When You're in the Seat

I flew Navy helicopters for six years. Aviation is relentless about information clarity. Before every flight you brief the mission. You know your aircraft's limits. You know what is normal, what is abnormal, and what needs action now.

When I turned that discipline on my own health, I found the opposite. I had years of lab results spread across clinics, a few specialists who had never met each other, and no single page that showed where I stood and where I was heading. So I built a spreadsheet of my own numbers, read the studies that applied, and brought a one-page summary to every appointment.

It was not my doctors' fault. How much of a visit is actually with the physician is hard to pin down, and the answer depends on who measured. In the latest national office-visit survey with time data (2019), the median visit included about 19 minutes with the physician (mean 21.3 in primary care); those minutes were taken from charts by Census field staff, and 32% were estimated because they were missing. ⁴ An analysis of electronic-record timestamps for 21 million primary care visits in 2017 found an average exam of 18.0 minutes, measured from when the physician opened the exam in the record, so it excludes check-in and the medical assistant's intake. ¹³ Studies that timed visits directly found less: in one study of 611 visits to 11 family physicians in 2003, face-to-face time averaged 10.7 minutes (13.3 including visit-specific work outside the room), and in two comparisons physicians' own reports ran longer than the observers' stopwatches. ¹⁴ These direct-observation studies are small and old. I found no recent study that splits a visit into check-in, vitals and physician time. And an observational study of physicians in four specialties found that nearly half of the clinic day, 49.2%, went to the electronic record and desk work, against 27.0% in face-to-face time with patients. ⁵ The people caring for me were doing their best inside a system that gives them little room to see the whole of me.

That is the problem this paper is about. It is not a story about bad clinicians. It is a story about who holds the whole picture, and about the fact that, today, the only person who could is the one with the least help doing it.

As a Navy pilot I used the term "situational awareness" constantly. It was a standing focus of our training and operations, and it came up on almost every flight and in almost every debrief.

In aviation, the researcher Mica Endsley described situational awareness (her papers write "situation awareness") in three steps: perceiving the elements in your environment, comprehending what they mean, and projecting what they will be soon. ⁶ Aviation regulators use the same term and the same three levels: New Zealand's Civil Aviation Authority guidance on situational awareness lays out Endsley's perception, comprehension and projection as the three hierarchical levels. ¹⁵

In one analysis of reports in NASA's voluntary Aviation Safety Reporting System that mentioned situational awareness, most errors were at the first step: 76.3% of the situational-awareness errors were failures at the first level: the information was missing, hard to see, not monitored, or misperceived. Only 3.4% were failures to project what happens next. A caution about that number: the reports are voluntary, so the study shows where errors cluster, not how common they are. ¹⁶

Medicine has looked for the same problem. In a review of 200 incident reports from one German anesthesia reporting system, two experts judged that 81.5% involved a situational-awareness error, most often in perceiving (38.0%) or comprehending (31.5%) the situation. ¹⁷ Researchers who looked for good ways to measure situational awareness among clinicians found 15 instruments across 25 studies and rated none of them as high-quality evidence. So I am not claiming that "improving situational awareness" has been proven to reduce harm in medicine. I am borrowing a clear way of thinking from a field that took information failures seriously early. ¹⁸

Here is the idea in plain words, applied to a person instead of a cockpit:

1. See it. Everything that is true about your health, in one place, including what is missing.
2. Understand it. What each number means for you, in context, with the reason behind it.
3. Look ahead. What is changing, what is due, and what to ask next.

Much of what follows is about step one: getting the information in front of the person who needs it.

This heading is my thesis, not a finding: no study tests it. What the studies below show is how many hands touch one patient's care.

Care is spread across many hands

In a large study of fee-for-service Medicare claims from 2000 to 2002, beneficiaries saw a median of two primary care physicians and five specialists in a single year, working in four different practices. ¹ A later study counted from the doctor's side: a typical primary care physician had 229 other physicians in 117 practices to coordinate with across their Medicare patients (fee-for-service Medicare, 2005). ¹⁹ These are old figures from fee-for-service Medicare, so they show the pattern then, not today's numbers. No single seat in the system sees everything.

