When her patient needed canakinumab for gout, Julia Buchfuhrer, DO, struggled to get prior authorization approval. The insurer denied her repeatedly.
Trying a different tactic, she used an artificial intelligence (AI) app to file the claim. It was approved in 24 hours. “Why this happened, I’m not sure. My guess is our AI understood what their AI system was looking for. It formatted the authorization exactly how they wanted it. It spoke their language,” said Buchfuhrer, founding rheumatologist at the JB Arthritis and Rheumatology Center in Downey, California.
Like other specialists, rheumatologists struggle with time management. Staff and physicians alike spend lots of time answering phones, responding to patient messages, and securing insurer approvals for expensive medications. All these factors can affect quality of life. A study of 128 participants attending the 2019 Rheumatology Winter Clinical Symposium found that 51% of respondents — and 54% of physicians specifically — met burnout criteria in at least one Maslach Burnout Inventory domain.
To improve practice efficiency, many rheumatologists are increasingly relying on AI to automate clinical documentation, reduce time spent on electronic health records (EHRs), and streamline workflows.
Heidi, Doximity (DoxGPT), and OpenEvidence are among the AI platforms with ambient AI scribe technology that are gaining traction in small and larger practices as physicians discover the benefits of using them. Paul H. Sufka, MD, a rheumatologist in Twin Cities Orthopedics in Minnesota, has sampled and used all these platforms. “As a techie, I’m willing to give anything a try,” he said.
Research has shown that using an ambient AI scribe leads to a moderate increase in weekly relative value units and patient encounters, without any increase in claim denials.
If a patient needs a personalized letter, Samy Metyas, MD, uses AI to draft it. When sending letters to other doctors, the AI can take the visit notes and convert them directly into professional correspondence, said Metyas, medical director of the Covina Arthritis Clinic in Covina, California, and clinical associate professor of rheumatology at the University of Southern California in Los Angeles.
“In a small practice like ours — with seven providers, including three physicians and four physician assistants — AI tools are especially useful for administrative tasks, documentation, scribing, and workflow management,” he added. They can provide insights into rare cases, assisting with drug approvals, and supporting insurance authorization processes.
AI tools also save his colleagues roughly 3-4 hours of note-taking every day.
Nilanjana Bose, MD, MBA, a rheumatologist with Lonestar Rheumatology in Houston, loves the versatility AI provides in drafting letters. You can modify the content and tone and make it sound very professional, she said. “It has significantly increased efficiency of our clinical staff, and they love it as well.” Additionally, an AI bot enables patients to submit documents directly into the EHR.
“In the past, the staff would have to download the forms and upload it, but now it’s automatically uploaded. This does save a few clicks for each form,” Bose said.
Small practices do not need a large, centralized IT system to use these tools, so they can be simple to implement, said Metyas, whose practice is waiting for EHR software updates to fully integrate AI features. Cost can be a consideration, “but the biggest advantage will come from full integration with our existing software,” he said.
How One Small Practice Uses AI
Zachary Fellows, MD, MPH, owner of Synergy Rheumatology and Wellness in Carlsbad, California, has come to rely on AI to streamline operations at his small practice. A custom-built system he developed automatically processes incoming faxes through Google Gemini — a large language model — to prioritize and schedule patient referrals more efficiently.
The system analyzes each fax to identify key details: where it came from, who sent it, and what type of document it is. It determines whether it’s a consult note, a referral, spam, or even a sales pitch and then prioritizes everything for the team.
For example, if it recognizes a referral, the AI will extract the relevant information and add it to a Google sheet. “My scheduler can access it immediately and call the patient. Often, we’re able to contact and schedule patients before we’ve even received the full consult notes from the primary care physician. As soon as authorization comes through, we’re already on the phone,” Fellows said.
Scribing tools also save time. “When we bring a patient into the room, we give them a quick heads-up and say, ‘We use an AI tool that listens to our conversation. It’s not watching us — just listening in the background.’ No one’s really had an issue with it,” Fellows said.
Using AI means he can relax and spend more time with patients instead of spending a large part of the visit trying to document and capture every detail. “I still jot down notes here and there to keep my thoughts organized, but all those small details from an hour-long conversation are accurately captured and stored in a note formatted exactly the way I like,” he said.
