Practical applications for machine learning and estimating algorithms: From research to health assessments

Over the last several years, there has been a myriad of published research utilizing machine learning and estimating algorithm methods. The application of these methods into research practice has provided more comprehensive insights on how complex environmental exposures impact associations with health outcomes. However, as these novel methods become easier to use, it is important to understand the underlying machinery, including the assumptions and limitations, as well as to be able to clearly and effectively communicate and interpret these results from research practice. With this growing body of literature lies the opportunity to integrate the knowledge that has been gained through research practice with health assessments, which in turn inform policy decisions. This presentation will provide a brief overview of the practical applications for machine learning in research practice and development of health assessments. 

Impact/Purpose

This abstract will be presented in an accepted symposium for the International Society of Exposure Science annual meeting. While individuals are exposed to multiple, simultaneous exposures from their environment, the effects of these complex exposures are difficult to quantify. As a result, environmental health policies have often focused on individual exposures. In recent years, statistical methods for data reduction and assessment of complex environmental exposures on health outcomes has become a rapidly advancing area of research. Machine learning methods (Bayesian Kernel Machine Regression) and estimating algorithms (G computation) have become favored methods for handling big data and analyzing complex environmental mixtures in epidemiological studies. These methods are powerful when utilized in the right context and provide meaningful findings that advance our understanding of how complex mixtures from environmental exposures impact human health. Studies utilizing these methods can be used to facilitate health assessments with respect to identifying environmental epidemiology studies and summarizing information to aid with policy decisions. However, a careful understanding of underlying workings of these approaches, including clearly defining assumptions and limitations of the methods, and appropriate interpretations, are needed in research practice to be considered for incorporation in health assessments and public health policy. We propose this symposium in order to connect the complex methods being used by academic researchers with the needs of environmental health policymakers. The speakers will provide an overview of methods currently being used in exposure science and epidemiology to analyze complex mixtures and a synopsis of what is needed in order to consider these research studies in informing environmental health policy.

Citation

Krajewski, A. Practical applications for machine learning and estimating algorithms: From research to health assessments. International Society of Exposure Science, Virtual, Virtual, August 30 - September 02, 2021.