In an article published in the journal Plos Climate, researchers from the USA conducted a comprehensive review of the literature describing the use of machine learning (ML) for predicting health outcomes from extreme weather events that are responsive to climate change.
Study: Machine Learning Unveils Climate Health Risks: A Comprehensive Review. Image credit: Roschetzky Photography/Shutterstock
Background
Climate change has diverse impacts on health, ranging from changes in temperature and air pollution to infectious diseases, food insecurity, and social disruption. Extreme weather events emerge as salient indicators of climate change, causing immediate and prolonged health issues like heat stress, respiratory and cardiovascular diseases, injuries, infections, and mental disorders. These effects vary among populations based on factors such as age, gender, socioeconomic status, preexisting conditions, and geographic location. Thus, there is a need for more precise and comprehensive methods to pinpoint and measure health risks among different groups and regions facing distinct climate scenarios.
ML is a subset of artificial intelligence (AI) that uses statistical methods to learn from historical data and make decisions. It can play an important role in predicting the impact of climate change or extreme weather events on the health of individuals or groups of people. Similar to its role in predicting diseases like coronary artery disease and breast cancer, ML-based tools can quantify and identify risks for individuals or populations. Additionally, ML-enabled devices have the potential to enhance existing systems by utilizing secondary data, including electronic health records. This approach allows customization to specific populations, improving disease specificity and addressing limitations in current emergency preparedness systems.
About the Research
In the present paper, the authors performed a scoping review to assess the potential of ML to predict outcomes of health from climate-sensitive weather events. This literature review aims to map the key concepts, sources, and gaps in using ML for predicting healthcare outcomes due to climate changes or bad weather events. The research followed the preferred reporting items for systematic reviews and meta-analyses extension for scoping reviews (PRISMA-ScR) guidelines and searched nine databases for relevant studies from inception to October 2022. These databases are Ovid EMBASE, Ovid MEDLINE, The Cochrane Library, Web of Science, medRxiv, bioRxiv, Google Scholar, Institute of Electrical and Electronic Engineers, and Engineering Villages.
The selection criteria for studies included those involving participants aged 18 years or older, examining extreme weather events such as heat waves, floods, and wildfires, measuring health outcomes such as death, hospitalization, or mental health conditions, and using ML-enabled modeling methods such as support vector machines (SVM), decision trees, linear regression, neural networks, non-linear exponential regression, logistic regression, spatiotemporal integrated Laplace approximation (INLA), or random forest.
The study excluded studies written in non-English language, commentaries, review articles, or editorials or did not include measures for model validation. Moreover, the authors screened the titles, abstracts, and full texts of the retrieved articles by two independent reviewers who checked the reference lists and cited articles from the included studies for additional sources. Finally, they extracted relevant data from the included studies using standardized templates and synthesized the findings qualitatively based on the type of extreme weather event.
Research Findings
The review identified 6096 records from the database search and seven studies from the full-text screening that met the inclusion criteria. Six studies predicted health outcomes from heat waves and one from flooding. The studies were conducted in the United States, Europe, South Korea, and China and used different study designs, data sources, and outcome measures. The health outcomes included mortality, morbidity, and post-traumatic stress disorder.
The ML techniques used were non-linear exponential regression, logistic regression, spatiotemporal INLA, random forest, decision tree, and SVM. The ML models were evaluated using various metrics, such as precision, accuracy, F1 score, specificity, recall, receiver operator characteristics curve, and precision recall curve. Furthermore, the models were validated using either internal or external methods, such as training and testing split, bootstrapping, cross-validation, or different data sets.
The outcomes showed that the use of ML algorithms to assess the health impacts of climate-sensitive extreme weather events was feasible and promising but also limited and heterogeneous. These models outperformed traditional statistical methods, such as linear regression or Poisson regression, in some cases. The models were able to capture the complex and nonlinear relationships between the exposure and outcome variables, as well as the spatial and temporal variations in the data. Additionally, these models also enabled the identification of the most important predictors and risk factors for health outcomes, such as temperature, humidity, air pollution, age, gender, and socioeconomic status.
The authors discussed the potential applications and implications of the ML models for predicting the health risks from extreme weather events influenced by climate. They suggested that ML models could be used to identify and quantify the risk to individuals and populations from specific threats, such as heat stress, dehydration, or respiratory infections. This helps guide proactive clinical and policy decisions, such as issuing heat alerts, providing cooling centers, or allocating resources. Moreover, it can help monitor and evaluate the effectiveness of interventions, such as heat adaptation strategies, emergency preparedness plans, or climate change mitigation actions.
Conclusion
Overall, the review summarized that the use of ML algorithms to assess the adverse health impacts of climate-sensitive extreme weather events is possible and promising. It pointed out the ethical and social issues that need to be considered when using ML models, such as data quality, privacy, security, transparency, accountability, and equity. The protocol has been registered in the Prospective Register of Systematic Reviews (PROSPERO) database.
The authors acknowledged several limitations and challenges in the literature, such as the limited number and diversity of studies, the lack of standardized data and methods, and the need for external validation and generalization. They recommended that future research should address these gaps and challenges, as well as explore other types of extreme weather events, health outcomes, and ML techniques. Moreover, they suggested that the development of data standards and frameworks for climate change and health, as well as the collaboration among researchers, practitioners, and policymakers, would help to ensure the robustness, validity, and usefulness of the ML models.
Machine Learning Cuts Costs in Solar Power Cooling