The Anatomy of the DRG System in Healthcare Part 1: Structure, Risk Measurement, Ratemaking, and the Grouper Algorithm

Wonderful people can be terrible patients. Some of us demand drugs we don’t need, and fail to take the ones we do. Others worry about minor ailments, but avoid checkups that would prevent more serious conditions. Still more of us suffer from uncontrolled, chronic diseases we ignore for far too long, requiring costly emergency interventions.

Managing the needs of high-cost patients can result in expensive and unnecessary procedures and prescriptions. Diagnosis-related groups (DRGs), such as the IR-DRG (International Refined DRG) used most commonly outside the U.S., were designed as one possible solution. 

Developed in the 1970s by Yale University, the DRG patient classification system seeks to encourage value-based care by standardizing payments to doctors and hospitals based on the disease being treated. The goal is to reward providers for positive patient outcomes, rather than for the volume of services provided as is typical under a traditional fee for service (FFS) system. 

Providers collect discharge claims and classify them into homogeneous, clinically meaningful groups. The DRG system then requires insurers to make reimbursements based on diagnosis code, irrespective of the underlying services or length of stay. So, for example, a typical DRG payment covers all charges associated with an inpatient hospital stay from the time of admission to discharge (though in some countries, the ambulatory family of benefits are also priced under the DRG system). Long-term care benefits, inpatient dental treatments, and certain consumable items would normally fall outside the scope of a DRG. 

When the system works well, it can improve patient outcomes while removing the incentive to pursue unneeded procedures. The system is also designed to promote the use of technological innovation to minimize a patent’s length of stay. While evidence suggests that the use of DRG systems often coincides with reductions in average patient lengths of stay, it can be difficult to ascertain whether this is due to medical intervention or the DRG system itself.

Use of DRG has grown throughout the U.S. starting in 1983 and spread to Australia in 1988 and throughout Europe in the 1990s. The DRG system was introduced in the Middle East by the Department of Health - Abu Dhabi (HAAD) in late 2010 in Abu Dhabi, and observers expect the system to reach Dubai in first quarter of 2020.Given its growing international popularity and cost containment potential, it is increasingly important for claims analysts and underwriters to understand the DRG structure, risk measurement methodology, ratemaking process, and grouper algorithms. 

Defining the DRG Structure

At a very basic level, DRG refers to a code, yet one must go much deeper to master this patient classification system. Although DRG system designs may vary, most reflect the following seven concepts:

• Base DRG code: System of digits signifying the following: the first two digits  refer to the Major Disease Category (MDC), the next three refer to the underlying diagnosis code (medical or surgical cases), and the sixth or seventh refers to the severity level of a given patient’s condition. 
• Weight:  Underlying resources defining the value of a particular diagnosis, as measured in units and often derived via algorithm.
• Base Rate: The common reimbursement rate set by a regulator, which will be multiplied by the “weight” to arrive at a measure of care required by a particular diagnosis.
• Cost: The total expense incurred for the services dispensed by the provider for a specific diagnosis. 
• Gap: The portion of cost borne by the provider that is higher than the DRG reimbursement. Often a provider absorbs the cost above the DRG reimbursement level as part of a “stop loss” arrangement up to a preset or “outlier” threshold.
• Marginal: An established percentage that applies to the amount above the DRG required payment and the gap that shall be reimbursed to the provider as outlier payment.
• Outlier payments: Expenses that fall within a simple formula: (Cost – DRG payment – Gap) x marginal. 

To understand how the DRG system classifies risk and assigns cost, consider the following typical patients (using United Arab Emirates dirham as currency):

(a) Patient 1 has a DRG with weight 1.5, incurred hospital cost => 145,000 AED 

• Gap= 50,000 AED
• Base payment = 8,500 x 1.5= 12,750 AED
• Outlier payment = (145,000 – 12,750 – 50,000 ) x 60%= 49,350 AED
• Total payment reimbursed = 49,350 AED + 12,750 AED = 62,100 AED

(b) Patient 2 has a DRG with weight 0.80, incurred hospital cost  => 16,000 AED 

• Gap= 50,000 AED
• Base payment = 8,500 x 0.80= 6,800 AED
• Outlier payment = (16,000 – 6,800 – 50,000 ) x 60%= 0 AED
• Total payment = 6,800 AED

As these examples illustrate, the principal focus of DRG ratemaking is to determine a base factor and/or means to “weight” the underlying value of a set of medical services. For instance, under the HAAD model the base factor is set to 8,500 AED and a grouper algorithm informs the calculation of a weight to arrive at a total payment. 

Understanding Risk Measurement Characteristics

The benefit of this calculation lies in its ability to deliver fair payment for a medical service while also achieving the lowest cost. Borrowing from classical risk theory, an efficient risk measure adheres to the following principles:

1. Monotonicity of risk measure 
2. Sub-additivity principle
3. Translation invariance principle
4. Scaling invariance principle

These same principles can be applied to the DRG system.

