Virtual Mentor Health

Health workers desire decision support to provide the highest quality of care using available resources. Faced with the most dangerous maternal-newborn emergencies, frontline health care workers must draw on knowledge and skills that may be distant memories or out-of-date while they work alone in remote facilities. Virtual Mentor is a hands-free decision-support chatbot that engages the health worker in an algorithmic, audible conversation through the most efficient and effective response to a crisis. Conversation scripts rely on global, national, and local guidelines and are crafted by teams of health workers representing every level of care within a system. At the bedside of every mother and infant, Virtual Mentor augments health care capacity by elevating each worker to the highest scope of practice and minimizes human errors associated with imperfect information recall and anxiety. We aim to reduce mortality by optimizing health worker action. 

About 300,000 women die each year during and following pregnancy and childbirth. The vast majority of these deaths (94%) occur in low-resource settings, and most can be prevented. The 3 complications that cause most maternal deaths are severe bleeding (mostly bleeding after childbirth), infections (usually after childbirth), and high blood pressure during pregnancy (pre-eclampsia and eclampsia). Severe bleeding, infection, and high blood pressure can all be treated with timely and correct medication administration and actions that support the mother's vital body functions. One million newborns die each year within the first 24 hours after birth, and 30% of these deaths could be prevented if health workers correctly complete the actions that help struggling babies to breathe. We aim to decrease preventable death by 30% among women and newborns within 24 hours of birth by placing Virtual Mentor with every skilled birth worker. 

The Virtual Mentor platform integrates proprietary algorithmic content, text-to-speech software, any off-the-shelf device that runs Android OS, and Bluetooth-connected speaker/microphone hardware. This integration is accomplished in unique configurations, depending on each context for use. For example, in the highest-level care facilities in Nairobi, Kenya, users access Virtual Mentor from a large, fixed smart screen that is connected to Wi-Fi or a data plan. In a frontline facility in rural Kenya, users access Virtual Mentor as an app downloaded to their phones (with credentials) that functions offline and requires intermittent syncs. With the Virtual Mentor application on, users engage in speech-only conversation to keep their hands free for direct patient care. The audible conversation is prioritized, but scrolling visual prompts are available on-screen to support the user if desired. Streamlined algorithms rely on simple user answers such as "yes," "no," and numerals (within ranges appropriate for the questions). Simple function commands include "next," "repeat," "go back," and "stop." On the back end, Virtual Mentor documents each use and its outcome with time- and geo-stamps. Virtual Mentor can support health workers in almost any clinical scenario, and is limited only by the rate of content creation.

Virtual Mentor directly benefits health workers and indirectly benefits consumers of health care services. Maternity care nurses in peri-urban Nairobi are the first group of health workers we approached for input. They prioritized postpartum bleeding as a common and deadly scenario for which they needed decision support. Together, we test multiple iterations of the solution through low-cost, high-fidelity emergency simulations. Through countless team-based content revisions, we prioritize efficiency and effectiveness of treatments according to patient status, word choice and statement structure that reduce cognitive burden for the target user, and reassurance for both the health worker and the patient. As we move to expand the use cases and contexts for Virtual Mentor, we expect to repeat this established research and development cycle to build a library of algorithms, and variations of algorithms most appropriate to the level and location, and language of the facility in which the health workers serve. At the request of the health workers who test Virtual Mentor, the next stage of product development will be to add maternal high blood pressure, infection during or after birth, and neonatal resuscitation to the content library. 

Virtual Mentor increases the effectiveness of frontline health workers by optimizing their actions during a maternal or newborn emergency. The product's value is in the content library, but once the content is created the solution is affordably scaled by user and by single use (session). By increasing health worker effectiveness, Virtual Mentor helps women, new mothers, and newborns access higher-quality care. We aim to deploy Virtual Mentor in frontline facilities in Kenya and request Solver support to progress responsibly through regulatory requirements, establish the brand in Kenya as a reliable decision support, and measure the impact of use on mortality.

