Part 2: Clinical Categorical vs. Regression Based Classification Systems

October 9th, 2013 / By Norbert Goldfield, MD, Richard Fuller, MS

In our last blog we discussed the fundamental aspects of patient classification systems, specifically categorical clinical and regression based models. While any classification system has its limitations, we’ve found that there are many advantages in using a categorical clinical model over a regression based model. Because a categorical clinical model does not rely solely on statistics and considers more variables than a regression based model, it allows flexibility in use and facilitates greater communication between hospital management and clinicians. Let’s discuss these advantages in more detail.

Facilitates Communication: It is easy for systems to create silos where financial (hospital management) and clinical (clinicians) aspects of healthcare do not “talk” to each other, creating the possibility of miscommunication. Since a categorical clinical model defines groups of clinically similar patients, a language is created that links the clinical and financial aspects of care. The importance of this communication value cannot be overemphasized. The language of the Potentially Preventable Readmissions (PPRs) methodology provides hospital administrators and physicians with a meaningful basis for evaluating both the processes of care and the associated financial impact. Communication between financial and clinical is absolutely critical to the success of readmission reform. The purpose of the reform (paying less for avoidable readmissions), logically, is to change behavior. The savings from avoided readmissions will be much larger than the savings from simply reducing the payment for an avoidable readmission. For this reason, facilitating communication between hospital management and clinicians is so important.

For example, Medicare’s DRGs revolutionized hospital management resulting in dramatic drops in length of stay and the use of ancillary services. DRGs are an example of a categorical clinical model, which are intended to predict resource use. The simple categorical nature of DRGs was critical to a hospital’s ability to respond in a constructive way to the incentives created by the Inpatient Prospective Payment System (IPPS). Indeed, CMS has emphasized the importance of the communications aspect of a categorical model like DRGs to the success of the IPPS:

“The success of any payment system that is predicated on providing incentives for cost control is almost totally dependent on the effectiveness with which the incentives are communicated…. Central to the success of the Medicare inpatient hospital prospective payment system is that DRGs have remained a clinical description of why the patient required hospitalization.”

Federal Register, May 4, 2001

Flexibility: The categorical nature of PPRs permits a separation of the computation of the relative severity adjusted probabilities of a readmission from the definition of the PPR categories. Such a separation is an inherent by-product of the categorical nature of PPRs and cannot be readily implemented in non-categorical systems, such as those based on linear or logistic regression. As noted by CMS the separation of the methodologies for developing the clinical model and the payment weights was a critical factor in the success and widespread adoption of the DRG system.

“The separation of the clinical and payment weight methodologies allows stable clinical methodology to be maintained while the payment weights evolve in response to changing practice patterns.”

Federal Register, May 4, 2001

Thus, the DRG clinical model has remained relatively stable, creating a consistent and powerful communication tool. However, the payment weights have fluctuated to reflect changing practice patterns and new technology. In a similar way, categorical clinical PPRs provide the user with a level of flexibility not possible with a regression model of any type. This includes:

1. Flexibility over geographic areas: With a categorical clinical model like PPRs, it is simple to adjust the model for use in different geographic areas. In essence, the categories remain the same, but the probability of a particular PPR can be different in, for example, rural areas. This can be done through a simple linear adjustment. This is done for rural hospitals under Medicare’s PPS. If the differences between two areas are not linear, a non-linear adjustment can be done by simply re-estimating the probabilities. (Note that these probabilities of a readmission are similar to the weights in a DRG based prospective payment system.)

2. Flexibility over time: Again, the categories can remain the same and the probabilities updated to reflect changes in practice.

3. Flexibility for a variety of outcomes: PPRs can be used with a variety of time windows (e.g., 14 days, 10 days, 60 days). The PPR definitions do not change when the dependent variable is changed, but the probabilities will change. A regression model would need to be completely re-estimated if the dependent variable is changed.

4. Flexibility for availability of data: Again, the PPR definitions do not change if a particular data element is not available, but the probabilities will change. A regression model would need to be re-estimated if and when an independent variable which was used in the original regression is not available.

Here at 3M we have a strong commitment to providing customers with an open, traceable methodology. Though each method had its limitations, the categorical clinical method’s ability to provide open communication and flexibility helped us determine it to be the best choice for identifying PPRs.

Richard Averill, MS, is the Director of Clinical and Economic Research for 3M Health Information Systems.

Norbert Goldfield, MD, is Medical Director for 3M Health Information Systems.