Lessons in confounding epidemiology: Household cleaning products, the microbiome and childhood obesity

At first, you wouldn’t think that cleaning products, the microbiome of a young infant, and childhood obesity had much to do with each other. Of course, anything related to the microbiome is likely generate interest and news outlets will happily pick up any new study about it. A new study made quite a splash here in Canada, although I suspect it probably made less of an impact in the United States. The article, published in the Canadian Medical Association Journal, found that exposure to household disinfectants was associated with higher BMI in young children, possibly because it was altering their gut microbiome. There was also the suggestion that eco-friendly cleaning products protected against obesity.

As you can imagine, such an association certainly made for catchy headlines. The idea clearly builds on the concept of the hygiene hypothesis. For those who are not aware of it, the hygiene hypothesis is the idea that the introduction of antibiotics, and our cleaner environment overall, has contributed to rising rates of allergies and asthma. While there is some truth to the idea, the reality is more complex. What is often overlooked in this discussion is the fact allergic diseases started becoming more common in the later part of the 19^th century, long before the development of antibiotics, and probably had more to do with people moving off of farms and into cities. Still it is true that avoiding potential allergens, like peanuts, has been shown to be counterintuitively counterproductive. In fact introducing peanuts early into the diet of high-risk infants can actually reduce the risk of developing a peanut allergy. We probably are better off being exposed to potential allergens early so that our immune system can better adapt to their presence.

So while not terribly likely, it is not impossible that antibiotics or disinfectants could have something to do with obesity by disrupting the balance our intestinal microbial flora. It is at least an idea worth studying. And while much has been written about the microbiome, it is important not to let our enthusiasm outstrip the science. Probiotics do have some uses, but they are not panaceas. But rather than getting into the broad concept of the hygiene hypothesis and the microbiome, which so many others have already written about, I wanted to look at this one specific study and explore a concept that is often brushed over very casually; the concept of confounding.

Confounding

Most people are aware of confounding in a general sense. It is the idea that a 3^rd variable is affecting the interaction between an exposure and an outcome.

Here, the causal relationships are indicated by the solid arrow, and the confounding relationship is indicated by the dashed arrow.

For example, people who carry lighters are at higher risk for developing lung cancer. This is true, as far as it goes, but it’s only true because people who carry lighters are also more likely to be smokers, and smokers are likely to get lung cancer. Thus, smoking confounds the association between carrying a lighter and developing lung cancer, i.e. it makes it seem as if lighters are related to lung cancer, but rather it is smoking that leads to both carrying a lighter and lung cancer.

There are numerous other examples in the literature. Hormone replacement therapy (HRT) was thought to be protective for cardiac disease in women until the Women’s Health Initiative randomized trial showed that it was not. Likely the many observational trials on HRT simply reflected that the women taking hormone replacement therapy were more health conscious and therefore had a lower risk of heart disease at baseline.

The study

So let’s examine this new study in detail. This study was part of a larger cohort study called the Canadian Healthy Infant Longitudinal Development (CHILD) study, which is being used to examine a myriad of conditions. Women were recruited during their second or third trimester of pregnancy. At 3 months post-partum they filled out a questionnaire, which you can see here, about which types of cleaning products they used in their homes. Researchers grouped the answers into three categories: disinfectant, detergent and other. They then gave them scores for whether they used these products not at all (0 points), less than once a month (1 point), monthly (2 points), weekly (3 points) and daily (4 points). They then summed these scores to get an overall point score for each study subject.

At about 3 months of age, they also got a fecal sample from their babies to test their gut bacteria. At 1 year and 3 years of age, they assessed the children again to measure their height and weight.

So to summarize the flow of the study was:

So what are the problems with this study? First, when you ask people to fill out a questionnaire, there is no guarantee that they will fill them out accurately. Apparently, researchers sent their research assistants into some homes to check whether the cleaning products that were reported in the questionnaires were actually present. They claim that there was high correlation between the self-reported questionnaire and the visual assessment of the research assistants, and cite a correlation coefficient r=0.32 (where zero means no correlation and 1 means perfect correlation). My experience has been that a correlation coefficient of 0.32 is not that great, and I wonder how accurately these questionnaires reflect the daily reality in the subject’s homes.

But setting that aside, this was not a randomized trial. Therefore we can easily imagine that the parents who used cleaning products regularly were probably different in some ways from the people who used them less regularly or not at all. In fact, the use of disinfectants was higher in one of the cities participating in the study (Edmonton) than it was in the other study sites. Also, disinfectant use was higher in households where infants were delivered by cesarean section, who received antibiotics during the delivery, and who were exposed to cigarette smoke in the home. Disinfectant use was lower in households were infants were breast-fed. In contrast, the eco-friendly products were used more often by mothers who breastfed their infants, and less often by women who were overweight pre-pregnancy.

So to summarize:

After analyzing the stool samples, researchers found that disinfectant use affected the gut microbiome, but detergents, eco friendly products, and other cleaning products did not. They principally saw an increase in Lachnospiraceae although there were differences in a few other bacterial species.

When it came to obesity outcomes at 3 years (the 1 year outcomes were not reported), children were more likely to be classified as overweight if their mothers were overweight before pregnancy, if they had a cesarean delivery, received antibiotics during delivery, were exposed to tobacco smoke in the home, and were not breast-fed.

So to summarize again:

At this point, it should be clear what is going on. Households with more disinfectant use were more likely to have overweight children at age 3 years. But households with more disinfectant use were also more likely to have a slew of other factors that were likely contributing children’s weight gain and the changes in their microbiome.

Remember that stool samples were taken at 3 months of age. At this point, before children start eating solid food, one can assume that a child’s microbiome will be determined by a few important variables like whether or not they had a vaginal birth, whether or not they were breast fed, and whether or not they received antibiotics in their as-yet short life.

The interpretation

So one interpretation of these results is that disinfectant cleaning products lead to weight gain and that eco-friendly products protect against weight gain by making changes to gut bacteria. But another interpretation is that changes in the microbiome were being caused by other factors that happened to be associated with some cleaning products.

There is no doubt that our environment plays a role in the rising rates of obesity. But I suspect much of that increase can be explained by our diet and our increasingly sedentary lifestyle. I am not certain if our cleaning products can explain that much of the current obesity problem in our society. I suspect that much of the effect size seen in this study can be explained by confounding, not causality.

With any non-randomized study confounding is a problem. It is incumbent on researchers to try to identify the variables that are tied to both exposure and outcome. And if researchers don’t do it, it’s up to you the reader to ask, “What else could affect this association?” We need to ask these questions because if we didn’t, we’d still be routinely giving women hormone replacement therapy.