Simplifying Prior Authorization
AI tools can facilitate difficult prior authorization and insurance-related tasks, as Buchfuhrer demonstrated with her rejected claim.
Rheumatologists prescribe many medications off-label. “There are rare autoimmune diseases with no [FDA]-approved treatments. A lot of the time I’ll send in a prescription fully expecting a denial so I can appeal it,” Sufka explained.
Doximity is an AI tool that’s useful for handling appeals, he continued.
“I can enter general patient information — what condition they have, which treatments they’ve already tried, and the severity of their symptoms — and it generates a strong appeal letter with appropriate citations. It does a good job now without making up sources, and I can send it directly to the insurance company,” Sufka said. His practice has been exploring new software and is planning a demo to see if it could integrate AI with the EHR to pull patient data automatically for prior authorizations.
Buchfuhrer uses a prior authorization AI tool called Tandem, which she says “has been a game changer” in streamlining the prescription process.
Previously, the pharmacy would initiate the authorization, and her staff would log it in to finish it, dealing with missing information, slow feedback, and confusion about where denials were sent. “Sometimes denials go to the patient, sometimes to the pharmacy, and occasionally to us. When they do come to us, I have to write and fax a lengthy appeal letter. It’s a very frustrating and inefficient process,” she said.
With Tandem, Buchfuhrer sends the prescription directly to the company. “We have HIPAA [Health Insurance Portability and Accountability Act] contracts in place, and the AI pulls information from my notes and the patient’s insurance details. It automatically creates a complete, accurate prior authorization based on my documentation,” she said.
She’s notified directly of approval or denial and can view the real-time status in a portal. When a patient asks about their medication, instead of saying, “I don’t know, let me call the pharmacy or insurance,” a process that can take up to 30 minutes, she can log in and know the answer within seconds.
“Even better, because I get information so quickly, we can act immediately. If there’s a denial, the AI generates a draft appeal letter that I can edit and send with one click. There’s no more reading through letters, typing responses, and faxing paperwork. It’s simple and efficient,” she said.
Another helpful feature is when a denial states that a medication isn’t on the preferred formulary and lists alternatives. “The system tells us which drugs are approved, and we can select one and submit it instantly,” she said.
“The very day I implemented Tandem, I had two prescriptions approved within 24 hours. That has never happened to me before — ever,” she added.
Large healthcare systems such as Kaiser or Sutter have entire teams dedicated to prior authorizations and pharmacies built into their system. In private practice, physicians pay staff directly and don’t have dedicated teams. “So this tool has a much bigger impact for us,” she said.
Mobile Apps Help With Patient Encounters…
AI can help with diagnosing patients and assessing uncommon or rare cases.
Metyas has found that OpenEvidence, an AI mobile app, is useful for getting the latest information and articles about certain diseases or medications. The app offers comprehensive, evidence-based information, drawing from published research articles. “When I want to learn more about a specific condition or clinical problem, I can enter the details and receive well-organized, reliable guidance. I can even input a patient case, and the app helps me work through it in a thoughtful way,” he said.
OpenEvidence is not designed to treat the patient or predict response to a drug, Metyas clarified. By expanding his approach to differential diagnoses, it helps him consider better treatment options for patients.
…and in Screening Candidates for Jobs and Fellowships
Jeffrey Curtis, MD, MS, MPH, a professor of medicine in the Division of Clinical Immunology and Rheumatology at The University of Alabama at Birmingham (UAB), uses AI to find and hire staff. “When I’m interviewing research coordinators, medical assistants, or other team members, I can upload their resumes and tell the system exactly what I’m looking for,” he said.
He might say, “Here are 6-10 resumes. Please review them based on these specific criteria which I provide and then summarize how well each candidate fits the role.” Curtis provides the job description, has AI rank the candidates, and then asks it to generate 5-10 targeted interview questions for him to ask. This leads to more informed decisions on who to hire, he said.
Similarly, AI can identify optimal candidates for rheumatology fellowships. UAB receives about 200 applications each year for fellowship positions. Each one is about 40 pages long, containing narrative information from their residency programs and letters of recommendation, plus roughly 2700 individual data points.
