By Deborah Borfitz

September 4, 2018 | From point-of-care decision support tools to targeted tumor sequencing tests, artificial intelligence (AI) is causing some shakeups within the guarded healthcare industry—not lowering the overall need for medical professionals but demonstrating promise in improving diagnostic and treatment precision and the overall quality and experience of care.

IDx-DR software for the autonomous detection of diabetic retinopathy, a leading cause of blindness, exemplifies the near-term potential of AI in shaping radiology, “simply because of the massive amount of data that already exists,” says Boston-based healthcare consultant and author Harry Glorikian. “Imagine the potential of an AI system that has been trained using every X-ray, CT, and MRI scan that has ever been taken along with corresponding diagnosis and patient outcome data.”

Virtual assistant companies such as Babylon Health and Sensely have been making waves with chatbots that address basic health-related questions, freeing up doctors for other tasks. But the technology may prove to be particularly useful in the field of mental health, says Glorikian, citing Woebot and Tess as two of multiple apps being used by people suffering from depression and anxiety.

Pharmaceutical companies mostly sit at the back of the adoption curve, Glorikian says, but Recursion Pharmaceuticals is a notable exception. Thanks to its AI-powered drug discovery platform, a potential treatment for cerebral cavernous malformation (a genetic disease) was recently granted orphan designation by the FDA and a phase I clinical trial is soon to be underway. Meanwhile, Boston-based FDNA created a phenotyping app using facial recognition software that to date has helped more than 100,000 patients with genetic conditions get properly diagnosed.

In oncology, developments around in vitro diagnostics are helping to better match patients to treatment, Glorikian continues. The Oncomine Dx Target Test of ThermoFisher Scientific and FoundationOne CDx of Foundation Medicine are both FDA-approved and CMS-reimbursed companion diagnostic tools employing next-generation sequencing technology—the former benefitting patients with non-small cell lung cancer and the latter those with all types of solid tumors.

Guardant Health has a “liquid biopsy” (Guarant360) blood test that completes tumor genomic profiling in under two weeks, and has demonstrated its utility in identifying therapeutic targets across different cancer types, Glorikian adds. Similarly, a research team based at Johns Hopkins University came up with a test (CancerSEEK) for earlier detection of eight cancers—some highly lethal with no current diagnostic screening test.

But overall, oncology has yet to see the “dramatic transformation” that AI promises, says Glorikian, pointing to last year’s public stumbles of IBM’s Watson machine learning system. “Part of the problem is that for some rare cancers or diseases we still don’t have enough data to put together a comprehensive training set.” The solution is more publicly available patient sequencing data and outcomes, which he foresees on the horizon.

Handling the Hurdles

AI will have an important role to play in patient compliance, perhaps building off what a few pioneering digital health companies are already doing. Pillsy and AdhereTech manufacture special bottle caps to track when users take their pills, which can be paired with a digital app that sends reminders through a smartphone or smartwatch to encourage medication adherence, Glorikian says. “AI can approach the problem from a completely different position by identifying patients who are likely to be nonadherent from the start and then companies can target interventions to them.”

The trick is getting enough of the right data to train an AI system to recognize those patients and identify the factors making them noncompliant, Glorikian adds. “Is it simply forgetfulness, so you can support the patient by giving them medication in a Pillsy bottle or helping them set up a reminder on their phone? Or is it something else, like the cost of medication or the inconvenience of taking it, which would require a different solution entirely?”

Google has multiple AI projects focused on healthcare, says Glorikian. These include DeepVariant, to help researchers build a better picture of a person’s genome, which could trickle into clinical practice over the next several years. One of the biggest barriers is the lack of interoperability between electronic health records and other IT solutions. “Health systems are struggling with the data deluge already and adding in genome sequencing data on patients will only make it worse.” Glorikian believes Google Brain’s Medical Digital Assist is on the right path by bringing this data to physicians at the point of care rather than increasing the administrative burden on them. Its goal is an AI-enabled system to handle physician notetaking during patient visits.

A Perfect Match for Telemedicine

In many ways, telemedicine is the ideal match for AI because it also aims to cut costs and diagnose illnesses faster and more accurately.

The diabetes sector has come the farthest with remote patient monitoring, according to Glorikian. “The integration of digital health APIs and continuous glucose monitoring is quickly changing how patients with diabetes manage their disease—and how doctors keep tabs on them.” Several companies are now taking on the challenge of mixing homecare telehealth and routine clinical practice.

Claris Healthcare recently launched a device (Claris Reflex) that monitors the activities and adherence to movement and icing schedules of patients recovering from a total knee replacement, sending that information wirelessly to their doctor, Glorikian notes. The innovation has “clear relevance” to orthopedic practices but might also be useful in the ongoing monitoring of patients after other types of surgical events and hospitalizations.

Basil Leaf Technologies has created a handheld, all-in-one device to monitor vital signs and analyze blood and urine samples that is being marketed to providers. “It’s not difficult to imagine this technology being used in the future by home health aides or even patients themselves, and the results sent to a doctor to monitor remotely,” says Glorikian.

*Editor’s Note: This article originally ran on Diagnostics World.