Chronic conditions are common. In a CDC analysis of 2023 survey data, 76.4% of US adults reported at least one of twelve chronic conditions counted, and 51.4% reported two or more. ²⁰ The CDC reports that about 90% of the nation's $5.3 trillion in annual health spending is for people with chronic and mental health conditions. ²¹

Information goes missing at the point of care

In a study of Colorado primary care visits in 2003 (published in 2005), clinicians reported missing clinical information in 13.6% of visits. ² In the cases where something was missing, clinicians judged the missing information "at least somewhat likely" to adversely affect the patient in 44% of those visits. That is a clinician's estimate, not a measured harm, and the study is two decades old. ²

More recent national data come from the people living it. In 2022, about one in five adults who sought care for a medical problem said they had to bring their own prior test results (such as an X-ray or MRI) to an appointment. ³ In 2024, about one in ten said they had to redo a test or procedure because earlier results were not available. ³ A separate 2016 survey found 19% of US adults reported that records or test results were not available at an appointment, or that tests were duplicated, in the previous two years. ²²

Fragmentation is also associated with worse outcomes, though these are associations in claims data, not proof of cause. Among 506,376 chronically ill, privately insured adults, patients of primary care physicians in the most fragmented quartile were more likely to have a departure from best practice (32.8% versus 25.9%) and had higher spending ($10,396 versus $5,854). ²³ One review of waste in US health care estimated that failures of care coordination cost between $27 billion and $78 billion a year, in 2019 dollars, a range that spans mixed-quality estimates. ²⁴

An honest counterpoint

I do not want to overstate this. A 2021 survey of older adults in eleven countries found that the United States did comparatively well on coordination of care, including a low rate of gaps in hospital discharge planning. ²⁵ The problems above are common, not unique to us, and not the same for everyone. ^{22, 25} The claim is not "the American system is the worst." The claim is that the person with the most to gain from a complete picture is the person least equipped to build one.

Who is coordinating?

The Agency for Healthcare Research and Quality defines care coordination as deliberately organizing patient care activities and sharing information among all of the participants concerned with a patient's care. ²⁶ Its own framework counts the patient among those participants, and notes that patients and families experience coordination failures particularly at points of transition. ²⁷

Among high-need adults who saw multiple doctors or services, fewer than half (43%) said, in a 2016 survey with a 7% response rate, that they had an informed, up-to-date person coordinating their care. ²⁸ I could not find a national figure for how many patients end up coordinating their own care, so I will not claim one. What the data do show is that about one in five patients with a visit said they had to bring prior test results themselves (2022). ³

Time. By survey and electronic-record measures, a primary care visit includes roughly 15 to 20 minutes with the physician, and direct-observation studies found less. ^{4, 13, 14} About 39% of the physician-seen visits in the 2019 survey were 15 minutes or less. ⁴ In one older study of videotaped primary care visits (392 visits, 1998 to 2000, elderly patients), a median of six topics competed for a 15.7-minute visit, with about five minutes on the longest topic and about a minute on each of the rest. ²⁹ I looked for a published source for the popular "seven-minute appointment" and did not find one, so I do not use that number.

The record. In the study that observed 57 physicians across four specialties, EHR and desk work took 49.2% of the clinic day; the 21 physicians who kept diaries reported one to two more hours of after-hours work each night, mostly on the EHR. ⁵ In one Wisconsin health system, EHR logs for 142 family physicians showed about 5.9 hours of an 11.4-hour workday in the electronic record per full-time equivalent, including about 1.4 hours after clinic hours. ³⁰

Diagnosis. The National Academies concluded in 2015 that most people will experience at least one diagnostic error in their lifetime. That is a committee conclusion, not a measured rate. ³¹ The same report names communication gaps among clinicians, patients and families, and care that crosses organizational boundaries, as contributors. ³¹ One estimate put outpatient diagnostic errors at about 5% of US adults a year, roughly 12 million people, with about half potentially harmful. That figure is an extrapolation. ³²

None of this is an indictment of clinicians. It is a description of the room the patient walks into.

