MedScen: Medical Scenarios

The solution: let's use our brains together to build the perfect software for doctors and nurses to easily use AI. The AI is a collection of simple and intricate logics that help the doctor in decision making. The AI asks for information about each patient by opening a Scenario based on the chief complaint. Each scenario is a collection of important questions that the doctor or nurse must ask from all patients in that context. The AI will then give recommendations on how to manage the patient. The AI makes sure the busy doctor or nurse don't forget important or rare diagnoses, tests and treatments. Each scenario is designed to ask all related questions, making sure the doctor does not forget them, and make it easy for the doctor to answer them by simply clicking the already provided options. 

How does this help with the measurement of the healthcare system? 

By bridging the gap between frontline healthcare workers and data collectors/analysts/policy makers.

Long version of the story:

Communication between different levels of healthcare system has always been lacking in 3rd world countries. Healthcare providers that work directly with patients are scrutinized by authorities that do not have a good monitoring system and have to supervise without access to details of conditions. 

In 3rd world countries, doctors that are sent to remote busy locations are usually newly graduated doctors who have relatively good knowledge of theory but have limited experience, therefore they tend to misdiagnose, forget important questions, and lack knowledge of available tests and treatments in those remote locations. They also lack understanding of the culture of the remote location and lack experience in "wording" sensitive questions according to each culture (aka tact; example: asking deeply religious people about sexual relationships). The AI in MedScen is capable of recognizing the scenario for each patient and is able to give individually-tailored recommendations.

In 3rd world countries, healthcare providers are not necessarily proficient in English language enough to keep up to date with latest international recommendations, they also lack the time to go through hundreds of newly published articles due to the high load of patients. The AI in MedScen is constantly updated by a team of experts and is able to provide the latest recommendations (example "according to 2022 AUA guideline on prostate cancer, the following is recommended for this situation: ... "). The software is also capable of adjusting the recommendation based on the resources and the culture of the location.

Some health related scenarios are extremely complicated and require a lot of time, research, and critical thinking by someone who is not mentally tired to solve. Healthcare providers in 3rd world countries are usually overworked and it is illogical to ask them to resolve such intricate situations in extremely short amount of time. The AI of the MedScen is capable of helping the healthcare workers with solving these complicated or rare situations, because the answer to those complicated scenarios is already in the system (based on previous similar experiences of other clinicians, or critical thinking of the software developers who think of those scenarios in advance and find the solution).

In 3rd world countries, the healthcare providers may not receive the new change in policy or guidelines as soon as they are released. For example during the time MJ worked at the ER in a remote location, the ministry of health created a new guideline stating that "Ceftriaxone should not be mixed with Ringer's serum." The authorities informed the first-line providers by only sending a fax to a few hospital administration offices, which was lost in between other paperworks. Solution: the MedScen software recommendations can be updated immediately and give the recommendation whenever the situation occurs; for example, as soon as the doctor clicks on Ceftriaxone, the new health minister guideline appears in the corner of the page in alarm color.

In 3rd world countries, new policies and guidelines are created by people who do not have detailed knowledge of different regions of the country. The MedScen software has a feedback loop that bridges this gap. MedScen's scenarios also improve data collection from different regions by: 

1. increasing the details that are asked from each patient, including details that doctors do not usually ask but are necessary for policy making

2. MedScen includes standardized definition of each medical term, making sure everyone has the same understanding of each clickable option

3. bypassing NLP (clicking pre-determined options instead of typing) removes the multiple translation stages and improves communication.

