番茄社区

A custom artificial intelligence (AI) app for insulin management advice helped patients with type 2 diabetes gain rapid glycemic control, a small randomized clinical trial found.

In titrating basal insulin, those who used a voice-based conversational AI application -- delivered over Amazon's Alexa at home -- more quickly achieved optimal insulin dosing (median 15 days vs over 56 days with standard of care), Ashwin Nayak, MD, MS, of Stanford University in California, and colleagues reported in .

AI app users also had significantly better insulin adherence during the 8-week trial (mean 82.9% vs 50.2%, respectively, P=0.01); reaped significant glycemic improvement (mean decreases of 45.9 vs 23 mg/dL, P=0.001); and were more likely to achieve glycemic control, or a fasting blood glucose (FBG) level under 130 mg/dL (81.3% vs 25%, P=0.005).

"The key insight here is that clinicians can leverage technology to help extend and augment care delivery in the home," Nayak told 番茄社区. "We introduced the concept of 'remote patient intervention,' which we think of as closing the loop on remote patient monitoring data by making a clinical decision in real-time, following a physician-prescribed protocol."

"Insulin dose management based on real-time blood glucose and medication adherence data reported by a patient is just one example, but you can imagine how this model of care delivery could be used in other diseases," he said.

Beyond glycemic benefits, users of the app also reported significantly less diabetes-related emotional distress than standard care (-1.9 vs 1.7 points in composite survey scores, P=0.03).

"We are currently working on making technology like this accessible to patients outside of research settings because we think it can really help patients in underserved areas who need high-touch care to get their diabetes under control," Nayak said.

The researchers developed their custom voice-based AI app and had it powered by Alexa (Amazon wasn't involved with the study). The software was equipped with titration algorithms by the and included emergency protocols to handle hypoglycemia and hyperglycemia.

Before making the app ready for patient use, the participants' diabetes clinician selected an insulin titration protocol. The technology did not allow the AI to independently decide the dose titration, the researchers pointed out.

The same app was then deployed by an Amazon smart speaker to study participants, who could interact via voice commands and short conversations with Alexa to assist them in at-home basal insulin titration.

Individuals randomized to the standard of care group had basal insulin titrated by their clinician and were instructed to fill out daily an online blood glucose and insulin log. They also received an Amazon smart speaker, which was set up with daily reminders to complete their log, but they did not have access to the tested AI app.

The study was conducted at four primary care clinics at one academic center from 2021 to 2022. Recruitment targeted adults with type 2 diabetes and an HbA1c over 8%. People using insulin pumps or who had technical barriers in the home were excluded, as were non-English speakers.

The 32 participants analyzed in the primary analysis averaged 55 years of age and nearly 60% were women. Mean HbA1c was 9.6% at baseline.

Nayak highlighted that the 81.3% of AI app users who achieved glycemic control had been highly adherent to the technology, logging data on 54 out of 56 days.

"Given that participants in the intervention arm were adherent to the device, we actually weren't that surprised that 81% achieved insulin dose optimization and glycemic control," he said. "We know the clinical protocols work, the challenge is providing patients enough support to follow them, which is where technology like this can be helpful."

Nayak added that his group was pleasantly surprised at how adherent participants were in terms of checking in with their Alexa device daily and following its instructions despite "very little communication" with them in general after enrollment.

"We were really happy to see that despite the lack of hand holding, participants in the intervention arm were interacting with their device almost 90% of the days they were followed," he said.

Among the limitations of the study was that except for data collected from electronic medical records, all other variables were self-reported.

Study authors also acknowledged that because the follow-up period was only 8 weeks, glycemic control was measured by mean FBG rather than HbA1c.

Comment