Software to Help Non-Native Speakers Talk to their Doctors

LiteraSeed provides a software application to improve communication between ESL (English-as-a-Second-Language) patients and providers in describing the patient's present symptoms and history. This in turn empowers medical providers to quickly diagnose and provide treatment, and enables equitable access to less costly and more effective healthcare services. 

Urgent/emergency care is fraught with communication challenges. New and inexperienced staff interact with patients whose cultural, educational and language barriers may prevent them from communicating effectively. 

KEY FACTS: 

• Miscommunication is a leading cause of medical error. 
• Medical error is the third leading cause of death in the US.
• Language barriers are a top reason immigrants do not seek healthcare. 
• Vulnerable groups (immigrants, refugees, expatriates/migrants) are at risk of miscommunication.

The motivation for this project is the tragic and preventable death of the founder's young relative while waiting to receive treatment in an Emergency Department. Improved communication could have resulted in the child receiving life-saving care.

There is no easy way to reduce miscommunication between ESL speakers and doctors during medical examination procedures. In fact, nine out of ten US adults have inadequate health literacy, and the share is higher among underprivileged populations [1]. Most people don’t have a medical background, so they don’t necessarily know how to describe their health problem or know what information to provide their doctor. This can be a life-threatening situation. Miscommunication is a leading root cause of medical error [2]. Nearly 250,000 people die each year due to preventable medical errors [3]. People who have language and cultural barriers have nearly a 50% higher risk and severity of medical error [4]. Language barriers are a top cause of impaired access to healthcare services for immigrants [5].

Obtaining a reliable History of Present Illness usually leads to the right diagnosis. However, patients often feel vulnerable when speaking about their health history. They sometimes don’t know what to say or how to say it, leaving critical information unspoken. Language and cultural barriers further exacerbate this. Improving communication at the first clinical encounter would prevent errors at the outset. Overcoming language barriers would enable equitable access to less costly and more effective healthcare

LiteraSeed will initially focus on medically underserved markets, such as refugees, economically disadvantaged, and those with English as a second language. This population features language and cultural differences that impact access to timely and effective healthcare. 

LiteraSeed’s initial clinical target is women’s health, specifically, pregnancy care, pregnancy-related complications, and predicting the likelihood of preterm labor complications, maternal mortality, and postpartum conditions. These demographics of women are at the highest risk for malpractice claims [6, 7]. We will conduct an IRB-approved proof-of-concept study with MIHS/RWHC that focuses on symptoms in the last month of pregnancy and up to six weeks postpartum. This will enable the testing of a known subset of symptoms perinatal women face. We will conduct focus groups and randomized control studies. We will work closely with the population and community organizations to design appropriately to their needs. 

The LiteraSeed team believes that by focusing the initial market on maternal health and newly arrived refugees and immigrants who come from nations where maternal mortality has been historically the highest may allow the study’s conclusions to be generalized globally. Once proven with this market, we will continue to expand to cover health conditions for children, men and women.

LiteraSeed provides a digital health platform that will help ESL speakers communicate with their doctors, allowing them to accurately describe: 

• their current health issues 
• relevant contextual information
• the seriousness of their condition

There are essentially three steps: 

1. Patients log on to the LiteraSeed web app.
2. Complete the History of Present Illness. 
3. Share the structured report with their healthcare provider. 

How it works: 

Patient-System - the user (i.e., patient) will interact with the user interface that will provide a question and a set of possible “helper” responses that is guided by algorithms at the backend. We intend to start with Bayesian classifiers, and over time make our algorithms more sophisticated. At the frontend, the interface combines plain text and visual aids. LiteraSeed's backend statistical inference system will interact with each patient to identify and provide the patient appropriate questions and offer "helper" answers. These questions aim to gather pertinent information on the patient's health, and any relevant background and contextual information that can lead to a speedy diagnosis. The user can then select from the "helper" responses to complete their answer, or they have the option to input free form text to add more specificity or a different response altogether. (As a future goal, they could also respond with image capture and our tool could analyze the image to provide feedback). In its current form, the interface is kept simple, presenting the user with a question about their health condition and images with “helper” text. The results are expected to be processed manually at the current stage. As a final step, a report is generated that can be sent to the healthcare facility in charge of the patient's care that would ultimately receive it through a provider portal. The system will also provide the patient guidance on their best next step. 

