Oops! Something went wrong while submitting the form.
June 5, 2024
Cardiovascular diseases are one of the most burdensome health threats globally, accounting for nearly 20 million lives lost each year1. One of the major challenges in addressing cardiovascular diseases is the complex range of factors that can lead to or exacerbate someone’s risk. These include natural and built environments, social and economic contexts, genetic factors, individual behaviors, and pre-existing conditions. Each of these factors not only affect cardiovascular disease but are also interrelated and context dependent.
In other words, cardiovascular diseases are a complex system.
Those confronting a complex system, e.g., country or city governments, policy makers, health systems, healthcare workers, and NGOs, often struggle to identify where to intervene to have the greatest impact. A lot of articles have been published on the impact of individual risk factors, e.g., income, on cardiovascular disease, but there are few tools that inform the prioritization across multiple interventions to maximize health.
Resources for interventions are limited, and optimizing their allocation requires an understanding of the entire system and the key leverage points within that system. This is both time-intensive and methodologically challenging.
The Novartis Foundation has started the AI4HealthyCities initiative to identify the key drivers of cardiovascular diseases and enable city governments to prioritize interventions with a potential for maximum efficacy. Collaborators in the initiative are already analyzing administrative, survey and other data sources. These efforts are needed to understand the local context and systems of cardiovascular disease specifically, but this requires multidisciplinary teams combing through mounds of data over many months to create a shared understanding of the system of a single disease or set of diseases.
But what if we could leverage advances in AI to generate a systemic understanding and representation of cardiovascular diseases, instantly leveraging the entire corpus of peer-reviewed scientific evidence? We set up a collaboration between System and the Novartis Foundation to do just that, and answer the question:
What are the most important social and environmental determinants of stroke, acute myocardial infarction (AMI), heart failure (HF), and end-stage renal disease (ESRD), and what roles do diabetes and hypertension play as potential mediators?
Health is a complex system, yet health knowledge is siloed. We only ever see a part of the whole, impeding research and care. At System, we think there’s a better way. What if, instead, knowledge was organized as one interconnected system? We’ve developed an architecture based on systems not on silos, that’s only possible today because of AI.
The System Graph consists of millions of statistical findings extracted using AI from scientific studies and expert-curated databases. This process runs daily on all newly published articles to ensure that the latest evidence is captured and represented. Once extracted, findings are assessed for accuracy, mapped to common ontologies, normalized, and linked together in a large-scale graph.
For example, the relationship between diabetes and stroke in the System Graph is backed up by 553 extracted statistical findings. We can also use the System Graph to start to identify potential causal pathways from the literature. For example:
The System Graph also categorizes millions of entities into a framework that includes biological, behavioral, environmental, and social determinants, all of which contain subcategories based on external sources, including AHRQ, CDC, and WHO.
Finally, similar to a meta-analysis or systematic review, evidence is filtered down based on quality, strength, and recency. For example, in certain analyses with the Novartis Foundation, the CVD system was filtered to only include relationships supported by a high number of sources, backed by a high number of significant findings, supported by one or more highly cited papers, with a median effect size that it is at least small (Cohen’s d > 0.2) and supported by evidence published within the past five years.
From a starting set of 2,513 determinants of HF, ESRD, Stroke, and AMI, a final list of 14 determinants was identified based on a number of studies supporting the impact of those determinants on one of our target outcomes. The results included environmental determinants such as air pollution, particulate matter, nitrogen dioxide, heavy metal exposure, noise level, temperature, and tobacco smoke exposure, and social determinants such as education level, employment status, socioeconomic status, income, neighborhood characteristics, geographic area, and marital status.
We also leveraged the System Graph to identify the most important mediators of these determinants and the various disease outcomes. Across the 14 key social and environmental determinants, there were many common biological mediators, including diabetes, hypertension, depression, metabolic syndrome, COPD and BMI, with diabetes, hypertension, and depression being the most common. Further, many of the social determinants, unlike the environmental determinants, were also mediated by behavioral and social factors.
For example, the relationship between education, employment, socioeconomic status, income, neighborhood characteristics and stroke were all mediated by behaviors such as smoking, alcohol consumption, physical activity, diet, and sleep, while education level and employment status were also mediated by insurance status.
In our AI4HealthyCities initiative, we aim to identify and address the key drivers of cardiovascular health to improve outcomes and inequities. System helps us to quickly synthesize vast amounts of published insights and understand and quantify relationships between determinants and outcomes. In combination with local data, this provides vital information to develop population health roadmaps.
Leveraging the System Graph, System and The Novartis Foundation identified 14 key social and environmental determinants of stroke, AMI, HF, and ESRD and potentially important mediating paths. The scale of the System Graph enables rapid identification of targets for intervention to improve health outcomes. Ultimately, this work will be used alongside other efforts, like the work being done across multiple city populations by the AI4HealthyCities team, to guide decision making over how best to allocate limited resources to improve cardiovascular health.
The success of this project demonstrates how healthcare companies, governments, public health organizations, and foundations can leverage System to develop a systems understanding of any disease at unprecedented speed and scale. Cardiovascular diseases are one of the most burdensome health threats globally, accounting for nearly 20 million lives lost each year. One of the major challenges in addressing cardiovascular diseases is the complex range of factors that can lead to or exacerbate someone’s risk. These include natural and built environments, social and economic contexts, genetic factors, individual behaviors, and pre-existing conditions. Each of these factors not only affect cardiovascular disease but are also interrelated and context dependent.
Filed Under:
Use Cases
