This article is based on a presentation at IBA’s Spring 2023 Conference on Healthcare Analytics by Keith Kilgore, Executive Director of Clinical Performance Improvement at IU Health and Joshua Sadowski, Senior Data Analyst for Quality, Safety, & Infection Prevention at IU Health Methodist/University Hospitals.
The data and the processes surrounding healthcare analytics are complex, which is clearly demonstrated by the analytics infrastructure employed by Indiana University Health and the projects taken on by the data analysts at IU Health. The infrastructure includes four main layers and four domains of tools, according to Joshua Sadowski, a Senior Data Analyst for IU Health Methodist & University Hospitals.
While the infrastructure is layered, Sadowski said that the purpose of performing analytics for healthcare professionals is simple: to help clinicians give patients better care.
“As people in informatics or support positions or analytics, at the end of the day, what we really want to strive for is supporting clinicians and making their lives easier,” Sadowski said.
Sadowski works in infection prevention, quality, and safety, supporting clinicians on the front line by giving them the information they need so they can improve quality and increase safe patient care. Sadowski began his presentation by explaining the purpose of data, which he defined as “a representation of the real world at some point in time, with error in it.”
Sadowski spoke in the first part of the “Healthcare Analytics Within a Large Indiana Health System” session during the Spring 2023 Institute of Business Analytics Conference on Healthcare Analytics. In his presentation, “Analytics Within a Clinical Environment,” Sadowski described the four layers of the data infrastructure at IU Health: the Source System, or databases; Source Mart, where data is loaded and copied into tables; the Integrated Layer, which creates a clean data; and Data Solutions, which allows end or business users to interact with the data easily.
Sadowski also gave an overview of the four domains of tools his team uses to sort through data:
- Clinical Research: The team uses primarily R & R Studio for this data, plus Encore to view patient enrollment for research studies, and REDCap as a secure database for storing research data.
- Operational Reporting: Sadowski’s team relies on Microsoft tools including Power BI, SSRS, and SSAS for operational reporting.
- Orchestration: To combine and organize data taken from the previous tools, the team uses Data Factory, Jams, and PowerShell.
- DevOps: To develop software and deliver data, they use Azure DevOps and Git.
Sadowski talked through three projects that his team has worked on, including work with computational phenotyping to monitor clinical pneumonia at one hospital and a project to decrease the spread of a multi-drug resistant organism that spreads easily within hospitals.
“Complex projects take time and thought,” Sadowski said. “Data in healthcare is very complex and not always clean, so all decision points that we make along the way matter. Take time with the data, really ensure that you’re cleaning and understanding the data, then using that data as you go out into the world.”
Sadowski’s third project involved developing a predictive model to support clinical decision making around hospital-acquired pressure injuries. The predictive model was successful at reducing injuries within the units, but Sadowski said that the team’s main concern was making the information easily accessible for clinicians.
“We care about what the dashboard looks like for the end user,” Sadowski said. “Does it create more questions than answers? Does it actually help them alleviate the clinical decision-making process, or does it cause them to stop and think more? Does it actually lower time and alleviate pain?”
Data and Healthcare Staffing at IU Health
The next portion of the session was given by Keith Kilgore, the Executive Director of Clinical Performance Improvement for IU Health. In his presentation, “Healthcare Staffing — How Much Work Is There?,” Kilgore discussed using real-time data to determine staffing needs on the front lines during the height of the COVID-19 pandemic.
Similar to Sadowski, Kilgore focuses on supporting clinicians, but from a different perspective. During one peak, a chief nurse executive approached Kilgore’s team for help solving nursing staff shortages. A hospital had 40% too few nurses, the chief nurse executive told Kilmgore, and was full of COVID-19 patients.
Kilgore’s first step when he was approached with this question was to determine what the real problem was. To figure out the main problem, Kilgore started with the nurses on the front line.
“In my role, I go right to where the people are,” Kilgore said. “It’s good to talk to the executives and to be in meetings, but where you really find out what’s happening with the work is at the front lines. We went directly to the nurses and started watching them, looking, walking, talking, and understanding exactly what they were doing.”
To figure out how many nurses were really needed on the floor, Kilgore explained that he didn’t need to be a nurse — he needed to have good observation skills. After some observation, his team realized that the nurses needed specific information to guess how the day was going to go. They built infrastructure to stream real-time data and merge disconnected systems like timekeeping, electronic medical records, and electronic resource planning.
Using this data, they realized that the way to determine nursing staff needs can be compared to a three-layer cake, with a baseline understanding that there are standard practices of care that nurses complete without documentation, and there is a different amount of care required on each unit and for each patient.
- Medical Layer: A majority of the tasks that nurses complete on the floor to take care of patients, including following doctors’ orders to administer medication and insert IVs. This does not include the standard practice of care.
- Patient Characteristics: This layer takes into account that individual patients have different traits, such as not speaking English as a first language, that can change the algorithm.
- Social Characteristics: These include people-centered dynamics that have an impact on labor for the unit, including managing visitors.
Kilgore and his team packaged this data set to determine the amount of time and number of nurses required for different units. What made this process different from a traditional data analytics project was that Kilgore engaged with the nursing team from the get-go before starting a change management initiative.
“By working directly with the nurses, they really realize that they are the ones designing this,” Kilgore said. “It’s a little bit like a customer-centric product delivery model. With data, you have to build it from the bottom up to make sure that anything matters at the top.”
Leave a Reply