Vol. 05 Issue 2, Spring 2000
Can you provide some historical perspective on the Nurses' Health Study?
Our study population began with 121,700 nurses over twenty years ago, and we are still getting a 90% response rate to our surveys. Over time, as the population has aged, there are more cases of breast cancer and other cancers. More numbers of cases gives us the ability to both refine our estimates for relations between lifestyle, environment and cancer, and also allows us to look at subsets of cases. For example in the case of breast cancer, we can start to look at those that are estrogen-receptor positive or negative. The other important piece for a long term study like this is the fact that we can, with repeated measures, start to look at people who change their behaviors over time, and determine whether that modifies their subsequent risk. One of the unique things about Nurses' Health is that we are going back to participants and updating exposure information. That is something that clearly isn't done in a retrospective study where you would normally just be getting one set of exposures measures; some of the other cohort studies have historically just had a baseline questionnaire, and looked at the risk of cancer over the next ten years.
What strengths lie in the cohort study design and what weaknesses?
Clearly the fundamental strength of the cohort or prospective design is that information is collected from participants when they are free from disease. In that setting you avoid the potential for bias in terms of recall of exposure, be it diet or activity or other factors, because you are getting the information from participants when they are free from disease and following them forward.
You could argue that for some outcomes there may be different diagnostic criteria or different attention to looking for the disease in a population like the cohort of nurses, but for a cancer outcome we are fairly safe. If you were doing a case-control study, you would be able to more rigorously apply diagnostic criteria, but in a study for cancer there is little debate for what is cancer and what is not. At the next level one could ask: are all the women having routine screening, or, are the women with bad diets less likely to be screened, and would that lead to some detection bias? A strength of the Nurses' Health Study in fact has been that because of prior medical training of the women in nursing, they have taken up healthy practices like mammography more rapidly than the population in general. That cuts down on the potential of detection bias. They have also more rapidly given up smoking than women in general. And while some say that this may cause a problem, I think in fact that it is helpful, because it lets us look at the benefits of quitting, which we've done. We're less dominated in our outcomes by smoking-related outcomes. Breast cancer and colon cancer are easier to look at in this setting.
In terms of limitations, like most of the cohort studies we have an underrepresentation of African Americans and other racial/ethnic sub-groups. It's a limitation of most all epidemiological studies. The Nurses' Health Study population is 96% white, and that reflects women who were being trained as registered nurses back in the '50s and '60s.
Who are the nurses in Nurses II and why the emphasis on nurses in general?
Nurses II is a younger cohort, comprised of women who were born between '46 and '64. Again these are registered nurses, and were chosen due to the quality of information we were getting in Nurses I. Part of the efficiency of a study like this is having accurate information, and our sense in setting up the study was that medical training would be advantageous. In addition to recording diets, if we were interested in hormones and other drug exposures the thought was that women who were trained in nursing would clearly pay attention to recording that. Also in terms of reporting their medical conditions, again the thought back in the mid-'70s was that women who were trained as nurses would be better informed about their health conditions than the average women. In the last 30 years or so that has changed, and the public in general is probably now more aware of their diagnoses than was the case in the mid-'70s when the design of this study was being finalized. When you look at media coverage and the way that the doctor-patient relationship has changed over time, there is just far more information out there. Think of "Our Bodies, Our Selves;" the fact that that was written tells you that things were not the same in the past as they are now.
Our knowledge of several possible risk factors in breast cancer development has been greatly strengthened by the Nurses' Health Study. What do you feel are the strongest conclusions? For breast cancer in particular, what do you still hope to find out? Are there areas that haven't been touched upon?
Alcohol and the risk of breast cancer, obesity and postmenopausal breast cancer, and use of postmenopausal hormones and postmenopausal breast cancer, would probably be the ones that are the strongest relationships. The lack of association between past use of oral contraceptives and risk of breast cancer is another clear thing that we documented. Documenting where there isn't a long-term health effect is also very important to have on the table.
We are still looking at components of the diet such as folate and glycemic load, and the patterns of postmenopausal hormones. That is an area that has changed fairly dramatically in the past ten years - the shift toward women with a uterus taking estrogen plus progesterone. In the '80s, whereas that might have been recommended, most women who were taking hormones were not doing it. Now it is clear that most are.
We also have a new effort underway to look at diet after diagnosis of breast cancer and how that might modify survival. We can talk to women about diet before diagnosis but no one really knows whether they are going to get breast cancer or not. But all the patients here who've got breast cancer come in and ask, "what can I do, what can I eat that will help me survive?" and no one has studied it. We've got nothing to tell them.
In Nurses II, with the younger population, we are recording more details about patterns of lactation, in order to have a more rigorous evaluation of breastfeeding and risk of breast cancer. The level of detail will be all the way to asking when solids were introduced to the infant, since that changes the demand on the body and hormonal exposures for the lactating mother. We hope to be able to refine our understanding of the relations; across many studies there is a strong suggestion that longer lactation leads to lower risk of breast cancer. The first generation of nurses actually didn't do a lot of breastfeeding. Population trends drive in part what we can and cannot look at in terms of breast cancer risk factors.
