RSNA 2016: How to apply business analytics to radiology

 - laptop chart data

There’s a wealth of usable data in every imaging practice, including performance benchmarks, scanner data and EHR records. Building a business analytics framework to extract and visualize data can be challenging, but use cases demonstrate impressive gains in productivity and value.

A Nov. 30 session at RSNA 2016 laid the foundation for implementing business analytics in imaging practices, taking attendees through the pipeline from data collection to visualization to enacting improvements gleaned from the analytics.

Data Collection to Clinical Corrections

“Business analytics is a collection of technologies, applications and practices that aggregates data from various systems, integrates them into one database, stores the data until it’s ready for analysis—and then the final piece is presenting the data to you, the decision-maker,” said Katherine P. Andriole, PhD, an associate professor of radiology at Harvard Medical School. “The hard part is making sure the data is in a unified and consistent format.”

This usable data will undergo the "Extract, Transform, and Load" process, otherwise known as ETL, according to Andriole. Information such as spreadsheets, text files, PACS or RIS data will be reformatted to fit the destination database, and the process must remain efficient as it scales with larger datasets.

Andriole used the ETL process for clinical scanner data from different sites to illustrate the procedure. Some sites used Microsoft Excel to track individual scans, while others used XML documents.

“The word ‘resource’ is what is used in the Excel file to indicate a CT scanner. In the XMl file, the name of the individual scanner is used,” said Andriole. “When we get to the transformation step, we have to tell the computer that resource and scanner do the same—unifying the variables.”

The next step is storing the data in the specialized format, usually in a data warehouse or data mart. A data warehouse is a safe, central repository of data that allows for a consistent reporting structure, providing links between disparate sources of information.

The third step is data analysis, typically accomplished using relational databases. However, Andriole believes that Online Analytical Processing (OLAP) databases are more valuable due to their flexibility.

“It’s a multi-dimensional analytic process. The databases are configured that you can have navigational hierarchical and relational databases,” she said.

The final step is visualizing and presenting the data, using dashboards, predictive models and more. Andriole stressed that data visualization should not be static; manipulation of the data during a presentation can reveal different viewpoints.

In addition, determining the goals of a business analytics system requires a department-wide effort. Everyone from radiologists and technologists to financial staff need to be involved in the decision process, setting goals and selecting key performance indicators (KPIs) that indicate if those goals are being met.  

“Once you have these, you present to your target personnel,” said Andriole. “Perhaps it’s a radiologist you want to speed up with their signature time. You present them with information about their colleagues around them, the target learns and modifies their behavior, and you keep repeating this process.”

Typical KPIs include operational efficiency measures such as report turnaround time or number of unread exams. Safety KPIs include wrong patient incidents, patient falls, or activations of the hospital’s critical result notification tool—indicating a patient’s death or serious illness occurring under their care.

Putting the Programming into Practice

Radiology applications of business analytics are almost endless, but Tessa S. Cook, MD, PhD, outlined three widely-applicable uses: radiation exposure monitoring, resident analytics and follow-up monitoring.

“Until we saw the [patient dose] numbers, we didn’t realize there was a problem,” said Cook, an Associate Professor of Radiology at the University of Pennsylvania School of Medicine.

After implementing procedural changes including real-time dose adjustments made by a radiologist, Cook was able to visualize the results.

“We had pre- and post-change doses with three different studies, showing a downward slope on the graph over time,” said Cook.

These KPIs used in the resident analytics project were a little different, according to Cook.

She recounted some residents had expressed concern about a lack of feedback, telling