Come see us.Precocity LLC
3400 N. Central Expressway, Suite 100
Richardson, TX 75080
Compliance requirements never sleep, and even world-class hospitals can have trouble keeping up with them. One such hospital has not only made it simple and frictionless for employees to report safety issues but also used the tool itself to increase employee buy-in, even with the onset of a pandemic.
To comply with regulations on delivering health care safely, a renowned medical research university and teaching hospital was using a feature of its electronic health records system to track safety issues in its facilities. The system already held all the information about patients, medications, equipment, and locations, so hospital employees used it to report problems like prescription errors, and “good catches” like cleaned-up spills.
But entering incident data into the software was a daunting process, whether for doctors, nurses, administrators, or maintenance staff. As important as it was to collect and organize safety data, it took too many minutes out of a workday to stop and report even a minor incident. The reporting requirement began to feel even more onerous as the COVID-19 pandemic stretched hospital crews thin.
After ten years of using the health records system, the hospital’s safety directors finally understood too many safety details might be going unreported. They turned to Precocity to develop a frictionless application that any of their 37,000 employees could use easily and quickly.
From the outset, Precocity worked with the safety directors to establish goals for the application:
The overarching goal of the project was to reduce friction and make it as easy as possible for users to help the hospital comply with health care standards for safety.
The hospital’s safety directors understood compliance, but they had never designed software before. The team leads at Precocity met with them in daily stand-ups, sometimes scheduling all-day working sessions to discover everything the app would have to address.
Precocity structured the project along agile guidelines, with sprints, end-of-sprint demos, and retrospective meetings. They created a continuous integration/continuous delivery (CI/CD) pipeline so hospital users could see progress daily and use app updates promptly. To ensure the hospital could maintain, extend, and support the app well into the future, Precocity specified long-life software development technologies that the client could maintain.
Working in as current a landscape as practical, Precocity provided full-spectrum engineering, including architecture, coding of front and back ends, testing, debugging, launching and integrating with the hospital’s AD. In short, they delivered everything needed to conceive, build, launch, and maintain a software application.
Besides developing the application, Precocity began to see opportunities to apply artificial intelligence (AI) and natural language processing (NLP) to the ten years of safety reports already in the system.
A big opportunity for automation was in the process of incident classification, after a report had been submitted. A team of reviewers had to manually classify and label incidents — fall, spill, wrong prescription — based on short phrases, paragraphs, or entire pages of free text. What if the application could automatically classify the submitted incident, then route it to the person responsible for dealing with it?
The hospital approved the idea, and Precocity started applying supervised machine learning to the ten-year-old database of users’ incident descriptions and reviewers’ labels. The result was a trained model the new app could use to analyze new incidents, assign one or more labels, and automatically route the incident report to the right person.
Another opportunity was further upstream at the time of incident reporting. What if the application compared every new incident to similar incidents in the past? For a given description made up of a few sentences or paragraphs of free text, what if the app showed users the ten most similar incidents from history? Users could then base their report on similar language instead of having to create it entirely from scratch.
Again, the hospital approved the plan, so Precocity applied document embedding to assign a numerical representation to thousands of existing incident descriptions. Now, as users open new incidents, the application performs a nearest-neighbor similarity search to compare the new embedding to the historical ones and ultimately show them similar incidents while they’re still filling out the report. The feature also aids in compliance by helping users include information that was useful in previous, similar incidents.
Achieving better compliance with AI
“It was a pleasure to work on such an impactful application. To be able to apply deep learning & natural language processing techniques, for such an important use case like safety reporting, was very gratifying.”
David Gillen, Chief Data Officer
The tip of the spear was to reduce the friction involved in getting hospital employees to take a few minutes and write the report. Streamlining the report screen made it easier to fill out and less daunting. Precocity’s innovations included making the application adapt to job roles and problem categories; for example, equipment problems don’t prompt for the same information prescription problems do. They also included populating drop-down menus with only those choices relevant to each user’s access and job purview.
But the eighteen-month development project went even further, leading to greater efficiency and safety while helping to avert a growing labor problem during a pandemic. Along the entire lifecycle of a safety incident, the application captures more safety data, reports it more easily, and saves more time than the legacy system did.
The AI-driven process of reviewing and labeling incidents ensured the right people received notification immediately when there was an incident in their area. The hospital accelerated that process by creating hundreds of separate roles and permissions so employees would receive notifications of only the incidents relevant to them, reducing inbox clutter.
Recipients of reports can add to or change the assigned labels, which continually refines the machine learning model. They can ask questions of the person who submitted the report; if the submitter requested anonymity, the application maintains it. Workflow in the application lets recipients assign the incident for resolution, then notifies stakeholders once the problem has been addressed.
The hospital calls the application “Hero” and uses the internal home page to promote heroic stories of good catches — the efforts of people who make the workplace safer. That encourages even more employee buy-in, making Hero much more than just a safety reporting tool.