- Artificial intelligence is playing an increasingly important role in healthcare analytics and will continue to dramatically influence the development of clinical decision support and business intelligence, says a new report by JASON, commissioned by the ONC.
Artificial intelligence, especially the branch of computer science known as deep neural networks or deep learning, has the potential to drastically improve the way clinicians and patients interact with data – if the industry focuses on cutting through the hype and implementing strong underlying governance and interoperability principles at the beginning of its development.
“The rapid digitization of health data through the use of heath information technology (health IT) in the United States has created major opportunities in the use of AI,” wrote a team of officials from the ONC, AHRQ, and Robert Wood Johnson Foundation in an accompanying blog post.
“Innovators and experts see potential in using digital health data to improve healthcare and health outcomes from the home to the clinic to the community. Yet, current AI is powered and limited by its access to digital data. With a range of health-related data sets, AI could potentially help improve the health of Americans.”
Health data is abundant, but it is also frequently locked away in siloes or proprietary formats that make it difficult to combine with external sources. Privacy concerns and competing business initiatives also tend to discourage organizations from sharing their data assets for research purposes, leaving AI and machine learning experts without the fuel they need to make the most of their algorithms and tools.
To better understand how to address these barriers, ONC asked JASON to lay out the opportunities, limitations, and potential use cases for AI in the next decade. JASON is an independent association of scientists and academics that has been advising the federal government on technical issues for decades.
“The new report…concludes that the broad advances in AI are significant and real,” states the ONC.
“The technology sector is witnessing what appears to be important new advances in AI that are bringing a new wave of interest for how it might shape the future of health and healthcare.”
The growing availability of data from networked smart devices and increased consumer comfort with ambient computing, mobile apps, and smartphone platforms are creating an environment that is ripe for artificial intelligence to augment healthcare both inside and outside of the clinic, JASON added.
Study after study from both academic institutes and private companies are also proving that artificial intelligence and machine learning have promising applications for clinical decision support within the traditional healthcare setting.
The report cites several intriguing projects that leverage AI as the basis for advanced clinical decision support, such as using imaging analytics to aid early screenings for diabetic retinopathy and classifying skin cancers based on a combination of images, biopsies, and expert input.
Deep learning may be a particularly effective way to deliver these enhanced decision support capabilities for cancers and other diseases. By mimicking the decision-making logic of the human brain and harnessing multiple layers of decision making that refine their conclusions based on the output of the previous layer, deep learning and neural networks offer an intricate and increasingly accurate way to classify images, text, and other clinical data.
In addition to focusing on traditional clinical data, AI researchers are increasingly integrating mHealth tools and devices into clinical trials to support data collection. Consumer-facing applications backed by serious analytics prowess are now available to monitor clinical factors like asthma management, Parkinson’s tremors, and heart conditions, among others.
“Revolutionary changes in health and health care are already beginning in the use of smart devices to monitor individual health,” observed the report. “Mobile devices will create massive datasets that may open new possibilities in the development of AI-based health and health care tools.”
Mobile applications and devices could provide a very rich source of data for large scale analytics, JASON says, but ensuring that the data is trustworthy, accurate, and useful no matter what its origin is a very broad and complex problem.
Caution is called for when developing and promoting consumer-facing applications and tools. Without an emphasis on standards and transparency, it will be very easy for companies to be less-than-scrupulous with their offerings.
“Consider an example for skin cancer detection,” the report says. “Computer-aided automated skin cancer detection was demonstrated on biopsy-proven clinical images and tested against 21 dermatologists. In parallel, online services already exist for remote dermatologist diagnosis of online-submitted images of skin moles.”
“We could imagine a scam service asking patients to submit self-taken skin mole images along with payment for an automated ‘quack’ diagnosis in return, one that did not actually use any validated classification scheme. More likely, the methods used by any one company may be hidden or obscure, meaning the user has no way to judge the soundness of the company.”
The healthcare industry will need to focus on delivering trustworthy and validated patient education resources that help them identify AI-driven applications that provide real utility.
JASON notes that the FDA has been closely monitoring the development of devices, applications, and their related business models, and may soon issue additional guidance on how to create useful tools that meet consumer and clinical standards.
At the same time, artificial intelligence could jumpstart the growing interest in crowdsourcing and citizen science, where the public contributes their knowledge or personal computing power to solve problems too large and complex for any individual organization to tackle.
“[Crowdsourcing] competitions could help accelerate new AI algorithm development, and an understanding of the biases and errors implicit to health data,” says the report.
“In cases where datasets are noisy, limited in number or scope, or otherwise not yet amenable to robust and autonomous computational processing, citizen science may be an approach to develop data sources.”
Collective contributions, whether from the public or from coalitions of academic and healthcare providers organizations, will be crucial for developing the critical mass of data required to support advanced AI analytics tools.
Developers and researchers need access to high-quality training data, which has been cleaned, validated, and labeled by experts, as well as unlabeled test data to help algorithms refine their skills.
“An aspirational goal for health and health care is to amass large datasets (labeled and unlabeled) and systematically curated health data so that novel disease correlations can be identified, and people can be matched to the best treatments based on their specific health, life-experiences, and genetic profile,” says JASON.
“AI holds the promise of integrating all of these data sources to develop medical breakthroughs and new insights on individual health and public health. However, major limiting factors will be the availability and accessibility of high quality data, and the ability of AI algorithms to function effectively and reliability on the complex data streams.”
Environmental and socioeconomic data is particularly scarce at the moment, limiting the ability of AI tools to accurately account for the significant impact of social determinants of health on the development of cancers and their interplay with genetic conditions.
“For many diseases, environmental exposures play a bigger role in health outcomes than genetics,” the report observes. “Yet, the amount of attention paid to environmental factors is a fraction of the attention that has been devoted to genetics.
With the goal of enrolling at least one million individuals across racial, ethnic, socioeconomic, and clinical lines, All of Us may represent a pathway to solving multiple problems for AI researchers.
The data will be standardized and accessible, clinical data will be collected for at least ten years to allow long-term monitoring, and the project prioritizes the integration of environmental and lifestyle data.
The nature of the project – a federal program with no incentive to limit data access to other organizations – may also help to reduce the traditional barriers of data sharing and interoperability.
Focusing on expanding access to high-quality and trustworthy data will be vital for the healthcare industry over the next few years, the report concludes.
Without laying a foundation of clean, accessible, complete, and multifaceted data governed by meaningful privacy rules and consumer protections, artificial intelligence may not be able to reach its full potential as quickly as could be hoped.
“Overall, JASON finds that AI is beginning to play a growing role in transformative changes now underway in both health and health care, in and out of the clinical setting,” the report concludes. “At present the extent of the opportunities and limitations is just being explored.”
As consumers and providers become more comfortable with data-driven computing, mobile devices and apps become more common, and data liquidity increases, healthcare organizations and consumers will soon be able to harness the potential of artificial intelligence on a much broader scale.