User:RoyBoy/sandbox/ABC studies

Criticism of media coverage

In an article entitled "Blinded by Science" for the Columbia Journalism Review, Chris Mooney argues that "balanced" coverage by the media of the ABC hypothesis, among other scientific hypotheses championed by "the religious right", is an example of how the scientific fringe manipulates public opinion by insisting on the illusory notion of journalistic "balance" instead of scientific accuracy. Mooney criticizes John Carroll (former Editor-in-Chief of the Los Angeles Times) for a rebuke Carroll made regarding an article written by Scott Gold about the ABC hypothesis for the L.A. Times. Gold's article covered the National Cancer Institute (NCI) workshop, and Carroll notes that when a scientific advocate (Joel Brind) for the ABC hypothesis is found:

“ It is not until the last three paragraphs of the story that we finally surface a professor of biology and endocrinology who believes the abortion/cancer connection is valid. But do we quote him as to why he believes this? No. We quote his political views.

Apparently the scientific argument for the anti-abortion side is so absurd that we don't need to waste our readers' time with it.

”

Carroll's concern is that Gold's article provides fodder to critics who claim that the L.A. Times has a liberal bias. Mooney writes in defense of Gold that:

“ As a general rule, journalists should treat fringe scientific claims with considerable skepticism, and find out what major peer-reviewed papers or assessments have to say about them. Moreover, they should adhere to the principle that the more outlandish or dramatic the claim, the more skepticism it warrants. The Los Angeles Times’s Carroll observes that “every good journalist has a bit of a contrarian in his soul,” but it is precisely this impulse that can lead reporters astray. The fact is, nonscientist journalists can all too easily fall for scientific-sounding claims that they can’t adequately evaluate on their own. ”

Responding to the criticism Carroll reiterated: "You have an obligation to find a scientist, and if the scientist has something to say, then you can subject the scientist’s views to rigorous examination."

Epidemiological studies

The majority of the results in epidemiology are calculated as a relative risk accompanied by a confidence interval. The significance of a reported relative risk can be contentious because of response bias, incomplete data, missed confounding factors, imprecise controls, or flawed statistical analysis.

Confounding factors

There are many confounding factors for breast cancer, increasing the difficulty to establish or reject a causal link to any particular factor. Genetics is a major factor that affects not only a woman's initial breast cancer risk but also her hormonal sensitivity, which in turn affects her susceptibility to a long list of socioeconomic and environmental factors. As Western society has modernized, environmental carcinogens, hormone replacement therapy (HRT), hormonal contraception, early menarche, and obesity have all increased while child rearing has been delayed and rate and duration of breastfeeding have decreased. Further, the number of women who have had an induced abortion has increased in recent decades while average has decreased; having given birth before age 30 is associated with a slight reduction in breast cancer risk. Scientific studies control for such factors using case-control methodology, matching each woman in a study who has had an abortion (case) with one or more very similar women with no abortion history (control).

The controversial nature of abortion may introduce a response bias into interview studies, especially for studies done in decades past when abortion was less socially accepted; the statistical significance of this bias has not been fully determined. In the late 20th century, there was some concern of an increase in breast cancer incidence. This was found to be partly due to longer lifespans and the development of better detection methods capable of finding breast cancer earlier.

Cohorts

Howe

The 1989 study by Holly Howe et al. at the New York State Department of Health examined young women with breast cancer in upstate New York (100/63 ABC cases/controls). The results indicated an increased 1.9 (1.2 – 3.0) relative risk for induced abortion and 1.5 (0.7 – 3.7) for spontaneous abortion. Although the study had 1451 breast cancer cases the number of individuals with an abortion history was low; consequently the confidence interval is quite large.

The authors believed that the study was inconclusive as fertility patterns were changing dramatically as a result of legal abortion and increased use of contraceptives. Further they did not have a complete reproductive history of younger women who may still have children affecting the results going forward, but Howe et al. concluded it raised new questions for continuing research as women's recorded contraceptive histories grew. Newcomb and Michels point out it examined only very young women and did not account for some confounding factors such as family history of breast cancer.

