A new Chinese meta-analysis compiles old data, bad data to try to minimize Abortion and Breast Cancer link
By Joel Brind, Ph.D.
It took just over a year—after a new meta-analysis of Chinese
abortion-breast cancer link (ABC link) studies corroborated the link
between having an induced abortion and a heightened risk of breast
cancer—for a new Chinese meta-analysis to attempt to throw cold water on
the link. In this instance, it’s a study by Dr. Jun Guo and colleagues
at university hospitals in Hubei and Beijing, China.
NRL News readers may recall my piece at the end of 2013, wherein I reported on the new meta-analysis by Dr. Yubei Huang, et al. of the Tianjin Medical University in China. (A meta-analysis is a study that pools the results of many studies, thus increasing the overall statistical power to find a significant result.)
Huang reported an overall 44% increase in breast cancer risk associated with induced abortion, even greater than the overall 30% increase I and my colleagues had reported in our 1996 meta-analysis. The study also showed a “dose” effect (more abortions resulting in even greater risk than does one abortion).
But what gave the Huang study even more clout was that it provided a compelling “meta-regression analysis” which demonstrated why certain studies in China (Shanghai, specifically) had failed to show the ABC link: abortion was so common in these populations the link was masked by the inability to find an appropriate comparison group of unaborted women. Moreover, in explaining this last phenomenon, Huang et al. quoted extensively from my 2004 published letter on the ABC link in China.
I had wondered just what form the backlash would take in the peer-reviewed medical literature—the prestigious journal Cancer Causes and Control in the present case. This is the journal in which both the 2013 Huang meta-analysis and the new 2015 meta-analysis by Dr. Guo and his colleagues appeared.
Interestingly, these were completely different groups of researchers, except for one: Dr. Hubei Huang—first author of the 2013 meta-analysis—also appeared on the by-line of the Guo study.
Most prominent in the Guo study is its conclusion: “The current prospective evidences are not sufficient to support the positive association between abortion (including induced abortion and spontaneous abortion) and breast cancer risk.”
So what new facts have the Guo group compiled to justify such exculpatory conclusions about abortion? None, essentially. The Guo meta-analysis is basically a rehashing of data from earlier worldwide prospective ABC link studies published between 2000 and 2008, sandwiched between two invalid Danish studies, one published in 1997 and one in 2013.
If there is nothing new, what justifies the authors’ ignoring the tsunami of strong ABC link data that has been flooding in from South Asia and elsewhere since 2008? It is the attempt yet again to make the case that only a certain kind of study “counts.”
As we have explained at NRL News Today, there are two kinds of studies. One is retrospective in design—i.e., based on data gathered at interview and/or on questionnaires from breast cancer patients v. healthy women. By contrast prospective data-based studies gather data before anyone in the study has been diagnosed with breast cancer. (Guo’s meta-analysis relies entirely on the latter.)
Retrospective data-based studies, it is argued, are unreliable due to reporting bias or response bias, terms with which NRL News Today readers may be familiar, owing their repeated use as justification to dismiss (without good reason) the overwhelming evidence of the ABC link for the last quarter century.
The response bias argument goes like this. There is a general tendency for women to underreport abortions at study interviews and/or on study questionnaires. But, it is theorized, unlike these healthy women, it is different with women who have been diagnosed with breast cancer. They are more likely to come clean and admit all their abortions.
Consequently, in a typical, retrospective case-control study, the case population (women who’ve had breast cancer) may appear to include more women who’ve had abortions, compared to the control population (women without breast cancer). This greater apparent frequency of abortion among cases produces a false finding of increased risk associated with abortion. Or so the argument goes.
But the trouble is, no valid evidence has ever been produced that there is any response bias in any ABC research –and plenty of evidence against it.
Oh yes, there was a study which claimed to show direct evidence of such response bias (a 1991 Swedish study). But it was actually based on the dubious notion that women with breast cancer actually make up abortions that did not happen (they called it “overreporting of abortions”). Why? Because the abortions did not appear on the computerized database of medical records, which database was presumed to be flawless and complete.
Never mind that the idea of “overreporting” was actually retracted in 1998 in the British Medical Association’s epidemiology journal. Nonetheless the response bias canard has been repeated often enough by respected medical research authorities to have established it as fact.
So that is the supposed advantage of all the studies compiled in Guo’s latest meta-analysis: They are based on prospective data only, so they are not subject to the possibility of response bias.
But there are troubles with the prospective studies that Guo et al. reviewed. They were methodologically so seriously flawed as to be invalid. I have published rebuttals of most of them—six of which were cited by Guo et al.
