subset of men participating in the Health Professionals Follow-up Study. The 35615 subjects were selected from a population of 51529 using several exclusion criteria: subjects whose reported dietary intake was far above or below the mean (presumably to exclude inaccurate data), those who left 70 or more items blank on the dietary questionnaire, those who had ever had any sort of cancer other than non–melanomatous skin cancer (the reason for which is not stated), a history of colon or rectal polyp, a history of ulcerative colitis, and a history of diverticular disease diagnosed prior to 1988 (an attempt to include incident cases only and to improve diagnostic accuracy). Subsequent analyses also excluded those with a history of gastric disease or duodenal ulcer. Other than the exclusion of cancer cases, these exclusions seem to be justified and should not seriously jeopardize generalizability. The rate of follow-up was 94%, which is outstanding.
Exposure to NSAIDs and acetaminophen was determined by a 1988 questionnaire that asked whether the individual currently used the agent 2 or more times per week. Analyses were also done for a subset of subjects who reported using 1 or both agents on both the 1988 and the 1990 questionnaires. There was no way for the investigators to determine actual frequency or dosage, however, which is unfortunate because a dose-response relationship can provide substantial reassurance about the validity of an association. Because the subjects were health professionals, one would assume that their reports of medication use were generally accurate.
The critical question relative to group comparability is: How might people who regularly use NSAIDs or acetaminophen differ from those who do not? Aldoori et al found, eg, that men who used NSAIDs were heavier, more active, more likely to smoke and to consume alcoholic beverages, and more likely to take vitamins. Most importantly, they were more likely to have had an endoscopic procedure. Those who took acetaminophen were also heavier, more likely to smoke, and more likely to take vitamins, and they were also more likely to have had an endoscopic procedure. We can only guess about other important differences. One might guess for instance, that users of NSAIDs or acetaminophen are more likely to take other medications and more medications than nonusers. They are probably more likely to see a physician regularly and to be examined and tested (including tests for occult blood in the stool). They may be more concerned about symptoms and therefore more likely to report them. These differences are potentially important since associations found between NSAID or acetaminophen use and any disease end point could instead be due to these other characteristics of the exposed group.
The disease end point chosen, symptomatic diverticular disease, is problematic. First, more than one third of the men in the age range of those in this study have diverticular disease, but only a fraction of them have been subjected to a barium enema, sigmoidoscopy, or colonoscopy and so remain undiagnosed. Second, the symptoms, abdominal pain, change in bowel pattern, and bleeding (including occult blood in the stool) can be the result of other conditions not excluded from the study population (eg, colonic angiomatosis, gastroesophageal reflux disease, Helicobacter pylori infection without ulceration, gingival disease, undiagnosed polyps, and NSAIDs themselves). In fact, less than 2% of individuals with diverticular disease ever bleed from a diverticulum,3 and the relationship between diverticulosis and abdominal pain is far from clear. Thus, anything that increases the probability that an individual will receive a lower gastrointestinal diagnostic procedure will automatically increase the detection rate of diverticular disease, and the finding of diverticular disease in the presence of symptoms does not necessarily mean that the symptoms are due to the diverticula even if a physician believes them to be. Aldoori et al did review actual records of a subset of patients to confirm that a procedure was done and that diverticula were found, but this falls far short of addressing this problem.
When evaluating an observational study, the order of assumptions should be first that the association is not real (ie, that it is due to chance, bias, or confounding), then that it may be real but is not causal, and finally that it is real and either directly or indirectly causal. The associations found in this study between NSAID and acetaminophen use and symptomatic diverticular disease, in my opinion, are interesting, but I have no idea what they mean. They may not be real. The 95% confidence intervals are wide and many of the lower limits are between 1.0 and 2.0. More likely they are real but not causal. Regular users of NSAIDs and/or acetaminophen are probably different from nonusers in many important ways, most of which have not been measured nor considered in the study design and analysis. It is most likely that some of these differences and not the use of the specific medications under investigation account for the associations that were found. It is possible that these medications actually stir up trouble in preexisting diverticula as the authors suggest. They might even be part of the causal chain leading to the development of diverticulosis, but the list of alternative explanations is far too long for us to know based on this study and the information referenced in the article. The data provided fall far short of satisfying Hill's criteria.
Finally, it is interesting to note how little is still known about the incidence of complications of diverticulosis. Much of the available literature is older than 15 years and practically none of it comes from primary care settings.
James W. Mold, MD
Department of Family and Preventive Medicine
University of Oklahoma Health Sciences Center
Oklahoma City
REFERENCES
1. Aldoori WA, Giovannucci EL, Rimm EB, Wing AL, Willett WC. Use of acetaminophen and nonsteroidal anti-inflammatory drugs. Arch Fam Med. 1998;7:255-260.
FREE FULL TEXT
2. Hill AB. Principles of Medical Statistics. 9th ed. New York, NY: Oxford University Press; 1971:309-323.
3. Bastidas JA, Gray GM, Gary M. Diverticulosis. Sci Am Med. [serial on CD-ROM]. 1997:chap 4; section XII.
RELATED ARTICLE
Use of Acetaminophen and Nonsteroidal Anti-inflammatory Drugs: A Prospective Study and the Risk of Symptomatic Diverticular Disease in Men
Walid H. Aldoori, Edward L. Giovannucci, Eric B. Rimm, Alvin L. Wing, and Walter C. Willett
Arch Fam Med. 1998;7(3):255-260.
ABSTRACT
| FULL TEXT
|