Harris

Another cohort study by Harris et al. (1989) was done looking at 49,000 women who had received abortions before the age of 30 in Sweden (65 ABC cases – compared with estimate of occurrence in the general population). The relative risk for women who'd given birth previous to the abortion was 0.58 (0.38 – 0.84), whereas women with no births had a relative risk of 1.09 (0.71 – 1.56). The confidence intervals did not establish statistically significant associations between breast cancer and different stages of reproduction, including abortion. The study reported an overall relative risk of 0.8 (0.58 – 0.99), making for a 20% reduced risk in comparison to "contemporary Swedish population with due consideration to age."

The study was funded by Family Health International, a pro-choice NGO and although the study started with 49,000 women there were fewer than 5,000 still in the study after 11 years. Harris made no adjustments for family history of breast cancer and the pill, and provides no explanation for a lack of a control group or why the study was limited to women with an abortion before 30 years of age. Brind contends correcting for either of these removes the 20% "protective" effect; and that the study did not account for the difference of nulliparous women in the cohort 41% in comparison to 49% in the general population. Possibly making the protective result about parity (childbearing) rather than abortion.

Melbye

A large, highly regarded ABC study was published by Melbye et al. (1997) of the Statens Serum Institute in Copenhagen, which had 1.5 million Danish women in the study's database (1,338 ABC cases, no controls used). Of those women, 280,965 of them had induced abortions recorded in the computerized registry, which was started in 1973 when having an induced abortion through 12 weeks was legal in Denmark. Although there was an observed increased ABC risk with "increasing gestational age of the fetus at the time of the most recent induced abortion"; the overall relative risk after statistical adjustment came to 1.00 (0.94 – 1.06), meaning zero increase in breast cancer risk. This led to the conclusion that "induced abortions have no overall effect on the risk of breast cancer." The Melbye study's conclusions have been used by many organizations, such as NCI, ACOG, ACS, RCOG and Planned Parenthood, as key scientific evidence of no ABC link.

Brind and Chinchilli had concerns about the Melbye study database as women in the study were born from 1935 to 1978, but the computerized registry of induced abortions only started in 1973. Melbye et al. responded that if the misclassified older women had their risk underestimated, it would be expected that the younger groups would have a higher risk. The statistically adjusted data indicated this was not the case.

However, the statistical adjustments made were another concern of Brind who argues that the Melbye study accidentally adjusted out induced abortion from the overall results. Instead of case-control matching, Melbye el al. decided to manually remove the many confounding factors that increased over time (e.g. smoking, late child bearing, etc.) and were raising breast cancer risk for younger women relative to older women (birth-cohorts). Brind believes finding exactly zero ABC risk was a consequence and red flag indicating ABC risk was removed along with the confounding factors. Melbye et al. found the point to be self-contradictory, considering Brind wanted birth-cohort matching, then argued against "taking birth-cohort differences into account." Brind has stated that he is against the use of just statistical adjustment and that standard case-control matching may more accurately account for birth-cohort differences.

Another letter to the editor from Senghas and Dolan questioned why a statistically significant result for induced abortions done after 18 weeks gestation was not specifically addressed in the results section of the Melbye study abstract. Melbye et al. explained even though they found the result "interesting and in line with the hypothesis of Russo and Russo, the small number of cases of cancer in women in this category of gestational age prompted us not to overstate the finding." The first section of Table 1 in the Melbye study:

Week of gestation	No. of Cancers	Person-Years	Relative Risk (95% CI) *	Multivariate Relative Risk (95% CI) †
<7	36	82 000	0.81 (0.58–1.13)	0.81 (0.58–1.13)
7–8	526	1 012 000	1.01 (0.89–1.14)	1.01 (0.89–1.14)
9–10‡	534	1 118 000	1	1
11–12	205	422 000	1.12 (0.95–1.31)	1.12 (0.95–1.31)
13–14	6	14 000	1.13 (0.50–2.52)	1.13 (0.51–2.53)
15–18	17	35 000	1.24 (0.76–2.01)	1.23 (0.76–2.00)
>18	14	14 000	1.92 (1.13–3.26)	1.89 (1.11–3.22)

* The relative risks were calculated separately for each of the five variables, with adjustment for women's age, calendar period, parity, and age at delivery of a first child. CI denotes confidence interval.
† Values were adjusted for women's age, calendar period, parity, age at delivery of a first child, and the other variables shown in the table.
‡ The women with this characteristic served as the reference group.