Reviewing that volume of information takes an enormous amount of time and effort. Even if a program director spent just 15 minutes on each application, it would take at least a week to screen all the candidates simply to decide who to offer interviews to, he said.
To address this, Curtis and colleagues built an AI pipeline that pulls in both narrative information and quantitative data, such as board scores. The system can efficiently prescreen applicants and suggest which candidates to invite for interviews. Presenting the findings at the American College of Rheumatology (ACR) 2025 Annual Meeting, Curtis and colleagues reported that the AI pipeline tool reduced the manual process of prescreening applicants from 50 to 2 hours in total.
The AI Assistant Will Speak to You Now
Ask any doctor what most annoys them about practice life, and they’ll likely say, “The ringing phones.”
Buchfuhrer’s practice is recruiting AI receptionists to address this problem. Patients often tell her, “I tried calling several times to cancel and couldn’t get through.” Sometimes the phones are quiet, and other times four patients call at once, she said.
There are simply not enough hands to handle all these calls, she said.
“The initial solution was hiring virtual assistants overseas, which I did, and it helped. But as your practice grows, call volume increases, and you need more staff,” she said. Studies show that about 50% of calls are scheduling-related: new appointments, follow-ups, cancellations, or rescheduling.
Her practice is about to go live with an AI receptionist, a measure that could reduce call volume to that 50%. If a patient asks something outside the bot’s scope, the call immediately transfers to a real person — either a virtual assistant or someone at the practice.
This way, no one gets stuck talking to a bot that can’t help them, she said.
Metyas uses the OpenEvidence app to assist with last-minute appointment cancellations.
When a patient cancels, the AI system can quickly fill that slot, which prevents lost revenue and improves efficiency. Cancellations are one of the biggest challenges for a practice, especially when the waiting list can be 3-4 months long.
Being able to bring in patients from the waitlist the same day or the next day makes a significant difference, Metyas said.
What the Future Holds for AI
As AI improves and integrates more deeply with the EHR, Buchfuhrer anticipates it will be able to handle more tasks such as medication refills. “For example, the AI could check if the patient was seen recently, if my notes support continuing the medication, and if labs are acceptable. That’s likely the future,” she said.
Sufka envisions a time when he can log into the EHR, and it immediately starts listening to him and the patient. “For now, let’s assume it’s mostly voice-based, and we just have a natural conversation. The system takes notes and either pulls in what I would typically order for that type of patient or makes suggestions,” he said. If he wants to make a change, he can say it out loud, and the EHR updates instantly. This could also apply to labs, imaging, even scheduling.
He might say, “Let’s follow up in about 3 months.” The AI-powered EHR system could respond with available dates, finalizing it right there in the room.
Curtis offered that it won’t be long before that same type of AI-EHR system might automatically review an EHR note, offer suggestions on alternative diagnoses, propose additional diagnostic testing, and make treatment recommendations.
Beyond the EHR, AI is also being incorporated into some interesting technologies, Curtis said. Combined with smartphone images from modern cameras, patients with rheumatoid arthritis can take a picture of their hands, and AI can detect whether joints are swollen. AI-enhanced ultrasound devices are being developed that no longer require a physician or medical technologist to perform the scan, nor to provide an interpretation. AI powers the machine itself.
Pairing AI with novel diagnostic tools such as ultrasound, MRI, and thermal imaging offers tremendous promise to standardize diagnostics, provide accurate results in real-time, and improve practice efficiencies. “AI has the potential to do far more than we’ve been able to achieve in the past — especially with advanced computer vision and sophisticated algorithms applied to medical imaging,” Curtis said.
AI Faces Some Roadblocks, Like Human Conversation
But there’s limits to what AI can do, Fellows emphasized.
There’s the “walled garden” problem, in which different AI and software solutions are unable to communicate with each other. Specifically, there’s an inability to integrate with existing systems such as EHRs. This lack of application programming interface or API access prevents direct document insertion into patient charts, which could significantly improve efficiency, he said.
