A number of studies have found high correlation between social factors and health. For example, in County Health Rankings: Relationships Between Determinant Factors and Health Outcomes, social determinants of health had the largest contribution to health outcomes:
- 40% SDOH
- 30% Healthy behaviors
- 20% Healthcare
- 10% Physical environment.
The study measured the health outcomes: premature death, physical health, mental health and low birth weights. If healthy behaviors are considered to be SDOH factors, the contribution is 70% (see Social Determinants and Health Behaviors: Conceptual Frames and Empirical Advances)
Organizations are recognizing the potential of AI to impact health outcomes based on SDOH data. AI has applications in managing healthcare outcomes, civic organization planning, developing economic and social policies and education planning.
Other “frontier” technologies combine with AI technologies to form solutions. For example, the Smart Cities Initiative defines frontier technologies as: artificial intelligence; Internet of things; digital twins; unmanned aerial vehicles/drones; wearable technologies; and virtual reality / augmented reality. These are interconnected and interdependent technologies important to solutions implementing sustainable development goals.
Open source data science software provides important benefits to civic base organizations and nonprofits. Open source increases access to scalable, innovative, cost-effective technologies to help organizations impact society and health in positive ways. Top open source technologies include:
- Development: TensorFlow, Scikit-learn and Torch
- Collaboration: JupyterLab and JupyterHub
- Operationalization: MLFlow
SDOH data offers great potential for useful AI models, but the universe of SDOH features is enormous and growing quickly. Social features can come for numerous domains and each domain can be served by numerous data sources. This can lead to difficulties in training machine learning models due to high dimensional data. The ‘Curse of Dimensionality’ presents a number of challenges when analyzing or visualizing the data to find patterns as well as when developing machine learning models.
The targets for SDOH model can also be challenging. Health issues and outcomes can be defined in many ways. And the populations for health issues may not always align with the populations used for features. If population information is not available it is necessary to compare the distributions of some general characteristics such as age, income, income, ethnicity, etc.
SDOH data and health data are highly sensitive. AI models must be developed in a way that respects:
- individual privacy;
- ethical principles and values that are shared by all;
- applicable laws and regulations.