4. the feedback loop inside the system improves quality of scenarios and enhances communication between policy makers, supervisors, analysts, and scientists with the frontline healthcare worker

Communication between different levels of healthcare system (frontline provider to management level) has been missing a proficient feedback loop in 3rd world countries. With MedScen, a frontline provider can always send message in the system directly on each scenario, for example: this question in this scenario is vague, or this question does not make sense, or the scenario is missing important questions, or this question must also include this option as an answer. MedScen constantly updates its scenarios and its solutions based on these feedbacks. If the system does not include a scenario, the frontline provider can always start a blank page and type on it. MedScen will later turn that into a scenario so that other providers can use the experience from this new scenario. For example, violence situation are each unique, it is best if the frontline providers type the details of the situation so that MedScen developers can learn from each scenario and find ways to avoid it in next situation, or provide suggestions to the clinic to avoid similar scenarios in future. Example of such scenario: some religious people become violent if a male nurse even suggests giving IM shot to their female relatives. Every shift that has no female nurse, MedScen can suggest to shift personnel to write about it formally on a flier and put on the wall to make it more official. 

An online platform that anyone from around the world can access to (medscen.com). The website is a simple software that can be downloaded for offline use as well. The software is a collection of hundreds of medical scenarios (including routine scenarios, complicated scenarios and rare scenarios). The user (healthcare provider) inputs patient information to the software by clicking well-defined medical terms. Each clickable option comes with very clear definition to make sure that ALL users have the same understanding of the options. MedScen is designed to bypass NLP by not allowing typing as much as possible. The software output is a matrix of TRUE/FALSE and numbers that can be readily analyzed by machine learning. Based on the clicked options, the software can suggest diagnosis, testing and treatment. MedScen makes sure that the healthcare provider asks all important questions from ALL the patients of each scenario, makes sure that diagnoses are not missed, important tests are not forgotten and helps the healthcare provider with finding the best treatment available in the area. The output can be turned into a standard "medical note" if the user requires (a text describing the patient condition).

MedScen also includes scenarios of healthcare violence, giving suggestions on how to use tact to ask sensitive questions from patients that are prone to violence (for example asking unmarried women about sexual relationships in deeply religious remote areas). MedScen also provides suggestion on how to handle violence when they occur and how to handle the situation.

We would like to ask designers of Solve-MIT platform to help us with designing our software. Thank you; All the buttons are in the right locations. Users find whatever they are looking for fast. Good job. This is the key feature of MedScen: Medical scenarios are complicated and may include a large number of questions asked from each patient with intricate answers; designing this software is more challenging than it sounds.

MedScen can literally help anyone asking for medical consult by improving communication in 3 different levels:

1- communication between patient and doctor

2- communication between doctor and the health system

3- communication between doctor and scientist 

The following are the benefits of improved communication:

Accuracy of diagnosis is increased. 

Doctors can request scientists to perform more research on specific questions. If a doctor feels that not enough research is done on a specific question, they can report it and request more research through the MedScen feedback loop. Budget for research will be better allocated through using an AI system to better understand the scientific need of the society; especially in the healthcare setting. Furthermore, MedScen allows for machine learning to answer some of these questions (will be explained later).

It is easier to find the best individually-tailored treatment using machine learning and AI. MedScen also helps doctors by making sure they do not miss diagnoses and helps them make the best clinical decision adjusted to the region's resources and what the patient needs. MedScen aims to reduce healthcare setting violence (especially ER waiting room) by improving communication and enhance quality of care for the patient.

MedScen especially helps doctors in the 3rd world countries because these doctors usually have a high load of patients that they have to diagnose and plan a treatment in very limited time (MJ had on average 2 minutes per patient). MedScen reduces the time needed for reporting by replacing "typing" with "clicking", also allowing the doctor to not forget important questions and diagnoses and thus reducing the mental burden and anxiety of the doctor. The software can also be used as a "scapegoat" by doctors that are subject to violence by patients who do not understand the purpose of some questions and tend to get offended if the question is too private (example: asking an unmarried woman if she can be pregnant is an extreme taboo in some areas and can lead to violence; the doctor can blame the software: "the machine asks this from everyone, sorry").

MedScen will eventually help medicine in general by allowing "pooling" of large good-quality data from different practices (through standardization of practices) and allowing machine learning to find logics that have been missed by human doctors.