Provider - Upon receiving the report, the medical clinic can then reach out to the patient to further clarify the patient's health issues and prepare the necessary treatment prior to the patient's arrival, if determined to be needed. 

The idea is to help patients provide the most accurate report on their health condition so that providers can more quickly and effectively diagnose and treat them. The images add additional context to aid people with limited language skills and/or limited time.

To reiterate, there are two components of the communication problem our solution aims to address: (1) medical literacy, and (2) language differences. 

LiteraSeed is rethinking how we engage and empower patients in obtaining high quality care and avoiding the risk of misdiagnosis and delayed treatment. We expect that the proposed digital health platform will significantly improve the efficiency of the communications process between patients and doctors and the effectiveness of its outcomes by bridging the gaps in care. Current solutions for history-taking are limited by time and communication barriers. These include clinicians interviewing the patient during a time-constrained clinical encounter and rules-based symptom-checkers that are narrowly focused. Unlike existing solutions, LiteraSeed helps patients circumvent communication, language and cultural barriers to accurately capture and document their current condition, background and personal information in their own words and language. Most people don’t have a medical background and words can have multiple meanings resulting in misinterpretation. LiteraSeed makes it easier to provide accurate and relevant health information by providing a simplistic and intuitive user interface (UI) that flexibly receives symptom input from the patient in a form and expression that feels natural to them, with a statistical inference system at the backend that interacts with each patient to identify and provide the patient appropriate questions and offer “helper” responses. As the patient provides inputs, the system will iteratively narrow down the health problem. The resulting report will then be used by healthcare administrators and clinicians to develop, expedite and optimize the care plan with an efficiency that can save lives. It also guides patients on their best next step.

There are three major components to our technology:

1. INPUT: at the front-end is an innovative user interface that will help the patient accurately capture and document her current condition as well as other background and personal information. The application provides visual aids for symptoms and support for (potentially multilingual) unstructured free form text input from patients describing their condition. NLP models and Topic Modeling techniques will be used to map users input to medical diagnoses. 
2. PROCESS: At the backend is LiteraSeed's statistical inference system (i.e. patient data diagnostic algorithms) that is guided by Bayesian and Machine learning algorithms that will interact with each patient to identify and provide the patient appropriate questions and offer "helper" answers. The system will iteratively narrow down the health problem. We will explore the use of a neural network-based inference system. 
3. OUTPUT: the system will then prescribe next steps to the patient and create a health condition report for the provider. 

Communication errors are a significant contributing factor of poor and sometimes devastating outcomes for patients, including preventable severe disability and death. Language barriers not only compound this problem but lead to less effective and more costly healthcare for patients. Common underlying factors in preventable medical errors include: 

1. lack of patients and families knowing and identifying warning signs and knowing the importance of seeking timely care, and
2. providers having inadequate information or training to diagnose and effectively treat the patient [8]. 

Our solution would help the patient accurately capture and document her current condition and background and personal information by identifying relevant information that could lead to an accurate diagnosis. Resulting actions include: (1) guiding the patient towards her next step; (2) being used by healthcare administrators and clinicians to develop, expedite and optimize the care plan. We could leverage existing clinical decision support to help complete this solution.

To support accessibility and ease of use, the user interface is being uniquely designed for low-literate and limited English speaking adults, meant to support quickly and accurately providing the right information based on validated tools and peer-reviewed studies. This would also help people in a time-limited situation. Furthermore, our solution aims to gather pertinent contextual information that can be key to an accurate and timely diagnosis. Eventually the crowdsourced data could be used to better inform the "helper" responses to further improve the system's effectiveness in soliciting this information to help the patient access and receive timely and appropriate care. 

Currently, we are in research and development. We plan to enroll at least 100 people in our proof-of-concept study at Maricopa Integrated Health System's (MIHS) Refugee Women's Health Clinic (RWHC), with potentially around 35 expectant mothers using the application, and comparing outcomes with a control and placebo group. 

In one year we plan to enroll at least 100,000 expectant mothers in our study across multiple healthcare sites, potentially more depending on calculations for statistical relevance to prove our technology is effective in meeting outcomes. 

In five years, we expect the technology to be available to consumers and health systems across the United States. At minimum around 4 million pregnant women each year could benefit from using our technology during pregnancy. We have plans to continue to expand to cover health conditions for children, men and women, which will allow us to positively impact the healthcare for millions of more individuals across the United States. Once proven impactful across the US, we can scale our solution globally. 