If we are going to ask patients to hold the picture, we should look honestly at how well the current system equips them.

• Understanding. At one academic medical center, among patients 65 and older discharged home after a heart attack or unstable angina, heart failure or pneumonia, 95.6% said they understood why they had been hospitalized, but only 59.6% accurately described their diagnosis in a follow-up interview. ³³ In a study of 140 patients or caregivers after an emergency department visit, 78% showed less than complete understanding (compared with the chart) in at least one of four areas, and patients perceived the difficulty only 20% of the time when it was there. ³⁴
• Health literacy. In the last national assessment, in 2003, 12% of US adults were rated proficient in health literacy, and 36% were at basic or below. ³⁵ I found no newer national measure.
• Medicines. In a review of 54 studies of patients leaving hospitals for the community, the median rate of unintentional medication discrepancies after discharge was 50% (across the 11 studies that reported it). ³⁶
• Results. A systematic review of tests in outpatient care found that between 6.8% and 62% of lab tests were not followed up, depending on the study. ³⁷ In one study of records from 23 primary care practices, 7.1% of abnormal test results showed an apparent failure to inform the patient or to document that the patient was informed (0% to 26.2% by practice). ³⁸

Note what these studies say about the patient's role. They do not say patients are the problem. To me they suggest that a spoken conversation alone is a fragile way to move important information. We think a written, current, checkable page could help; that has not been tested.

And patients do want the information. In a 2022 survey of portal users at four academic medical centers (18.8% responded), 95.7% of those who answered preferred immediate release of their results, including 95.3% of those whose results were not normal. ³⁹ Among people who used a portal in 2022, 90% used it to look at test results. ⁴⁰

Federal law has changed more than most people realize.

• You have a right to your records. Under HIPAA a provider must act on your request for your records within 30 days, with one possible 30-day extension, and in electronic form if the records are held electronically and are readily producible in that form. ⁷ For electronic records, you can also ask for a copy to be sent to a person you designate, with a signed written request. ^{7, 8}
• Blocking is prohibited. Since April 2021, rules under the 21st Century Cures Act bar providers, health IT developers, and health information networks from practices they know (for developers and networks, know or should know) are likely to interfere with access to your electronic health information, unless an exception applies. ⁴¹
• Many health plans must let apps in. Since 2021, Medicare Advantage, Medicaid, CHIP and federal-marketplace plans must make your claims and some clinical data available to apps you choose. ⁴²
• Most of us have a portal. In 2024, 65% of US adults were offered and used a patient portal in the past year (up from 25% in 2014). ⁹

But look at what happens after the door opens. In the same 2024 national survey, 59% of adults had records or portals at more than one organization, and only 7% used an app to bring them together. ⁹ Among people who used a portal in 2022, only about one in five used it to send information on to someone else. ⁴⁰ The records are increasingly available to those who use portals. They are not assembled, and they are not pointed at your next decision. Doors are open, and the rooms behind them are still separate.

In early 2023 we wrote a business concept for Phi Longevity. Looking back, it was a plan built around the delivery side: a coordinated group of practitioners, one shared back office for scheduling, billing and records, and economies of scale so that clinicians could focus on care instead of administration. The problems it listed were real: small practices without scale, services that were hard to reach, and records that were not integrated, which forced clients to be "the central point of records management." But look at the order of those ideas. Even while naming the patient's burden, we were solving for the efficiency of the people serving them.

In 2024 we ran a small pilot. We are not going to describe results here, and there are no in-person programs today. What I can say is what it taught us. The care was only half of the work. The other half was coordination: every person on the team needed the same complete, current picture, keeping that picture together was manual and constant, and the one person in every room was the client. We watched people turn into their own care coordinators, carrying folders between appointments and telling their story again to each new clinician.

