Mapping global research on climate and health using machine learning (a systematic evidence map)

Climate change is already affecting health in populations around the world, threatening to undermine the past 50 years of global gains in public health. Health is not only affected by climate change via many causal pathways, but also by the emissions that drive climate change and their co-pollutants. Yet there has been relatively limited synthesis of key insights and trends at a global scale across fragmented disciplines. Compounding this, an exponentially increasing literature means that conventional evidence synthesis methods are no longer sufficient or feasible. Here, we outline a protocol using machine learning approaches to systematically synthesize global evidence on the relationship between climate change, climate variability, and weather (CCVW) and human health. We will use supervised machine learning to screen over 300,000 scientific articles, combining terms related to CCVW and human health. Our inclusion criteria comprise articles published between 2013 and 2020 that focus on empirical assessment of: CCVW impacts on human health or health-related outcomes or health systems; relate to the health impacts of mitigation strategies; or focus on adaptation strategies to the health impacts of climate change. We will use supervised machine learning (topic modeling) to categorize included articles as relevant to impacts, mitigation, and/or adaptation, and extract geographical location of studies. Unsupervised machine learning using topic modeling will be used to identify and map key topics in the literature on climate and health, with outputs including evidence heat maps, geographic maps, and narrative synthesis of trends in climate-health publishing. To our knowledge, this will represent the first comprehensive, semi-automated, systematic evidence synthesis of the scientific literature on climate and health.