Verisk Health DXCG Medical Classification System - Version 7
Verisk Health DXCG Medical Classification System - Version 7
Verisk Health DXCG Medical Classification System - Version 7
Structural Summary
Introduction
This document describes the Medical Classification System as reflected in 2008-2009 for the seventh revision. The medical classification system represents a significant enhancement over the preceding version released in 2000. Enhancements made in Version 7 include improvements in clinical specificity, updates to capture changes in medical treatment practices and coding, distinctions made between hierarchies of conditions for concurrent and prospective risk adjustment and predictive models, addition of chronic and non-chronic flags, and the accommodation of large sample sizes that improve accuracy. The international business of healthcare has generated a constant stream of data that document the diagnosis and treatment of disease, the management of health and wellness, the occurrence and treatment of injuries, external causes of morbidity and mortality, and the generation of new life. Coding systems such as the various versions of the International Classification of Diseases (ICD) are used to document conditions, diseases, health status, and causes of mortality. These standardized coding systems are a vital tool for healthcare practitioners as well as managers of the healthcare system. The ICD-9 and ICD-10 morbidity coding systems consist of over ten thousand codes that document minute details related to a persons condition. This level of detail or granularity is vital in the inpatient, outpatient, and physician office settings. However, this richness becomes cumbersome when attempting to report trends, describe the treatment of conditions, or quantify present and future risk. For example a single disease process may require dozens of codes to describe the specific manifestations of the disease. In order to track prevalence and activity related to this disease, the many codes may need to be aggregated to fewer or one level of detail. The Medical Classification System was initially developed to predict the resources needed to treat patients so that a provider or health plan could be fairly compensated for the expected risk of managing a person or group. The clinical and statistical validity have been established and its uses have grown to include predictions of many diverse measures and outcomes for a wide range of business applications. This white paper documents the seventh revision of the DxCG Medical Classification System. Rationale for Updating the Medical Classification System There are many reasons that necessitate periodic overhauls of medical classifications systems. These reasons include the following: ICD-9 codes are updated annually. As new codes accumulate over time, it is difficult to maintain the clinical homogeneity of the existing grouping of health conditions.
-2-
Disease prevalence changes. For example, the fraction of Americans who are overweight or obese has increased in recent years (CBO Testimony 2008). Obesity is a major risk factor for diabetes, cardiovascular disease, hypertension, stroke, osteoarthritis and certain forms of cancer; moreover, obese persons incur greater healthcare costs. New medical technologies and therapies may change the cost and utilization patterns for a health condition. (Cutler and McClellan, 2001). The technological change in the treatment of cataracts from the late 1960s through the late 1990s is a good example. In the late 1960s a cataract operation was an intensive procedure involving three nights in a hospital. Since the late 1990s it is a routinely performed operation in an outpatient setting in under half an hour, with a lower likelihood of complications. Larger research databases are available. In the earlier version, we were forced to group conditions together in order to have a sufficient population for an HCC. In 2005, the benchmark database for a Commercial population we use in our research has grown to over 13 million individuals. With this larger database, we have been able to add a significant degree of clinical specificity without compromising statistical robustness of the methodology. Recent changes to certain ICD-9 codes allow more clinical specificity. For example, beginning in 2005, chronic kidney disease (CKD) codes specify the stage of the disease and obesity related codes specify the body mass index (BMI) level. Providers change diagnostic patterns. For example, diagnosis rates of depression doubled after Prozac-like drugs became available, and cataract surgery was performed much more frequently as the procedure improved (Cutler and McClellan, 2001).
When we consider this rapidly changing healthcare landscape, the need to periodically re-work the system from the ground up becomes apparent.
Basic Structure
The DxCG Medical Classification System consists of 4 levels of detail. They are as follows. DxGroups: This is the most granular level of detail. DxG are collections of ICD codes that are clinically similar. There are currently 1,010 DxG, with three additional categories reserved to document invalid or missing data. Two of the 1010 DxG are reserved for conditions coded only in the ICD-10 classification system. In the new Version 7 classification system each ICD code is assigned to one and only one DxG.