Nurses II also has the 10 to 14 year old children of the nurses participating in a four-year follow-up looking at diet, physical activity, weight gain, weight control practices, uptake of cigarette smoking, etc. We are pursuing a range of hypotheses there. It won't be too long until this population moves into the period of having benign breast biopsies and that in itself will be informative.
Considering the evidence that breast cancer risk is likely greatly influenced by exposures and conditions early in life, possibly even before birth, can you comment on the strength of some of these associations and the ways that Nurses' Health is contributing to our knowledge in this area?
We've got lots of hypotheses and because of the way we have conducted our research to date, there isn't a whole lot of direct evidence, most of it is indirect. For example, we know that age at menarche is important, and we know that age at first pregnancy is important, so it's not a great leap of faith to say that things happening between menarche and first pregnancy might also be important. We know that around the world there are many studies that show that height is related to breast cancer and other cancer risk, but not many people have studied the determinants of height. If you look generation by generation, we are on average taller than our grandparents; what else has happened? The number of childhood infections we got was much lower than our grandparents, and our diets have changed. But no one has really dissected what part of diet might be related either to the height change or adult cancer risk change.
With regard to in utero exposures, Dmitrios Tricholpoulos proposed a while back that hormone levels that one is exposed to in utero might vary according to the mother's age, weight or weight gain during pregnancy. And, given that the breast is sensitive to hormones, maybe there is then genetic damage being done even in utero that is a consequence of exposure to estrogens during that early growth phase. Trying to take that hypothesis and test it has been slow. We've been able to ask women how heavy they were at birth, which they usually have to go ask their own mothers (although it seems that women are more likely to know how heavy they were when they were born than men are). At some level there is a fair excitement about the potential role of exposures from conception up to first pregnancy as being important, but the actual amount of information out there is still exceedingly small. It would be wonderful if we could find that some component of diet, be it folate or fiber or something else, during that period is protective against cancer. We can have some impact on what kids are eating in the home, perhaps more easily than a population-wide change by the time that they are, for example 25-50. And, the fact that there are school lunch programs means that if one could get beyond the politics of menu-setting, there is that potential of change outside the home, too.
Some of the reported results of the study with regard to breast cancer have sparked discussion about the study's ability to capture "real" breast cancer risk factors; for example, whether or not there is a connection between dietary fat intake and breast cancer risk. Debated questions here include whether the "low-fat" levels studied in Nurses' Health were low enough to make a difference, and the closely related issue of the significance of animal-based foods in the diet. Can you comment on the dietary fat issue?
When we came up with our first non-finding of the hypothesis testing for a relationship between dietary fat intake and breast cancer risk, people questioned why the finding was null. The fact that we are not now the only ones failing to find a prospective relation - there are eight other studies - to me is some affirmation of our finding. The level of commitment to the hypothesis - not wanting to accept that there is no relation - is somewhat surprising. There is lots and lots of debate that has helped people think through ways of analyzing this issue, but I think that the reality is that adult fat intake is probably not a major issue in terms of breast cancer. One of the notions we have is that the very low fat diet in the US is typically high in carbohydrate and in the US that tends toward simple sugars, and maybe something like the glycemic load is in fact having an adverse effect. There is a counter-balancing, if you will. I think we can look at some of those things, but the bottom line still comes back to the low-fat diet as it's available to most people in the US isn't the solution that a plant-based diet might be. At the moment that's not the way that people are getting to the low fat.
Another example is in the area of organochlorines in blood samples of participants; Nurses' Health found no significant differences in blood levels of chemicals tested between women with and without breast cancer. Questions here include whether the blood is the appropriate site to test this question and whether timing of exposure is critical. Is Nurses' Health pursuing the questions of exposure to environmental chemicals further?
The findings come from nurses' blood collected in 1989 and values there do not predict subsequent risk. I think part of the problem in studying this has to do with which environmental contaminant we are dealing with; some of them are across the board going down and for some there are still pockets in the country where there may be some substantial exposure. But the sense we have is that again, this is not an area that has been as rewarding at explaining breast cancer as thought when the initial hypothesis was set forth. We have another updated analysis that is under review at the moment that is basically showing the same results. That is as far as we have gone with breast cancer. But yes, because we collected it in 1989, we have blood levels in middle-aged and older women, and again, one might ask, is it your exposures in adolescence that matters? These are issues that, across the board, are very hard to disentangle.
What do results mean to the individual woman? We are learning more about genetic predisposition to breast cancer. Is genetic information collected in this study? How can knowledge gained from Nurses' Health contribute to useful information for women of varied genetic backgrounds?
