Getting to the bottom of hospital finances: Florida Hospital, an experienced data miner, stretches its extensive use of data mining to deliver financial results that will improve its bottom line - Data Warehousing/Mining

Health Management Technology,  August, 2003  by Alex Veletsos

• E-mail
• Print
• Link

In the late 1990s, the IT department at Florida Hospital in Orlando, Fla., was approached by IBM to pioneer the use of business intelligence (BI) solutions--specifically, data mining--in the healthcare industry. We took IBM up on its offer, and after detailed consideration, we began researching specific BI initiatives.

In addition to demographic clustering and association-based analysis, we considered predictive modeling. This style of modeling is popular among data analysts and applied statisticians, and it is usually approached as a sequential process in which data is first segmented and predictive models are then developed for those segments. The modeling is used toward the assessment and discovery of hidden relationships in the data.

At the time, using predictive modeling for the identification of previously unknown trends and correlations was quickly becoming popular in data-intensive industries such as financial, retail and manufacturing. When IBM approached us, it was a no-brainer, considering the large amount of patient data that resides in our hospital. We were pleased to be presented with an opportunity to maximize the use of our data for predictive accuracy.

Setting the Stage

After conducting various research and testing efforts, we launched two pilot studies using IBM's Intelligent Miner for Data to study the factors associated with readmission for patients with congestive heart failure and patients suffering from stroke. The successful data mining analyses resulted in 100 percent ROI for the cost of the software in one year's time. Because of the value of the BI tools, we expanded our use of data mining technology to other pertinent hospital projects, applying it most recently to the analysis of financial data.

The healthcare industry--specifically, hospitals and clinical organizations--is often plagued by unpaid bills, collection agency fees and outstanding medical testing costs. As a matter of fact, hospitals typically end up paying 30 percent to 50 percent of recovered bad-debt revenue to outside collection agencies. Because of these factors, Florida Hospital is focused on ways to reduce accounts receivable and bad-debt balance.

Using IBM Intelligent Miner and DB2 database software, we recently implemented an advanced financial data-analysis project. The project's goals, based on predictive modeling, are twofold: to reduce bad debt and ultimately increase revenue, and to identify causation factors for increased MRI or CT scans and ultimately decrease expenses. The IBM software used in the project allows us to dig up answers to questions we previously thought were unavailable.

The implementation of business intelligence software has offered us a distinct competitive advantage. Through implementation, we have been able to identify trends and correlations that were previously invisible--with minimum investment. The benefits are invaluable, so looking for new ways to utilize this software has not been a tough decision to make.

The Project, Part 1

Part 1 of the project, as mentioned, is the use of predictive modeling for the reduction of bad debt. The study was initially completed in 2001, and the model was refreshed again three months ago. It includes data from approximately 2,400 patients, including both inpatients and outpatients.

Here's how it works. First, the IT department carves out a segment of the DB2 database for the predictive modeling software to go after. Once patient data is added, the predictive modeling software goes after that segment of the database and provides the patient financial services (PFS) department with a list of patients, from the most to the least likely to pay the bill.

The answers found are based on a variety of data variables, including credit factors, demographic information and previous organizational payment patterns. Each of these variables aids the financial department in assessing who of the thousands of patients will most likely avoid bill payment. In turn, Florida Hospital is able to predict who is most and least likely to pay in the future, and may be able to save the 30 percent to 50 percent in collection agency fees on a large portion of its self-pay receivables.

The potential of using this predictive model is approximately $200,000 a year in savings. Of course, the hospital's PFS and IT staff will continue to refine the data model to obtain more information and discover new elements of data toward future cost savings.

The Project, Part 2

Part 2 of the financial data analysis is the use of association analysis and demographic clustering in an effort to identify reasons and causes of increased per-patient test volumes. Florida Hospital is not alone in the quest to assess the origin and causation of increases in costly testing procedures. While increased testing volumes may indicate increased expenses that may not be fully reimbursed, they may also impact inpatient length of stay (LOS). We have focused the second part of the financial analysis project on MRI and CT scans, two very expensive tests to administer and maintain.