One of the areas enjoying some of the highest degree of AI innovation is health, taken here to mean both life sciences and healthcare delivery. Health is a huge market and is attractive to AI researchers given the tremendous volume and variety of health data being produced constantly, as well as the potential for improving care. At the same time, the health industry possesses unique challenges to overcome before meaningful progress can be achieved.
The scale of data surrounding people’s health is constantly increasing, both in depth and in breadth. AI algorithms today can leverage a wide variety of sources, including:
Patient records: Health encounters, ranging from routine immunizations to emergency surgeries, are now digitized and stored in Electronic Health Records (EHRs).
Administrative records: Massive sets of claims from providers to payers tie together patient characteristics, diagnoses, prescriptions, and procedures. Social determinants of health (SDoH): Sources such as the Census Bureau offer a complementary lens into patient health journeys by highlighting demographic and socioeconomic factors.
Internet of Things (IoT) streams: Wearables and smartphones provide constant and real-time signals via sensors that track vital signs such as heart rate, activity level, and blood oxygen saturation. Genotypes: Entire genetic maps, and hence knowledge of molecular predispositions, of patients are now routinely available due to the markedly reduced cost of sequencing the human genome (down from roughly $100 million in 2001 to less than $1,000 in 2019).
Omics: Large-scale molecular fingerprints at the cellular and person levels, including, for example, libraries of proteins (proteomics), metabolites (metabolomics), and lipids (lipidomics), are made possible by the lower costs of data and processing power, coupled with advances in biochemistry and related sciences.
Research and Development (R&D): Academic and commercial entities drive biological innovation and breakthroughs every day in the areas of fundamental biology, drug development, and translational applications, and in doing so generate large amounts of scientific data.
This multidimensional slew of data, coupled with modern developments in statistical theory, programming, and computational hardware, means that it is a truly exciting time to be applying AI to questions of health.
But while there are many exciting developments underway and on the horizon for AI in health, it is important to realize that AI is not a panacea. It is important to realize that there are some very real challenges to accessing the data required for, or developing predictive models about, biomedical problems.