Doctors Using Doubao in the Clinic

On October 25, in a cardiology consultation room at a hospital in Xuzhou, Jiangsu, a doctor opened Doubao while listening to a patient describe their condition. In the end, the patient refused the prescription the doctor wrote.

After the matter drew attention, the hospital where the doctor works verified it and provided a piece of information that the patient's family member's camera had missed: the doctor was checking a new drug, found it unsuitable for this patient, and did not issue any new prescription that day.

According to industry interviews and media citations, Doubao's average daily health-related consultation volume is estimated at the level of 20 million visits (ByteDance has not officially disclosed an exact figure), equivalent to moving the traffic that once went to "Baidu Health search" into a chat box.

According to a survey of 647 medical workers released by DXY, 98.5% of doctors had used AI in their work in the past three months, 70.8% used it at least several times a week, and 81.4% had used Doubao.

In those few seconds when the consultation room paused, a communication that had originally belonged only to the doctor and the patient was intruded upon by an invisible, intangible third party, and the traditional doctor-patient relationship was quietly changed.

Among Dr. Hong's patients was a 52-year-old middle-aged woman with an 8-year history of rheumatoid arthritis who had long been on oral methotrexate tablets, with her clinical symptoms well controlled.

That visit was because the patient said she had experienced frequent and urgent urination in recent days. After a series of outpatient tests, the results showed a bacterial infection, and Dr. Hong considered it a simple urinary tract infection.

Dr. Hong said that at the time his first judgment was to prescribe symptomatic compound sulfamethoxazole. "This is one of the first-line drugs for the disease, and it is cheap and effective," Dr. Hong told Xianwei Gushi.

Halfway through writing the prescription, Dr. Hong suddenly remembered that this patient had previously used methotrexate tablets, and from memory it seemed that using these two drugs together might involve a certain antagonistic reaction.

Image | The package insert for methotrexate

"Methotrexate is mainly secreted and excreted through the renal tubules, and trimethoprim in compound sulfamethoxazole also goes through this route, so in theory they compete in excretion and raise the blood concentration of methotrexate. But how great is the clinical risk? Should it be avoided as much as possible, or can it be used with close monitoring of blood counts?"

In the past, in such situations, Dr. Hong would consult the pharmacopoeia to confirm. But that day the pharmacopoeia happened to be in the inpatient department office, not at hand, so Dr. Hong instinctively took out his phone and entered into Doubao: "methotrexate compound sulfamethoxazole interaction bone marrow suppression."

Doubao quickly gave several lines of prompts. One of them was very clear:

Compound sulfamethoxazole can significantly increase the blood concentration of methotrexate; combined use raises the risk of bone marrow suppression, and in severe cases can cause pancytopenia; combined use is recommended to be avoided;

Another prompt was that if combined use is necessary, blood counts and methotrexate blood concentration need to be closely monitored, and dose reduction should be considered.

There was also one she still remembers to this day: this interaction also exists in low-dose regimens; it is not only high-dose chemotherapy that requires vigilance.

So Dr. Hong deleted compound sulfamethoxazole and switched to another antibiotic. "I did not use it to prescribe or make a diagnosis; this tool is more like a portable pharmacopoeia."

"Look, doctor, Doubao really did say so"

Unlike Dr. Hong, Dr. Luo in the respiratory department has recently been most troubled by patients repeatedly using results they searched on Doubao to "interrogate" him.

What left the deepest impression on him was a middle-aged male patient in his fifties this year.

The patient discovered an approximately 6 mm pure ground-glass nodule in his right lung during a physical examination organized by his company.

Dr. Hong looked at the other party's CT examination report, which showed that the nodule had clear boundaries, uniform density, and no solid component.

In Dr. Hong's diagnosis, according to the consensus on the diagnosis and treatment of pulmonary nodules, for a pulmonary nodule of this degree, regular follow-up and periodic repeat CT are the first choice.

But the patient first handed over his phone. On the screen was a Doubao conversation: "Ground-glass nodules may be early lung adenocarcinoma; early surgical resection is recommended."

"Look, Dr. Luo, I checked, and Doubao says it may be cancer. I want to just cut it out."

If this had been a few years earlier, Dr. Luo would certainly have said to the patient, "Whatever they say online, you believe it?"

Dr. Luo pulled up the patient's chest CT, located the nodule, began explaining the current situation to the patient in detail, and finally even guided the patient to revise the question asked to Doubao so that it fit his own condition more closely.

The patient made a change, and the answer on the screen changed, with the conclusion roughly the same as Dr. Luo's recommendation.

