2011 | 2010 | 2009 |2008 | 2007 | 2006 | 2005 | 2004 | 2003

Breastfeeding and leukemia: Another reason to blame moms or to be suspicious of breastfeeding mania?
Rebecca Goldin, PhD and Cindy Merrick, November 11, 2011
While some have called breastfeeding a magic bullet for pediatric maladies, STATS provides an in-depth analysis of the claimed health and financial benefits of nursing.


dinosaurBreastfeeding, which provides infants with many immunological benefits, is being elevated by some to the status of magic bullet for pediatric maladies. To read a 2010 report by Harvard Medical School professors Bartick and Reinhold, one finds oneself wondering what infant blessing the act of breastfeeding doesn’t claim to bestow. While researchers continue to explore whether breastfed children experience lower risk for a host of diseases, such as leukemia, Bartick leapfrogs this process and goes straight to the payoff: the many lives and millions of dollars in treatment for this disease that would be saved solely by the compliance of that contingent of inadequately-breastfeeding moms. In the following analysis, we explore both the claimed benefit and the asserted financial benefit that would be realized if 90 percent of women were to nurse their babies.

Breastfeeding: a “protection” against leukemia? Leukemia, a cancer of the blood and bone marrow, is the most common form of cancer in children, striking about 3,500 young people under the age of 20 in America per year. Though the number seems high, the absolute risk of leukemia is low: about one in 25,000 kids is diagnosed with leukemia yearly. The good news is that while the incidence rate of leukemia remains relatively flat, mortality has dropped significantly in the past 20 years. About three quarters of leukemia diagnoses are Acute Lymphoblastic Leukemia (ALL), for which the 5-year survival rate for children under age five is approximately 90 percent. For the most significant other form of leukemia in children, called Acute Myelogenous Leukemia (AML), the five-year survival rate for children under 15 is about 60 percent.

The causes of leukemia are not well understood. The biggest risk factors include having Down syndrome, having a sibling with leukemia, and having a history of radiation treatment for other cancer. It has been hypothesized that leukemia in some people may be the result of exposure to a specific bacterial or viral source. In theory, passive immunity given by breast milk could have a preventive effect. For this reason, a possible association between breastfeeding and a lower risk of diagnosis of childhood leukemia continues to be explored.

In 2007, the Agency for Healthcare Research and Quality (AHRQ) published a systematic review of current research on the relationships of various infant and childhood maladies with breastfeeding. Their report included “four systematic reviews or meta-analyses that examined the relationship between breastfeeding and childhood leukemia,” chiefly referencing two recent reports which studied the possible association. One of the reports was a literature review by Guise, et al, which was performed in 2005 on behalf of the Office of Women’s Health in the US Department of Health and Human Services. The other was a meta-analysis authored by Kwan, et al, published in Public Health Reports by the National Institutes of Health in 2004. AHRQ pooled the results of their four sources, in making their own recommendations regarding the association between breastfeeding and leukemia.

In aggregating results from multiple sources to generate “new” statistics, researchers must consider the quality of the data from each source, and account for the fact that some sources will have used better data-gathering and reporting techniques than others. For example, if one study controlled for the socio-economic status (making adjustments to the data which “account” for the fact that poorer women are more likely to use formula than wealthier women), but another study did not, a meta-analysis has to find a way to account for this difference when combining the studies. Aggregated data can be very powerful because the more data we have, the less likely we will observe a spurious result that is not present in the whole population. However, if all the studies involved are biased in the same direction, a meta-analysis will not correct that bias. This is one of the reasons that meta-analysis are so useful, yet remain controversial.

In their literature review, Guise, et al, found that of 10 quality-rated studies, six were found to have poor enough quality that they were discounted from further consideration. Two were rated as “fair,” and two as “good.” Of these four, one “good” and one “fair” found a significant protective effect of breastfeeding for over 6 months (up to one year) against diagnosis for ALL, and the other “good” and the other “fair” did not. Similarly, of the two “good” studies, one found a duration effect (lower risk with longer breastfeeding), and the other did not. Guise’s conclusion, overall, was that “there are few high-quality studies to inform an important question for parents as to whether it is possible to reduce the risk of childhood leukemia by breastfeeding, and those few studies disagree.”






