Pros And Cons Of Predictive Modeling Approaches

Adjust font size: A A A | Printer-friendly version

November 3, 2004

Predictive risk modeling methodologies in health care are used to identify high-risk patients before they become high-cost patients. Since the initial interest of CMS in the 1980s, predictive modeling has grown up quite a bit. Most of this maturation is due to vast improvements in the quality of electronic medical data; these improvements have led to creativity among predictive modeling experts.

There are several ways to categorize types of predictive models, according to Lori Weyuker, president of Weyukergroup, vicechair of the Health Section of the Society of Actuaries, and presenter at Learn How to Apply Predictive Modeling Techniques to Predict Healthcare Claims and Hospital Revenues -- one of HFMA's 2004 Fall Seminars to be held December 8-9 in Chicago. Two leading categorizations are medical data-based models and prescription drug-based (Rx-based) models. In comparing these two models, there are several pros and cons of each to consider.

Predictive Power

Predictive models based on full medical data, that is, medical data from all sites of service, have proven to have the highest predictive power. (Predictive power is the ability of the predictive model to "predict" healthcare consumption for a given population.) Many statistical tests show that these medical data-based models are 50 percent more predictive than Rx-based models.

Current Data

In the arena of data availability, the Rx-based models have the advantage because prescription data, drawn from prescription claims, are typically available as electronic data within two months. In general, the more current the data, the more it reflects the current disease burden of a given population. On the other hand, medical data, usually drawn from inpatient and outpatient hospital claims or office visits, can take much longer to become electronic data that is available for administrative use. Medical data from office visits or outpatient hospital use can take six months before becoming electronically available, and inpatient hospital data often takes 12 months or more before becoming accurate electronic data.

Ease of Obtaining Input Data

Again, Rx-based models have the advantage. Obtaining prescription data from a given population is a much easier task than obtaining medical data from that same population. A typical non-Medicare population may have four prescription claims per insured person per year. On the other hand, this same population will have three office visits per insured person per year, plus outpatient and inpatient claims. Trying to obtain medical data from these three different types of claims that reside in disparate databases is a complicated and sometimes messy task. Obtaining prescription data, which usually requires only one database for a given insured population, is relatively easy.

Choosing a Model

There are other factors to think about when deciding which type of predictive model is optimal for a given application. Some of these factors include:

Ability to intuit the results of the predictive model's output
Ease of understanding the mechanics of the predictive model
Clinical validity of the predictive model
Stability over time of the predictive model results
Selection of a model that minimizes the ability to game results

Whichever method you choose, predictive modeling adds important decision support in service line planning, risk management, and managed care contracting.

SOURCE:

HFMA's 2004 Fall Seminar Learn How to Apply Predictive Modeling Techniques to Predict Healthcare Claims and Hospital Revenues to be held December 8-9, 2004, Chicago. Presented by HFMA and the Society of Actuaries.

Additional Resources

Accounting and Reporting by Institutional Healthcare Providers for Risk Contracts, HFMA's Principles & Practices Board Statement Number 11.
Managing Risk in Managed Care Contracts, a prerecorded HFMA and Society of Actuaries two-part audio webcast held March 2 and 16, 2004 (scroll down the screen for course titles and dates).
"Risk Management in Long-Term Care" (This book is offered in alliance with the American College of Health Care Administrators.)
"Financial Strategy for Managed Care Organizations: Rate Setting, Risk Adjustment, and Competitive Analysis" (This book is offered in alliance with the American College of Healthcare Executives.)

If you have questions or comments about HFMA Wants You to Know, contact editor Laura Noble.

featured sponsors