Volume 32, Number 6, November/December 2000
Studying the Health Effects of Induced Abortion
A great deal of research has been conducted on the question of possible adverse health effects of induced abortion. Unfortunately, much of it has serious methodological problems or is irrelevant to today's conditions. A careful analysis of the literature established minimum methodological standards for research examining the effect of induced abortion on subsequent pregnancies.1 These criteria eliminated studies that were subject to recall bias, that evaluated outmoded abortion procedures, that commingled spontaneous and induced abortions and that used inappropriate comparison groups.
A digest that appeared in a recent issue of Family Planning Perspectives [After abortion, Danish women's odds of preterm delivery are doubled, 2000, 32(4):200] describes two research studies—one involving the impact of induced abortion on the duration of subsequent pregnancies2 and one on the effect of abortion on low birth weight3—that violate the last of these standards. The researchers compare first births among women who have had an induced abortion with second and subsequent births among women who have never had an abortion. It is well known that first births are more risky than later births, and it is impossible in these studies to distinguish the risk associated with a first birth from the risk resulting from a prior abortion.
In the case of low birth weight, the data presented in the original study suggest that the identified risk may be completely unrelated to abortion. While 5.0% of first births among women with one prior abortion were low birth weight (Table 3 of the article),4 the same was true of 5.2% of the first births among those in the control group (Table 1 of the article).5 The analysts find an elevated risk when they compare women whose first pregnancy was terminated with women having their second or subsequent birth, among whom the rate of low birth weight is 3.5-4.7%.
In fact, women with a prior abortion were more likely to smoke; when this is taken into account, the data suggest that a previous abortion may have been somewhat protective against low birth weight when the first birth occurred within six months of the abortion. However, a multivariate analysis would need to be carried out for this hypothesis to be tested properly.
The analysis focusing on premature and postterm delivery suffers from the same methodological weakness, although the difference between first births and subsequent births is not as great as it was with low birth weight. Even if an appropriate comparison showed increased risks associated with abortion, however, the finding that women who had had an abortion were at heightened risk of either preterm or postterm delivery would be unconvincing because of the absence of controls for other risk factors. For example, the dataset contains no measure of smoking status, although the authors recognize that smoking is correlated with having had an abortion.
What the researchers have shown is that induced abortion does not protect against the risks of delivery as much as a prior birth does. This finding has little practical importance, since few women look to abortion to reduce the risk of their first birth. The way in which the studies were presented, however, would lead most readers to conclude that having an abortion increases the risk of an adverse outcome associated with their first birth. Such misleading reports can cause unnecessary anxiety to patients and can also have political repercussions. (Already, antichoice groups in the United States, pointing to one of these studies,6 have argued that state laws should require physicians to warn patients that an induced abortion would increase the risk of prematurity in future births.) Investigators studying the potential complications or side effects of politically sensitive procedures such as induced abortion have a special duty to be certain that their research approach is free of bias and is presented in a way that is not conducive to misinterpretation.
Stanley K. Henshaw
The Alan Guttmacher Institute
1. Hogue CJR, Cates W and Tietze C, The effects of induced abortion on subsequent reproduction, Epidemiologic Reviews, 1982, 4:66-94.
2. Zhou W, Sorensen HT and Olsen J, Induced abortion and subsequent pregnancy duration, Obstetrics & Gynecology, 1999, 94(6):948-953.
3. Zhou W, Sorensen HT and Olsen J, Induced abortion and low birth weight in the following pregnancy, International Journal of Epidemiology, 2000, 29(1):100-106.
4. Ibid., p. 103.
5. Ibid., p. 102.
6. Zhou W, Sorensen HT and Olsen J, 1999, op. cit. (see reference 2).
The authors reply:
While we controlled for gravidity in our study, not for parity, it is possible that this does not provide the best adjustment for the background risk associated with reproductive history. Whether adjustment for parity will provide an unbiased adjustment is, however, also open to question.1
The association seen between parity or gravidity and reproductive failures may reflect selection rather than risk. Reproduction is being planned and controlled partly using the reproductive history, which unfortunately may destroy the ability to make unbiased observational analyses, regardless of the adjustments made. It is possible, perhaps even likely, that these forces of selection act differently for women who choose an abortion than for women who do not. It was for this reason that we chose to study only the first part of the reproductive career for the cohort members. In light of these limitations, conclusions must be drawn with caution; we stress that our results were seen only for subgroups of the population, and that the overall risk was small and detectable only in a very large study. Most other studies have been too small to detect a risk of this magnitude, including those that we conducted in China.2
In our study, we actually also compared results with the first pregnancy among controls and found similar results for preterm birth but not for low birth weight, as Henshaw has indicated. Although these analyses were made for women of the same parity (nulliparous women), the results may still be subject to bias, since controls were selected only if they had a subsequent pregnancy. Unlike Henshaw, we are not convinced that this comparison is better that the one we make in the tables.
We also performed sensitivity analyses to estimate the possible effect of confounding by smoking. Our results indicate that it is unlikely that smoking explains all of the excess risk seen in the studies.3 A fully adjusted analysis should include data from the entire pregnancy period, on the use of contraceptive methods, about exposure to all confounders and on the desire for a given family size. It is possible that in such an analysis the results may change in either direction.
Danish Epidemiology Science Centre
University of Aarhus
1. Olsen J, Options in making use of pregnancy history in planning and analysing studies of reproductive failure, Journal of Epidemiology and Community Health, 1994, 48(2):171-174.
2. Zhou W, The importance of first trimester induced abortion for subsequent pregnancies: cohort studies in Denmark and Shanghai, China, PhD thesis, Institute of Epidemiology and Social Medicine, Report 23, University of Aarhus, Aarhus, Denmark, 1999.