Number of studies




There are not many studies with good quality

Protective effects from nursing more than 6 months?


1 yes
1 no

1 yes
1 no, but close to statistical significance

There may be a protective effect, but it is not consistent among high quality studies. The study that did not find an association did find a difference in leukemia rates (favoring nursing) but it did not reach statistical significance.

More protection for longer nursing?



1 yes
1 no

It is not clear that the more you nurse the more benefit you get.

In contrast to the Guise report, the meta-analysis by Kwan, et al, found statistically significant effects from both long-term and short-term breastfeeding against ALL, and from long-term breastfeeding against another form of leukemia, Acute Myelogenous Leukemia (AML). The Kwan meta-analysis has the shortcoming of not having used data quality assessment methods as Guise did, opting rather to be inclusive of a large number of studies without considering the quality of the data.

If the poor quality of the study does not bias the result (the relationship between leukemia incidence and nursing) then a meta-analysis including many poor quality studies could still shed light on the question. Some of the randomness associated with low accuracy would “cancel out” with enough people included. However, if the studies’ poor quality tends to favor a conclusion in one direction, then such a meta-analysis would be biased in the same direction. One such bias is socio-economic status. Kwan et al claim to have found a way to adjust for the confounder of socio-economic-status, which some of the included studies neglected.

The AHRQ report supports Kwan’s method of analysis by noting that “potential for confounding in each study was considered in the meta-analysis,” quietly putting to sleep the question of why the high quality studies did not show the same results.

Guise, et al, disagree with AHRQ’s reasoning, and criticize the Kwan methodology for rather specific reasons. To conclude that the protective effect of breastfeeding is the same for two different types of leukemia is, to Guise, an indicator of bias, as is the fact that there is no duration effect for nursing less than a year. There are additionally other differences among compared groups which are not controlled for in the Kwan meta-analysis, aside from the issue of socio-economic status. In particular, Guise points out that the differences between mothers who breastfeed longer than six months and mothers who breastfeed are manifold: “These groups of mothers also differ significantly by age, prepregnancy BMI, income, education, race, and gravidity [the number of times a woman has been pregnant]. Risk for ALL may be associated with [more than] 1 of these factors, and the effect may not be removed entirely by adjusting for SES [socio-economic status].”

In the final analysis, AHRQ concludes that, minimally, there is a .80 odds ratio in favor of breastfeeding for six to 12 months against a diagnosis of ALL. This is a “middle ground” between Kwan’s claim that the effect is stronger, and Guise’s statement that the conclusions are, at best, “conflicting.”

So far, the debate about the relationship between nursing and leukemia includes very few high quality studies, with weak results. Since leukemia is so rare, the only reasonable study design is case-control, but such studies fall victim to certain kinds of bias, like recall bias. The best we can do is turn to large databases, which increasingly have detailed information on the disease, as well as surrounding factors, for the child who has developed leukemia (as well as for the controls, who have not).

The Cost of Leukemia If breastfeeding were to reduce leukemia rates, however, it may be good public health policy to promote breastfeeding, regardless of how the benefits compare to other possible actions to reduce risk. Perhaps this is a motivating force being the Bartick et al 2010 missive about the costs incurred by “inadequate breastfeeding.” The authors assume a causal relationship with scientific certitude, and attempt to put a dollar value to it all.

The authors compute the excessive per patient expenditures, as well as what they consider to be preventable deaths, attributable to women not breastfeeding, citing the report described above by the AHRQ as their source for the protective effect of breastfeeding. The AHRQ report results in two categories: short-term breastfeeding duration, which they termed “less than or equal to 6 months,” and long-term, “more than 6 months.” The reason is obvious – differences among breastfeeders and non-breastfeeders tended to be greater for those who breastfed. In fact, for short-term duration, AHRQ reported no statistically significant effect for ALL. They reported a long-term effect for ALL, with odds ratio .8, and conceded an association between breastfeeding and AML, without citing an odds ratio.

