OBJECTIVE: To describe Iran's hospital activity with Australian Refined Diagnosis Related Groups (AR-DRGs). METHOD: A total of 445,324 separations was grouped into discreet DRG classes using AR-DRGs. L(3)H(3); IQR and 10th-95th percentile were used to exclude outlier cases. Reduction in variance (R(2)) and coefficient of variation (CV) were applied to measure model fit and within group homogeneity. RESULTS: Total hospital acute inpatients were grouped into 579 DRG groups in which 'surgical' cases represented 63% of the total separations and 40% of total DRGs. Approximately 12.5% of the total separations fell into DRGs O60C (vaginal delivery) and 28% of the total separations classified into major diagnostic category (MDC) 14 (pregnancy and childbirth). Although reduction in variance (R(2)) for untrimmed data was low (R(2)=0.17) for LOS, trimming by L(3)H(3), IQR, and 10th-95th percentile methods improved the value of R(2) to 0.53, 0.48, and 0.51, respectively. Low value of R(2) for AR-DRGs within several MDCs were identified, and found to reflect high variability in one or two DRGs. High within-DRG variation was identified for 23% of DRGs using untrimmed data. CONCLUSION: Low quality and incomplete data undermines the accuracy of casemix information. This may require improvement in coding quality or further classification refinement in Iran. Further study is also required to compare AR-DRG performance with other versions of DRGs and to determine whether the low value of R(2) for several MDCs is due to the weakness of the AR-DRG algorithm or to Iranian specific factors.
OBJECTIVE: To describe Iran's hospital activity with Australian Refined Diagnosis Related Groups (AR-DRGs). METHOD: A total of 445,324 separations was grouped into discreet DRG classes using AR-DRGs. L(3)H(3); IQR and 10th-95th percentile were used to exclude outlier cases. Reduction in variance (R(2)) and coefficient of variation (CV) were applied to measure model fit and within group homogeneity. RESULTS: Total hospital acute inpatients were grouped into 579 DRG groups in which 'surgical' cases represented 63% of the total separations and 40% of total DRGs. Approximately 12.5% of the total separations fell into DRGs O60C (vaginal delivery) and 28% of the total separations classified into major diagnostic category (MDC) 14 (pregnancy and childbirth). Although reduction in variance (R(2)) for untrimmed data was low (R(2)=0.17) for LOS, trimming by L(3)H(3), IQR, and 10th-95th percentile methods improved the value of R(2) to 0.53, 0.48, and 0.51, respectively. Low value of R(2) for AR-DRGs within several MDCs were identified, and found to reflect high variability in one or two DRGs. High within-DRG variation was identified for 23% of DRGs using untrimmed data. CONCLUSION: Low quality and incomplete data undermines the accuracy of casemix information. This may require improvement in coding quality or further classification refinement in Iran. Further study is also required to compare AR-DRG performance with other versions of DRGs and to determine whether the low value of R(2) for several MDCs is due to the weakness of the AR-DRG algorithm or to Iranian specific factors.