Image from Impact Report Volume 21, cover designed by the Duke Clinical Research Institute. Any use of the terms ‘Bed View’ or ‘Bed Watch’ is purely descriptive and does not imply affiliation with, endorsement by, or reference to any trademarked product or company.
A brief history
In 2018, Dr. Jason Theiling responded to a Duke Institute for Health Innovation call for proposals with an intriguing idea: What if we could use machine learning to predict whether patients were ready for general hospital admission or intensive care?
We wanted to get this information into the hands of the people who needed it most. DIHI and Emergency Medicine identified the Bed Czars and Bed Control team as the perfect place to start. But before integrating anything into the EHR, we needed to test a simple, working prototype with them.
The need: a real-time visual showing each patient’s bed status, the model’s predictions, and other key information at a glance.
The solution: Rapidly-updating color-coded patient beds, beyond Epic and in the hands of Duke clinical and technical developers. DIHI Bed View was born.
While we were validating the model, COVID-19 hit. Soon we asked, “What if this same tool could instantly show us who had positive COVID symptoms and tests, and where they were across the health system? What if we needed to track ECMO machines and ventilators in real time?”
Bed View quickly scaled to include data from all Duke hospitals and became viewable to a small group of COVID response leaders. (Impact Issue 21 p.24-25)
Since then, Bed View has been quietly working behind the scenes, helping DIHI test and validate real-time models (models that update every 5, 15, or 60 minutes) for hospitalized patients, and models that helped us test ways to identify specific patient groups. All uses were IRB-approved.
Examples
1. The Bed View was a workflow tool when using DIHI’s mortality-risk-predictor-model for advance care planning via a transitions of care toolkit. Each bed holding a patient at high mortality risk was color-coded red. Then, nurses reviewed the Bed View to identify patients to attend to, plan how to resource the hospital floor, and locate the patient for care.
2. The Bed View was used to support a pilot of Acute Hospital Care at Home. The Bed View dashboard included an algorithm identifying candidates for Hospital at Home, and color-coded those patients accordingly within the Emergency Department and Internal Medicine units. When there was a new candidate, the leading physician was notified and the clinical team could attend to the patient at the right place and time.
3. The Bed View, with an event paging system, ensured that a behavioral emergency alert system could be applied within real-time workflows. The real-time behavioral emergency outcome label, the predictive model, and secure paging technology were used to alert the BRT nurse, prompting a chart review and bedside assessment as needed to identify and intervene in imminent behavioral emergencies.
2026
Ideas that have risen include:
- To help solve delayed patient placement decisions with real-time bed availability visualization by using the color-coded length of stay feature to identify beds likely to discharge soon.
- To help solve unbalanced nurse-to-patient ratios across units with live acuity monitoring by using the risk color codes and department-level patient counts.
- To help solve inefficient discharge planning workflows with prioritized patient identification by using the length of stay filter and clickable patient list generation feature.
- To help solve ED boarding and hallway overcrowding with pool bed visibility by using the Pool tab that shows all patients in shared spaces like ED halls and periop rooms.
- To help solve preventable device-related infections and complications with system-wide device visibility by using the text overlay feature to instantly identify all patients with central lines and Foley catheters for daily necessity reviews.
Precedence
Admittedly, the “Bed View” visual is not a new concept. DIHI’s Sepsis Watch’s red and black cards idea inspired the colored bed rectangles. An Epic user can identify a patient bed location after several clicks or can view patient location in tiles or rows with a General Electric (GE) Healthcare “Hub.” However, to our knowledge, identifying patients’ places, care timing, or care procedures can be done by few other products at such a low cost. Furthermore, DIHI’s on-site and ready to develop and test specifically to Duke Health’s needs.
- https://www.bedwatch.com/bed-control
- https://www.achalahealth.com/advancedbedmanagementsystem
- Use of a novel electronic maternal surveillance system to generate automated alerts on the labor and delivery unit
- A Successful Model for a Comprehensive Patient Flow Management Center at an Academic Health System
- An Electronic Dashboard to Monitor Patient Flow at the Johns Hopkins Hospital: Communication of Key Performance Indicators Using the Donabedian Model
Architecture Overview
Before 2023, the need for timely access to a wide breadth of clinical data and model results, such as active infections, anticipated discharge destination, and mortality risk – presented a significant technical challenge. Extracting a consolidated view of this information for admitted patients required complicated Structured Query Language (SQL) joins and aggregations that spanned many tables and databases. Complex user demands, materials, and code caused the initial implementation of the Bed View dashboard to be resource-intensive and response times to be unacceptable – minutes in some cases.
The DIHI data engineering team needed an organized and optimized data mart to provide quick snapshots of any admitted patient‘s clinical status. Previously, DIHI housed the Bed View dashboard source data in the DIHI Pipeline databases (Active and Archive) and several model results databases. The goal was to update this data mart every five minutes – so the Bed View dashboard could deliver near real-time insight.
Extracting the source data and updating the data mart every five minutes required a parallel processing approach that an engineer could scale up as they identify new data subject areas for extraction and inclusion in the Bed View dashboard.
The following narrative and diagram depict the current architecture we implemented to ensure the Bed View data mart is updated every five minutes. The Bed View dashboard now consistently presents refreshed results in seconds.
Bed View Data Mart Extract-Transform-Load Data Flow
- Every five minutes “near” real-time ADT data is extracted to determine admitted encounters/patients and their current bed: (location > department > room >bed).
- For each data subject area of interest, such as active infections, the list of encounters is split into equal chunks and a message for each chunk is placed into a work queue.
- Workers process queued messages. For example, they extract the active infections for the chunk of encounters and update the Bed View data mart. These workers run in parallel and can be scaled up. The 2024 implementation had 16 workers.
This parallel worker paradigm allows simple, less resource intensive SQL queries to extract data in parallel and to load results into the Bed View data mart in parallel. The parallel extracting and loading minimized the window required to refresh the data mart every five minutes.
Conclusion
We foresee Bed View becoming a standard tool Duke Health employees use to ensure they meet daily access, quality, and cost goals. Any bedside care provider can view analytics at the patient level at their current time and place while, in one view, maintaining the perspective of the floor, service line, and hospital. A manager can view their whole area or unit for a quality assessment but drill down to the beds driving the failure or success. Any employee could view benchmarks set from the past and see how the last five minutes of work today will compare to them. The “Bed View“ dashboard makes it possible to understand the current state better and implement improvements on the spot.