Other sections listed age at induced abortion, number of induced abortions, time since induced abortion, and time of induced abortion and live-birth history. There was an indication of a relative risk of 1.29 (0.80–2.08) for 12–19-year olds (relative to 20–24 subcohort), and a protective effect 0.74 (0.41–1.33) for women with an induced abortion before and after their first live birth (relative to induced abortion after 1st live birth subcohort); both results were statistically insignificant.

Michels

A study by Michels et al. (2007) from the Harvard School of Public Health containing 105,716 women (233/1,225 ABC cases/controls) concluded with a relative risk of 1.01 (0.88 – 1.17) "after adjustment for established breast cancer risk factors." Some of the results lead the authors to stipulate: "Although our data are not compatible with any substantial overall relation between induced abortion and breast cancer, we cannot exclude a modest association in subgroups defined by known breast cancer risk factors, timing of abortion, or parity." This modest association was mostly not statistically significant.

Further cohort studies

Several other recent prospective cohort studies have also found little evidence of a link between induced abortion and breast cancer. A study of 267,361 European women (746/2,908 ABC cases/controls), published in 2006, found no significant ABC risk. Another 2006 study involving 267,400 women (872/771 ABC cases/controls) in Shanghai found no evidence of an ABC link. The Shanghai study also noted that women who had an abortion were at a significantly decreased risk of uterine cancer.

Meta-analysis

Beral

In March 2004, Beral et al. published a study in The Lancet as a collaborative reanalysis on Breast cancer and abortion. This meta-analysis of 53 epidemiologic studies of 83,000 women with breast cancer undertaken in 16 countries did not find evidence of a relationship between induced abortion and breast cancer, with a relative risk of 0.93 (0.89 – 0.96). Organizations and media outlets referenced the Beral study as the most comprehensive overview of the ABC evidence.

Brind maintains that like meta-analysis this study is subject to selection bias, which he believes is reflected in the removal of 15 published, peer-reviewed studies with positive ABC results for "unscientific reasons"; and including 28 unpublished studies that outnumber the remaining 24 peer reviewed studies. Beral refers to the Lindefors-Harris response bias study as an explanation for higher ABC risk found in interview based studies, however Brind notes in 1998 that Lindefors-Harris conceded their initial conclusion may have been unsound.

Brind

Brind et al. (1996) conducted a meta-analysis of 23 epidemiologic studies. It calculated that there was on average a relative risk of 1.3 (1.2 – 1.4) increased risk of breast cancer. The meta-analysis was criticized for selection bias by using studies with widely varying results, using different types of studies, not working with the raw data from several studies, and including studies that have possible methodological weaknesses.

The strong reaction to the study particularly in Britain and the United States prompted the editor-in-chief Stuart Donnan to write an editorial. In it he notes:

However, in the light of recent unease about appropriate but open communication of risks associated with oral contraceptive pills, it will surely be agreed that open discussion of risks is vital and must include the people – in this case the women – concerned. I believe that if you take a view (as I do), which is often called 'pro-choice', you need at the same time to have a view which might be called 'pro-information' without excessive paternalistic censorship (or interpretation) of the data.

The Royal College of Obstetricians and Gynaecologists (RCOG) in March 2000 published evidence-based guidelines on women requesting induced abortion. The review of the available evidence at the time was "inconclusive" regarding the ABC link. They also noted "Brind's paper had no methodological shortcomings and could not be disregarded." However, in 2003 the RCOG concluded that there was no link between abortion and breast cancer.

Interviews

Interview (case-control) based studies have been inconsistent on the ABC hypothesis. With the small numbers involved in each individual study and the possibility that recall bias skewed the results, recent focus has switched to meta-analysis and record based studies which are typically much larger. Included are a few interview studies of note.

Daling

Janet Daling from the Fred Hutchinson Cancer Research Center headed two studies on the ABC issue looking at women in Washington state. The 1994 study (845/961 ABC cases/controls) results indicated an associated relative risk of 1.5 (1.2 – 1.9) among women who had given birth before having an abortion. This was reflected in higher risks for women younger than 18 or older than 30 years of age who have had abortions after 8 weeks' gestation. Their conclusion emphasized that although the evidence suggested the possibility of a correlative relationship, their findings were not consistent enough to establish one.

