Benefits of AI in Healthcare and Data Sources
AI has already brought marvellous advancements in the healthcare sector but awareness is on the rise around the risks associated with it as well. In this series of posts, we are concerned with the ethical and legal compliance issue of protecting privacy and safeguarding health data used in machine learning when developing and using AI in healthcare.
Part 1 covers the benefits of AI in healthcare and the sources of health data. Part 2 addresses the risks associated with noncompliance, the potential misuses of health data facilitated by AI, as well as the challenges around privacy-preserving techniques. Finally, Part 3 sets out various attacks launched against AI models and the data used for their training.
The Benefits of AI in Healthcare
AI is revolutionizing medicine by venturing into uncharted territories and addressing significant global healthcare challenges. One notable example is AlphaFold, an AI-powered algorithm for protein structure prediction, which achieved a breakthrough in 2021 by solving the longstanding protein folding problem. This achievement, which had impeded critical advancements in biology and medicine for half a century, opens up new avenues for research and drug development.
Similarly, innovations like InSilico Trials enable pharmaceutical companies to conduct simulated clinical trials for drug discovery on expansive population models. These trials offer greater control and flexibility while mitigating resource constraints, ultimately facilitating the creation of more effective and safer drug products.
A use of AI that seems less headline-worthy at first glance but has made a big difference for healthcare practitioners is AI-powered clinical documentation software.
The administrative workload that comes with being a healthcare practitioner involves clerical tasks unrelated to direct patient care but it contributes significantly to physician burnout, a problem underscored by a survey from Athenahealth showing over 90% of physicians experience regular burnout.
Developed by companies like Microsoft’s Nuance Communications, Abridge, and Suki, ambient clinical documentation technology allows consensual recording of doctor-patient visits, which are then automatically transformed into clinical notes and summaries using AI. This innovation aims to alleviate the administrative burdens faced by doctors, enabling them to prioritize meaningful engagement with patients.
The momentum behind ambient clinical documentation technology is further underscored by substantial investment and rapid adoption rates across the healthcare sector. Abridge, for example, has notably progressed through a $30 million Series B funding round to a substantial $150 million Series C round within a short span, reflecting strong market confidence and the perceived value of their technology in addressing critical healthcare challenges. This financial backing has enabled Abridge to expand its reach across 55 specialties and support 14 languages, demonstrating the technology’s broad applicability and potential for global impact.
These developments, coupled with the reported adoption by several hundred health organizations and the integration of ambient clinical documentation into major electronic health records systems, signal a transformative phase in healthcare, where AI significantly enhances the clinician’s ability to focus on patient care.
Sources of Healthcare Data
To train AI systems such as AlphaFold, InSilico Trials, or ambient clinical documentation systems relevant data is key, and lots of it. We discuss in Part 2 of this series what privacy issues arise around the data acquisition, collection, and processing. As a first step we shall consider the possible sources for the data AI models require for training.
Health data originates from a multitude of sources within the healthcare ecosystem, reflecting the comprehensive nature of patient care and medical research. Primary sources include electronic health records (EHRs), which contain detailed information about patients’ medical history, diagnoses, treatments, and laboratory results.
Additionally, medical imaging technologies, such as X-rays, MRIs, and CT scans, produce vast amounts of diagnostic data that contribute to patient assessments and treatment planning. Wearable devices and health tracking apps further augment health data collection by continuously monitoring physiological parameters, activity levels, and lifestyle habits.
Furthermore, public health agencies compile population-level health data through surveillance systems, registries, and surveys to monitor disease trends, track outbreaks, and inform public health interventions. Collectively, these diverse sources of health data form the foundation for clinical decision-making, medical research, and public health initiatives, driving advancements in healthcare delivery and improving patient outcomes but also potentially serving as fodder for machine learning models.
Conclusion
Having conducted a brief survey of the beneficial uses of AI in healthcare and the data sources that enable the training of AI models and also Safeguarding health data used in Machine Learning, Parts 2 and 3 dive into the risks associated with noncompliance with data protection laws and regulations as they apply to AI development and use, challenges that we face regarding privacy-preserving techniques, as well as various privacy attacks.
