The risk of premalignant and malignant pathology in endometrial polyps

At the time of writing a paper on the above topic (1), our data extraction was performed using manual counting. When we computerized the data recently, we found that there was a minor difference in the figures reported in the tables. In this letter we provide the correct data and discuss its implications on our original conclusion. In the paper (1), we had concluded that ‘In abnormal uterine bleeding, hyperplasia was but cancer was not more common in women with endometrial polyps than in those without’. We reached this conclusion because cancer was not more frequent among women with polyps compared with women without (2/62 vs. 6/186 or 3.2% vs. 3.2%). This relationship remains essentially unchanged in the revised data as the frequency of cancer among women with polyps compared with women without is similar (3/62 vs. 6/186 or 4.8% vs. 3.2%). Our conclusion about hyperplasia was reached because it was more frequent in women with polyps than in women without (7/62 vs. 8/186 or 11.3% vs. 4.3%). The result following computerization of the data shows that hyperplasia remains more frequent in women with polyps than in women without (6/62 vs. 9/186 or 9.7% vs. 4.8%). Therefore our conclusion remains unchanged. Our conclusion may, however, be challenged on the basis of variation in P-values associated with the revised data set shown in corrected Table I. We would strongly rebut this challenge on the basis that the significance level used in hypothesis testing is based on an arbitrary dichotomy and it is not as informative as the confidence interval (2). Consequently, it may be misleading in observational studies such as ours. In fact, Sterne et al. (3) have recommended that p-values should not be used in observational studies and 90% confidence intervals around odds ratios should be readily accepted. Based on our originally reported figures in Table I (11.3% vs. 4.3%), the odds ratio was 2.8 (90% confidence interval 1.1–6.9). This result was significant at the borderline of the traditional cut-off value of p < 0.05. Based on our revised data (9.7% vs. 4.8%), the new odds ratio is 2.1 (90% confidence interval 0.85–5.2). This result is not significant at the traditional cut-off p-value. However, it shows a significant trend towards an association between hyperplasia and polyps. Furthermore, the true values are more likely to lie near the center of the interval than at the extremes, i.e. the true point estimate is more likely to be > 1.0. Thus, we believe that the evidence is suggestive of a clinically significant relationship that merits consideration. This trend has been confirmed in an independent data set collected in our clinic subsequent to our paper involving more than 790 women. Hyperplasia was found more frequent in women with polyps than in those without (14/95 vs. 38/695 or 14.73% vs. 5.46%, p = 0.0006, odds ratio = 2.98, 90% CI = 1.72–5.17) while there was no significant difference with regard cancer (1/95 vs. 19/695 or 1.05% vs. 2.73%, p = 0.33, odds ratio = 0.37, 90% CI = 0.07–2.06). With regard to our original Table II, we concluded that compared with no estrogenic exposure, HRT and tamoxifen exposure increased the odds of having polyps with hyperplasia or cancer but this result was not statistically significant (odds ratio 2.6, 90% confidence = 0.68–9.18, p = 0.28). In light of the corrections now identified and shown in corrected Table II, this conclusion remains unchanged as the corrected figures revealed an odds ratio of 1.6 (90% confidence intervals = 0.48–5.5, p = 0.5). Thus, these minor changes did not affect our original conclusion. In the interest of scientific integrity we have brought the errors identified subsequent to publication through this letter. In particular, when systematic reviews are carried out in the future the provision of correct data would enhance the value of such secondary analyzes. We do not believe that the original work (1) falls into the category of retracted articles (4) and it could continue to be cited as valid work in the biomedical literature along with the discussion highlighted in this letter.