The second larger study Daling conducted in 1996 (1,302/1,180 ABC cases/controls) found that abortion was associated with a relative risk value of 1.2 (1.0 – 1.5). The study also found a significant relative risk of 2.0 (1.2 – 3.3) for nulliparous women with an induced abortion at less than 8 weeks gestation. Daling et al. concluded that:

“ There was no excess risk of breast cancer associated with induced abortion among parous women. These data support the hypothesis that there may be a small increase in the risk of breast cancer related to a history of induced abortion among young women of reproductive age. However, the data from this study and others do not permit a causal interpretation at this time; neither do the collective results of the studies suggest that there is a subgroup of women in whom the relative risk associated with induced abortion is unusually high. ”

Daling et al. examined the possibility of response bias by comparing results from two recent studies on invasive cervical cancer and ovarian cancer. The results argued against significant response bias. However, Rookus (1996) study noted that patients with cervical cancer may report differently than breast cancer patients.

Further interview studies

A 2001 study (1,459/1,556 ABC cases/controls) conducted in Shanghai, China by Sanderson et al. from the University of South Carolina and South Carolina Cancer Center at Columbia concluded that there was no ABC link and that multiple abortions did not put one at greater risk. Since induced abortion is common, legal, and even mandated by the government in China, the recall bias was minimized. Brind has argued that the same factors that make the Chinese study ideal for reducing recall bias also makes them inappropriate for comparison to the West. Specifically, with China’s strict population control, the vast majority of the abortions in the Chinese study were done after the first full-term pregnancy. This differs from North America.

The Istanbul University Medical Faculty published a study in 2009 (742/930 ABC cases/controls) of outpatients from clinics, authored by Ozmen et al. it found a 1.31 (1.13 – 1.53) increased ABC risk. The authors point out various potential biases such as selection, information and even hospital admission bias may have impacted their results. They believe the large pool of patients available to them and the resulting large size of the study "provided reasonably stable risk estimates."

Response bias

Response bias occurs when women intentionally "underreport" their abortion history, meaning that they deny having an abortion or claim to have fewer abortions than they actually had. This can happen because of the personal and controversial nature of abortion, which may cause women to not want to provide full disclosure. Women in control groups are less likely to have serious illnesses, and hence have less motivation to be truthful than those trying to diagnose their problem. When this occurs, it artificially creates an ABC link where none exists. Three major studies have been published examining abortion response bias.

An editorial by Weed and Kramer focused on how Brind's meta-analysis dismissed bias as a factor. The editorial cites the Lindefors-Harris response bias study that used a "registry-based gold standard to show that healthy women consistently and widely underreport their history of abortion." Weed and Kramer considered this compelling evidence there could be systematic bias within the studies included in the meta-analysis. However, subsequently the Lindefors-Harris conclusion was quietly retracted in 1998. Weed and Kramer believed a causal conclusion was a "leap beyond the bounds of inference" and concluded:

Because bias impedes our vision and is subject to sound inquiry, we are far from reaching a scientific "limit". Indeed, after this excursion into the issue of abortion, bias, and breast cancer, it seems our future has as much to do with human behavior as with human biology.

A review of ABC studies was conducted by Bartholomew in 1998. It concluded that if studies least susceptible to response bias are considered, they suggest there is no association between abortion and breast cancer. Chris Kahlenborn, M.D., a pro-life researcher and specialist in internal medicine, observes in his book Breast Cancer: Its Link to Abortion and the Birth Control Pill that if report bias were a significant factor in interview-based studies, then:

... thousands of other studies in medicine might now be deemed 'worthless.' Every time one had a disease or 'effect' that was caused by a controversial risk factor (i.e., one of the causes), the study might be considered invalid based upon 'recall bias.'

Lindefors-Harris

The Lindefors-Harris (1991) study (317/512 ABC cases/controls) was the first major study to examine response and recall bias. It used the data of two independent Swedish induced abortion studies, and concluded there was a 1.5 (1.1 – 2.1) margin of error due to recall bias. However, eight women (seven cases, one control) included in this error margin apparently "overreported" their abortions, meaning the women reported having an abortion that was not reflected in the records. It was decided that for the purposes of the study, these women did not have abortions.

The 1994 Daling study examined the findings on overreporting of the Lindefors-Harris study and found it "reasonable to assume that virtually no women who truly did not have an abortion would claim to have had one." In 1998 the co-authors of the Lindefors-Harris study acknowledged this in a letter:

We are not surprised to find some Swedish women confidentially reporting having had induced abortions during the period 1966–1974 that are not recorded as legally induced abortions. It is plausible that such induced abortions are more susceptible to recall bias than induced abortions performed within the legal context in Sweden.