So how do the authors get around the flaws I had documented in the studies they compiled? They don’t really.
Rather, they just bury the key caveat to their conclusions deep in their discussion section: “Thus, our result of non-significant positive association between abortion and breast cancer based on cohort (i.e., prospective) studies could be expected, especially when most early cohort studies suffered from potential methodological flaws as argued by Brind and colleagues (6 articles cited).”
But if these studies Guo et al. included in their meta-analysis were so flawed, why were they not excluded? That is a very good question, especially for the Danish studies that bookend the Guo collection.
The first such study is the 1997 study by Melbye, et al. which found no association between abortion and breast cancer. It was so large—1.5 million women—that because it found no increase in risk its inclusion into just about any meta-analysis overpowers any finding of increased risk
But NRL News Today readers may recall that the Melbye study was the worst of the worst. For example, 60,000 women who’d had an abortion were misclassified as not having had any abortions.
That fact alone renders their conclusion of no association invalid. And to the discredit of Guo et al, they should have excluded the Melbye study from their analysis under their own exclusion criterion of having “incomplete data of interest.”
The 2013 bookend is a Danish study by Bräuner et al., which looked at breast cancer incidence among older women, starting at age 50, who had been cancer-free until that age. They were then followed up for 12 years for the development of breast cancer.
Brauner et al. found no evidence of an ABC link, but the breast cancer patients in her study who’d had any abortions had had their abortion an average of 28 years before cancer diagnosis. So while the Bräuner study is not invalid, it is a very limited finding. It really only applies to women who have survived cancer-free for decades after abortion.
But Bräuner et al. leveraged this limited finding to the broad general conclusion: “Our study did not show evidence of an association between induced abortion and breast cancer risk.” (I had also published a critical letter re: the Bräuner study, which was notably not cited by Guo et al.)
Notice that the Bräuner conclusion is rather similar to the overall Guo conclusion. What’s going on in both cases is the reporting of flawed, limited data which doesn’t really show anything, but highlighting the conclusion as if the search for evidence were valid and thorough.
But it is enough to throw cold water on the growing public perception that there really is an ABC link. Just don’t drink the water: it isn’t clean.
Editor’s note. Joel Brind, Ph.D. is a Professor of Human Biology and Endocrinology at Baruch College, City University of New York; Co-founder of the Breast Cancer Prevention Institute, Somerville, NJ; and a frequent contributor to NRL News and NRL News Today. As he does each year, Dr. Brind will speak at the annual NRLC convention in New Orleans.
Joel Brind, Ph.D.
NRL News readers may recall my piece at the end of 2013, wherein I reported on the new meta-analysis by Dr. Yubei Huang, et al. of the Tianjin Medical University in China. (A meta-analysis is a study that pools the results of many studies, thus increasing the overall statistical power to find a significant result.)
Huang reported an overall 44% increase in breast cancer risk associated with induced abortion, even greater than the overall 30% increase I and my colleagues had reported in our 1996 meta-analysis. The study also showed a “dose” effect (more abortions resulting in even greater risk than does one abortion).
But what gave the Huang study even more clout was that it provided a compelling “meta-regression analysis” which demonstrated why certain studies in China (Shanghai, specifically) had failed to show the ABC link: abortion was so common in these populations the link was masked by the inability to find an appropriate comparison group of unaborted women. Moreover, in explaining this last phenomenon, Huang et al. quoted extensively from my 2004 published letter on the ABC link in China.
I had wondered just what form the backlash would take in the peer-reviewed medical literature—the prestigious journal Cancer Causes and Control in the present case. This is the journal in which both the 2013 Huang meta-analysis and the new 2015 meta-analysis by Dr. Guo and his colleagues appeared.
Interestingly, these were completely different groups of researchers, except for one: Dr. Hubei Huang—first author of the 2013 meta-analysis—also appeared on the by-line of the Guo study.
Most prominent in the Guo study is its conclusion: “The current prospective evidences are not sufficient to support the positive association between abortion (including induced abortion and spontaneous abortion) and breast cancer risk.”
So what new facts have the Guo group compiled to justify such exculpatory conclusions about abortion? None, essentially. The Guo meta-analysis is basically a rehashing of data from earlier worldwide prospective ABC link studies published between 2000 and 2008, sandwiched between two invalid Danish studies, one published in 1997 and one in 2013.
If there is nothing new, what justifies the authors’ ignoring the tsunami of strong ABC link data that has been flooding in from South Asia and elsewhere since 2008? It is the attempt yet again to make the case that only a certain kind of study “counts.”