MedScen also serves policy makers and supervisors: it asks extra questions that doctors usually do not need to ask but are important for policy making (example: if a patient did not get the best treatment, was it because of high price? Was it because treatment was unavailable in that region? Doctors usually do not ask or report these information). The data output of the software allows for accurate machine learning with low bias because all questions are asked from all patients and the data is not subjective (no NLP, plus standard definition of all options). Complex medical terms and concepts are usually lost in the process of NLP but a click-based software bypasses that problem.

After graduating from medical school, MJ was sent to a remote under-developed region to work as the only primary clinician in an emergency department (ED) with 500-600 patients per day load. MJ was graduated from top medical school in a well-developed region of the country, therefore, from the beginning of working in the new region, MJ got a well-deserved culture-shock: patients could not communicate what they wanted with MJ because they had different presumptions; they also spoke MJ's language as their second language and had different understandings of some words. The extremely high patient load was too much for a newly-graduate. What they teach to medical students does not prepare them for nuances of practice in a remote under-developed region with limited access to almost everything. MJ wished she had a software that helped her with: understanding dialect differences of the region, working as a scapegoat for intimate questions, reminding a novice doctor of dangerous or rare conditions which may be missed in middle of huge load of angry impatient patients, and helping the doctor with tact by wording questions and recommendations better adjusted to the region culture.

SS is a professional software developer and interface designer. AAE is a healthcare data scientist with years of experience in machine learning for health related problems. AAE has been mentoring MJ for the past five years. LOR and FD are doctor and nurse and they joined the team because they understand the importance of this software and would like to provide their experience in building the medical logic of the system.  There are other people who wish to be added to the team however there seems to be a problem with the website (the + sign disappeared). The creator of Mindify (a click-based EMR software) wishes to be added to the team to bring his years of experience with this type of software. We have doctors and nurses who wish to be added to the team to bring their own experience to different scenarios and help us improve the questions and the language.

MJ is one of very few medical doctors who have treated thousands of patients in primary care setting in a 3rd world country AND has been training in data science and the art of data collection in the past 7 years. She has deep understanding of data that is lost in the steps of data collecting and gathering in medical settings due to the complexity of healthcare systems and the human body.  

Furthermore, our team has been working with different software developers in the past years, testing their electronic medical recording (EMR) products and has been applying some changes to them. Therefore we understand the strengths and shortcomings of each product. After carefully analyzing their products, we believe that building the software from zero is the best solution. Making sure the mistakes of those companies are not repeated in MedScen.

After serving in the remote region, MJ and AAE created a software called SmartER which is an intelligent triage system designed to involve nurses better in the triage of patients by distributing the load of triaging hundreds of patients in a short amount of time between the doctor and the triage nurse. SmartER can be incorporated into MedScen and is technically an accumulation of intricate scenarios shown in one-page software that will need very limited education to operate.

The idea of this software sounds simple but it is actually a huge leap in the field of medicine. I require a team of reputable experts to 

1. Teach doctors about the importance of this software; one of the challenges of the software is convincing doctors that AI is NOT going to replace human doctors; rather it improves their practices and takes off a huge load off their shoulders.

2. Convince data scientists and policy makers to stop making predictions from low-quality data and force doctors to change their practices based on those biased models.

3. We need a team of doctors from different specialties to each give their opinion on the quality of the software and its output

4. Most importantly: This software can work to its fullest if it becomes the ONLY software for medical scenarios used by every practice or at least the majority of them. Allowing pooling different practices to generate big-data for machine learning. Therefore we need the reputation of an internationally well-known institution to advertise it and educate people and doctors about its necessity.

5. The design of this software is challenging. The location of each button is crucial and should not change once it is introduced because there will be too many buttons, too many questions and human eye must get used to them. Designers of MIT Solve challenge website have done a great job so I would like to ask for their expertise.

