Patient portal messages can train AI to support personalized care in endocrinology (and beyond)

AI tools that help inform patients are highly useful, but those that directly delve into patient data for communications cues are risky. That’s according to a small group of physicians asked by Stanford researchers to judge precision AI tools developed from a large dataset of diabetes patients’ expressed needs.

Corresponding author Jiyeong Kim, PhD, MPH, and colleagues describe the project in NPJ Digital Medicine. 

The researchers first created a tentative set of AI algorithms for a discrete use case: crafting tailored care for patients with diabetes. They trained AI models on more than 324,000 messages posted in patient portals over an 11-year period by more than 11,000 patients. The team applied multiple prompt-engineering techniques for querying the models. 

Next the investigators had five physicians, all endocrinologists, evaluate the tools. The main study objective was gauging specialized physician perception of usefulness vs. risk in the tools, all of which were intended to help individualize care.  

In their study’s discussion section, Kim and co-authors note their findings may offer “essential ramifications for developing potential AI tools for precision patient care for diabetes and beyond.” They present a number of takeaways from the work. Here are highlights. 

1. AI tools that physicians consider ‘highly useful’ help with clinically undemanding tasks. 

Examples include tools that expedite administrative processes (e.g., drafting templates of authorization letters) and aid in patient education (e.g., creating educational materials for common lab inquiries and customizable responses for glucose monitoring and pump usage), the authors report. They remark: 

‘These AI tools may help clinicians with timely interaction and streamlined support, and in an efficient manner.’ 

2. AI tools deemed ‘risky’ directly handle patient data. 

Examples include synthesizing patient data for message triage by urgency and topic.  

‘Our work contributes to offering critical ramifications for the development and advancement of potential AI tools for precision care.’

3. A holistic approach that accounts for all relevant and essential data could be ‘ultimately optimal.’ 

AI tools can prepare evidence-based yet easily understood educational materials on the impact of meal timing and nutrient composition with real-life examples to empower patients, the authors point out before adding: “This may help them understand those important relationships and have more autonomy in dietary management.” 

Future studies might ‘further develop AI tools that can craft meal plans for individuals with specific dietary restrictions to enhance tailored patient care and embrace diversity.’

4. Appointment-related issues were very common. 

“This highlights the need to reinforce the existing scheduling system,” Kim and co-authors state. “Perhaps a real-time interactive assistant could soon triage scheduling queries and efficiently schedule patient visits in the patient portal.”

‘Moreover, this AI-enabled conversational agent may especially help those with limited proficiency in direct scheduling, including older adults and non-English speakers, which could narrow the digital divide and possibly reduce related health access differences.’ 

5. Real-time resource updates could be worth pursuing further.

Examples might include policy changes and medication supplies. “During the widespread shortage of [certain diabetes] medications over the past few years, patients were exposed to the risks of falsified medications and unreasonable consequences,” the authors note. More: 

‘AI-enhanced prediction models based on past medication use and supply could help avoid future imbalances of supply and demand, creating an environment for equitable medication access.’ 

6. An interdisciplinary educational approach is important.

This was so for the study cohort, the authors point out, because “the combination of good glycemic control, medications for diabetes and osteoporosis and a dietary and lifestyle-based approach is optimal for bone disease care in patients with diabetes.”  

‘AI-driven automated Q&A for common queries for bone imaging and surgery and a decision-support tool for clinicians to optimize MRI referrals could be further refined for development.’

7. Proactive mental health screening or telepsychiatry-based support could be helpful.

As needed, AI assistance “may step in as a frontline symptom screener along with contemporaneous efforts,” the authors write. More:

‘Future efforts could focus on targeted mental health support by harnessing AI tools for women as a start, because women were found to use telehealth for mental healthcare more than men.’ 

In their conclusion, Kim et al. underscore that they found AI-powered analyses “able to comprehensively voice patients’ needs and concerns.” More:

‘Demonstrated AI tools could expand the scope of AI’s use in tailored patient care that is not limited to diabetes.’

The study is available in full for free.