As we have explained at NRL News Today, there are two kinds of studies. One is retrospective in design—i.e., based on data gathered at interview and/or on questionnaires from breast cancer patients v. healthy women. By contrast prospective data-based studies gather data before anyone in the study has been diagnosed with breast cancer. (Guo’s meta-analysis relies entirely on the latter.)
Retrospective data-based studies, it is argued, are unreliable due to reporting bias or response bias, terms with which NRL News Today readers may be familiar, owing their repeated use as justification to dismiss (without good reason) the overwhelming evidence of the ABC link for the last quarter century.
The response bias argument goes like this. There is a general tendency for women to underreport abortions at study interviews and/or on study questionnaires. But, it is theorized, unlike these healthy women, it is different with women who have been diagnosed with breast cancer. They are more likely to come clean and admit all their abortions.
Consequently, in a typical, retrospective case-control study, the case population (women who’ve had breast cancer) may appear to include more women who’ve had abortions, compared to the control population (women without breast cancer). This greater apparent frequency of abortion among cases produces a false finding of increased risk associated with abortion. Or so the argument goes.
But the trouble is, no valid evidence has ever been produced that there is any response bias in any ABC research –and plenty of evidence against it.
Oh yes, there was a study which claimed to show direct evidence of such response bias (a 1991 Swedish study). But it was actually based on the dubious notion that women with breast cancer actually make up abortions that did not happen (they called it “overreporting of abortions”). Why? Because the abortions did not appear on the computerized database of medical records, which database was presumed to be flawless and complete.
Never mind that the idea of “overreporting” was actually retracted in 1998 in the British Medical Association’s epidemiology journal. Nonetheless the response bias canard has been repeated often enough by respected medical research authorities to have established it as fact.
So that is the supposed advantage of all the studies compiled in Guo’s latest meta-analysis: They are based on prospective data only, so they are not subject to the possibility of response bias.
But there are troubles with the prospective studies that Guo et al. reviewed. They were methodologically so seriously flawed as to be invalid. I have published rebuttals of most of them—six of which were cited by Guo et al.
So how do the authors get around the flaws I had documented in the studies they compiled? They don’t really.
Rather, they just bury the key caveat to their conclusions deep in their discussion section: “Thus, our result of non-significant positive association between abortion and breast cancer based on cohort (i.e., prospective) studies could be expected, especially when most early cohort studies suffered from potential methodological flaws as argued by Brind and colleagues (6 articles cited).”
But if these studies Guo et al. included in their meta-analysis were so flawed, why were they not excluded? That is a very good question, especially for the Danish studies that bookend the Guo collection.
The first such study is the 1997 study by Melbye, et al. which found no association between abortion and breast cancer. It was so large—1.5 million women—that because it found no increase in risk its inclusion into just about any meta-analysis overpowers any finding of increased risk
But NRL News Today readers may recall that the Melbye study was the worst of the worst. For example, 60,000 women who’d had an abortion were misclassified as not having had any abortions.
That fact alone renders their conclusion of no association invalid. And to the discredit of Guo et al, they should have excluded the Melbye study from their analysis under their own exclusion criterion of having “incomplete data of interest.”
The 2013 bookend is a Danish study by Bräuner et al., which looked at breast cancer incidence among older women, starting at age 50, who had been cancer-free until that age. They were then followed up for 12 years for the development of breast cancer.
Brauner et al. found no evidence of an ABC link, but the breast cancer patients in her study who’d had any abortions had had their abortion an average of 28 years before cancer diagnosis. So while the Bräuner study is not invalid, it is a very limited finding. It really only applies to women who have survived cancer-free for decades after abortion.
But Bräuner et al. leveraged this limited finding to the broad general conclusion: “Our study did not show evidence of an association between induced abortion and breast cancer risk.” (I had also published a critical letter re: the Bräuner study, which was notably not cited by Guo et al.)
Notice that the Bräuner conclusion is rather similar to the overall Guo conclusion. What’s going on in both cases is the reporting of flawed, limited data which doesn’t really show anything, but highlighting the conclusion as if the search for evidence were valid and thorough.
But it is enough to throw cold water on the growing public perception that there really is an ABC link. Just don’t drink the water: it isn’t clean.
Editor’s note. Joel Brind, Ph.D. is a Professor of Human Biology and Endocrinology at Baruch College, City University of New York; Co-founder of the Breast Cancer Prevention Institute, Somerville, NJ; and a frequent contributor to NRL News and NRL News Today. As he does each year, Dr. Brind will speak at the annual NRLC convention in New Orleans.
Posts
No comments:
Post a Comment