|
|
|||
|
|
|
|
|
|
||||
|
Background |
Health Data Desk is a suite of point-and-click, button-driven software tools to help government regulatory agencies, insurance companies, HMOs, and healthcare providers find and explore patterns and exceptions in hospital electronic billing claims records. Using both public and private databases, the top-level displays identify hospitals with anomalous medical records, coding/billing patterns, inefficient and/or clinically inappropriate treatment patterns by clinical product/service lines. The software also identifies "financial and clinical "best practice" institutions by region. Drill-down tools focus the analysis on a specific subset of records. An analysis of Medicare public data using Health Data Desk is leading to the recovery of hundreds of millions of dollars by HCFA and the Department of Justice from selected hospitals throughout the country. Health Data Desk's approach to the analysis of healthcare fraud is informed by leading-edge statistics and data analysis. This focus is differs from traditional, accounting-based approaches. Rather than examining each tree in the forest of records, Health Data Desk begins by viewing the forest as a whole, identifying likely areas of fraud, and then drilling down to examine individual records and sets of records. This approach can identify patterns of fraud and increased expenses that cannot be detected at the individual case level of analysis, greatly increasing opportunities for recovery and cost reduction. Health Data Desk is built on Data Desk, Data Description's award-winning Exploratory Data Analysis software program. Data Desk is a proprietary data management, analysis, and display system unlike any other available. The methods we employ grow out of the discipline of Exploratory Data Analysis (EDA), first proposed by Dr. John W. Tukey of Princeton University. EDA is a philosophy of data analysis as well as a collection of techniques and methods. The EDA philosophy emphasizes letting the data lead the analysis rather that restricting analyses to preconceived hypotheses and patterns. Dr. Paul Velleman, the founder of Data Description, Inc. and designer of Data Desk, earned his doctorate with Dr. Tukey and has pioneered EDA methods on computers. EDA methods are a form of "Data Mining" (although they predate the field that is commonly called "data mining" today). They differ from many data mining methods by involving a human domain expert in the analysis process. It is a pleasant dream of data miners that the computer, left on its own with a large amount of data, can discover important patterns and relationships. Occasionally, that may happen. But experience has shown that human experts bring so much ancillary knowledge and experience to the table that it is unwise to exclude them from the process. EDA methods combine graphics, model building, rule-based record selection, outlier detection, and traditional statistical methods. Each technique is chosen to suit an appropriate point in the analysis. For example, graphical methods are especially effective for providing an overview of a large amount of data. The data analyst can easily focus on a subset of the data by interacting with the graph (for example, by clicking on parts of it.) Tables and summaries are appropriate once a specific question has been identified. Rule-based case selection can identify anomalous records that deserve further attention and analysis. Statistical models help one see past the inevitable variation in the data to compare records to what we might expect them to show based on other available data. Health Data Desk combines these methods with an easily accessible graphical interface. This makes it easy for domain experts to work directly with the data pursuing patterns and anomalies with the greatest opportunities for recovery and improvement. The Hospital Inpatient tool suite of Health Data Desk contains the Overview Analysis Module, DRG Analysis Module, Per Diem Analysis Module and Readmission Analysis Module. |
 Overview Analysis DRG Analysis Per Diem Analysis Readmission Analysis |
|||
|
|||||
|
|
|||||