We have a three step process to bringing our technology to market: 

1. Launch a prototype of the application within a small-scale controlled proof-of-concept study with MIHS/RWHC. The study will help to determine the efficacy and feasibility of our solution to result in 10% fewer delayed or missed arrivals to the planned labor and delivery facility. This will provide validation to engage additional partners in a multi-site study. 

2. Launch a beta application within a larger scale, multi-site study over a 12-18 month duration to collect data on efficacy over a more significant population sample (e.g., at a minimum of 100,000 participants), and building the platform to include content covering a wider range of maternal health conditions. 

3. Publish results of the study in a peer-reviewed medical journal. Use the data from the study as a selling tool to rollout to health systems across the United States, and expand to include other health conditions for children, men and women.

Over the next 5 years, the team envisions that the LiteraSeed platform will be a well-established player in the healthcare space for its breakthrough approach to overcoming language and cultural barriers and bridging the gaps in patient-provider communication to dramatically impact patient care and outcomes. We will continue to improve our innovative user interface and machine-learning backend and the initial studies will help us to gather the data to realize this vision. Following this three step process, LiteraSeed expects to be in hundreds of hospitals and clinics across the United States.  

Current identified barriers are both technical and financial. The primary financial barrier is the need to raise funding to help us build and overcome the technical challenges.  

Specifically, the technical challenges are related to: (1) design of the user interface that receives image, text or conversational speech-based inputs, and (2) architecting and training of the machine learning system, which will take the form of a neural network based inference system. This would output a health condition report to the provider and offer guidance on the best action for the patient.

We are planning to secure financial support through grants and investments, as well as strategic partnerships. We're applying to SBIR grants and are forming a strategic partnership with a local health system to conduct research and testing of our technology solution. In the interim, we are bootstrapping. 

We have a team that can build the technology and inform the diagnostic algorithms, and we will hire a data scientist to help us with the machine learning algorithms.

One full time, two contractors. 

LiteraSeed’s extended team consists of highly capable, fast-moving clinicians, researchers, software architects, engineers, developers, and subject matter experts who leverage their expertise in healthcare, software development and human subjects research.

Aziza Ismail has spent the past two years performing research and field work to identify the healthcare challenges this project seeks to address. She is leading product design, user experience research, project management and business strategy. She has established a network of potential key partners.

Prarthana Mathur is a seasoned technology architect possessing a computer science engineering foundation and an IT acumen reflecting 20 years of progressive experience solving business problems. She is experienced in providing solution architecture and design applying digital, big data/machine learning and cloud solution space. She is leading LiteraSeed’s software architecture and engineering efforts. 

Nada Rizk has a Master’s degree in Clinical Research and Epidemiology from Stanford University and has held several roles as a research coordinator. She will be designing and implementing clinical studies to inform our diagnostic algorithms. Furthermore, she is well connected to several clinics that can test and utilize our technology and is directly involved with day to day patient interactions. 

LiteraSeed plans to hire 1-2 full stack web developers and a data scientist.

We are forming a strategic partnership with a local municipal hospital (Maricopa Integrated Health System - MIHS) and its Refugee Women's Health Clinic (RWHC). RWHC has collaborated with LiteraSeed during early development, including in the design of a research strategy to test the digital platform with patients from the refugee clinic. 

For the LiteraSeed platform, healthcare consumers are the primary end-users, while institutional partners are the paying customers. We’ll start by providing consumers access to the platform free of charge. We anticipate that healthcare insurers and/or institutional partners will eventually provide reimbursements and/or pay for their members to have access to the platform at a cost of $1-10 per visit per patient once our model is proven. 

Our hypothesis is that once the LiteraSeed platform is proven to help patients identify the symptoms of high-risk complications sooner and communicate them to their healthcare provider better the goals of the platform can be achieved. Proving the idea with this initial market of pregnant women will help LiteraSeed demonstrate value worthy of reimbursement from public-private healthcare insurers, particularly state Medicaid agencies, which will also support health systems embracing the platform as paying customers.

Upon proving efficacy, user adoption and retention (or product-market fit), we will target health systems as paying customers by offering backend integration on a subscription basis, where clinicians and other frontline staff become end-users on the platform. 