Several mobile applications in this space are meant to provide didactic instruction, in the format of an online course. A well-known example is the Safe Delivery app, which provides text and video instructions for managing obstetric complications. Users can work through the course content at their own pace and complete quizzes to check their knowledge. Another example is Bodhi Health Education, a mobile nurse training app that uses gamification and hypothetical patient cases. These self-paced learning apps provide health workers with content, but they are not useful during hands-on patient care (or hands-on practice). 

Other mobile apps offer an interactive checklist approach to complex clinical cases, such as the care of sick infants. These apps are text-based and require the health worker to be visually and manually engaged with app, holding the tablet or phone while they enter data and tick boxes with their fingers.

Virtual Mentor is the first application to guide the health worker, in real time, through the correct response to the patient's presentation, in a voice-only conversation. While the health worker's hands are administering medications, applying pressure, massaging, and inserting catheters, Virtual Mentor provides decision support without requiring any manual manipulation of the device. Virtual Mentor's conversational design keeps the health workers engaged in a timely series of interventions that begins with diagnosis and proceeds systematically through assessments and treatments, encouraging the health worker to "push through" fear, forgetfulness, and distraction to resolve the patient's problem.

At this time, Virtual Mentor relies on open-source text-to-speech (TTS) engines, and English is the first language we have tested for usability in Kenya. As Virtual Mentor Health grows, we will experiment with TTS for other languages and leverage machine learning algorithms to optimize speech recognition in each new context.

As of June 2020 Virtual Mentor has been piloted for usability with 16 nurses and midwives in Kenya. In one year, we project that 400 nurses and midwives will have used Virtual Mentor as a training tool while we gather more usability and effectiveness data, and that a focused group of 6 midwives will be actively using Virtual Mentor in a clinical care pilot study. With solid safety data from study in actual patient care, we project that in 5 years 10,000 nurses, midwives, and other frontline health workers will be using Virtual Mentor as a decision support tool during real emergencies.

Virtual Mentor Health needs a fully-developed sales model. Because the market for health care worker decision support tools in low-income country contexts is not fully understood, we do not yet have a convincing case for capital investment. We need support from Solve MIT to develop our business case.

Co-founders: 4

Software developers (contractors): 4

We are a team of perinatal health workers, providing care to mothers in the United States and Kenya. We have health worker training, public health research, and implementation experience in many countries in Africa, Asia, and the Americas.

We currently partner with Kenyatta University in Kenya to conduct ethical studies in that country. In the near future we will partner with University Research Co.,LLC (URC), University of California San Francisco (UCSF),

Population Council, and (Management Sciences for Health (MSH) on implementation research activities in Madagascar.

Solve can help Virtual Mentor Health develop a business plan for sustainable, for-profit growth.

We hope to partner with a business strategy team that understands the market for decision support tools across all global context but specifically in Africa and low-income countries in Asia.

Virtual Mentor augments the scope and capacity of frontline health workers, increasing their effectiveness even in remote areas and without immediate access to a physician consultant. Through a speech-only conversation with Virtual Mentor's chatbot, frontline health workers proceed calmly and systematically through the most up-to-date treatment algorithms for an emergency. 

With the Health Workforce Innovation Prize, we will take careful but rapid steps to test Virtual Mentor in actual patient care. First, we will build out the suite of maternal and newborn emergencies that Virtual Mentor will support in a maternity (labor and delivery) unit. Next, we will complete a small pilot with expert midwives at referral facilities who are unlikely to make clinical decision errors regardless of Virtual Mentor's performance. After a proof of safe use in actual patient care, we will execute the first clinical trial in frontline maternity facilities to test Virtual Mentor's effectiveness in preventing perinatal injury and death. 

We plan to carry out these activities in Kenya, but we are open to other geographies if we find partners who want to test Virtual Mentor in their frontline facilities.