Unlike all other medical technologies, artificial intelligence (AI) represents an active participant in the patient experience. AI will provide a data-based opinion to influence and sometimes automate the decisions of the health care professionals involved.
AI for providers: Medical imaging
Michael Recht, chair of radiology at NYU Langone, is spearheading a partnership between NYU Langone and Facebook’s AI research group. The goal of this partnership is to reduce the time it takes to conduct an MRI by a factor of 10. The project, called FastMRI, could enable MRIs to replace X-Rays, avoiding radiation exposure and increasing a provider’s visibility into what ails a patient earlier in the treatment process.
AI for pharma: Drug discovery
Michael Frank, director of R&D strategy within Pfizer’s Worldwide R&D Group, helps Pfizer leverage machine learning (ML) to accelerate drug discovery. On average, it takes 14 years and $1.6 billion to develop a new drug, with just a 10 percent success rate after entering human clinical trials.
When ML is applied to the process, it simulates and predicts the binding affinity and potency of certain compound combinations, making the process of discovering a new drug 10 times faster. Treatments take less time to get to market and at a fraction of the cost, providing solutions to previously untreatable patients.
AI for payers: Waste reduction
The Institute of Medicine estimates the United States wastes $1 trillion each year on health care spending — one-third of what it spends. Mark Kanner, lead data scientist at Aetna, focuses on how AI can reduce the errors, fraud, and abuse that result in that wasted money.
With 39 million members and 5,700 hospitals representing 1.2 million health care professionals in Aetna’s network, the volume of claims data is significant. Beyond the sheer volume, there’s structured and unstructured data in varying formats, and all of it requires millisecond-to-millisecond analysis.
It’s easy to imagine how certainty of which claims should be paid, and in what amounts, within such a complex system is difficult to come by. Data scientists like Kanner are using this trove of data to train ML models that can recognize if a claim should be paid as requested.
These models can ask questions like:
- “Has there been an unexpected number of patients with this same diagnosis?”
- “Is the submitted dollar amount higher than claims for similar procedures?”
- “Does the member live too far away from the provider for this claim to be legitimate?”
Teaching machines to answer those questions will eventually enable that $1 trillion of waste to be repurposed toward improved patient outcomes.
A word of caution
AI bias is real and it happens when the training data fed to a ML model accidentally provides the AI with a biased view of the world. Since the stakes of any new medical technology are patient lives, health care professionals must take great care to build “explainable AI” in order to avoid handing over the reins of the caregiving kingdom to a collection of biased, and potentially fatal, machines.
That said, I’m confident the industry will navigate these AI waters successfully and provide patients with a level of care far more effective than where we are today.