Fellows also “draws the line in the sand” on AI call centers. “Part of our practice’s success has been distinguishing ourselves from these large hospital systems where it’s difficult to get a human on the line that listens to you or knows you. If you’re talking to an AI, there’s no human connection behind that. It’s a poor substitute for the human front desk or the human medical assistant,” he said.
AI assistants can save time, but they’re not perfect, Buchfuhrer acknowledged. Patients can get upset when they realize they’re not speaking to someone in the office. With AI, people may worry: “Is this even going to help me?”
What’s critical is that the bot sounds human and quickly recognizes when it can’t help — then transfers the call to a real person, she said.
Some studies challenge the notion that AI is a time saver. One JAMA Network Open survey of 122 physicians that was conducted in 2023 found that AI-generated drafts for patient messages submitted to the EHR inbox did not shorten physician response time and reply length. While the response to AI was mostly favorable, “the doctor still had to read the message, review what the AI composed, and in many cases, revise the draft AI response,” Curtis said.
While it reduced cognitive burden and often drafted more empathetic responses than the provider might have written themselves, it provided no time savings in responding, he added.
AI cannot replace human doctors in areas such as empathy, trust-building, and complex decision-making, Metyas emphasized. “I do not believe AI can replace the trust built between doctors and patients. This is especially important for vulnerable populations, including elderly patients, individuals with cognitive impairment, trauma patients, or those from diverse cultural backgrounds,” Metyas said.
While it can assist in tasks such as analyzing x-rays, AI requires human oversight. It can’t perform physical exams or make independent medical decisions. “We still perform thorough exams, which can reveal important findings patients may not mention, such as rashes, an enlarged liver, or leg swelling. These discoveries can change the direction of diagnosis and management. AI cannot replace this hands-on assessment,” Metyas said.
Ultimately, a doctor should always review AI generated results, sources told Medscape Medical News.
“When we are drafting a letter, we are still going to look at the letter to make sure everything looks good before we send it out,” Bose said.
Curtis likes to think of AI as a helpful intern — someone who’s sometimes right but always needs to be fact-checked. “When I create AI prompts, I focus on chain-of-thought prompting. I don’t just want an answer — I want to see the reasoning behind it. I want the AI to explain why it reached that conclusion and break its thinking into logical steps.” Traceability is also important, he added.
“I want to understand what the sources of information were that AI relied on to reach its conclusions, how credible those sources are, and if the data it was trained on reflect the type of patients I see in my practice,” he said.
Sometimes, when you ask for this kind of explanation, the AI catches its own mistakes or flawed logic. One of the newest approaches in AI, so-called “agentic AI” has the capacity to solve problems autonomously. “This approach transcends how most people use AI, which is to merely ask it to perform single tasks that require constant human prompting,” Curtis said.
The goal of AI is to enhance professional work, Fellows said. “I don’t think we’re here to replace humans, but if we can augment what we do and do better, I think that’s where AI really could shine,” he said.
Metyas reported being a speaker and consultant for AbbVie, Amgen, Janssen, UCB, Sanofi, Mallinckrodt, and ANI. Curtis reported having received honoraria or consulting fees from AbbVie, Amgen, Aqtual, BMS, GSK, Janssen, Eli Lilly, Novartis, Pfizer, Sanofi, Scipher, Sensimetrica, Setpoint, and UCB and research grants from AbbVie, Amgen, Aqtual, BMS, GSK, Janssen, Eli Lilly, Novartis, Pfizer, Sanofi, Scipher, Setpoint, and UCB. Curtis also reported leading several real-world evidence data coordinating centers, including the ACR RISE Registry, the PatientSpot patient registry (formerly ArthritisPower), and the Excellence Network in Rheumatology to Innovate Care and High-impact research practice-based network. Buchfuhrer reported being a speaker and a consultant for AbbVie, Amgen, Johnson & Johnson, and Exagen. She also reported being a paid member of the medical policy committee for Cardinal Health. Fellows reported being a speaker and consultant for Eli Lilly and a member of the California Rheumatology Alliance Board of Directors. Bose reported being a member of the Medical Policy Committee for Cardinal Health. Sufka had no disclosures.
Jennifer Lubell is a freelance medical writer in the Washington metropolitan area.
<
Leave a Reply