Epidemiology estimates population average risks and is very good at predicting the future burden of disease in the population. If we know that 30% of the population is currently smoking cigarettes, and their age, we can predict how many cases of lung cancer there will be, how many heart attacks there will be. But we are very poor at predicting which of the smokers actually will get lung cancer, which of them will get a heart attack. But we can get very close to the exact right number of cases. Even with breast cancer, I have seen it said and I in fact would agree, that like smoking and lung cancer, we know what causes breast cancer, it's just we cannot predict at the individual level who will get breast cancer. The milieu of female hormones clearly drives breast cancer risk. We can contrast the rates in China and Asia with the US, and when you take into account the different reproductive patterns, you explain more than 50% of the difference between rates in China and rates in the US. What I'm saying comes back to the fact that the results of epidemiologic studies are definitely applicable to the population, be it one thousand women or ten thousand women. What they mean to the one woman in ten thousand trying to interpret them is harder. We say drinking two alcoholic drinks a day increases risk of breast cancer by 50%. But if an individual woman cuts down her alcohol, she hasn't cut down on her risk by 50%. If a thousand women cut down their alcohol then the number of cases of breast cancer in that thousand women will be reduced accordingly. But most of them weren't going to get breast cancer anyway, and most of them still won't get breast cancer when they've stopped drinking. The challenge is how you move from the population level to the individual level and I actually don't think that genetics is going to help there. Like all these dietary exposures, there is going to be a myriad collection of genetic factors that predispose to cancer, apart from the very high risk BRCA1 that affects 5% of breast cancer. Otherwise I actually don't think that all the genetic detail is going to help very much.
One can take epidemiologic findings and say that they have a clear implication for the whole population; for example, if we say low folate in the diet increases risk of colon cancer. Then, if the FDA recommends an increase in folate fortification to the food supply, that is going to lift the folate levels in the whole population, so the population level finding is translated to a population level change. What the translation is for individuals gets to be a whole lot harder unless it's something that we can see everyone changing. I would rather we frame messages like, "everyone should increase their level of physical activity," than "if you walk for half an hour a day you have lowered your risk of disease A, B, or C." The latter unfortunately implies that if you walk for half an hour, you won't get the disease. We're not promising that, we are talking about lowering the chances. We're trying to do a better job talking about risk from the academic end.
Have the ongoing results of this study changed your own thoughts one way or the other with regard to how much an individual can do to actually prevent cancer? How about we as a society; what are the barriers to risk reduction on a large scale in the US?
Both the results and trying to teach about cancer prevention have definitely pushed me to think more and more about the ways we can translate our findings to societal level changes. As we talk about increasing physical activity we start to ask, what are the barriers to people being out there walking? Is it violence on the streets? Is it cars running you over? Why are our towns designed so that you can't go to the store unless you've got a car? Seeing our results and thinking about how they translate back to behavior changes helped me start to think about the level at which we have to think about a change and what the barriers might be.
Think of breastfeeding, and my analogy there would be changes made with regard to smoking. Say breastfeeding has a proven impact on breast cancer risk, with higher breastfeeding lowering risk (and the same applying to ovarian cancer). If we actually applied an OSHA-type regulation to something like breastfeeding we can say that in fact employers have to provide lactation facilities for women upon return to work because the disease burden from not breastfeeding is so high. You can do the arithmetic and show that, but what are the issues? In this country women return to work early after childbirth compared to Europe, where they probably get six to nine months off, which in itself would facilitate longer breastfeeding. When a woman returns to work and wants to continue breastfeeding, she will need facilities to be able to pump. Many employers would consider that a distraction, but in fact the cost effectiveness data show that an infant who is fed breast milk has fewer infections, which means the mother is going to take less time off work to take her kid to the doctor. Changing the workplace to facilitate breastfeeding would be wonderful; we changed the workplace around tobacco exposure because your smoking would influence my health; your breastfeeding is not going to influence my health, so it's harder to get the externalities all lined up. But one needs to ask, if we could change the workplace around the health impact of passive smoking, which is small but detectable, why can't we do the same around breastfeeding?
Dietary change is complicated by all the marketing forces that span so many different levels, be it the junk food advertising that's aimed at kids when they are watching TV, through to the way food is packaged. Then, in the inner cities, access to supermarkets and fresh fruits and vegetables becomes an issue that has a real impact on having population-wide changes. In Boston the mayor has sponsored farmer's markets in every neighborhood during the summer, so maybe that has an impact on fruit and vegetable consumption in the summer. But what happens to the other nine months of the year in Boston, where you can go four or five miles and a couple of bus changes to get to the supermarket? You can get to some of these structural barriers pretty quickly when you start to think about changes in diet. The literature on tobacco uptake is informative. The adult message that is put out by the industry is that everyone makes his or her own informed choice about smoking. But rather it is probably a learned behavior, with a variety of forces and influences, including what is deemed to be culturally appropriate. Our diet follows that same model, with a variety of forces and influences. We must identify them and harness them to promote wellness.