Physician shortages and the rise of AI health chatbots

Patients who can’t get in to see a clinician are less willing to sit and wait. Faced with backed-up schedules, understaffed units and rising out-of-pocket costs, many are opening a browser or an app instead, asking artificial intelligence (AI) tools to explain their symptoms, lab results and treatment options before they ever step foot in an exam room.

That behavior is colliding with a workforce that is already stretched. Hospitals and clinics are struggling to fill shifts, experienced nurses are retiring, and the acuity of hospitalized patients keeps climbing.

As gaps widen, clinicians are seeing more people arrive having already “checked with ChatGPT” or a similar tool, sometimes reassured and sometimes convinced they have a very different diagnosis.

Rosemarie Aznavorian, D.N.P., RN, CENP, CCWP, CCRN, executive vice president of client services and chief clinical officer at MedPro Healthcare Staffing, sees both sides of that equation. As a longtime nurse leader, she works with health systems that rely on supplemental and international staff to keep beds open and services running. At the same time, she hears how patients are using AI to fill in what they perceive as access and information gaps.

Medical Economics sat down with Aznavorian to learn more about how staffing shortages are shaping patient behavior, why more people are turning to generative AI for health advice, and what risks that poses when self-diagnosis replaces in-person assessment.

The following transcript was edited for style and clarity.

To start, could you introduce yourself and MedPro?

Certainly. I’m the executive vice president and chief clinical officer at MedPro Healthcare Staffing. We provide supplemental staffing to facilities, whether they’re standalone or part of health care systems, as well as freestanding labs.

We primarily have two service lines within our organization. The first is what facilities are very much used to, which is domestic travelers – for example, 13-week traveling nurses or 13-week traveling radiology technicians who come in and help supplement hospitals when they have a need.

The other service line is international staffing, which is primarily what I oversee. We bring in foreign-educated professionals – nurses, medical technologists and physical therapists – from other countries who meet United States criteria for clinical practice, to help provide a long-term workforce solution versus an every-13-week turnover. The goal is for them to join as part of the core staff when their assignments are complete.

From what you’re seeing, how directly are staffing shortages translating into longer wait times or lost continuity of care for patients?

Significantly.

There are a couple of pieces here. The first is if you’ve reviewed, or are interested in, the Institute of Medicine report on the future of nursing — the first one was done back in 2010 and the second one was done five years later — they‘ve really painted a picture of what was coming that would significantly impact patients receiving care.

There are a number of reasons. Nursing school enrollment is significantly lower, so we’re producing fewer nurses coming out of programs. A lot of us are aging out and getting ready to retire, and the pandemic increased the number of early retirements. When those experienced nurses leave, they take with them all their experience in caring for acutely ill patients.

At the same time, patients are getting sicker and more acute, which means they’re waiting longer to access care. When hospitals are not staffed the way they should be — based not only on volume but also acuity — nurses are stretched very thin, and that creates a few risks.

First, there can be missed care: care that wasn’t able to be delivered because there wasn’t time, or because a sicker patient needed that nurse’s attention. If nurses are being asked to work overtime, workforce fatigue can set in, which can also lead to medical errors.

And then, of course, wait times increase — in the emergency department, waiting for care, waiting to go to the operating room for surgery, waiting longer to get tests done. All of that immediately impacts the level of care provided.

On top of that, hospitals are under financial pressure. Insurance reimbursement is less optimal than what hospitals would like, and the cost of care continues to rise. All of those factors together have created a perfect storm.

When appointments are delayed because of shortages, are physicians actually hearing patients say they’ve turned to tools like ChatGPT Health instead for medical advice?

They’re doing a number of things to avoid going to their physician’s office if they can’t get a reasonably timed appointment, urgent care visit or emergency department visit. They’re using AI to self-diagnose, which many times can lead to misguided information.

It might tell them their issue is not cardiac-related when it could be a cardiac-related issue, because so many signs and symptoms mimic one another. That requires a clinician to do a clinical assessment, because sometimes patients present differently than what the algorithm suggests.

The advantage of AI is that it can raise questions that help educate patients about what they should be asking their physicians or care providers. That can support a full, detailed clinical assessment, and help them understand potential treatments and complications.

So, there are positives and negatives to it.

Where are physicians most often encountering patients who say they’ve used these tools? Is it so-called “hospital deserts,” rural settings and safety-net clinics, or is it in larger outpatient practices, as well?

Actually, it’s all of those.

In rural communities, many times patients are underserved in terms of care providers and feel the financial impact of receiving care. They might be uninsured or underinsured, or they might not be able to afford copayments, so they wait and wait and wait. That’s one component.

The other component is in larger facilities, because people often wait until it’s almost too late to be seen by a physician or urgent care, and they’re sick enough that they have to go through the emergency department, which overwhelms the ED.

From a staffing perspective, many emergency departments are tightly staffed. Critical care units are tightly staffed. Operating rooms are tightly staffed. All of that compounds the access problem.

Are there particular specialties or care settings where staffing shortages are more strongly pushing patients toward AI-based care?

I wouldn’t say facilities or specialties are pushing them toward AI. It’s more about how easily accessible AI is now and how common it’s become.

Patients are self-diagnosing, which is not always the best thing to do because, as I mentioned, symptoms can mimic each other and you can get misguided information from tools like ChatGPT.