That is the lesson that changed the plan. Optimizing delivery efficiency makes the pipeline better. It does not change who is standing at the end of it. If the person at the center of the picture is the only constant, the system should be designed around that person: their record, their questions, their time, their decisions.

This is not a new idea. In 2001, the Institute of Medicine defined patient-centered care as care "respectful of and responsive to individual patient preferences, needs, and values," and listed it among six aims for the health system: safe, effective, patient-centered, timely, efficient, and equitable. ⁴³ The same report set rules for redesign that read like a description of situational awareness: patients "should have unfettered access to their own medical information" and the system "should anticipate patient needs, rather than simply reacting to events." ⁴³ It described patients' stories of fragmented care in which relevant information is lost, overlooked, or ignored. ⁴³

That was a quarter century ago. A landmark 2003 study (care in the late 1990s) found that adults in 12 US metropolitan areas received 54.9% of recommended care. ⁴⁴ That study is old, and I did not find a newer national figure.

What does an honest case for patient access look like? Let me give you the evidence, including what it does not show.

• When patients were invited to read their clinicians' notes, 99% of those who opened a note and answered the survey wanted it to continue, and between 59% and 77% (depending on the site) said it helped them feel more in control. The survey response rate was 41%, and the benefits were self-reported. ⁴⁵ Seven years later, in a survey of 136,815 invited portal users (21.68% responded), 72.62% of the note readers who answered rated reading notes "very important" for taking care of their health. ⁴⁶
• Patients who read notes also spot errors. Among 22,889 readers, 21.1% reported a mistake, and 42.3% of those called it serious. These were perceived mistakes, not verified against the chart. ⁴⁷
• Beyond that, the evidence that access improves health is thin. In one health system, patients invited to read notes had somewhat better adherence to blood-pressure medicines (79.7% versus 75.3%) but not to cholesterol medicines. ⁴⁸ A 2013 review found the evidence that portals improve health outcomes insufficient; a 2021 update found studies on health outcomes generally favorable but evidence on utilization and efficiency unclear. ⁴⁹

The honest summary: patients value access, and they may be able to help catch errors. Whether access alone improves health is not established. That is exactly why a pile of records is not the goal. The goal is a picture a person can use.

Using the three levels:

Level 1, see it. One place for everything: results from the health systems you connect, documents you upload, and devices. The page keeps track of where each result came from. And it shows what is missing, because a gap you cannot see is the most dangerous kind. (In the aviation reports, failing to perceive information was the most common error. ¹⁶)

Level 2, understand it. Each finding is explained in plain language, with the guideline or published research behind it so that you, or your clinician, can check it.

Level 3, look ahead. What is trending, what is due, what to ask at the next visit. You can share the same one-page summary with any clinician you choose, through a link you create, so the next conversation starts from the same page.

This is what the Phi Longevity Care Team Brief is designed to do: connect your records (today through MyChart at participating Epic health systems, or by uploading files), bring them together, and produce one page for your care team, with citations to clinical guidelines and published research. We are not claiming this has been proven to improve health outcomes. It has not been evaluated in a trial.

Many people are already using AI for health questions. In a May 2026 KFF poll, 29% of US adults said they use AI chatbots for health information and advice at least monthly, up from about 17% in June 2024. ⁵⁰ In a separate KFF poll earlier in 2026, 41% of those who used AI for health information said they had uploaded personal medical information such as test results or doctors' notes. ⁵⁰ (The two polls ask different questions, so they should not be compared.)

What AI does well. In a study of 195 questions from a public online forum, a chatbot's answers were preferred over physicians' answers in 78.6% of ratings, and rated empathetic far more often (45.1% versus 4.6%). Those physician replies were short forum answers, and the model was an early ChatGPT version (the 2022 release), so this shows tone and thoroughness, not clinical accuracy. ¹¹

What it does poorly, in the lab and in life.