"AI is not wrong; sometimes it tailors its answers to the person asking. For the same question, when we doctors ask, it gives targeted guidelines; when a patient without a medical background asks, it may be a broad, general answer," Dr. Luo explained to Xianwei Gushi.

But this approach still did not reassure the patient, because a neighbor he knew also had a pulmonary nodule, and surgery showed it was cancer.

To further reduce the patient's anxiety, Dr. Luo had to break down and thoroughly explain the information he knew to the patient:

At present the probability of malignancy is not high, and follow-up is the more reasonable choice; having surgery now would remove a lifetime's worth of lung function in exchange for what may be a nodule that "never needed to be removed in the first place."

Even with such patience, the patient was not persuaded on the spot, and then pressed Dr. Luo further: "If the nodule keeps growing during follow-up, then what?"

That day in the outpatient clinic, Dr. Luo spent nearly half an hour on this patient.

"If every patient comes in holding content they searched on Doubao to discuss with me, then overall medical efficiency will become slower and slower," Dr. Luo said.

In the same consultation room, the same tool, in the hands of doctors and in the hands of patients, triggers completely different trust responses.

The "pretty words" in the case records

Not long ago, a hot image also appeared online: a note was posted at the door of a hospital outpatient clinic that read, "Patients who have already obtained a diagnosis from Doubao, please do not seek advice from us again."

Image | An image circulating online

Speech-to-text, structured first drafts, and reminders for missing items can significantly compress documentation time, and the standardization of medical records is also more stable.

"The AI in the medical record system is indeed getting stronger," Dr. Chen said. After some hospitals launched large-model medical record systems, writing efficiency increased severalfold, and the rate of Grade A medical records rose to over 97%.

In addition, previous media reports noted that after a hospital in Wuhan put AI voice medical records into use, doctors saved nearly three hours per day on average, and the medical record pass rate rose to 99%.

Dr. Chen told the younger doctors in his department about his experience handwriting medical records when he was a resident. In his time, during daily ward rounds, changes in a patient's condition had to be recorded personally by the attending physician—even if that record ultimately consisted of only one line.

For example, if the patient only said at the bedside, "I had a fever of 39 degrees last night," AI can quickly use the context to fill in the patient's onset acuity, fever pattern, and accompanying symptoms, each looking quite convincing.

"Our quality inspection system is also using AI, but at this stage, in medical record quality inspection, what AI checks is whether the medical record structure is complete and whether the logic is self-consistent," Dr. Chen said. "As for whether this sentence actually reflects the patient's situation in the consultation room, AI cannot determine that."

"No matter how standardized a medical record is written, if there is one false statement in it, it is a false medical record. The quality control score will not tell you this; when something happens to the patient, a lawyer will tell you."

The regulatory line has in fact already been drawn: the Detailed Rules for the Supervision of Internet Diagnosis and Treatment, implemented on a trial basis in February 2022, clearly states that prescriptions must be issued by the treating physician personally, and the use of artificial intelligence and other automatic prescription generation is prohibited. Since then, Beijing, Shanghai, Fujian, and other places have followed up one after another, with consistent standards.

Health and efficiency should not appear on opposite ends of the scale

In the several hospitals we visited, doctors all said that from a personal perspective, the use of AI is not a monster; the existence of AI tools such as Doubao, Yuanbao, and Qianwen is actually helpful for doctors to look up information, and the key is that doctors themselves must use their professional ability to make effective judgments.

Take merely looking up medications as an example: when a doctor encounters an uncertain detail in clinical work, stopping to "consult the books" is itself nothing new.

Dr. Luo also found a way to get along with patients who use Doubao to seek medical care.

"Now doctors, patients, and AI feel like a triangular relationship, and to make this relationship stable, doctors' explanations are needed. If doctors do not explain in enough detail, patients will go find answers themselves. And the direction they find may not lead to what is correct. In the final analysis, one still needs to be strong oneself."

But he still firmly believes that writing medical records has never been clerical work; it is a training ground for clinical thinking.

Only as young doctors repeatedly deliberate over how to summarize the chief complaint, which details to retain in the history of present illness, and in what order to arrange the differential diagnosis can they possibly gradually grow a path of their own.

(At the request of the interviewees, except for public figures, all persons appearing in this article use pseudonyms. The medications and diagnostic results appearing in the text are only for the needs of drafting this article and cannot be used as any auxiliary diagnostic information.)

This article comes from the WeChat public account "Xianwei Gushi" (ID: xianweigushi), author: Tiantai Wuyao, published by 36Kr with authorization.