In the field of clinical diagnosis, the ability to efficiently and accurately extract relevant information from patient history can significantly aid healthcare professionals in providing appropriate care. Recently, advanced language models like Google BARD and GPT-4 have shown potential in this area. To compare their performance, we ran the same prompt and HPI sample text through both models, asking them to list diagnoses and main problems. Here, we present the results and analyze their respective outputs.
Prompt and HPI Sample Text:
The prompt: "List out all the diagnosis and main problems from this clinical HPI"
HPI Sample Text: Source - https://www.med.unc.edu/medclerk/wp-content/uploads/sites/877/2018/10/hp2.pdf
Ms. ___ is a 47 year old African American female with Crohn's Disease, DM, and HTN who presented to the ED after two days of severe abdominal pain, nausea, vomiting, and diarrhea. She stated that on Wednesday evening after being in her usual state of health she began to experience sharp lower abdominal pain that radiated throughout all four quadrants. The pain waxed and waned and was about a 4/10 and more intense than the chronic abdominal pain episodes she experiences periodically from her Crohn’s disease. The pain was sudden and she did not take any medications to alleviate the discomfort. The abdominal pain was quickly followed by two episodes of partial diarrhea and soft stool that was tan in color with no signs of blood. Her abdominal pain continued and she developed nausea and then vomited six times that evening before going to sleep. Overnight her abdominal pain worsened and she stayed in bed for most of the day on Thursday. She had nausea again all day but had no other episodes of diarrhea or vomiting that day and did not eat anything for fear of vomiting. She was able to drink water and keep it down. By late Thursday night, her pain had intensified to a 10/10 and she called 911 and was brought to the ER by ambulance from her home in Burlington.
Ms. ___ also stated that she had just ended a three week course of prednisone four days ago, which she had started about a month ago at 60 mg and tapered herself down over a few days by 10 mg. She began the course of prednisone last month because she felt as if she was about to have a Crohn's flare at the time. Ms. ___ was last hospitalized at UNC for Crohn's disease exacerbation in March 2007. She denies any recent hemoptysis, constipation, hematochezia, melena, and changes in her bowel habits since Wednesday. She has been compliant with taking her medications for Crohn's and has been stable on her mesalamine, mercaptopurine, and omega -3-acid supplement.
Upon arrival to the ED, Ms. ___ was put on IV fluids, given fentanyl 50 mcg IVP, phenegran 12.5 mg diluted with 10 mL NS IVP, and Mg sulfate IVP. Radiological images were obtained through an abdominal CT scan, ultrasound, and 2V XR. Ms. ___ was not given any other narcotics for her pain because of a past violation of a pain contract after a positive toxicology screen for cocaine resulted in her discharge from her family medicine provider and due to suspicions that she was narcotic-seeking.
Results:
Google BARD provided the following diagnoses and main problems:
Crohn's disease
Diabetes mellitus (DM)
Hypertension (HTN)
Abdominal pain
Nausea
Vomiting
Diarrhea
Prednisone withdrawal
Crohn's disease flare
Main problems included severe abdominal pain, diarrhea, nausea and vomiting, worsening abdominal pain on Thursday night, and a past violation of a pain contract due to a positive toxicology screen for cocaine.
GPT-4's output included:
Crohn's Disease
Diabetes Mellitus (DM)
Hypertension (HTN)
Severe abdominal pain
Nausea
Vomiting
Diarrhea
Recent prednisone use
Pain management
Radiological images
Past violation of pain contract
Comparison and Unique Highlights:
Both Google BARD and GPT-4 successfully extracted relevant diagnoses and main problems from the HPI sample text. They identified common diagnoses such as Crohn's disease, DM, HTN, abdominal pain, nausea, vomiting, and diarrhea. However, there were some differences in their outputs.
Google BARD uniquely highlighted "Prednisone withdrawal" and "Crohn's disease flare" as additional diagnoses, which might be relevant for the patient's current condition. Meanwhile, GPT-4 focused on other aspects like "Recent prednisone use," "Pain management," and "Radiological images" that were not identified by Google BARD. Both models mentioned the past violation of the pain contract, an important factor in the patient's medical history.
Google BARD's Expansion Beyond Diagnosis:
In addition to providing a list of diagnoses, Google BARD went a step further and offered more in-depth information on the main problems faced by the patient. This additional information can be valuable for healthcare professionals in understanding the severity and complexity of the patient's condition. Key points highlighted by Google BARD include:
Abdominal pain that is severe, radiating throughout all four quadrants, and waxes and wanes.