Slightly misleadingly, Bartick states that, “the AHRQ report noted [an odds ratio] of .8 and .85, respectively, for 6 months of any breastfeeding and the development of acute lymphocytic leukemia (ALL) and acute myelogenous leukemia (AML).” However, AHRQ did not do a meta-analysis for AML. The report instead noted that Kwan maintained such a relationship and that Guise found the AML-breastfeeding link inconclusive.

Perhaps more disingenuous is Bartick et al’s conclusion: They advocate that the medical system could avoid these costs by having women breastfeed exclusively for at least six months. All the evidence she cites, however, concerns some breastfeeding for six months. In other words, even if we believe the most aggressive assertions about the relationship between breastfeeding and leukemia, we should encourage women to maintain some breastfeeding for six months to reduce risk. There is no evidence from any of these analyses that six months of exclusive breastfeeding is better than partial breastfeeding.

In their enthusiasm to promote breastfeeding, Bartick et al also exaggerate the costs, pricing direct costs at $136,444 per child in 2007 dollars. In citing the average cost of treating ALL, they reference a paper by Rahiala, in which the amount $103,250 is given in dollars from the year 2000. Bartick converts this to 2007 dollars by way of a Bureau of Labor Statistics’ (BLS) web page giving Consumer Price Index conversions, somehow along the way inflating the cost given by their own source by at least $12,000 per patient per year. The BLS web page converts $103,250 (2000) dollars into $124,320 (2007) dollars.

On the other hand, if we assume that this relationship exists, what can we say of the cost to society? In other words, once we monetize the cost, we can spread this cost across the population of babies who received ”inadequate” breastfeeding in the first six months of life, and ask how much the medical expenses and monetized deaths “cost” per non-breastfeeding child. It is an absurd exercise, but one that is demanded by the Bartick premise, that monetizing the costs (including death) can make the argument that women should breastfeed.

We took the total dollars lost due to the purported increase in leukemia attributable to insufficient breastfeeding according to the Bartick analysis ($135,408,459) and divided it by the number of children who were inadequately breastfed according to the same analysis. According to Bartick, 87.7 percent of children were not breastfed exclusively for 6 months. The Centers for Disease Control reports that there were 4.14 million children born in 2005, resulting in 3.63 million children who were not breastfed exclusively for six months. This amounts to 135,408,459/3.63 million = $37.30 per child.

If instead we go with the aggressive AHRQ conclusion that the excess costs are due to babies who are not partially breastfed for at least 6 months, then only 57.1 percent, or 2.36 million children, were inadequately breastfed, resulting in a cost of 135,408,459/2.36 million which equals $57.35 per child. Finally, if we place the costs squarely on parents who never breastfeed, these costs should be divided among the 25.9 percent of children who are never breastfed, or 1.07 million children. In this case, the cost is 135,408,459/1.07 million or $126.28 per child.

These costs are less than the cost of one month’s supply of formula and diapers. When Bartick’s argument is brought back to the level of reality – of each individual woman’s choice to breastfeed or not – the idea that the monetary cost of risk due to leukemia essentially loses meaning. Monetizing the decision is far less helpful than discussing the weight of the evidence about the health issue. If it were a purely monetary decision, perhaps even more women would feel that the trouble of nursing is not worth the purported savings. Instead, women should make their decision using their feelings and circumstances, along with the scientific evidence regarding the risks.

To give a matter of comparison, this supposed difference in risk is far, far lower than the risk of driving: the risk of death to a child due to a traffic accident not involving alcohol (according data published by the Centers for Disease Control) is about four times as high, though it also includes traffic deaths of babies 0-1 years old, whereas the leukemia data include only children older than 1.

The evidence for a protective effect of breastfeeding against childhood leukemia is still inconclusive. And while there is no harm done – and much good accomplished – by continuing to encourage women to breastfeed their infants for the known positive effects, should we take our encouragement so far that we end up blaming a mother’s breastfeeding decision for a child’s leukemia diagnosis?


Technorati icon View the Technorati Link Cosmos for this entry