The healthcare community has been buzzing since the public release of chatGPT last November, and now, only months later, numerous applications of the latest generative AI models are already in use in healthcare settings. New applications include medical record summarization, transcription and recording of doctor-patient conversations with automated clinical note generation, improved billing and claims processes, and generation of referral and authorization letters. So far, generative AI applications automate and improve the efficiency of existing administrative and care-adjacent processes (to reduce costs), but don't actually change the healthcare that providers deliver.

The launch of Dr. A.I.™ 

HealthTap has now deployed a generative-AI solution that actually helps improve the process of care. By leveraging the medical knowledge inherent in the latest and biggest large language models (LLMs), the new Dr. A.I.™ interviews patients, functioning as a physician's assistant, spending the time and gathering the information that the doctor needs prior to each patient visit. Dr. A.I. is integrated in the flow of patient care on HealthTap, so that people who schedule visits in the virtual primary care clinic now can supercharge their visit experience by engaging with the virtual physician assistant before seeing the doctor. 

The new Dr. A.I. gathers relevant information that their doctor will need prior to their virtual consultation. Dr. A.I. considers the age and gender of the patient, and asks medical interview questions based on their chief complaint or reason for visit. Each question is part of a natural conversation, with a back and forth interactive conversation. Dr. A.I. explains what it is asking and why, probing all the areas that might illuminate the underlying problem a patient has. 

When it has completed taking the history, Dr. A.I. summarizes and organizes the patient history for the doctor, so they can review the information in a quick moment prior to the consultation. As a result, doctors can more easily and quickly hone in on the most important and relevant concerns, and still have time to address the real world issues that are so important for doctors and patients to discuss. 
Dr. A.I. launched at the end of September this year, and we are already seeing the results. Doctors are leveraging the Dr. A.I. summary as the core of their clinical history documentation, adding to it when needed to complete the picture in ways that only a human doctor can do. Patients are delighted they have the time to discuss more than just the mundane basics of their symptoms and the proposed treatments. In a preliminary review of performance, 93% of cases reviewed showed that the Dr. A.I. patient summaries included helpful information for the doctor prior to the virtual consultation. 

A sample medical interview and clinical note

Here is an example of a recent Dr. A.I. pre-visit consultation on HealthTap, which demonstrates the detail and thoroughness of the medical history that Dr. A.I. took for this problem. 
The patient is a 69 yr-old male who has a history of mild hypertension treated with lisinopril 20 mg po daily. He says the reason for his visit is "Having trouble sleeping through the night. Getting up too many times" 

Here is the Dr. A.I. medical interview: 
"Hello! I understand that you're having trouble sleeping through the night and you're waking up multiple times. Let's gather some more information to help your doctor understand your situation better."

Dr. A.I. then follows with a series of questions designed to identify the cause of the sleep disturbance. 

• Can you describe the nature of your sleep disturbances? For example, do you have difficulty falling asleep, staying asleep, or do you wake up too early in the morning?

• How long have you been experiencing these sleep disturbances?

• Do you experience any symptoms such as snoring, gasping for air during sleep, or have you been told that you stop breathing while sleeping?

• Have you noticed any changes in your urinary habits, such as a weak or slow urinary stream, difficulty starting or stopping urination, or a frequent urge to urinate?

• How many times do you typically wake up to go to the bathroom during the night? And have you noticed an increase in your overall daily urination frequency?

Dr. A.I. asks additional questions to confirm the lack of other symptoms, including pain on urination, changes in color, smell or consistency of urine, fever, night sweats, fatigue, lower back pain, pelvic discomfort, blood in urine or semen, burning, changes in appetite or weight loss, and changes in mood or energy level. 

Finally, Dr. A.I. asks about prior medical conditions and medications, if the PSA was measured, specific questions about family history of sleep disorders or other medical conditions, and whether or not they have tried any methods to improve sleep, such as adjusting sleep environment or practicing good sleep hygiene. 

Upon completing the interview, Dr. A.I. thanks the patient for completing the interview and explains that their doctor will get the results before their visit. 

Here is the summary that the doctor received for this patient prior to the virtual visit: 

• The patient is a 69-year-old male who has reported a new health concern of having trouble sleeping through the night. He wakes up multiple times to go to the bathroom, approximately three times a night, and this frequency is increasing.

• He has noticed a weak and slow urinary stream. He has not noticed any changes in the color, smell, or consistency of his urine and does not experience any discomfort or pain while urinating.

