What Patients Actually Want From Healthcare AI (And What’s Missing)

The Pattern in the Chart

In 2009, after another cardiac workup that resolved before the test results came back, Danny van Leeuwen was diagnosed with multiple sclerosis. He had had it for 25 years.

The pattern was in the chart the entire time. Twenty-five years of episodes that kept getting read as possible heart trouble because his father had died at 45 of a second heart attack. Every clinician along the way was being reasonable. No clinician along the way was looking at the timeline. “The pattern of what was going on was in my records for 25 years, but nobody had synthesized it,” Danny told us.

He is now 17 years into life with MS, 50 years into a nursing career that started in 1976 as the first male public health nurse in Western Massachusetts, and well into a parallel career as a patient advocate sitting on technical panels at CMS, AHRQ, PCORI, and the National Academy of Medicine. He hosts the Health Hats podcast, where his shorthand for himself is patient, caregiver, nurse, advocate, informaticist. We invited him on this week not just because of his healthcare AI patient experience, but because all five hats are now using AI.

Why Patient Data Access Isn’t the Same as Usable Data

When Danny first heard Dave deBronkart’s “Gimme My Damn Data” campaign more than a decade ago, his response was: “Watch what you wish for. You’ll be trying to drink dirty water from a fire hose.”

The metaphor is, if anything, kinder than what actually happens. Last December, Danny started a campaign to collect his medical records from his primary care practice going back to 2011. Two months later, a four-pound box of paper arrived at his door. Not chronological. Not digital. From his other care system at Mount Auburn, he requested the last three months of records and received 296 pages of redundant, non-searchable PDFs. He had nine appointments in that window.

The patient who knows what to ask for, who has spent a career on outcomes panels and informatics committees, who runs his own healthcare podcast, still cannot in 2026 obtain his own data in usable form. “Access to data and access to usable data are really different,” he told us. We have not solved this.

Pattern, Question, Answer, Decision

What Danny actually wants from AI is something narrower and more useful than the panel-discussion version of patient empowerment. He wants patterns. From patterns, he wants to formulate better questions. From better questions, he wants better answers. From better answers, he wants to make decisions he can live with. “That’s the job: formulate better questions, seek better answers, make better decisions. And AI is the tool that I try to use to do it.”

A few examples of what that looks like in practice. Danny keeps a personal health spreadsheet — symptoms, mobility, mood, blood pressure, falls, weight, time spent playing his horn. He drops it into a Claude project, sometimes with journaling notes, and asks for patterns. The model has told him “have you thought about seeing a physical therapist” when his fall count crept up. It flagged a mood change that he had attributed to the disease itself, which his neurologist later listed as an allergy to the neuropathy medication he was on. It read ten years of neurologist notes and surfaced the disability scale his neurologist had been quietly tracking, which Danny then asked his neurologist to put at the top of every note so he could find it.

His rule of thumb for AI output: “Sleep on it and check it again.” The first read always feels amazing. The second read finds what the model missed.

The Three T’s and Two C’s

Danny evaluates every digital health technology against the same five-letter rubric. He calls it the Three T’s and Two C’s. Time, Trust, Talk, Control, Connection.

Time, in his framing, is not the clock. It is what fills the clock. “The clock isn’t the enemy. It’s the wrong things filling the time.” A tool earns Time when it gives clinicians presence back and lets patients prepare instead of catching up.

Trust takes longer to earn than most AI vendors expect, and “you can’t shortcut trust in the use of any tool.” Most digital health tools have a trust deficit, not because they are untrustworthy, but because the users have not had the time to develop trust in them.

Talk is the real conversation that decisions get made in. AI can help prepare for it and help process it afterward. It cannot replace it.

Control is the patient’s standing inside a clinical encounter. “If I’m feeling like an ant ready to be crushed, I’m not making good decisions.” Tools that hand control back to the patient pass this test.

Connection is the human lifeline. “When somebody greets you when you cross a threshold, that’s a connection.” AI can extend connection by helping patients find communities and by being available at 3:00 in the morning. AI cannot manufacture connection.

What the Data Misses

Three dimensions never make it into the record. Pain, fear, and cognition. “Pain changes what you can do and what you can decide. Fear closes your heart. It closes your mind. So when you’re scared in a clinical encounter, you’re not making good decisions. You’re just saying yes to end it. And cognition is, you know, it varies. Like, I can absorb better at 10:00 in the morning on a good day compared to 3:00 in the afternoon when I’m spent.”

Until those three dimensions are captured, every AI tool built on the chart alone is reasoning from a partial picture. The patient knows it. The clinician often knows it. The chart does not.

What This Means for Builders

Danny’s closing advice for clinicians, system builders, and developers came in one sentence: “You need to have patients, caregivers, and practicing partner clinicians in the design. They need to be there from the beginning.”

The rest was for patients. Use it. Find a buddy. Talk to your clinician about it, and pay attention if they will not. If you can, mentor someone else. AI literacy spreads the way other literacies do, which is to say, slowly, person to person.

We will close with the line Danny offered as a corrective to his own framework. “It isn’t first about the data. It’s first about what about life and what about the things that are important to people, uh, patients, caregivers, and the clinicians that they partner with, and how can AI help them?”

Listen to the full conversation.

You Might Also Enjoy

• S1E10 with ePatient Dave deBronkart — Danny references Dave’s “Gimme My Damn Data” campaign directly in this episode; Dave’s foundational case for patient data access is the conversation Danny’s “drink dirty water from a fire hose” caveat is in dialogue with.
• S1E30 with Amy Price — Danny credits Amy with teaching him how to query AI, how to be skeptical, and how to ask questions from different angles; her thinking on AI literacy is the practitioner version of what Danny demonstrates patient-side.
• S1E15 with Adam Rodman — Adam’s counterintuitive finding that human-in-the-loop can worsen conclusions comes up directly in this episode and pairs with Danny’s “I don’t care what the study says, keep humans in the loop” stance for a real tension worth sitting with.