The results of the models predicting index admission length of stay highlight the importance of severity measures (Table 3-24, Table 3-25, and Table 3-26). The results using independent variable Set 1 with the APR-DRG severity measures (Table 3-24) indicate that beneficiaries in APR-DRG severity level 4 (extreme severity) have a mean length of stay that is 12.1 days longer than that of beneficiaries in the lowest APR-DRG severity level. Similarly, in looking at the regression model predicting index admission length of stay using the MS-DRG severity measures (Set 2, Table 3-25) beneficiaries in the MS-DRG MCC severity level had a length of stay 4.8 days longer than that of beneficiaries in the MS-DRG No CC level. When the HCCs were added to the multivariate models as severity measures (Set 3, Table 3-26), we see that after controlling for MS-DRG severity levels, beneficiaries with particular comorbid conditions are likely to have longer lengths of stay in the acute setting. For example, beneficiaries with HCC 79 Cardio-Respiratory Failure and Shock have an index admission length of stay 3.6 days longer than that of beneficiaries without this HCC.
Indicator variables for acute DRGs were included in independent variable Sets 1 and 2 (Note that these variables were not included in Set 3 because of the high degree of correlation between the DRG and HCC indicators). In Set 1, only beneficiaries with DRG 210 had significantly longer length of stay (Table 3-24). Beneficiaries admitted to acute hospitals for DRG 209, DRG 089, and DRG 127 had shorter lengths of stay compared to those of beneficiaries admitted for other diagnoses. The results from Set 2 showed that both beneficiaries in DRGs 209 and 210 had significantly longer index admission lengths of stay than those of beneficiaries in other DRGS (Table 3-25).
The impact of age on index admission length of stay varied across the sets of independent variables. When using the APR-DRG severity measures and the MS-DRG severity measures (Sets 1 and 2, Table 3-24 and Table 3-25), increasing age was associated with increasing index admission lengths of stay. However, when using the MS-DRG and HCCs (Set 3, Table 3-26), only the coefficient on the age 75 to 84 variable was significant indicating that beneficiaries in this age group had a length of stay 0.13 days longer than that of beneficiaries younger than 65.
Results of the impact of census region was similar across models; beneficiaries in the Middle Atlantic had slightly longer index admission lengths of stay compared with beneficiaries in New England. Beneficiaries from the East North Central, West North Central, Mountain, and Pacific census division had shorter index admission lengths of stay than did those in New England. Since the models controlled for patient differences, this seems to reflect local practice pattern differences.
Supply of services did not have a significant effect on index admission length of stay with the exception of the supply of IRF beds using independent variable Set 3 (Table 3-26). In this model, an increase in supply of IRF beds was associated with a shorter index admission length of stay. In looking at the characteristics of the acute hospitals, both urban location and government-run hospitals were associated with longer index admission length of stay.
Across all three sets of independent variables, the presence of any PAC subprovider was associated with an increase in index admission length of stay. For example, in the results using independent variable Set 3 (Table 3-26), beneficiaries discharged from an acute hospital with any PAC subprovider had index admission lengths of stay that were 0.135 days longer than those of beneficiaries discharged from acute hospitals without subproviders. The presence of any colocated post-acute provider was not significant in predicting index admission length of stay.