• He has not noticed any changes in his appetite or weight recently. The patient has hypertension and is currently taking lisinopril.

• The sleep disturbances have been progressively worsening over the last year. He sometimes feels cranky due to the sleep disturbances.

• He consumes a lot of coffee during the day and usually has a glass of wine or two with dinner. He has not tried any methods to improve his sleep.

• The patient does not experience any other symptoms such as night sweats, fever, or unexplained fatigue during the day.

• The patient expressed a concern about possibly having undiagnosed prostate cancer. He has not experienced any other symptoms such as lower back pain, pelvic discomfort, or blood in his urine or semen.

Note that Dr. A.I. has organized the answers to the interview questions in a logical way to make the diagnosis clear. During the interview, Dr. A.I. is designed to not provide advice to the patient nor suggest possible causes of the patient’s symptoms, even if the patient directly asks such questions while texting in the interview. Although the patient's question to the AI about "could this be prostate cancer?" wasn't answered during the interview, Dr. A.I. highlighted this concern for the doctor to address. 

After a live virtual consultation with the patient, the doctor concluded that this patient most likely is suffering from gradual onset benign prostatic hypertrophy. Based on the relatively mild symptoms, The HealthTap Medical Group doctor suggested medication management, with a plan for a non-urgent consultation with a urologist for PSA and/or rectal examination, stool FIT, and follow up for progression of symptoms. 

What's next

It turns out that behind the scenes, Dr. A.I. also can generate a pretty good differential diagnosis, and suggest treatments that address the patient's main concern (in this case, regarding sleep disturbance). However, on HealthTap, we do not show Dr. A.I.'s suggested treatment advice or its assessment of differential diagnosis, either to the doctor or the patient. With continued evolution and further validation of the medical accuracy of its operation, we expect in future to leverage these abilities too. 

Treatment suggestions

For reference, here are the treatment recommendations that were internally generated for this case, showing how effective the LLM already is at making helpful recommendations: 

• Consider reducing caffeine and alcohol intake, especially close to bedtime, as these can disrupt sleep patterns.

• Practice good sleep hygiene, such as maintaining a regular sleep schedule, creating a quiet and dark sleep environment, and avoiding stimulating activities before bed.

• Consider seeking a consultation with a urologist for further evaluation of urinary symptoms.

• Discuss with your doctor about getting a PSA (Prostate-Specific Antigen) test to rule out prostate cancer.

Differential diagnosis list

And here is the differential diagnosis list, along with Dr. A.I.'s stated rationale for each. 

Benign Prostatic Hyperplasia

• The patient's age, gender, and symptoms of increased urination frequency and weak urinary stream are consistent with this condition.

Sleep Apnea

• Although the patient does not report typical symptoms such as snoring or gasping for air, his age and sleep disturbances could suggest this condition.

Insomnia

• The patient's difficulty staying asleep throughout the night could be indicative of insomnia.

Prostate Cancer

• The patient's age, gender, and urinary symptoms could suggest this condition, although he does not report other typical symptoms such as lower back pain or blood in urine.

Overactive Bladder

• The patient's increased urination frequency and nocturia could suggest this condition.

Urinary Tract Infection

• Although the patient does not report typical symptoms such as pain during urination or changes in urine, his urinary symptoms could suggest this condition.

Bladder Cancer

• Although less likely, the patient's age and urinary symptoms could suggest this condition.

Diabetes

• The patient's increased urination frequency could suggest this condition, although he does not report other typical symptoms such as increased thirst or unexplained weight loss.

Chronic Kidney Disease

• Although less likely, the patient's age and urinary symptoms could suggest this condition.

Interstitial Cystitis

• Although less likely, the patient's urinary symptoms could suggest this condition.

Conclusion

Dr. A.I. has already demonstrated a remarkable ability to function as a virtual physician assistant, with its use limited to information collection and summarization. We look forward to continued evolution of generative AI solutions and to clinical validation of the differential diagnoses and recommended treatment advice. 

We previously demonstrated that showing doctors a list of potential diagnoses improves their consideration of alternative explanations and increased diagnostic accuracy[1]. In a similar vein, we expect that showing doctors suggested treatment alternatives will also lead to better and more thorough treatment plans. 

[1] Doctors’ diagnostic accuracy with and without diagnostic decision support AMIA 2020 Virtual Clinical Informatics Conference