Diarrhea that is partial and tan in color with no signs of blood.
Nausea and vomiting that is six times on Wednesday evening and none on Thursday.
Abdominal pain that worsened on Thursday night and became a 10/10.
A past violation of a pain contract after a positive toxicology screen for cocaine resulted in her discharge from her family medicine provider.
Furthermore, Google BARD also provided a treatment plan, which can be beneficial for medical practitioners in formulating appropriate interventions for the patient. The treatment recommendations suggested by Google BARD are:
Continue IV fluids.
Give fentanyl 50 mcg IVP, phenegran 12.5 mg diluted with 10 mL NS IVP, and Mg sulfate IVP.
Continue to monitor her vital signs and pain level.
Administer medications as ordered.
Provide emotional support and reassurance.
Discharge her to home with instructions to follow up with her primary care physician.
To be fair GPT4 would also provide this information with an additional follow on prompt of "Provide a treatment plan for this". The results back included:
Based on the provided clinical HPI and the diagnoses identified by both Google BARD and GPT-4, a comprehensive treatment plan for this patient can be proposed as follows:
Pain management: Administer fentanyl 50 mcg IVP and phenegran 12.5 mg diluted with 10 mL NS IVP to manage the patient's severe abdominal pain. Regularly assess the patient's pain level and adjust the medication as needed, considering their history of violating a pain contract.
Fluid and electrolyte management: Continue IV fluids to prevent dehydration due to vomiting and diarrhea episodes. Monitor the patient's fluid intake and output, and electrolyte levels to maintain balance.
Gastrointestinal symptoms management: Provide antiemetics (e.g., ondansetron) to control nausea and vomiting, and antidiarrheal medications (e.g., loperamide) if diarrhea persists.
Monitor vital signs: Continuously monitor the patient's vital signs, including blood pressure, heart rate, respiratory rate, and temperature, to detect any abnormalities and ensure stability.
Crohn's disease management: Ensure the patient continues taking her prescribed medications for Crohn's disease (mesalamine, mercaptopurine, and omega-3-acid supplement) and consider consulting a gastroenterologist to evaluate the need for adjusting the treatment plan, especially in light of the recent prednisone withdrawal.
Diabetes and hypertension management: Confirm that the patient continues to adhere to her medication regimen for diabetes mellitus and hypertension. Assess blood glucose levels and blood pressure regularly and adjust medications if necessary.
Imaging and laboratory tests: Analyze the results of the abdominal CT scan, ultrasound, and 2V XR to identify any potential complications or underlying issues. Conduct additional laboratory tests, such as complete blood count, blood chemistry, and inflammatory markers, to monitor the patient's overall health status and disease progression.
Emotional support and patient education: Provide emotional support and reassurance to the patient during this challenging time. Educate the patient on the importance of medication adherence, lifestyle modifications, and routine follow-up appointments with healthcare providers.
Discharge and follow-up: Once the patient's condition has stabilized, and her symptoms have significantly improved, discharge her with clear instructions on medication management, self-care, and potential warning signs that warrant immediate medical attention. Schedule a follow-up appointment with her primary care physician and, if necessary, a gastroenterologist to monitor her condition and make any adjustments to her treatment plan.
This comprehensive treatment plan aims to address the patient's immediate concerns while also considering her ongoing chronic conditions. By addressing each aspect of her health, the plan can help the patient achieve better overall well-being and reduce the likelihood of future complications.
Conclusion
This expansion beyond diagnosis by Google BARD showcases its potential in not only identifying relevant medical conditions but also providing useful insights into the patient's main problems and suggesting treatment plans. This added value can assist healthcare professionals in making well-informed decisions and delivering more comprehensive care.
In conclusion, while both Google BARD and GPT-4 demonstrated impressive capabilities in extracting diagnoses and main problems from clinical HPI, there were subtle differences in their outputs. These distinctions might play a role in how healthcare professionals choose to use and rely on these language models in the future. It is essential to continue evaluating their performance and improve their accuracy in different clinical settings to further enhance their utility in medical practice. The added insights provided by Google BARD, such as the main problems and treatment recommendations, demonstrate the potential of these advanced language models in revolutionizing patient care and medical decision-making.
Commentaires