Treating Asthma with Big Data

It is critical to apply scientific rigor in interpretation when using healthcare data, so-called "big data", and the computational tools available to analyze that data, according to an article authored by Danielle Belgrave, PhD, of the Department of Pediatrics at the Imperial College in London, UK, and colleagues.

The authors argue that one of the most challenging aspects of treating asthma is identifying the endotypes that underlie the condition. They further suggest that the heterogeneous nature of asthma is the reason for variable responses to therapeutic agents, such as corticosteroids. Identifying specific endotypes, along with biomarkers for them, will allow for greater personalization of treatment.

Many genetic studies, expected to be game-changers, have, say the authors, “delivered neither meaningful clinical diagnostic tools nor useful insights into disease pathogenesis.” They note four factors that could contribute to the inconsistent findings in genetic studies to date: gene-environment interactions, gene-environment correlations, and epigenetic mechanisms, and the use of aggregated definitions.

However, studies that have focused on specific asthma subtypes, with recognized genetic variants, have shown greater promise. The authors point out the success of a genome-wide association study focused on a subtype that is associated with the gene CDHR3. In that study, two birth cohorts were compared and those with CDHR3 were at greater risk of asthma, but “there was no association with an aggregated definition of ‘doctor diagnosed asthma,’” say the authors.

Studies into the impact of environmental factors on the development and severity of asthma have also yielded inconsistent results. Multiple studies have shown that different phenotypes have different environmental associates in childhood wheezing. The authors say that better-defining outcomes of interest will lead to more precise and individualized medicine.

Research into biomarkers has been, overall, more successful than genetic or environmental studies, with four specific T2 mechanisms having been identified. However, the authors note “biomarker identification for asthma and allergic diseases [is] still in its embryonic stages,” and add that no biomarkers pinpointing the underlying processes that cause disease have been identified.

The phrase “big data” can refer to both the large amount of healthcare data being generated and to the complex nature of the data. Even with advanced computational tools and the potential of machine learning, which could predict outcomes and suggest optimal treatments, the authors suggest there is an inherent bias in such a quantity of data. Additionally, the authors suggest that the quantity of data can make it difficult for scientists to know what they are looking for, or what questions they should be asking.

The problem, according to the authors, is that “data-driven hypothesis-generating approaches to understanding disease are overshadowing traditional hypothesis-based research via carefully constructed questions and observations.” They advocate for a combination of the two approaches.

One approach to harnessing some of the power of big data may be through using birth cohorts, which allow researchers to observe and explore the development of disease over time. The authors use the Study Team for Early Life Asthma Research (STELAR) as an example. STELAR combines the data gathered in five birth cohorts in the UK in order to learn more about asthma over time.

The authors conclude that the current challenges in treating asthma can best be met through an integrative approach which combines “big data” with “big reasoning” and encourages cross-disciplinary collaboration among researchers. The article, "Disaggregating Asthma: Big Investigation vs. Big Data", has been published online ahead of press in the Journal of Allergy and Clinical Immunology.