With the eight alleged overreporting women removed, the error margin was reduced from 50% to 16% which severely limits its statistical significance. Brind believes the remaining 16% could have resulted from the Swedish fertility registry – where women were interviewed as mothers – which could have increased their tendency to underreport, given that a mother might not want to appear unfit.

Rookus

The Rookus (1996) study (918 ABC cases/controls) compared two regions in the Netherlands to assess the effect of religion on ABC results based on interviews. The secular (western) and conservative (southeastern) regions showed ABC relative risks of 1.3 (0.7 – 2.6) and 14.6 (1.8 – 120.0) respectively. Although this was a large variance, Brind et al. pointed out that it was attained with an extremely small sample size of 12 cases and 1 control.

Rookus et al. supported their finding with an analysis of how much recall bias existed with oral contraceptive use that could be verified through records. It corroborated the bias, but Brind's et al. letter argues that it only indicated response bias between the two regions, not between case and control subjects within regions. Rookus et al. responded by noting that there was a 4.5-month underreporting difference between control and case subjects in the conservative Catholic region. This was indirect evidence for a reporting bias since women's comfort levels with reporting oral contraception are theoretically higher than induced abortion. Rookus et al. also acknowledged the weakness in the Lindefors-Harris response bias study, but emphasized that more controls (16/59 = 27.1%) than case patients (5/24 = 20.8%) underreported registered induced abortions. They concluded that asserting a causal ABC link would be a disservice to the public and to epidemiological research when "bias has not been ruled out convincingly."

Tang

A study by Tang et al. (2000) (225/303 ABC cases/controls) done in Washington State found controls were not more reluctant to report induced abortion than women with breast cancer. Their results were that 14.0% of cases and 14.9% controls (a difference of −0.9%) did not accurately report their abortion history. They do note likely underreporting occurring in certain sub-groups of women; such as older women in a Newcomb study reporting abortions prior to legalization, and a predominantly Roman Catholic population in the Rookus study.

Spontaneous abortion

Studies of spontaneous abortions (miscarriages) have generally shown no increase in breast cancer risk, although a study by Paoletti concluded there is a "suggestion of increased risk" 1.2 (0.92 – 1.56) after three or more pregnancy losses. Some argue that this apparent lack of effect of miscarriages on breast cancer risk is evidence against the ABC hypothesis, and pro-choice advocates have claimed it is proof that neither early pregnancy loss nor abortion are risk factors for breast cancer.

One of the problems with comparing miscarriage to abortion is the issue of hormone levels in early pregnancy, a key point because the ABC hypothesis rests on hormonal influence over breast tissue development. While it is true most miscarriages are not caused by low hormones, most miscarriages are characterized by low hormone levels. Kunz & Keller (1976) showed that when progesterone is abnormally low a miscarriage occurs 89% of the time. Advocates of the ABC hypothesis argue that, given the association of most first trimester miscarriages with low hormone levels, spontaneous abortion is not analogous to an induced abortion.

Studies which show a protective effect

Several studies indicate that having an induced abortion prevents breast cancer. Examples include:

1. The meta-analysis by Beral et al., described above, found an odds ratio of 0.93 (0.89 – 0.96), indicating that having an induced abortion was associated with a 7% reduction in the risk of getting breast cancer. In selecting studies for inclusion, Beral et al. removed 15 studies with positive results citing response bias and other concerns.

2. A Scottish record linkage case-control study by Brewster et al., published in the Journal of Epidemiology and Community Health, reported an odds ratio of 0.80 (0.72 to 0.89) for induced abortions, indicating that having an induced abortion was associated with a 20% reduction in the risk of getting breast cancer.

3. The cohort study by Harris et al., described above, found an odds ratio of 0.80 (0.58 – 0.99) finding a 20% reduction in the risk of getting breast cancer. However, it failed to adjust for several confounding factors or explain the lack of a control group.

4. A Serbian study by Ilic et al., published in the journal Tumori, reported an odds ratio of 0.47 (0.25-0.90), indicating that women in this study who had induced abortions were less than half as likely to get breast cancer as women who did not have induced abortions. This study was small (168 cases and 171 controls) and was limited to women with children.