Data scientist who work with healthcare data have faced the challenge of models that work well on the "test set" but in practice fail to show predictability. Most of these data scientists believe that the problem will be solved by changing the type of machine learning, or tuning the hyperparameters of the model. However, Mehrsa believes that the problem rises from inaccurate data. Most data scientists use insurance data (because they are large enough for ML), however, insurance data is not gathered for the purpose of medical diagnosis and treatment, it is prone to errors and bias, and lacks details. There are click-based EMRs currently available in the market (e.g. Mindify / CLICKS International). However, MedScen is the only software that is scenario-based. There have been attempts at improving NLP on medical notes (cTAKES, MED-BERT), however, they still fail in complex medical concepts, and they fail when doctors make typing errors or forget a dot (".") at the end of sentence (a common typing error when typing fast). There is also a lack of standardization between different doctors and their practices; different hospitals train doctors in different ways and even medical terms do not always have the same meaning (example: ask 100 urologists what is the meaning of "voiding dysfunction"). Therefore, the output of even the most perfect NLP on medical notes can still not be pooled together.

While most healthcare data scientists focus on model hyperparameters and improving NLP, MedScen is focused on scenario-based collection of data with pre-determined questions and well-defined answers for each patient condition. MedScen also has an efficient feedback loop that improves the scenarios, medical terms, and suggestions based on the culture/language/location. The data that is collected from MedScen does not need different levels of cleaning (information is usually lost and bias is formed in those levels).

MedScen is also unique in its focus on standardization of practice and medical terms. Each clickable option in the software comes with a clear definition that includes 3 parts: 

1.What the option "is"; including whether a special context needs to be present for the doctor to click this option.

2.What the option "is not" (conditions that are similar and are commonly confused).

3.Further explanation: an example or an image.

Having clear definition is a crucial step before any machine learning, making sure that ALL users have the same understanding of each question and their options. Language and dialect are extremely important in collecting data for machine learning.

Categories in MedScen are not forced AFTER the patient visit by someone who has not seen the patient. Rather, categories are chosen at the time of the visit when the patient is present and is able to provide additional information to choose the correct category. The category is chosen by a data collector who is a subject expert (doctor or nurse) and understands each category and their purpose.

MedScen can work with other EMRs that are already available in the hospital system because the output of MedScen can be turned into a text and copied and pasted in the hospital's EMR software. Therefore, hospitals do not need to change their entire systems. MedScen is also free. These make MedScen unique because other EMR systems charge large amounts of money for installation and maintenance. All other EMR systems keep patient data in their own database, which subjects them to hackers and leaking of sensitive patient information. So those systems require a large team of security experts and multiple data warehouses for back up. MedScen does not require such complex backend support because it does not record anything (at the moment, the recording database can be added later if necessary). Currently MedScen is only an intelligent mediator to standardize practices and data reporting as well as helping practitioners solve complex situations.

Our goal is to finish all the routine scenarios of Emergency Medicine, Internal Medicine, General Surgery, Gynecology & Obstetrics, and Pediatrics by the end of year one. At the same time we aim to perform conferences and talks in different university hospitals and introduce MedScen to the attending physicians and training doctors and nurses so they get familiar with the system, understand its necessity, and come to a common ground with the medical terms (making sure all doctors agree with the definitions).

By the next five years we aim to make MedScen a routine practice for medical doctors, routinely taught in med schools and all practitioners would need to pass short courses (one or two online classes) to make sure they all understand how MedScen works, how to send feedback through the system, how categories must be chosen and most importantly WHY categories are necessary (some knowledge of data science is necessary for medical practitioners). Ideally by the end of five years we are able to pool large data with details and confounders included to make accurate machine learning models, tailoring medicine for individual patients and their unique disease condition.

