Field note: Using AI in Healthcare

Dad James Cummings on using AI to help his rare-disease kids' doctors

James Cummings is a Connecticut dad who burst into my life (and many other advocates’ and gurus’ lives) a year ago and has been doing everything in his power to rearrange the health data universe to help him help his kids. I hope he’ll be sharing more of his #PatientsUseAI journey as time goes by. In this post you’ll see why.

Watch for his positive outlook against overwhelming odds and his participatory medicine approach: patient and family as empowered, empowering partners for their clinical care teams.

In this first post he illustrates one of the #PatientsUseAI use cases: Organizing Information for Action. It would be possible to quibble with details of what he says (please do, in comments!) but I do not want to constrain this man’s voice!

I am a father to two children with separate rare diseases. My 5-year-old son has a hyper-rare bone marrow disorder called Diamond Blackfan Anemia (DBA) (5-7 in 1,000,000 live births). His 3-year-old younger brother was born with VACTERL association (1 in 10,000 to 40,000 live births) among multiple other chronic conditions. He was recently diagnosed with neurological tumors and profound hearing loss in both ears.

With such severe and rare conditions, I’ve spent an incredible amount of time inside hospitals and meeting new healthcare specialists. I am grateful for the lifesaving clinical care and exhaustive efforts from their talented teams, but there is one incredibly important realization I had: 

No one has more abundant time
and aligned intentions
to dedicate to positive outcomes
than patients and caregivers themselves. 

Through no fault of their own, practitioners are overworked and under-supported, juggling increased patient loads, administrative duties, and their own personal lives.

In this first post I’ll try to paint the picture. This is what I clearly see as an arriving evolution of how to use AI to help our first massive challenge — aggregating and organizing our records.

Power of participatory patients 

As an undiagnosed and rare disease dad, I’ve learned the value of patient energy. Rare and undiagnosed disease communities are advanced participatory patients and caregivers. Underserved, they are highly motivated to commit 24/7 worry to finding answers and contributing solutions any way they can. Despite not having any formal medical education or experience, this participatory behavior is a soon-to-be universal paradigm for all patients and caregivers. 

Priority need: Complete and organized personal longitudinal health records 

The best way I can contribute to our highly-trained doctors is to organize all available facts. There is no better utilization of patient energy than aggregating their entire and accurate health data, which includes electronic health records (EHR) from all the various providers, genomic data, and all their contributed data such as wearables, social, family history, and tracking and monitoring devices.

Having such so-called longitudinal health records (LHR) for both my sons has made provider encounters much more productive. LHRs help clinicians get a clear understanding of my sons’ health profiles and history to diagnose and prescribe treatment sooner.

But, bad news: the current record sharing process leaves data out

EHR portals like Epic’s MyChart (used by 150 million patients across the US) feature a tool called Share Everywhere that allows providers to view the data. However, after five years, I’ve learned that when I grant doctors access with tools like Share/Care Everywhere, the data they receive is often basic and fragmented or incomplete.

For Example:

My 3-year-old is treated by six different health systems (four of which use Epic); the shared data is extremely limited compared to all the data the source provider had recorded.

His multiple abnormalities have caused us to meet dozens and dozens of new physicians. In his current case of identifying the cause and treatment for his newly discovered neurological tumors, we were expedited to one of the country’s leading pediatric neurosurgeons at a hospital that uses Epic. Prior to the elevated encounter, I granted EHR access from all my son’s other provider portals, shared MRI scans with radiology reports, and participated in a telehealth call. 

As we waited to meet the surgeon, the physician assistant (PA) met us in the room to conduct her interview. As she began asking questions, I saw her struggling to remember previous discussions, review Epic dashboard, and reference the radiology report. After I had to inform her that my son had passed several newborn screening tests, I knew the EHR transfer she received was missing important information!

Enter ChatGPT 

Fortunately, prior to our visit, I had loaded GPT-4 with data I had gathered through varying health information management (HIM) departments and MyChart download requests. At this point, all my requests were delivered in PDF format (approx. 1,000 pages); I collated a chronological history (including selective clinical notes and radiology reports) the best I could into a homemade LHR inside my Google Drive.

My instruction prompt for GPT-4 was to review (and respond “in layman terms”): 

• Text summary of the LHR I had assembled

• PDFs of pertinent clinical notes and test results

• MRI report. 

GPT-4 generated a concise response summary, encompassing my son's medical history, relevant data for previous encounters, and analysis of the MRI. On its own, it also pointed out specific observations such as his newborn screening results, hearing test history, 72-hour ECMO (portable bypass) treatment, drug therapeutics and toxicology (many of which were missing from the EHR transfer).

Aggregating and including granular data (not included in Epic’s Care Everywhere export) into my son’s LHR enabled GPT-4 to suggest his hearing loss is not congenital (from the newborn and recent audiology results) and offered ECMO as a possible cause. Without this detailed information, the summary would have been less precise.

Immediate effect on our visit

When I shared GPT’s summary and assessment from my phone with the PA, it appeared her clarity and situational awareness immediately accelerated. The PA rapidly absorbed the information while reading the summary, referencing her Epic dashboard then turning back to the phone incessantly. I felt like light bulbs flashed in her head as she tapped into the LHR “Matrix”. 

