DocScribe | Devpost

THE PROBLEM

In 2019, the World Health Organization found that medical errors are a leading cause of death and injury, amounting to a deficit of 37.6 to 50 billion USD loss through added health care, disability, and lost productivity. These errors are defined as the failure of a planned action to be completed as intended or the use of a wrong plan to achieve an aim. [1]

DocScribe focuses on tackling both definitions of error by removing the variability between the doctor and the patient. Potential points of variability could include:

Verbose language from doctor
Complacency with understanding from patient
Internal assumptions and biases that doctors and patients make
Triage for patients
Burnout in doctors ## WHAT IT DOES Before visiting a doctor, a patient is required to submit their patient history documents as a singular PDF, which is saved in DocScribe's system for the doctor to access likely before the patient's and doctor's appointment.

At the start of the appointment, the doctor and patient will engage in a standard conversation about concerns and symptoms. During this, DocScribe will instantaneously transcribe all of this information and provides further insight upon medications and diagnoses that may be undercovered from human mistakes.

This insight has many implications. It can range from a simple transcription of what happened during the appointment to analyzing and validating the responses that both the doctor and patient make. For example, if the patient contradicts themselves by saying they have migraines daily, but later admits to only having migraines occasionally, DocScribe will flag this as a concern and will prompt the doctor to question the patient about the inconsistency. In the same light, if the doctor prescribes a medication without providing reasonable side effects, DocScribe will alert the patient to question if this medication will be right for them. To make sure that all questions are answered properly, clarifying questions must be answered before the doctor can progress to the next screen.

After the appointment is complete, the doctor has access to a private screen, which has holistic evaluations of their performance, predictive improvements based on trends that can be made for future appointments, and an updated patient history PDF. This PDF is in the same format as the entrant PDF, which allows DocScribe to adapt to fulfill both the doctor's and patient's inconsistencies in appointments. In the long-term, these three tools will independently improve a doctor's performance for appointments.

Having summaries is important for the patient. On average, after a week of the visit, 15% of patients erroneously recall and 36% need a prompt to recall information, so having a summarized, easily accessible, and mutable source is key to information security and adaptivity. [2]

WHY IT IS BUILT THIS WAY

Similar to the approach to the problem, DocScribe was created to solve specific issues. There are several issues in the current process between patients, doctors, and researchers. In 2013, Invokana hit the market and was considered a revolutionary drug. However, there wasn't a lot of research to determine that it was a drug with a major downside, and it resulted in thousands of direct amputations, ultimately leading to this drug being black-boxed. Since DocScribe utilizes AI, it is adaptive and is constantly updating to match whatever research is happening in the field so that a mistake like Invokana doesn't happen again. [3]

In a similar light, Avandia was a drug that was intended for patients with diabetes. However, there were conflicting studies on whether it raised heart problems or whether it only had positive effects. Since there wasn't a holistic databse to this information, different doctors prescribed different medications, but later research justified the increased heart problems to be legitimate. This drug was also black-boxed, and DocScribe could have been transparent with the conflicting evidence rather than doctors only evaluating one side of the story. [4]

DocScribe has an individualist, engine mentality. For the best functionality, previous PDF reports will be used to maximize efficiency and minimize errors during each visitation. This means that each patient can be assured that they are receiving care that fits them rather than a one-size-fits-all solution.

DocScribe is not a replacement for a doctor. This allows patients and doctors to maintain autonomy over the decisions that they make. It is merely a tool that listens into conversations to not only identify plausible mistakes, exaggerations, or missed information during appointments but to also help the doctor understand which information to clarify with their patient before concluding their appointment to significantly decrease the chances of medical errors being made.

DocScribe's last screen, Reports, is also very crucial to helping the doctor improve their diagnosis and understanding of a patient's symptoms and conditions with the AI-generated feedback coming from the original Appointment transcript.

LASTING IMPACTS

The purpose of this project is to really narrow down the chances of medical errors made during appointments and medical diagnosis, between both the patient and the doctor. As stated before, the goal is to not only narrow down on the deficit of added healthcare costs and lost productivity, but also to ensure that patient safety is secured and prioritized, and while it does not guarantee that medical errors will be erased, it significantly brings them down.

FUTURES/NUANCES/HARMS

We focused on creating a minimum viable product (MVP) during HackHarvard. This leaves room for growth. The below are sample regions for improvement:

Q: Will DocScribe work if the doctor cannot be present (i.e. post-operation)? A: Doctors will supply DocScribe with supple information, and DocScribe will become more authoritative in the one-on-one interaction with the patient.

Q: How will DocScribe work with patients that don't speak English, patients that are deaf, and non-verbal patients? A: This information will be stored in the PDFs under the Patient Info tab. This would require modifications to the LLM prompts, utilizing a webcam feed for sign language, and a textbox with pop-up notifications respectively.

Q: How will DocScribe improve the long-term quality of doctor reports? A: DocScribe will store the improvements from the PDFs over an extended period time, analyze common trends that the doctor messes up at, and provide predictive solutions to improve the quality of future visits.

Q: DocScribe's current system implies that a patient will see the same doctor throughout future visitations. What if a patient sees a different doctor? A: DocScribe will divide patient reports and doctor reports. This will add better adaptations to the model during the visit.

Q: What if patients are not comfortable with sharing details about their life to an AI? A: DocScribe will use aggregates (for example: if cholesterol runs in your family, you have a heightened chance of heart failure) to allow patients to be informed while maintaining anonymity.

A couple of implications are that there is a microphone that picks up both the doctor and the patient and that both are understood by the model.