Adjusting MedScen suggestions to local culture and resources requires at least one year of active feedback loop between frontline users of MedScen and programmers of MedScen. The one thousand scenarios at the end of year one are those that generally match to majority of settings. Once MedScen is established in different locations, programmers will use the feedback they receive based on IP location of the user or the name of their facility. Using this feedback loop, MedScen scenarios, wording of questions and suggestions will be adjusted to location of IP address of users. Example: People in this area commonly use the word "Phlegm" to describe any type of secretion from the body, use the word "mucoid secretion" instead to ask for texture of secretion.

By the end of year one we aim MedScen to have at least one thousand scenarios in at least two languages (English and Portuguese). MedScen must also include ALL emergency scenarios by the end of year one.

By end of year one we aim MedScen to be taught in at least one reputable medical school.

Adjusting MedScen to different cultures and resources will require the feedback loop to be active for at least one year in different locations, so by the end of year five we aim to adjust the 1000 scenarios to at least 20 locations with different cultures and resources.

Ideally MedScen should record information of each patient in order to follow them up and improve its future recommendations based on patients' outcome. This requires each patient to have a unique ID, and the software must have access to important registries such as death records and hospital and ICU admissions. This is the five year aim of MedScen because it requires government approval and help.

Our team members have talked to multiple medical doctors about the idea of MedScen and they all call it The Future while asking "why is this not already a thing?!". Scenario based medicine is the best type of medical practice and many doctors already use something similar in their routine practice. For example our Urogynecology team uses google forms. Some doctors have printed forms with fill-in questions. MedScen has the goal of unifying all doctors from around the world by standardizing scenarios and allowing pooling of intelligence, because human body is similar around the world. At the same time, MedScen adjusts to each region's specific resources, dialect, and cultural norm. Currently doctors use international platforms like www.medscape.com to consult each other about difficult cases. MedScen improves this exchange of intelligence by bringing wisdom of older doctors to novice doctors. 

The United Kingdom started the QResearch project in 2010 due to the lack of good data for policy making (https://www.qresearch.org/). The QResearch uses a computer system called EMIS and has link to important registries such as death records and hospital and ICU admissions. The QResearch is not similar to MedScen and EMIS is not click-based. But the aim of QResearch is the same as MedScen in that it aims to improve "quality" of patient data to improve decision making and policy making. For example the QResearch provided a better understanding of the true mortality rate of COVID-19.

https://www.human.health/platf... was created with unclear goals but seems to have a similar aim to MedScen: individually-tailored medicine. They do not have a clear strategy like MedScen and they are not scenario-based.

Simple online AI system that includes all medical scenarios adjusted to the location of the user.

Software improves communication between patient and doctor by providing pictures and videos for the patient so they can communicate their symptoms better;

for example

"Dizziness" may sound like a simple symptom, but patients usually have trouble explaining their symptoms to the doctors;

The cause of dizziness can be something benign or something cunningly lethal.

All of these pathologies will be described by the patient as "Dizziness"

In these situations it is crucial for the ER doctor to ask for specific details of the symptoms.

However, if the doctor does not speak the main language of the area, or even just the dialect, there will be communication problems; 

example: cantonese speaking doctor going to mandarin speaking rural areas of china will most definitely have trouble understanding details of dizziness and may find difficulty differentiating benign dizziness over a red flag symptom of an underlying potentially lethal disease.

The core of MedScen is: communication

Ancestral technologies: the feedback loop in the MedScen allows doctors to report traditional medicine that seem to work well. For example, a doctor can report that "a herbal medicine that is commonly used in our area" works very well for this scenario. MedScen developers will use this knowledge to fund formal research and prove/disprove the theory while testing the side effects of the herbal medicine.

At the moment MedScen does not record any data. However, if necessary, the data can be recorded if unique patient ID is added to it and high-security data warehouse is purchased. MedScen will maintain the right to its data and if necessary will sell de-identified data to pay for its expenses. However, MedScen aims to remain a not-for-profit organization with the main goal of using AI for individually-tailored medicine for all.