Initially, we are applying for SBIR funding, and plan to partner with health systems to raise additional funding to support clinical studies. Once we have gathered initial efficacy data, we will raise investment capital and also work with sponsors who can reimburse use of the platform for our users. From the earliest release of our application, LiteraSeed will reach out to institutional partners to sponsor the app for our users. As our model becomes proven and the number of users to the platform grows this will unlock other potential revenue streams. 

As a health tech startup, we want to find the right mix of clinical, healthcare and technology partners and mentors to help us develop and deliver a solution that is truly effective in positively impacting healthcare for patients, providers and health systems.  

We need technical experts to guide us and provide feedback on technology development, and to help us recruit the best and most fitting talent. 

We need medical experts and healthcare providers/clinicians to use and provide feedback on the solution and its usefulness in a clinical setting. 

We need healthcare-focused organizations with data and knowledge of how to measure outcomes and evaluate success to help us determine if this is a good solution to further guide solution development. 

We need legal support and guidance to help us properly set up partnerships.  

Healthcare and relevant healthcare advocacy organizations

Organizations that have a common mission to reduce medical error. As a starting point, organizations looking to reduce maternal deaths and improve maternal outcomes. Organizations focused on improving health outcomes for the medically underserved, including refugees and limited-English speakers. 

AI is necessary for our machine-learning based inference system which will help determine the right questions to ask the patients and help provide "helper" responses to solicit accurate data on their present health condition. It will also be useful in creating a symptom-to-condition weighted scoring and deduction model. These system approaches largely involve models of Fuzzy Logic, Artificial Intelligence (AI), Neural Network (NN), Support Vector Machine (SVM), Genetic Algorithm (GA), Evolutionary Computing (EC) or a hybrid application of all these techniques with an appropriate reasoning mechanism. The proposed system will be implemented as a rule-based expert system leveraging Machine Learning techniques and will aim to provide scalable integration with new diagnostic AI algorithms and data sources via APIs and feed the results back into its scoring model. As a novel application it will be able to listen into community health update API, apply density estimation techniques to identify the outbreak of a disease in an area and be able to apply that weighted factor to the diagnosis for a patient.

The prize will allow us to gather the resources and talent to build this AI-enabled platform and gather initial data to train our machine-learning models. 

As a health tech solution with a focus on medically underserved patients, community engagement is vital to the successful development of our solution. We will use the prize to engage with focus groups and study participants to test the effectiveness of the application. We will conduct on-the-ground research to observe, listen to and interview the beneficiaries. 

Our initial clinical target is women's health, particularly maternal care and pregnancy-related complications. The prize will help us in our work with maternity care organizations and Ob/Gyn doctors and related care providers (e.g., mid-wives and doulas) to gather data, design experiments, and set key metrics to measure the success and effectiveness of our solution to improving outcomes for women who are pregnant and up to 6 months postpartum. 

Our solution is AI-based and healthcare focused. We believe the impact will result in significant savings and increased productivity to the US economy. Inefficiencies resulting from miscommunication, which include misdiagnosis and delayed treatment, total USD 750 billion in avoidable costs per year. Obtaining a reliable “History of Present Illness” usually leads to the right diagnosis. Getting this data from the patient right from the beginning of the clinical encounter will have a ripple effect in patient safety and cost savings in part by setting the right trajectory of care from the start. 

We will ensure that data is sourced, maintained and used ethically and responsibly by building and maintaining a (HIPAA) compliant database. The sensitive data captured will be secured by applying encryption at rest as well as during transmission by using industry standard security protocols. Adequate access controls will be applied to ensure data is not compromised to unauthorized users. Only required data will be stored in the database. Users will be able to provide consent before storing their data. 

We are women-led and the first generation of the solution is focused on improving maternal care for women and girls. Our solution is a life-enhancing, innovative approach to connecting patients with their healthcare providers throughout their pregnancy and postpartum periods. In the event that a high-risk complication presents, the application will alert the healthcare providers who have known and provided care to the mother throughout the pregnancy and/or labor and delivery so that they can appropriately respond and help arrange transportation for the patient to get to the hospital best-suited to treat her condition. This will help to ensure quality healthcare to vulnerable populations, particularly those with language barriers and who are economically disadvantaged. While technology is instrumental to helping us bridge these gaps, it is incomplete without medical providers. Partnering with medical providers is an important part of our strategy and solution.