Are DRGs Homogeneous
With Regard to Resource Consumption?

Patient classification systems claim to have defined "homogeneous" case groups. This is the reason why they are accepted to be used in prospective payment systems with flat case rates.

This study uses data published on the Internet of the systems HCFA-DRG, GHM, AR-DRG, HRG and data obtained from the producer of the system RDRG. The coefficents of variation (CVs) of the groups of the different systems are compared. Because of different populations, methods of cost calculation and trimming rules, only rough results have been evaluated.1

The counts of cases belonging to groups with CVs > 50% related to length of stay (LOS) or charges, i.e. to groups with a great dispersion, exceed 80% clearly almost throughout all systems compared (exception: GHM). Many systems even show figures above 90%.
The counts of cases belonging to groups with CVs > 100% related to length of stay (LOS) or charges, i.e. to groups with a very great dispersion, fall to 8% – 25% (exceptions with higher percentages are ARDRG-4.0 and HRG 3; both with data not trimmed).

The CVs found are very high. From a statistical viewpoint it must be stated that most case groups in most DRG systems cannot be said to be homogeneous with regard to costs or LOS.

The continuing DRG refinement process which can be observed in the development over the last 20 years has not led to satisfying homogeneity concerning resource intensity. Other techniques have to be developed to gain more homogeneity (e.g. multidimensional approaches, additional classification criteria).2

The tables below list the quotas of case numbers in case groups with great dispersion. As indicators of great dispersion within case groups, the numbers of cases belonging to groups with coefficients of variation (CVs) > 50% and > 100% are taken.

The figures in the tables below are comparable only to a limited extent since the selection from the universe of the cases does not correspond (trimmed/untrimmed, including/excluding day cases) and since the standard for comparison differs (duration of stay/invoiced amounts/costs). One thing, however, is clear: the values that have been found are much too high throughout.

The quota of cases in case groups with CVs of over 50% should be as low as possible. These figures indicate, however, that this quota is very high, i.e. the dispersion within most case groups is very high, and there is a lack of homogeneity.

As to the HCFA-DRGs, 99% of all the cases are in case groups with a variation coefficient of over 50%. The values displayed by the GHM system, in which 76% of all the cases are in case groups with a variation coefficient of over 50%, are only a little better. (Table 1)

Only once a distinction is made according to cases with variation coefficients of below and above 100% do we receive a picture that is a bit more expressive. However, a variation coefficient of 100% and more already indicates a very high degree of dispersion. (Table 2)

In the HCFA-DRG system, the quota of cases in case groups with a variation coefficient of more than 100% is as little as 25% when the dispersion of the charges is measured (and 18% when the dispersion of the lengths of stay is measured). Only in the GHM system (and, partially, also in the RDRG system) does this approach yield reasonably low values. However, the underlying GHM data were trimmed, which usually ensured better values.

It is striking but logical that cases in specific case groups disperse distinctively less than cases in container groups. It was only in the British system that this situation was not observed.

The analysis of the RDRG system revealed the effect which severity refinement had on the variation coefficients (Table 3). Whereas 97% of all the cases are in base groups with an charges-related variation coefficient of over 50% when grouped into Base-RDRGs, the quota of these cases decreases to 82% when grouped into single RDRGs. (When these cases are analysed according to length of stay, the values are slightly higher, namely: 99% when grouped into Base-RDRGs, and 88% when grouped into RDRGs.)

The lowest dispersion is observed by the case groups of the class "D" CC, which subsume cases with inessential comorbidities or complications. Even so, as many as 64% and 74% of the cases are in the case groups of the class "D" CC with a variation coefficient of over 50% as related to charges and length of stay, respectively.

The more serious the comorbidities or complications, the stronger the dispersion within the groups becomes.

An analogous picture results from the observation of the case groups with variation coefficients of below and above 100% (Table 4): 17% of the cases grouped into Base-RDRGs and 8% of the cases grouped into RDRGs are in case groups with an charges-related variation coefficient of over 100%. (When analysed according to length of stay, the quotas are 22% and 8%, respectively.) Here, too, dispersions decrease with the decreasing severity of the comorbidities or complications: both in terms of charges and of length of stay, 20% of the cases of class "A" CC and 7% of the cases of class "D" CC are in case groups with a variation coefficient of over 100%.

There is an argument whereby a 100% homogeneity cannot be achieved within patient categories since there are always differences in efficiency, and that it is precisely the function of patient classification systems to provide incentives for the improvement of economic efficiency. This argument is legitimate. However, it must not blind us to the fact that, apart from a lack of economic efficiency, there are a number of other factors that impair quantifiable homogeneity. The more or less great cost divergences that can be ascertained in a PCS-based analysis of cases are only the visible tip of the iceberg of causes which are unknown at first. One of the causes of such differences between planned and actual costs may – as mentioned above – be economic inefficiency. Another important cause, however, may be an insufficiently differentiated patient classification system. Anteceedent causes may be the poor encoding quality or inadequate calculation methods for planned and/or actual costs. (Table 5)

HCFA-DRG data sources:

• Overview: http:// cms.hhs.gov / AcuteInpatientPPS / FFD / list.asp
• HCFA-DRG16 data 1998/99:
  http:// www.hcfa.gov / stats / aoborf00.zip

For this study the analyses of untrimmed Maryland data of non-pediatric cases of 1997/98 received from Health Systems Consultants, Inc. was used.

AR-DRG data sources:

• Overview: http:// www.health.gov.au / casemix / report / hospmor7.htm
• AR-DRG4 data 1996/97 (public hospitals):
  http:// www.health.gov.au / casemix / report / tables / tablea1.xls

GHM data sources:

• Overview:
  • http:// www.le-pmsi.org / telechargeables / mco /
  • ftp:// ftp.le-pmsi.org / pub /
• Cost Weights, Costs, and Reference Costs GHM 1999:
  http:// www.le-pmsi.org / telechargables / mco / echelles / xlisar99.zip

HRG data sources:

• Overview: http:// www.doh.gov.uk / hes / standard_data / available_tables / hrgs /
• HRG3 data 1996/97:
  http:// www.nhsia.nhs.uk / phsmi / casemix / d_loads / 967v3_ns.zip
• HRG3 data 1997/98 (trimmed):
  http:// www.nhsia.nhs.uk / phsmi / casemix / d_loads / 978tri_e.zip