Q&A: Algorithm provides ‘holistic view’ of how air pollution contributes to asthma in kids

Exposure to certain air pollutant mixtures appeared significantly associated with more severe asthma in children, according to results of a machine learning-based analysis published in The Journal of Clinical Investigation.

These air pollutants included acrylic acid, which is used in adhesives, plastics and floor polishers; ethylidene dichloride, a solvent used for plastics, oils and fats that also is used as a degreaser and fumigant; and hydroquinone, which is commonly found in cosmetic and health products.

Air pollutants linked to serious adverse health effects — air toxics — are known to impact asthma development when exposed to children early in life. Many studies have investigated the effects of individual air toxics on children’s asthma, although the impact of multiple air toxics in combination has not been thoroughly studied.

“Like many scientists, we wanted to provide a more comprehensive picture of how air toxics contribute to childhood asthma,” Gaurav Pandey, PhD, assistant professor of genetics and genomic sciences at Icahn School of Medicine at Mount Sinai and a senior study author, said in a Mount Sinai press release. “Traditionally, for technical reasons, it has been difficult to study the health effects of more than one toxic at a time. We overcame this by tapping into the power of machine learning algorithms.”

Pandey and colleagues developed a novel machine learning algorithm, Data-driven ExposurE Profile (DEEP) extraction, to evaluate the impact of early childhood exposure to several air toxics in conjunction. They compared residential pollutant exposure in the New York metropolitan area to the asthma outcomes of 151 children participating in Mount Sinai’s Airway in Asthma study.

Findings showed that pollutants such as trimethylamine, which is used in waterproofing, may impact asthma severity alone; whereas others, such as acrylic acid, increased severity alone and in air toxic combinations. Some pollutants, such as toluene and cobalt compounds, impacted asthma severity only in air toxic mixtures.

“Our results show how breathing individual and combinations of pollutants may lead to poor asthma outcomes,” Supinda Bunyavanich, MD, MPH, MPhil, professor of pediatrics and genetics and genomic sciences at Mount Sinai, associate director of the Jaffe Food Allergy Institute and a senior study author, said in the release. “We hope that having a more comprehensive, holistic view of air pollution may one day be able to reduce the chances that children will be burdened by asthma.”

Healio spoke with Bunyavanich and Pandey to learn more about the study and the implications of their findings.

Healio: Can you describe the challenges asthma poses to children?

Supinda Bunyavanich

Bunyavanich: Asthma is the most common chronic disease in kids, affecting around 8% of children in the United States. It leads to symptoms such as chest tightness, wheezing, coughing and feeling short of breath, and it can really impair life. Kids can feel like they can't exercise as much, or they may even feel unable to go to school. Among kids with asthma, about 60% miss school each year because of asthma symptoms, and a third of adults who have asthma miss work because of their asthma symptoms, so it really is a burden on life. Related to that, there are millions of doctor's office and ED visits, and even sometimes death, related to asthma, costing the U.S. more than $56 billion a year. Asthma affects a lot of people of all ages, from all walks of life, and has real impacts on daily life.

Healio: Why is it important to identify mixtures of air pollutants that can affect asthma development?

Bunyavanich: Many studies have looked at air pollutants and how they may be related to health outcomes including asthma. A lot of the prior studies have targeted specific pollutants but, in real life, when you walk outside you breathe in a mixture of air pollutants, not one at a time. It's important to think about exposure to pollutants in combinations and also how they interact with one another in the ambient atmosphere.

Although it affects so many people, there's still a lot of about asthma that we don't yet understand. Finding out more about asthma, what its biomarkers are and what potential best treatments might be, has been a topic of research here at Mount Sinai. As part of this, we recruited a cohort of children with and without asthma who we are studying to determine better ways to care for them.

Healio: How did you create the DEEP extraction algorithm and how does it work?

Gaurav Pandey

Pandey: The goal here is to identify these combinations because air pollution is made up of hundreds of chemicals that we know of. This is a very challenging mathematical problem, because 100 pollutants will give rise to 2¹⁰⁰ subsects. That becomes an exponentially large space to look for these pollutants, so we clearly can’t [examine the combinations] brute-force one by one.

We decided to look at algorithms in machine learning that can actually identify these combinations. Decision trees is a class of algorithm that has this capability by using a straightforward model to capture these combinatorial effects of the variables in your data.

In our case, the data are largescale high-dimensional pollution profiles from Environmental Protection Agency’s National Air Toxics Assessment (NATA) database. We used a much more sophisticated version of these decision trees called the extreme gradient boosting algorithm, or XG boost algorithm, to derive hundreds of trees from which we tried to identify the combinations of pollutants that occur more frequently than others. If multiple trees are capturing the same pollutant combination, then they are more likely to be associated with the outcome.