• Monotonicity of risk measure: This principle states that a higher risk should be compensated with more capital than a lower risk. In a DRG context, provider services involving more resources and complexity are compensated with higher payments than services involving lower resources or reduced complexity.
• Sub-additivity: This risk principle states that adding two correlated risks shall not lead to the addition of the underlying capital for each separate risk due to diversification effect. In a DRG context, adding two services which are highly correlated, such as a re-admission for the same Inpatient treatment within 30 days, should not lead to a double payment for provider services.
• Translation invariance: Adding a non-correlated cost to a primary diagnosis should lead to adequate compensation to the provider. For instance, if a patient with congestive heart failure is also treated for pneumonia, the DRG system should fairly compensate the provider for the services dispensed, considering that they are not correlated.
• Scaling invariance: Increasing hospital inpatient activity under a particular diagnosis should not lead to additional compensation.

Demystifying Risk Pricing within the DRG System

DRG weights are derived actuarially, either from base statistics or via customized grouper algorithms. After collecting cost information from hospital databases or a hospital information system, claims can be classified by DRG code and Major Disease Classification (MDC). Typically, the following key fields are completed:

(a) Total number/volume of activities 
(b) Total cost incurred for the entire treatment
(c) Severity level 

After a claim is classified, it can be evaluated for inappropriate codes, errors, or outliers or include vital services which have not been captured under the hospital cost information system. Typically, the outlier threshold would be set to around 95% Value at Risk (VaR). Next, weights can be derived through the DRG grouper algorithm, with the “weights” being equivalent to a “parameter” estimation under a regression model. Testing follows to ensure lower co-morbidities are assigned lower weights (monotonicity of reimbursement). Upon review (one- to three-year basis), all weights should be updated to maximize the correlation between costs and weights and, at the same time, adhere to the following constraints:

(a) Adjust for difference in credibility of DRG weights
(b) Keep severity levels in line with underlying risks
(c) Ensure that the overall case mix is constant – before and after DRG implementation

A DRG system should be updated periodically based on hospital costs submitted by providers. The above risk pricing process is based on weights inferred from nationwide hospital statistics. Very often, however, the DRG system is ‘imported’ from another country and can be subject to a cost differential adjustment; testing and review are paramount in these cases.  

Explaining the DRG Grouper Algorithm 

When deriving DRG weights using available hospital cost data, commercially available grouper models are always preferred initially. Doing so enables an organization to:

• Save considerable time and effort that would otherwise be required to generate a specialized model.
• Scale activities more easily and adapt more readily to new diagnoses. 
• Benefit from models that have been independently validated by reputable institutions such as the Society of Actuaries.

Typically, a statistical model such as the logistic regression or a generalized linear model is fitted to underlying data to produce the parameters that link every risk factor. The following factors are typically employed:

• Classification of patients into major diagnostic categories: This determines the part of the human body that is subject to the DRG classification.
• Classification of diseases and related health problems based on international norms: Typically, the ICD-10 classifications are used.
• Medical intervention classification list
• Demographics and characteristics of patients. This reflects the discharge date as well as other factors such as age, sex, occupation, and residence.
• Cost data from hospital information system

The DRG system is far from alone in its reliance on grouper algorithms to assign value to risk. In fact, the advent of DRG has unleashed a wave of innovations. Multiple alternatives have emerged to model inpatient medical events, including: 

• The Adjusted Clinical Group, developed as a case mix adjustment measure for ambulatory setting in the United States.
• The Chronic Illness and Disability Payment System, a model designed to adjust capitated payments relating to disabled and temporary assistance for needy families.
• Clinical Risk Groups, which are similar to DRGs but instead of quantifying resources relating to a particular hospital inpatient activity, CRGs cover the total resources in relation to a particular individual.
• Diagnostic Cost Groups, otherwise known as the Hierarchical Condition Category, cover not only inpatient treatments but also, ambulatory or daycare activities.
• The Episode Treatment Group illness classification and episode building algorithm identifies clinically homogeneous episodes of care. This not only covers inpatient treatment but also, all related costs which are associated with a particular medical episode.
• Drug Grouper Models incorporate drug-based risk adjustment models or therapeutic class groups.

If there is a single commonality behind these models, it is an attempt to deliver greater efficiency by the more accurate calculation of risk and a stronger connection between payment and treatment. 

Indeed, DRG coding itself succeeds when it closely links hospital reimbursement payments to underlying services or provider activities and attains efficiency in the provision of hospital services, without compromising on the level of quality of care. But success is not assured. 

In the second installment of this two-part series, RGA will explore governance and risk mitigation approaches and what can go wrong in DRG implementations.

For more on digital health data and mortality risk management,  contact us.