Machine learning and artificial intelligence has been hyped as a panacea for improving healthcare delivery. Traditional machine learning and AI algorithms typically require massive amounts of data for training and optimization. However, many problems in healthcare become "small data" problems when the predictions are focused on individual patients and providers.
For example, 52 percent of hip replacements are performed by low-volume surgeons who take on fewer than 10 cases per year. As a result, it's hard to model individual provider performance for the health system's network and devise a treatment plan. The same problem rings true in the emergency department; 65 percent of ED visits are related to issues outside of the top 10 most common complaints, so modeling individual pathways for the best care is challenging.
"Analytics on massive online data that work for e-commerce and internet advertising won't work in a hospital setting," said John Guttag, chief technology officer of Health at Scale Technologies and distinguished professor of computer science at Massachusetts Institute of Technology. But technology can help. Health at Scale, a San Jose, Calif.-based company led by current and former faculty with ties to MIT, Stanford, Harvard and University of Michigan, is pioneering new classes of machine learning and artificial intelligence technologies to improve the predictive power of small data sets in the healthcare setting. The company runs some of the largest deployments to date of healthcare machine learning, serving millions of members in live production settings for leading health organizations.
At an Oct. 11 executive roundtable at the Becker's Hospital Review 5th Annual Health IT + Revenue Cycle Conference in Chicago, sponsored by Health at Scale, John Guttag — along with Mohammed Saeed, MD, PhD, CMO of Health at Scale; and Ilan Rubinfeld, chief quality officer of Detroit-based Henry Ford Hospital — joined a packed audience of healthcare leaders to discuss how hospitals can use advanced predictive machine learning and artificial intelligence to spark precision health delivery.
The small data obstacles
Precision health delivery — a care strategy that leverages individual patient variability and factors differential healthcare system and provider performance and capabilities to optimize care delivery — holds the promise to match patients with the best possible treatments and providers at the right time. However, when focusing on an individual patient and provider, the data sets specific to these patients and providers are often small. This challenges the use of traditional analytics and conventional machine learning, which rely on the availability of large amounts of data for training and inference. In this setting, core advances in machine learning are needed, with the ability to exploit information in datasets that are small but hold valuable information about individual patients and providers.
The solution
The small data challenges in realizing the true potential of precision health delivery led Health at Scale’s team to focus on developing a new class of machine learning and artificial intelligence analytics that address the specialized needs and challenges of small data problems in healthcare. The company’s scientific and clinical leadership discussed how Health at Scale's proprietary small data machine intelligence is different from traditional analytics and conventional machine learning in its ability to enable robust predictive and personalized fine-grained inference of optimal care decisions from healthcare small data.
"The key innovation of making accurate predictions from small data sets enables us to build and deploy personalized models that factor a new patient's unique medical history and characteristics, along with the historical performance of providers in the patient’s preferred geography in caring for prior patients like the new patient," Dr. Saeed said. "Personalized predictions driven by machine intelligence can rank providers in a region and recommend the top providers mostly likely to drive improved outcomes for a specific patient. Unlike existing approaches based on quality scores and rankings, we view predictive personalization as an integral breakthrough in improve care delivery. For example, two patients who may need cardiologists have their own unique health characteristics and needs, and should be matched to different top providers based upon each patient’s medical characteristics, and the providers’ past performance."
By using platforms and applications that integrate advanced artificial intelligence and machine learning, hospitals and health systems can significantly reduce hospitalizations and emergency room visits while leading to lower total costs of care over the long term. Dr. Saeed presented the results from two recent studies conducted by Health at Scale evaluating the ability of precision health delivery guided by specialized machine learning to potentially drive improved outcomes for nationwide Medicare beneficiaries. Both studies, looking at the use of precision navigation to match patients to providers and precision interception to match patients to early prevention initiatives demonstrated opportunities for substantial reductions in hospitalizations and emergency room visits. Through machine intelligence-based precision health delivery, Health at Scale is on a path to fundamentally transform how patients choose treatments and providers; much in the same way as eharmony has changed the game in online dating by making it more predictive and deeply personalized.
"This is where we need to go [in healthcare]," Dr. Rubinfeld said. "I absolutely think we need to be able to link a precision view of the patient to a precision view of an intended intervention. This is an essential need as we continue to push for the best possible outcomes and quality of care for those we serve."
Conclusion
Precision delivery is a new dimension to precision medicine, but is challenging to implement given small data challenges. Health at Scale's advanced machine learning was developed specifically to address healthcare delivery challenges in matching patients with the right care by the right provider at the right time, and promises to be a transformative innovation in healthcare.