Authors

Inspiration

The Problem No data set currently exists that can track coronavirus exposure and spread within hospitals, and hospitals lack the bandwidth to create it for themselves.  It is urgent that we create the data necessary to optimize tradeoffs between quarantining exposed healthcare workers and maintaining the capacity necessary to treat unprecedented surges of critically ill patients. In areas where the pandemic is reaching its peak, hundreds of healthcare workers have been exposed to COVID-19, but we cannot afford to quarantine all of them for 14 days at a time.

The Current State  Through interviews with frontline workers and COVID Command Center leadership, we discovered vulnerabilities in the current process for identifying team members who may have been exposed to COVID after a patient or co-worker tested positive.  Someone, typically a charge nurse, has to do a manual chart review and “ask around” to figure out who was involved in the care of a COVID positive patient or who was on a team with a COVID positive provider. Charge nurses must manually create the list and manually notify relevant team members. This approach cannot scale, cannot give us insights into patterns of exposure and spread, and takes staff away from mission-critical patient care.

Solution and Implications Automated clinician coronavirus exposure tracking will play an important role in helping health systems and public health curb the pandemic. By tracking clinician exposure, we will be able to: 1) Test and quarantine healthcare workers with the highest levels of exposure.  2) Empower public health officials to monitor healthcare workers with high exposure and provide guidance on preventative measures. 3) Create a registry of healthcare workers and patients who have tested positive for COVID-19,  then recovered, and are presumed immune.  4) Uncover ways to optimize staffing and scheduling to minimize exposure.

Fortunately, prior to this crisis, the Surefhir team had already built a care team tracking AI for care coordination and attribution. In a world of team-based care, it is very difficult to keep track of who has been involved in care and who is responsible right now. This care team tracking tool is the foundation of our Infection Exposure Tracker. Surefhir is a portfolio company of Jumpstart Foundry, and is a proud member of the MATTER community.

What it does

Our Infection Exposure Tracker works like a detective: it creates a log of who provided care to infected patients, who worked side-by-side with an infected clinician, and who was in the vicinity of infected patients and clinicians.   First, the tracker surveilles chart and flags patients who have tested positive for an infectious disease, in this case, COVID-19. Then, the tracker looks back to see which clinicians have been involved with that patient, their duration of their involvement, what they did. The “lookback period” can be modified as new information becomes available about COVID-19’s incubation period, but we assume it is 14 days. This action produces a “care team timeline” that tracks how care team membership changed over time. Each clinical task (a procedure, physical exam, med administration, order, etc) has an associated “exposure factor” per disease. Some clinical tasks, such as doing a procedure of physical exam, necessarily involve close contact with a patient for many minutes. Other clinical tasks, such as progress notes and orders, do not require extended periods of time at the bedside. In the case of COVID-19, spending time in a patient’s room increases the chances of contact with virus-containing droplets. A total exposure score is calculated for each clinician based on the cumulative duration and closeness of contact with infected patients.  When a clinician tests positive for COVID, administrators can look to see which clinicians they have collaborated with recently thanks to the care team timeline feature. We would access COVID-19 test data either through direct administrator input, or through linking clinicians' MRN to their NPI.  The tracker also keeps a registry of patients and healthcare workers who have tested positive for the virus, who are then presumed immune after they have recovered.  Once serological tests for COVID-19 antibodies become widely available, this status can be confirmed.  This registry will help hospitals make staffing decisions to reduce exposure for healthcare workers without immunities, as well as aid public health agencies in their surveillance efforts.  Hospital administrators will see a ranked list of clinicians, from highest to lowest exposure,  which can be filtered by service line. The dashboard also displays insights into both exposure, infection, and recovery rates.

How we built it

Part 1: Patient-clinician exposure     Our CTO, Alok Mathur co-founded Novo Innovations, a successful health information exchange company that sold to Aetna for $500M. We have extensive experience in leveraging distributed computing to build scalable, robust systems that interface with electronic medical records. Our Infection Exposure Tracker is in essence a custom health information exchange with data visualization and analytics capabilities layered on top. We built a node that receives Hl7 2.x messages on diagnoses, orders, procedures, notes, vitals, ADTs, and medication administrations. From these messages we uncover 1) who has been diagnosed with the disease 2) who is on their care team and 3) the actions of each clinician. Even if a diagnosis is not yet documented, we can potentially look at lab results and vitals as well to determine if a patient has coronavirus. Each action is assigned an exposure score, which is weighted based on the amount of close contact with a patient it requires. Our current exposure score methodology is based on interviews with clinicians on the duration and degree of close contact involved in each clinical action. We intend to nuance and validate this methodology through input from experts in infectious disease experts, epidemiology, and environmental health and safety.