Feeding my son’s LHR into GPT-4 prepared me so well to consult the provider that it took minutes to reach a level of understanding that may have otherwise taken the entire encounter or even days. The PA was empowered to leverage her education, experience and talents to make the encounter efficient and productive to deliver the best care possible. 

It’s not over, but we have surged forward.

My son's neurological masses and hearing loss are still without a known cause and treatment; the ending to the story is not yet written. Even if the result is “we don’t know,” compiling his LHR and distilling it with artificial intelligence (AI) helped us get there faster, preventing delays in piecing information together anecdotally and avoiding improbable/potentially unnecessary rabbit holes.

Going forward, I will keep my son’s LHR updated and repeat this process for the many future introductory encounters he will have. 

The integration of comprehensive data into a single accessible location represents a significant innovation. Without such LHRs, large language models (LLM) like GPT-4 cannot effectively speed up clinicians' comprehension.

It was a fascinating technological and patient anthropology exposition

 Here’s the future I see for this changing world:

• Humans are evolving to contribute to their own health outcomes. Rare/undiagnosed communities are the leading “real world evidence” of this inevitable universal behavior. Aggregating their data into a LHR is the most vital and fundamental contribution patients and caregivers can make.

• Despite vocation and talents of providers (legal and logistical limitations [and yes, even financial interests of systems and EHR vendors]), the data aggregation role lies (and will always lie) on the responsibility of consumers. Nothing or nobody else has: time, energy, wholly-aligned intentions, and funding to do it, nor the awareness of what everything means in the global perspective.
  
  Ensuring a comprehensive health data set, including not only EHR, but genomic, non-clinical, social determinants of health (SDOH), family history, wearable and monitoring data is collected (and up-to-date) starts/stops with the patient/caregiver. (I hope to say more about this in the future.)

• Deep learning LLMs are most useful when fed all-encompassing data sets. AI will provide nominal relief for providers in the form of documenting patient visits, completing billing templates, dealing with insurance companies, and working with EHRs, but until we feed it comprehensive LHRs, its power to help us understand, diagnose and treat is limited. 

Lesson learned: health data stagnates unless it’s organized and accessed

It’s surprising (and contrary to perception) that the medical industry is not the leading sector to capitalize on new technology. We collect data at a tireless pace with projections of 180 zettabytes (that’s 180 quadrillion terabytes!) by 2025 and healthcare is now the fastest growing sector with more than 1/3 of all data.

When patients and caregivers aggregate and grant access, AI can help: medically illiterate consumers begin to understand, empower providers to perform more effectively, recommend diagnoses and prescribe treatment from an individual LHR, and 10x, 100x, 1,000x... research advancements for rare/undiagnosed diseases

Things I Wish…

For any health IT nerds reading this, here are some extra points I think are important. We need:

1. Easier way for patients to obtain complete EHR data across various sources. Currently, requests from portals and HIM departments produce PDF forms; FHIR endpoints are not providing all available info; complete raw electronic health information (EHI) exports are not easy to procure

2. True interoperability, not “moment-in-time” interoperability. LHR is a manual process and is not automatically refreshed as time goes by

3. More aligned structure between the various EHR sources to make LHR more easy to organize

   1. More robust adoption of HL7® FHIR® standards 

This growing patient behavior will drive legislation. We need more policy to unlock the future by preventing data blockers. We need standards for meaningful technology use and to fuel AI models. 

The next plateau of health innovation starts with the patient

Both my boys have had relatively hard first years. Parents of chronically ill children rely heavily on clinicians for solutions, but we must recognize our capacity to contribute value where possible. 

My wife and I are far from the first family to make this realization. Pioneers like “Doc Tom” Ferguson, e-Patient Dave, Regina Holliday, “Mighty Casey” Quinlan, Susannah Fox, and many many others have paved the way for legislation and emphasize the importance of patient voices and participation. And that’s not to mention the many patient families who’ve done what I’m doing by hand, without AI.

After our second son was born with a separate diagnosis from his brother, but suffering from the same lack of data aggregation, data organization and use of machine learning (ML), I have indefinitely put my career as a digital product designer aside to dedicate my remaining time on this planet to help the future of medicine happen sooner. 

Participatory medicine has changed our lives. 

Besides like-minded patient advocates and participatory medicine leaders, I entrenched myself into the ecosystem including legacy systems, policy/regulatory landscape, big tech and data standards of Health IT. I have an unquenchable thirst to meet and discuss anyone involved in healthcare modernization. 

I often hear “how harrowing” my personal journey is, but I’m overwhelmed with a sense of purpose/duty to help all of humanity. I feel like the luckiest man in the world. 

Generative AI gives us an absolutely unprecedented way to help our doctors help us. Gathering the data is a lot of work, but I am so happy to be able to help — something the doctors themselves have neither the perspectives nor the resources to support. But I do.

Drop me a note if you would like my perspective, or make a public comment here to start a conversation.

LinkedIn: https://www.linkedin.com/in/cummings216/

Twitter: https://x.com/_Cummings_