We had information from this cohort of children in the Airway in Asthma study, and we used the ZIP codes of their residence to connect them with the NATA air pollution profiles. Once we had these pollution profiles and the asthma outcomes that Dr. Bunyavanich has collected, we ran the XG boost algorithm to identify those combinations. A combination of cohort data, air pollution data and machine learning ultimately led us to the combinations that we described in the paper.

Healio: What did your study reveal about the relationship between specific air toxics and asthma?

Bunyavanich: We showed that there are some air toxics that are individually associated with asthma outcomes, but there are also many combinations of them. Looking at air toxics individually wouldn't show a relationship between exposure and outcomes such as needing a daily asthma medication, having to go to the ED or being hospitalized for asthma. But looking at them in combination, you see that specific air toxics together can lead to those bad outcomes.

Healio: Did any of your findings surprise you?

Bunyavanich: I was surprised that there were several combinations that stuck out, many of which had acrylic acid as what we call the “primary branch point,” meaning it is the first condition of exposure that, in combination with other air toxics, leads to these adverse asthma outcomes. Acrylic acid is very common in the environment, so this heightened our awareness about combinations that we are exposed to on a regular basis.

Pandey: Our observations are surprising because of our approach. The individual associations were known, but as a combination they were not. We have some hypotheses about how they might be acting together. One of the chemicals stabilizes acrylic acid in the environment, so instead of it dissipating, it actually persists. Whatever bad effect it has may be exacerbated because it can't escape the environment. We believe this needs further follow-up in terms of trying to figure out the mechanisms through which these combinations work together to cause that worse outcome.

The DEEP algorithm itself is like an outcome of the study. With this work, we also want to encourage environmental health researchers and clinicians to start looking at these mixtures holistically as groups of chemicals rather than individually, in terms of air pollution but also water pollution and other kinds of mixtures that we are exposed to. Fortunately, through machine learning in other areas, we now have tools through which we can drill down into these mixtures and try to understand how the components are affecting us.

Healio: Can these findings be used to help prevent the development of asthma among children in the future?

Bunyavanich: We certainly hope that findings like those from this study will help toward that goal. A main take-home point of this study is that air pollutants should be monitored together, and perhaps regulated together, because they work in combination. In terms of thinking about public health and public policy, it's important to think about air pollutants holistically so that they can be managed together. Our study really highlights those combinations that might be higher on the priority list to think about and to regulate more closely.

Healio: What research would you like to conduct next in this area?

Pandey: We would really like to see these results get replicated in another cohort to make sure that we are not postulating something that is specific to the New York City area. I also would love to see the DEEP algorithm and its variants being applied to other cohorts to study the environmental relationships to more diseases, such as cognitive impairment or fatty liver disease.

We also are interested in understanding the mechanisms through which these combinations affect downstream asthma outcomes. Exactly what molecular pathways or other entities are actually getting disrupted because of these combinations is still unclear, which is why Dr. Bunyavanich has been collecting molecular data in her cohort to try to build these maps of how air pollution affects these asthma outcomes.

Bunyavanich: Our common goal is to understand asthma as deeply as possible. Not only environmental factors that impact it, but also how do those interact with genetic and personal factors? How can we better tease out that web of relationships that is so important to people getting asthma and experiencing exacerbations? This is one part of the picture. It'll be important to think about how it links to other molecular forces at play.

Healio: Is there anything else you would like to add?

Bunyavanich: The Mount Sinai cohort we used has a large catchment from the New York metropolitan area. We have captured data from children who live in the city and the greater tristate area. We have a diversity of subjects in our study and hopefully many of the findings that we found are relevant to other urban-centered areas that include suburban and outer areas. The diversity of our cohort also helps with the generalizability of what we found.

Pandey: The team that worked on this project itself also was very diverse in terms of personal and academic backgrounds. This was a group multidisciplinary effort, with experts in environmental health, machine learning people and asthma/allergy. This study would not have happened unless people from these diverse areas came together. This is a good example of multiple disciplines working together to understand things related to health, which is happening a lot more, and it needs to happen even more going forward.

Reference:

Childhood Asthma Study Uncovers Risky Air Pollutant Mixtures. https://www.mountsinai.org/about/newsroom/2021/childhood-asthma-study-uncovers-risky-air-pollutant-mixtures. Published Oct. 6, 2021. Accessed Oct. 22, 2021.

For more information:

Supinda Bunyavanich, MD, MPH, MPhil, can be reached at Icahn School of Medicine, Mount Sinai, 1 Gustave Levy Place, Box #1498, New York, NY, 10029; email: supinda@post.harvard.edu.

Gaurav Pandey, PhD, can be reached at Icahn School of Medicine, Mount Sinai, 1 Gustave Levy Place, Box #1498, New York, NY, 10029; email: gaurav.pandey@mssm.edu.