Part 2: Clinician-clinician and clinician-patient exposure By linking clinician’s NPI to their MRN, we will know if a clinician tests positive for the virus. (We will also make a manual-input failsafe.) When this happens, we look at the last two weeks of data to construct their patient panel and discover 1) which clinicians worked on the same patients at the same time and 2) which patients might have been exposed.  Administrators can pull up a list of these teammates and patients in order to take appropriate measures. There is also the matter of secondary spread-- clinicians who did not directly collaborate or round together, but who shared the same working spaces on the same days.  Because we can easily extract patient panels for each clinician for each day, we can trace who was working on the same unit at the same time as an infected clinician.

Challenges we ran into

The main challenges we ran into involve the limitations of how APIs are implemented in current EMRs. Neither FHIR nor EMR proprietary APIs were well suited to our use case, since they are geared towards developing patient-facing apps or clinical decision support tools. They are meant to provide diagnostic and administrative data about a patient, not their care-team. There is no efficient way to capture which patients tested positive for COVID given an entire patient census. In fact, most vendors do not implement an API to obtain a system-wide census. They are set up so we must query each patient’s chart individually. We cannot access orders, notes, other clinical activity across the system without knowing which specific patient and encounter one wants. This led us to scrap the idea of building a API-based system from scratch, and instead we built off of our existing architecture to pull HL7 messages.

Accomplishments that we're proud of

1) We have extensive experience in distributing PHI  from hospital and lab systems to other hospitals, outpatient settings and public health agencies. We know we can use HL7 2.x to obtain the data we need to track patient infection and clinician exposure. We built an HL7 2.x node and tested it on synthetic data we got from our pilot partner health system.  2) We discovered that the current state of exposure tracking was asking charge nurses to do chart reviews and “ask around” to figure out who was exposed to an infected patient. This is not going to scale once the pandemic approaches its peak! 3) To our knowledge, we are the only system that automatically tracks care team membership and staff exposure rates. 4) To our knowledge, we are the only group close to developing a registry of patients and health care workers who have recovered from COVID-19.

What we learned

There are important public health use cases for this data, and we can leverage our experience building health information exchanges to make it readily available to public health agencies.
Armed with data on exposure and infection rates in healthcare workers, public health agencies could develop and provide data-driven guidance to healthcare workers who have been exposed. Public health could also take the lead on monitoring exposed health care workers for symptoms and take measures to prevent community spread.  We also learned that nobody else seems to be working on a federated registry of healthcare workers and patients who are presumed immune to COVID-19. This registry would not only help healthcare workers optimize staffing, it would also enable more people to go back to work and help restart the economy.

What's next for Clinician Coronavirus Exposure Tracker

Development Plans Prior to a market launch, we will work with experts to improve and validate our exposure score accuracy. One way to do this would be to train our AI on exposure scores calculated by experts. This hackathon has helped us make connections with epidemiologists and infectious disease experts that we will leverage in our next phase of development.  Through the ONC, we will connect with public health leaders to define how to interface with their systems to deliver actionable data on COVID-19 exposure, infection, and recovery. One potential outcome of a collaboration with public health would be a system to monitor exposed healthcare workers using existing text message notification capabilities. This would take some pressure off of health systems to figure out how best to keep exposed healthcare workers and their communities safe. This hackathon inspired a new priority: leveraging the data we collect to build a COVID-19 immunity registry. While our COVID-19 immunity registry will initially focus on healthcare workers, we will develop a plan to expand it to include patients. As serological testing becomes more widely available, we will capture results of COVID-19 antibody testing for this registry.

Implementation Plans Here’s what it would look like for a health system to implement the Infection Exposure Tracker. We would integrate with their “support environment”, not necessarily production. This tracker only requires once a day updates, taking feeds from the support environment does not risk the performance of the production system, which is often an objection that prevents integration projects from moving quickly.  It would take three man-days for a typical hospital IT department to set up a server to send us copies of the HL7 messages of interest. All PHI will be stored within the hospital’s secure environment, whether on-premise or in the cloud.

Projected Timeline April - Exposure score improvement and validation, identifying strategic partners in public health and health systems. May- First tests in non-production environment.  June- Market launch. Publish on EMR marketplaces, starting with Epic, which covers more than 28% of the nation’s hospitals, and 58% of hospitals with more than 500 beds.

Try It out

Hackathons

Technologies

hl7

Devpost Software Identifier

253067