MedScen team is currently an international team of University of Campinas (Brazil) professors and students and New South Wales (Australia) students.

Problem with the website: the challenge does not let us add new team members. (The + sign disappeared).

We plan to hire MIT Solve challenge team if possible. The design of this platform shows that they are experienced. So Solve team's diversity will ideally be added to our team. We also read the MIT policy of diversity and we agree with the terms. We will incorporate it when hiring new team members.

MedScen will eventually focus on medical practice in different locations and their nuanced difference. For example, if we are currently focusing on MedScen for rural areas of Philippines, we will require consultation from doctors and nurses who have worked in those areas long enough to teach us about the nuanced differences of the region of their practice, people's culture and dialect variations that may generate confusion in communication between patient and the healthcare provider.

MedScen's diversity is therefore dependent on the region it is focusing on, at any given time. If we are currently focusing on Nigeria, we will be hiring as many Nigerian healthcare professionals as we can from different regions of Nigeria. If we are currently focusing on Indonesia, we will be hiring as many healthcare professionals from Indonesia as we can. We will be also hiring doctors and nurses that did not grow in the regions they practiced (doctor without borders maybe), because they have been dealing with cultural and dialect differences and they already know the nuances.

MedScen is currently a free online AI that provides logic to doctors and nurses. The feedback loop in the system is supposed to enhance its logic and the quality of its questions by one year. After one year MedScen can be introduced to governments and major universities as an intermediate in their information system. MedScen will improve communication in the information system by asking the right question and making sure that the answer is well-defined, not vague, and follows peer-reviewed internationally accepted definitions. The feedback loop will constantly improve the questions and their answers and bring the authorities closer to frontline workers.

Details of the product's benefit for each level of costumer is provided below.

Product provided to patients: individually-tailored medicine. Currently doctors use the result of RCTs (Randomized Clinical Trials) to make decision for patients. Those RCTs pool patients together based on race, sex and common comorbidities to compare one intervention to placebo. RCTs however, lack information to make decision for individual patients with unique needs and disease characteristics. Most RCTs are aimed at expensive treatments that benefits the funder of RCT. They are also mostly held in high-tech facilities in richer areas because they are expensive. Furthermore RCTs pool people together and usually include relatively healthy younger patients compared to real-world patients who are usually older, more frail and with more comorbidities. The AI in MedScen along with its unique way of acquiring accurate patient information allows it to perform machine learning on pooled data from different areas, adjusted to confounders and unique characteristics of the individual, and local resources. This will allow MedScen to provide individually-tailored management suggestions.

Product provided to frontline healthcare provider: an AI system that reduces medical errors, reduces time for reporting, reduces tension, anxiety and violence in the healthcare facility using its pre-determined scenarios that include pre-determined questions with clickable options, pre-determined order of questions (more private questions are asked later with pre-determined wording adjusted to culture). The AI suggests the most up-to-date locally-adjusted testing and treatment options.

Product provided to healthcare systems: MedScen allows standardization of practice, allows for pooling of good quality data from different practices with details of condition and confounders reported. The feedback loop allows better supervision and better communication between different levels analytical team. The AI also reduces healthcare cost by finding individually-tailored treatments that does not require multiple trial and errors (currently doctors randomly choose a medication and constantly change it until they find the best medication) 

Healthcare data is extremely valuable, especially data that includes epidemiological information or customer needs. MedScen currently do not need financial support because it does not record data. However, if data needs to be recorded then the cost of the software will increase and MedScen will need to sell de-identified data to big pharma companies. However, we believe that the governments should be the negotiator of selling the data, as big pharmaceutical companies eventually use the epidemiological data to ask health systems to pay for their products.

Pfizer is willing to pay up to 300,000 USD for information regarding epidemiology of metastatic prostate cancer in latin America. Given that this information is required routinely and from different diseases, we can presume that selling data from MedScen resources will be a stable source of income. However that requires MedScen to store patient information.