Healthcare data is growing, increasing the need for business intelligence tools to optimize efficiency while consolidating large data sources.
Real time healthcare data is growing at a pace and volume that outpace humans, creating a need for business intelligence (BI) tools that can optimize operational efficiencies while minimizing large sources of data into small meaningful pieces of information, according to leading experts at RSNA 2016. “Business analytics (BA) is a collection of technologies and functions, both hardware and software, that enable the exploration of data to gain insights that can, then, inform business decisions,” shared Katherine Andriole, PhD, FSIIM, associate professor of radiology from Brigham and Women’s Hospital, noting it can be used to automate things, optimize processes, leverage data for a competitive advantage, and sometimes forecast the future. “Because it is capable of handling large amounts of information unstructured, BA can present and visualize data in a way you can interpret dynamically in real time.” Often used interchangeably, Andriole said, “business intelligence asks questions of the data by changing the visualization, whereas business analytics is really a subset of the overall BI framework.” It uses stronger statistical and quantitative analysis to explain associations, asks why something happened, receives an explanation, and offers the opportunity to avoid repeating. “It’s not just documentation,” she said. Where does the data come from, and how is it used? There is an incredible amount of useful data available to radiology professionals through a variety of system databases, including financial, research, electronic medical records, RIS, and PACS systems. “By combining data from these disparate systems one can visualize trends, draw correlations, and detect gaps,” said Andriole, ultimately offering operational decision-making through metrics and reporting, as well as clinical and research knowledge. “The good news is there is information everywhere, and it resides in different systems. But, that is also the bad news,” she said, acknowledging the need for these kinds of technologies to understand data that is coded differently and coming from different locations. Using a four-step process, creating data that is in a unified, consistent format is the most difficult part of BA. 1. Aggregation- Information is collected from many different databases.2. Integration-Using a three-step Extract, Transfer, and Load (ETL) system, data is extracted from sources, transformed through a cleaning process that reformats and standardizes data using a data integrity and verification process, and, then, loaded into a targeted system efficiently.3. Analyzes- Unified data is stored in a data warehouse where it is ready for analysis, often using an online analytical process (OLAP) that mines the data.4. Presentation- Ability to share information in the most useful, consumable way. When using BA tools within your organization, Andriole said, “it is important to understand what metrics will represent what you are trying to do and be able to validate it.” Setting department goals, including all relevant parties, and setting key performance indicators (KPI) that reflect progress are effective in creating change. “Start with low hanging fruit,” she said, noting metrics around access to advanced imaging, order signing, patient falls, and ANCR are good starting points. Well thought out strategies ensure the best use of data “Information is not enough – it is the insight it can provide,” said Luciano Prevedello, MD, MPH, Division Chief Medical Informatics, Ohio State University Wexner Medical Center, agreeing BA offers guidance through turbulent times to manage risks, improve quality, and anticipate change. The timing is now, given the shift from fee-for-service to value-based care, he said, to use it within both private practices and academic centers as a high-efficiency, low-cost value. In offering key strategies to ensure overall success when using BA, Prevedello said there are some pre-requisites to consider prior to beginning, including quality checks in place around unstructured or messy databases prior to data mining, using data that is unique and offers differentiation from the competition, and governance within each organization related to access and privacy of the data. “Information is power. There is a natural maturity level that happens around the analytical process,” said Prevedello, explaining organizations will make wrong assumptions or predictive analysis with vendor-based solutions if the quality checks regarding data have not taken place. “There has to be someone overseeing that component.” Machine learning can also be used to verify clean data and do additional cleaning processes involved in predictive BA. The highest level of maturation within business analytics systems is predictive analysis. “There are also a bunch of assumptions embedded in a predictive model as with any other statistical model,” he said, noting an algorithm or model that may work today may not have relevancy tomorrow. It is important to continually revise models as the future changes. Once the organization embraces BA fully and can use predictive capacities, it has reached its final level of maturity. “Depending on the level of complexity at your site, at least some of the simple analytics can be used on a daily or monthly basis,” Prevedello said. Simple to more complex, BA can use pivot tables within Excel spreadsheets to determine the most commonly-used protocols, use natural language processing, such as open source, to create more mineable data from the clinical decision support tools, and, at the highest complexity, predicting the future with machine learning algorithms. Identifying delays in CD importation and increasing efficiency related to timely, accurate uploads using RapidMiner and deep learning was one example Prevedello used to demonstrate predictive BA. Identifying patients with an elevated risk for readmission within 30 days using a three-tiered stratification system (low, medium, high) and, then, embedding that information into the electronic medical record for every patient, allowing for specific, targeted interventions depending upon level of risk of the patient, is another type focused on outcomes. For example, at the population health level, predictive analytics using historical patient data can create predictions around breast health imaging, including underserved neighborhoods, access issues, and communication. “All of these techniques are very useful, and organizations should be familiar with them to be able to use them at a time that works best for you,” Prevedello said.