A physician-specific small language model based on text recognition of keywords. It triggers user-defined assessment/plans for medical diagnoses. The user saves on time and error.
Thank you
When the patient comes into the hospital with some sort of ailment. It is up to the doctor to come up with the reason why you are ill and the best therapy to fix it. That is the assessment and plan.
During the 1st peak of the covid19 pandemic our medical center had a severe shortage of hospitalists to treat the volume of patients in need of admission for respiratory failure. Psychiatrists, pediatricians, orthopedic surgeons and others with limited infectious disease experience volunteered to help.
Our hospitalist department created paper templates based on the typical covid admission workflow for these doctors to use under supervision. The doctor shortage was ameliorated here and lives were saved.
This experience highlighted for us the opportunities of a web-based language model platform specific to physicians. This whole process could have been streamlined and improved with the use of a simple language model platform. Our platform has shown to reduce typing, physician error, likely ameliorating physician burnout in local testing.
Type or paste keywords from hpi (sob, chf, etc.) into the dashboard text field. User defined assessment/plans populate based on keywords. Save various reminders in the plan (c.diff treatment vs other bacterial colitis), treatments, links to EBM, as you see fit.
Assessment/plans are easily shareable within the Phrasefire community. Updated assessment/plans based on EBM at a large research instituation can easily be shared with docs at a remote medical center.
Doctors spend more time typing into the EMR than with our patients. This leads to fatigue and burnout. Phrasefire cuts into that incessant typing.
https://www.medscape.com/viewarticle/868421
Yes according to this clinic study doctors now use 55% of time in the EMR, only 27% of the time with our patients. that is sad! https://www.acpjournals.org/doi/10.7326/M16-0961?articleid=2546704
Forbes summarizes it well.
https://www.forbes.com/sites/brucelee/2020/01/13/electronic-health-records-here-is-how-much-time-doctors-are-spending-with-them/?sh=2f3865a35172
21.5 minutes saved per 10 patients admitted or seen in clinic. Assuming 7 on/ 7 off hospitalist schedule and 182 day work year this equates to 3905 minutes = 65 hours = 2.7 days/year
Assessment subsection contains 5.16 sentences
on average (Table 1)
https://aclanthology.org/2021.acl-long.384.pdf
5.16 * avg sentence length (words) 12-17 words per sentence in scientific articles, similar subject matter to medical note =
61.92 - 87.72 words in the sentence. Typically assessment/plans are much longer than 12-17 words.
https://www.aje.com/en/arc/editing-tip-sentence-length/
https://www.elsevier.com/connect/writing-a-science-paper-some-dos-and-donts
88 words * avg typing speed https://www.ratatype.com/learn/average-typing-speed/
88 words 88/*41 wpm for avg computer typist = 2.15 minutes per note saved
In a 10 pt admission day, that is 21.5 minutes saved. Assuming 7 on/ 7 off schedule, 182 day work year this equates to 3905 minutes = 65 hours = 2.7 days/year
in summary:
Average length of Assessment/Plan
https://aclanthology.org/2021.acl-long.384.pdf
Extrapolated average assessment/plan length from the average sentence length in scientific documents
https://www.aje.com/en/arc/editing-tip-sentence-length/
Average typing speed
According to this study, medical error could be prevalent enough to be the third highest cause of death in the U.S.
Medical errors are an under-recognized cause of death.
https://www.hopkinsmedicine.org/news/media/releases/study_suggests_medical_errors_now_third_leading_cause_of_death_in_the_us
Each error's median cost is $939 as of 2009. We extrapolated this data from the David and Kaplan study to the largest healthcare provider in Southern California at 4.7 million members. This hospital system likely sees up to 14100 medical errors yearly at a cost of $13,239,900/year. Cautious estimates of a 10-20% reduction in hospital-based error using PHRASEFIRE suggest a savings of 1410 - 2820 errors yearly, saving approximately $1,323,990 - 2,647,980.
A healthcare system serving 1 million patients would likely save between 141-282 errors yearly at a cost of $132399 - 264798.
https://www.sciencedirect.com/science/article/pii/S1098301512042660
The website works with mobile devices, but is optimized for desktop use. That is where the bulk of our typing data is in input at our hospital.
Patient sensitive data is not input into our system. We instruct users to paste the HPI without any patient identifiers such as age, sex, name, location. Any data input into the system is used to trigger a display of user-generated assessment/plans. It is not saved or analyzed in any way.
Users are encouraged to save EBM articles directly in the assessment/plan. Reminders, notes, treatment considerations are good candidates for this.
By having evidence-based medicine articles directly linked to the assessment/plan as well as the ability for the attending to set specific plans for diagnoses, it is likely that this is the end or at least a reduction in the practice of “pimping” known all to well in our medical establishment.
.png)
.png){kind=link}