Our main finding is the strong association between symptoms of gastroesophageal reflux and the risk of esophageal adenocarcinoma, regardless of the presence of Barrett's esophagus. An association, although weaker, was also found for adenocarcinoma of the cardia, but not for esophageal squamous-cell carcinoma.
The strengths of the study include the population-based design, the uniform classification of tumors, and the complete and rapid case ascertainment, which enabled us to conduct personal interviews with all subjects. To avoid the problem of reverse causality, we restricted our analysis to symptoms that had begun at least five years before the interview. Patients with an esophageal tumor may be more inclined to remember previous esophageal symptoms (including reflux) than are other subjects. But because the patients were unaware of the histologic subtype of their tumors and of the possible differences in the pattern of risk among the subtypes, the effects of any recall bias should be similar for squamous-cell carcinoma and adenocarcinoma. For this reason, the totally negative finding with regard to squamous-cell carcinoma should allay concern about recall bias. Moreover, the clear dose–response relation points to a biologic effect rather than to bias.
We adjusted for tobacco smoking, body-mass index,13-15 and other possible confounders. None of them had a serious influence on the risk estimates. The effect of gastroesophageal reflux on the risk of esophageal adenocarcinoma thus appears to be independent of other variables. Moreover, associations of the magnitude found here are not plausibly explained by confounding. A nonparticipation rate of 27 percent among controls could have introduced bias. However, nonparticipation by patients and controls is unlikely to be linked to reflux.
Symptoms of reflux are considered to be fairly accurate, but not perfect, indicators of gastroesophageal reflux disease.11,16-19 The misclassification should be random, hence leading to underestimation rather than exaggeration of the association with the risk of cancer.20
Some misclassification of tumors according to site was unavoidable in some cases as a result of anatomy that had become deranged by the tumor. To evaluate the contribution of misclassification to the moderate association we found between reflux and risk of adenocarcinoma of the gastric cardia, we reanalyzed our data after applying more conservative criteria for cancer near the gastroesophageal junction (69 tumors were located within 10 mm of the gastroesophageal junction in endoscopic, surgical, and histopathologic measures). The point estimates were unchanged, indicating the robustness of our findings. The differential association with the symptoms of reflux implies that adenocarcinomas of the esophagus and of the gastric cardia are different diseases, contrary to recent suggestions.15
Multiple comparisons involving three groups of patients and one control group inflated the risk of a type I error. However, the strength and consistency of the observed association and the clear dose dependency make an effect of chance unlikely.
As compared with the only previous epidemiologic study of gastroesophageal reflux and esophageal adenocarcinoma or adenocarcinoma of the gastric cardia,10 we found a considerably stronger association. Our more extensive data on symptoms of reflux and our ability to distinguish accurately between adenocarcinomas of the esophagus and those of the gastric cardia may explain the discrepancy.