Preventive care is probably the most critical piece. If patients can maintain their health by seeing their primary care physician — or, in underserved communities, their nurse practitioner — that would help them get ahead of problems.

Many rural areas have underdiagnosed hypertension, diabetes and other chronic conditions that don’t show obvious signs and symptoms until they’re already impacting activities of daily living. If those aren’t picked up early, patients are more likely to turn to AI when they finally feel unwell.

How worried should physicians be about patients using GenAI tools as a substitute for access to care, rather than a supplement to it?

We should be worried, and we need to be acutely aware.

Patients will come into their physician’s office or the emergency room and say, “I checked ChatGPT and I’ve done X, Y and Z, and this is what it’s telling me it is.” If the diagnosis is actually different, that can cause some dismay, because they’re going to receive care they weren’t expecting.

We need to be aware of that and not be judgmental when we’re talking to patients. That’s really important, because they’re trying to find information about their clinical problem so they can either manage it at home and avoid an emergency or urgent care visit, or at least understand it better.

So the key is not to be judgmental when you’re asking questions.

A recent OpenAI report says that 5% of all messages sent to ChatGPT over the last year were related to health care. Does that reflect unmet clinical needs, patient frustration, simple curiosity, or a mix?

Probably one of all three — you hit the nail on the head.

There’s curiosity about their own care. Since the advent of electronic health records, results may be posted to the portal or lab site before the provider has interpreted them. Patients see a lab result and start asking, “Is this in a normal range? What does this mean?” and then they go down the rabbit hole.

They also look for alternative methods of treatment. If they’re told they need their gallbladder removed, for example, they may ask: What does that mean? Are there different ways to manage it? Is it a full operation? Can it be done robotically? Is there anything I can do with diet or medication to avoid surgery?

So there is an education component, which is good — you have a more educated patient when they come in. But it can be detrimental if they only use AI as their guidance.

What risks does this trend pose, especially when patients act on AI advice before they’ve ever seen a clinician?

Misdiagnosis is a big one. Patients can self-diagnose, and it may be the wrong diagnosis. They may try over-the-counter treatments that don’t work or are contraindicated for their condition, and that can cause side effects.

As I mentioned, AI can be a good educational tool. Nurses — who have been rated the most trusted profession for 15 or 20 years running — need to help patients take the helm. Patients are in charge of their care, but we need to provide them with information so they can make appropriate decisions for themselves.

Looking ahead, do you see staffing solutions as one of the most immediate ways to slow the shift toward AI health advice, or is patient reliance on these tools now baked in?

Staffing is really, really critical.

If hospitals don’t have the number of nurses they need based on acuity, that’s a problem. It’s not just about ratios. Some states have mandated ratios — one nurse for a certain number of patients — but more importantly, it’s about how sick the patients are.

You might have 10 patients on a unit who are all getting ready to be discharged, so there’s less care to deliver. On another unit, you might have patients who are sicker, post-op, just transferred out of the ICU or on telemetry. That’s going to require more nurses.

If you have 10 patients, four of whom are telemetry patients and six of whom are getting ready to be discharged, you might need fewer nurses because of the acuity mix.

But if patients are treating themselves before they get there, or attempting to, they’re going to be sicker when they show up. Then you’re going to need more nurses to care for them because of how acutely ill they are.

Is there anything else you want to share?

It’s important for facilities to look at health care staffing companies as a consultative arm and see how we can help.

I’ve been a nurse for 47 years and was a chief nursing officer for a very large health care system in Texas for 10 years, so I’ve been on that side of the house. I actually used MedPro as part of my long-term workforce solution before I came here.

In one of my publications, I coined what I call the 80/10/10 rule. About 80% of a hospital’s staff should be their own — full-time, part-time, per diem or float pool.

Another 10% should be from the international side, because those nurses and medical technologists are with you for 36 months, so they become part of your core staffing and are dedicated to the mission, vision and values of the organization.

The remaining 10% should be those 13-week assignments we talked about, to cover things like family and medical leaves, a new physician and service line or a new tower that needs to be staffed while you hire permanent staff.

If all the stars align, 80/10/10 works. But if a facility has a significant number of vacancies and only 65% of their own staff, they have to ask: How much international staff do we need for our long-term solution, and how much do we need short term?

Health care staffing companies can partner in that. Part of the reason I came to work for MedPro when I moved back to Florida is that we are clinician-led and clinician-run. Our CEO is a nurse. There’s myself and my counterpart, Patty Jeffrey, RN, the other EVP — we’re all nurses, and we’ve all been in leadership positions in hospitals. We understand the financial pressures, the acuity and the operational decisions that have to be made, and the need for flexibility with staff coming in and out.

Facilities should really look at staffing companies in a consultative way, rather than only thinking, “I don’t want to spend the money on staff who aren’t my own.” What’s the risk if you don’t have enough nurses or medical technologists? You won’t have the revenue because you can’t provide care for the patients.

It can be a vicious cycle: Do you spend the money to have additional staff, or not? If you don’t, what happens? Missed care, risk-management issues, falls, poor patient outcomes, increased morbidity and mortality.

Accrediting bodies like The Joint Commission, CMS and state surveyors will look at the level of care being delivered. There’s a lot at risk if you’re not providing the appropriate amount of staff when you can.