OBJECTIVES: Diagnosis Related Group (DRG) systems aim to classify patients into mutually exclusive groups of patients, with the patients in each group having the same expected length of stay (LOS). We examined the ability of current classification variables to explain LOS variation between DRG-like Diagnosis Treatment Combination (DBC)s for ten episodes of care in the Netherlands, including breast cancer, stroke and inguinal hernia repair. Additionally, we assessed the predictive ability of some other classification variables. METHODS: For each episode of care, the relevant DBC codes of all hospitalizations in 2008 were identified and all available determinants that may serve as classification variables were acquired from the national database. Ordinary least squares regression was used to examine the predictive ability of these classification variables. RESULTS: The current classification variables are not sufficiently distinct to classify patients into mutually exclusive groups of patients. ICU admissions and hospital type may serve as valuable classification variables. Additionally, episode-specific variables may improve the Dutch grouping algorithm. CONCLUSIONS: Although it may not be feasible in the short term, grouping algorithms would benefit greatly from the introduction of classification variables tailored to the needs of specific episodes of care. A first step would be to focus on 'general' classification variables meaningful for specific episodes of care.
OBJECTIVES: Diagnosis Related Group (DRG) systems aim to classify patients into mutually exclusive groups of patients, with the patients in each group having the same expected length of stay (LOS). We examined the ability of current classification variables to explain LOS variation between DRG-like Diagnosis Treatment Combination (DBC)s for ten episodes of care in the Netherlands, including breast cancer, stroke and inguinal hernia repair. Additionally, we assessed the predictive ability of some other classification variables. METHODS: For each episode of care, the relevant DBC codes of all hospitalizations in 2008 were identified and all available determinants that may serve as classification variables were acquired from the national database. Ordinary least squares regression was used to examine the predictive ability of these classification variables. RESULTS: The current classification variables are not sufficiently distinct to classify patients into mutually exclusive groups of patients. ICU admissions and hospital type may serve as valuable classification variables. Additionally, episode-specific variables may improve the Dutch grouping algorithm. CONCLUSIONS: Although it may not be feasible in the short term, grouping algorithms would benefit greatly from the introduction of classification variables tailored to the needs of specific episodes of care. A first step would be to focus on 'general' classification variables meaningful for specific episodes of care.