-3-
Condition Categories This next level of granularity is commonly used to develop risk adjustment and prediction models. The 1010 DxG are collapsed into 394 Condition Categories that summarize the conditions relevant for making predictions. This number of conditions empirically captures the right balance between having specific conditions and maintaining an adequate number of individuals with each condition to make reliable predictions. Condition Categories are the most used and most rigorously tested level of the Medical Classification System. In almost all instances, each DxGroup is assigned to one and only one Condition Category. There are a few rare exceptions. For example Type I diabetes with renal manifestation activates a Diabetes with renal manifestation Condition Category and a separate Condition Category indicating that Type I diabetes is present. This dual assignment ultimately creates parsimony in the Diabetes Condition Categories, and appropriately deals with the multiple pieces of information present in the original ICD diagnostic code. Related Condition Categories This new level of detail for Version 7 represents a need to report condition prevalence on a less granular level without rolling the detail up to the body system level. For example some body systems such as musculoskeletal can be broken up anatomically into different body regions. Condition Categories can evolve through disease pathways that collectively represent a larger, coherent clinical picture. The 117 Related Condition Categories offer the opportunity to report and focus on these groups of similar conditions. Aggregated Condition Categories The highest level of grouping, and the lowest level of detail or granularity, offers the opportunity to focus on morbidity at the body system level. Virtually every healthcare system from coding to medical specialization recognizes a need to view conditions at this level. Aggregated Condition Categories then track the activity of clinicians who focus on 31 specific domains in the healthcare space.
-4-
Condition Hierarchies
One of the most important features of the Medical Classification System is its clinical hierarchical structure at the Condition Category level. This structure provides the framework for risk and predictive model development. Depending upon the number of times an individual interacts with the healthcare system, there may be one to many documentations of a particular disease in administrative data. In a year period of time, there may be multiple instances of acute condition documentation and multiple opportunities to document the existence of a single chronic condition. Within a disease pathway, the level of morbidity within a period of time may wax and wane, resulting in multiple levels of a disease pathway being documented. In order for a Medical Classification System to be useful, it must be reasonably insensitive to coding variation and care-seeking behavior. It must also be capable of describing the most significant manifestation of a condition. Condition Categories can be treated as logical variables which are flagged when one or more codes within the category are registered in a period of time. Traditionally that time period is a year so as to filter out potential seasonal variations. Once all Condition Categories are registered for an individual in a base year, then the hierarchical structure can be applied. The figure below represents part of the Diabetes Hierarchy in a Concurrent setting.
-5-
Figure 1: Partial Diabetes Hierarchy If a patient has diabetes with autonomic neuropathy, such a condition would be registered by Condition Category DMC.40 (Diabetes with Neurological or Peripheral Circulatory Manifestations). If that same patient was admitted because of diabetic ketoacidosis, Condition Category DMC.20 would be registered. Within any hierarchy, a disease higher in the severity spectrum trumps one lower in that spectrum. In this case, only DMC.20 would be activated after applying the hierarchy. This is illustrated below.
-6-
Condition Categories which have the hierarchies activated in such a manner are called Hierarchical Condition Categories, or HCC. From the standpoint of medical management and reporting, the user may want to know all the Condition Categories which have been registered for an individual. But from a predictive modeling or morbidity assessment point of view, the Hierarchical Condition Categories are a better representation of what matters. For example in this case it is important to know that the patient has neurological manifestations of his diabetes, even though from an expected cost standpoint the salient feature is the episode of ketoacidosis. If a provider chooses to code other lesser manifestations of this disease, it will not change the expected cost for the patient in a risk model that uses the Medical Classification System.
Clinical Interactions
Every clinician knows that age and disease profile can affect the severity of a condition and the resources needed to treat it. For example, congestive heart failure is much more difficult to treat when the patient has diabetes. The risk associated with acute renal failure is dramatically different from an infant to a child to an adult. A medical classification system must be capable of reflecting these co morbidities and complications in some way if it is to be used in risk and predictive models.
-7-
Rather than subdivide Condition Category groups into amorphous complicated entities, the approach chosen to account for age and disease profile was to layer it on later in model building or specific reporting applications. This choice reflects a desire to retain transparency, and create flexibility on the reporting and modeling front.
Summary
The Medical Classification system is a principles-driven approach to documenting the conditions of an individual at multiple levels of detail. The chronic vs. non-chronic code separation and the hierarchical structure of the Condition Categories create the foundation for multiple reporting and model building activities. The system taxonomy and the logical structure enhance ease of use, and make this the medical classification system of choice for healthcare assessment.
References
Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project (HCUP) (2008). http://www.hcup-us.ahrq.gov/toolssoftware/chronic/chronic.jsp. Accessed May 2009. Congressional Budget Office (CBO). Growth in Health Care Costs. CBO Testimony by Peter R. Orszag before the Committee on the budget United States Senate. January 31, 2008. Cutler DM and McClellan M. Is Technological Change in Medicine Worth It? Health Affairs. 2001 (September/October); 20(5): 11-29.
-8-