• In a randomized study of 1,298 members of the public across ten medical scenarios, the language models tested alone identified the relevant conditions in 94.9% of cases, but people using those same models did no better than a control group that used their usual sources. ¹²
• Most published evaluations of large language models in health care (2022 to early 2024) did not use real patient data: in a review of 519 studies, 5% did. ⁵¹
• AI models fabricate references. In 2023, 47% of ChatGPT-3.5's medical references in one test were fabricated; in a 2026 test of ChatGPT-5 on orthopedic guideline prompts, 7.13% of 2,736 references were fabricated and 49.34% were bibliographically accurate. ¹⁰

What these findings share is a pattern: the model can be knowledgeable and the experience can still fail. The authors of the 2026 study point to a breakdown in the exchange: users gave models incomplete information, and models' correct answers did not always reach users, who often chose poorly among the options offered. ¹² Our bet is that a tool working on your own records instead of the open internet could narrow that gap. It is a bet, not a finding. I did not find a randomized trial showing that AI working on a person's own longitudinal records improves outcomes. (I searched PubMed on October 9, 2026, and reviewed the studies cited here.) ^{12, 51}

So the design rules we hold ourselves to are narrow.

1. Your records, not the internet's. The analysis works on the records you connect or upload, not on a general web search.
2. Checkable. Recommendations carry citations to clinical guidelines and published research that you or your clinician can look up.
3. Honest about gaps. When something is missing or too old to score, the page says so, and the action is to find it, not to guess.
4. Questions, not directives. For prescription medicines, it suggests a question to ask your clinician. It does not replace one.
5. You decide who sees what. Nothing is shared until you choose to create a link. You send it yourself, it expires, and you can stop it with one click.

Under HIPAA you can ask a provider, in a signed written request, to send an electronic copy of your electronic record to someone you designate. ⁸ That means you are not stuck choosing between a closed system and a free-for-all.

Our own commitments are simple. We store records in HIPAA-eligible systems under a signed business associate agreement with Google Cloud. We do not sell your health information, and we share it only with the service providers that run Phi Longevity and the people you choose. You can read the details in our privacy policy, our consumer health data privacy policy and our trust page.

I would rather tell you what is unproven than have you find out later.

• We do not know whether giving people better situational awareness improves health outcomes. Reviews of patient portals disagree on health outcomes and are unclear on utilization. ⁴⁹
• We do not know whether AI working on a person's own records helps or harms them in real life. I did not find a randomized trial that answers that. ^{12, 51}
• Many of the numbers in this paper are old, drawn from specific populations (Medicare claims, Colorado practices, one health system), or self-reported. Where I could, I said so.
• Fragmentation findings are associations. Sicker patients may see more clinicians, and these studies may not fully separate that from the effect of fragmentation itself. ²³
• Situational awareness is an analogy borrowed from pilots. The instruments to measure it in medicine are not mature. ¹⁸

These are our suggestions, not medical advice.

1. Ask for your records. You have a right to them, in electronic form if they are kept electronically. Providers generally have 30 days. ⁷
2. Use your portal and look at your results. Most people who use a portal already do (90% of portal users viewed test results in 2022). ^{9, 40}
3. Keep one page. A single current page with your conditions, medicines, recent results and your questions is something you can hand to any clinician. It is the thing I did with a spreadsheet.
4. Ask "what is missing?" At each visit, ask what tests or records would complete the picture.
5. Check any AI answer against a source. If a tool cannot show you where a statement comes from, treat it as a question, not an answer. ¹⁰

The patient is the only constant. That is my thesis, not a slogan: every clinician sees a slice, every system holds a piece, and the one person in every room is you.

Aviation treats information failures as something to design against: the briefing, the instruments and the habits exist so the person in the seat can see. Health care can offer the person in the seat the same thing.

So here is what I would ask you to do. Get a handle on your whole set of records, in one place. Then get a second read on what it all means, so that you walk into your next appointment knowing what to ask. That second read is there to guide your conversation with your doctor, not to replace it.

That is what Phi Longevity is built to help with: bring your records together and get a one-page brief, with citations to clinical guidelines and published research, to take to your care team.