How a family doctor uses AI in his practice
The use of artificial intelligence (AI) has become indispensable in everyday life. Artificial intelligence is also firmly established in the medical field, be it in dermatology, oncology, or radiology.
The use of AI in general practice is a new development. Family doctor Professor Wolfgang von Meißner from the Spritzenhaus practice in Baiersbronn in the Black Forest demonstrates how this can work. He has been testing AI in his daily work for several weeks. He reports that the change has been enormous.
The practice has been using an app for several months now, allowing patients to make appointments or order prescriptions. It has now added an AI language model, also known as a Large Language Model (LLM), which can be used in a variety of ways. "It records patient conversations and summarizes them intelligently."
Voice system answers callPatients can also call our MedicBot and order a prescription. The voice system answers the call, records it as an audio file, and the AI simultaneously transcribes the conversation. The MedicBot, as the AI is called at von Meißner, provides a summary of the conversation.
The AI can also access data from the PVS, such as patient data, long-term diagnoses and the medication plan, and prepares the order for the doctor," explains Daniel Teigland from MediTech, who developed the practice app with AI together with general practitioner von Meißner.
This prescription order then goes to the signature pile in the PPS, the prescription can be approved by the doctor, and the patient can redeem it. "The app allows the patient to track every step that takes place in the practice in real time and the status of their prescription order," says Teigland. An employee to answer the call is no longer necessary.
MedicBot prepares prescription ordersThe MedicBot checks the prescription order in advance to see whether the medication and dosage are suitable for the patient or whether the caller may not have spoken clearly and needs to be readjusted.
"The MedicBot prepares everything so that at the end, the doctor can check and approve the medication one last time. If something isn't right or hasn't been understood, the MedicBot will let you know," says Teigland.
Family doctor Wolfgang von Meißner in his practice at the Spritzenhaus in Baiersbronn in the Black Forest.
© Ines Rudel
The second application of AI is recording doctor-patient conversations in the practice. Von Meißner no longer "hides behind his computer" and types the findings during the consultation with the patient. He has his smartphone on the table and records the conversation via the practice app.
The advantage: The doctor can look his patient in the eye, listen carefully, and even perform a physical examination. Von Meißner: "I'm closer to the patient again, thanks to AI."
“The AI can remember everything”When using AI, doctors are forced to question their own self-image, says von Meißner. Because AI listens better. "As a doctor, I only understand what I want to understand, I unconsciously categorize patients, and after a long day, I don't listen as well anymore."
AI is completely different. So it can happen that I read the AI's summary and ask if the patient really said that because I can't remember it. The AI can remember everything and forgets nothing, like a kind of co-pilot for the doctor."
During such a doctor-patient consultation, the MedicBot could in the future intervene and advise the doctor on which direction to follow up. The AI can also provide targeted advice. However, for this, it currently needs more data, says von Meißner.
Summary for patientsA summary is provided after the consultation not only for the doctor, but also for the patient after their visit. "The summary of the consultation that the AI creates for the patient can already be a significant relief in the daily work of the general practitioner," says von Meißner.
“Patients provide positive feedback because they receive clear instructions on how to take a new medication or which symptoms they should report back for.”
The project is being scientifically supported: "We are a founding member of the Supraregional Health Service Research Network (SHRN) of the Institute of General Practice at the University of Lübeck and are cooperating on the topic of AI with the Department of General Practice and Health Services Research at the University of Heidelberg."
Von Meißner is convinced that the family doctor who grows old in the countryside with his patients, who sits at the regulars' table and knows about the affairs in the village, will no longer exist.
Do “multi-practitioner practices” belong to the future?In the future, there will be an increasing number of "multi-practitioner practices" with physician assistants (PAs), VERAHs, MFAs, and multiple physicians. These will know less about their patients. AI can complement and support this, ensuring that the patient sees the right practitioner at the right time.
AI is a practice employee who bears no responsibility but can, for example, read and summarize an entire patient file in a short time. No doctor can do that in a few minutes. And the AI learns along with you if the summary needs to be improved, says von Meißner.
Non-medical staff can prepare anamnesis with AI"As a doctor, I can ask the AI a question and get an answer from the file, not just a list of a searched word," says von Meißner. In the future, the AI will also be able to remember from the data that the patient, for example, had a colonoscopy a year ago and had polyps removed.
If the patient returns to the doctor's office a year later with abdominal pain, the AI can provide the doctor with crucial information. An AI could also know that the patient's brother died of stomach cancer several years ago and thus alert the treating physician to the patient's concerns, says von Meißner.
Another advantage is that the AI not only uses the existing patient data from the PVS, app and conversation, but also writes new data in a well-structured form into the patient file, which can then be accessed, says Teigland.
In addition, non-medical staff could also prepare a high-quality medical history using AI in the future. It's also conceivable that the preliminary consultation with a bot could take place at home, Teigland says.
A major problem with LLM is the traceability of the programs. With the MedicBot, which von Meißner plans to use in his practice, every step is transparent and understandable, even for laypeople, says Teigland.
"We mustn't leave the field to large companies. All professional societies should join forces, and there should be no secrets in AI," says von Meißner. "When training AI, it's important that we use open source data and not data from the pharmaceutical industry. We want to define how models are trained."
“AI is extremely stupid”It's not just about feeding the AI with guidelines or scientific studies, but with concrete conversation examples from which it can learn. And these have to be many different examples. That's time-consuming and expensive. "Because AI is extremely stupid."
Biases must be considered when training AI in the data. "The internet and the data base tend to be misogynistic and racist." This is something to keep in mind, says the family doctor.
As soon as the AI is capable of suggesting specific and regular diagnoses, it must be certified as a medical device. Currently, the MedicBot is a documentation support tool, not a diagnostic tool. "The speed at which we are moving with AI is rapid," says von Meißner.
When using AI, it's important that the data is secure. "All of our data is either stored on private, decentralized servers or directly in the practice," says Teigland. "We have a comprehensive security concept."
A different level of flexibilityThe app is end-to-end encrypted. Practice staff and doctors can work without a VPN. "With the app, the practice is in the doctor's pocket. It's a whole new level of flexibility," says von Meißner.
In the future, AI will support every area of the practice. For example, if a doctor is absent and appointments need to be rescheduled, patients can be called automatically and appointments can be reassigned to other doctors in the practice. There's no need for an MFA to call patients, as everything is handled from the system.
“AI will not replace doctors, but at some point patients will no longer go to doctors who do not use AI,” says von Meißner.
Arzte zeitung
