With the simultaneous hype and concern surrounding ChatGPT and Med-PaLM, it is easy to get somewhat lost in the enthusiasm and anxiety around generative artificial intelligence (AI) tools. Indeed, the last two installments in this column have focused on these generative AI tools.
Readers are encouraged to check out the May/June and July/August issues of ComputerTalk for more information on generative AI. We will remain in the AI space for this issue, but our focus is different. Much of the talk surrounding generative AI centers on potential uses, both in and out of the healthcare domain.
Generative AI Applications in Healthcare
While there are a few current examples of generative AI applications in practice, much of the conversation is speculative in nature. In this column, we are going to explore existing AI tools that are marketed in the United States.
The U.S. Food and Drug Administration (FDA) is responsible for medical device approval and maintains a list of devices along with their approval status. Medical device classification is a combination of anticipated risks to the consumer, intended use, and indications for use. We are not going to delve into classification categories and codes, but readers should be aware that there is a structured, systematic method to applications for approval and ultimate device classification.
The COVID-19 pandemic was life-altering for many across the world, including people in this country. In healthcare (and other industries, like higher education), it can be argued that the pandemic hastened the rate of change for the acceptance of remote models of care. Prior to the pandemic, remote patient monitoring using devices largely from consumer electronics companies was slowly gaining steam, primarily under the digital health moniker.
However, resistance and questions around data validity and integration were common, and frankly, reasonable to ask. The pandemic forced flexibility in both policy and practice to enable continued care during quarantine periods and for those who could not travel.
Air Next
While its emergence cannot be directly attributed to the pandemic, NuvoAir’s Air Next diagnostic spirometry device highlights the convergence of remote monitoring and artificial intelligence. The device targets those with COPD (chronic obstructive pulmonary disease) in an effort to allow at-home lung function monitoring with heightened accuracy. The company frames its services as “virtual first,” meaning that in-person care interactions are a secondary focus.
Air Next is also available for remote diagnosis and treatment of those with asthma. For those use cases, artificial intelligence enters the picture in clinicians’ assessment of the data collected by the device.
However, the pandemic enters the conversation when clinical trials are involved. The pandemic impacted clinical trials at varying levels, including completely halting all operations for many trials. This highlighted the need for remote monitoring of clinical trial participants. In clinical trials focused on respiratory conditions, spirometry data is critical to assess outcomes and patient response to interventions.
A quality assurance process known as “over-reading” is common for clinical trial spirometry data. Over-reading involves a secondary review of spirometry data. The manual nature of over-reading can lead to extended delays in assessing data quality. To address this limitation, NuvoAir partnered with ArtiQ, whose software automates the over-reading process through artificial intelligence. The availability of real-time data regarding spirometry data allows clinical trial participants to immediately perform another reading.
KardiaMobile
Looking at another clinical condition, we turn to the heart. AliveCor is well-known in the digital health world. In fact, its heart rate monitor for the iPhone received FDA clearance in 2012. Today, the company’s focus remains cardiovascular health through the KardiaMobile personal electrocardiogram (ECG) device. The KardiaMobile allows patients to record an ECG anytime they desire, using an FDA-approved device. In 2021, AliveCor expanded the list of approved uses of its personal ECG device to include the detection of QT interval prolongation.
As a refresher, the heart produces a routine, repetitive pattern of electrical pulses in the process of stimulating a regular heart rate. The ECG measures this electrical activity in the heart. The presence of a prolonged (lengthened) QT interval in the ECG is indicative of a potentially very serious arrhythmia that could lead to death.
A variety of factors can cause a prolonged QT interval, including medications. For our pharmacist readers, we know these medications are found across many of the most commonly used drug classes, including antibiotics and antidepressants. With AliveCor’s six-lead KardiaMobile personal ECG device, patients have a tool to record ECG data, which can then be interpreted by the AliveCor QT Service.
The AliveCor QT Service is the actual tool that received FDA approval in 2021 for QT prolongation detection, using data recorded by the KardiaMobile 6L (6 lead) ECG. This interpretation of ECG data is enabled by AI algorithms. These algorithms were trained on over 750,000 ECGs from more than 250,000 patients.
Oxevision Vital Signs
As we wrap up, we will briefly look at a third AI product, from Oxehealth in the UK. Have you ever been a patient in the hospital? Did you find it ironic that those caring for you encouraged you to get rest (for optimal recovery), but you were frequently awoken during the night to record basic vital signs?
Certainly, those vital signs are important and necessary, but quality sleep in the hospital has been elusive during my hospital stays. While I cannot confirm that experiences such as mine were the driver, it seems logical that Oxehealth’s Oxevision Vital Signs product is intended to address scenarios like those previously described.
Oxevision uses infrared cameras to noninvasively measure pulse and respiratory rate, without interacting with the patient. In fact, the clinician can remotely access the data without the need to enter the patient’s room. Oxevision uses AI-based algorithms to assess the images collected by the cameras, alerting clinical staff of potential concerns. Oxevision is not intended for longitudinal monitoring, but for spot-check measures of no more than 15 seconds. I have no planned hospital stays, but I definitely will be on the lookout for Oxevision the next time I am a patient.
The FDA has approved more than 500 AI-based devices to support patient care and management. We touched on a few, select examples in this column. If you have a device that you find particularly compelling, please email me. CT
Brent I. Fox, Pharm.D., Ph.D., is a professor in the Department of Health Outcomes Research and Policy, Harrison School of Pharmacy, Auburn University. He can be reached at foxbren@auburn.edu.