These analyses provide important information for understanding who uses PAC services, how their likelihood of using PAC services differs by certain characteristics, and which factors are most important in predicting hospital length of stay, probability of PAC use, relative probability of PAC site of care choices, hospital readmission rates, and average episode payments. Age and severity of illness factors were important in all the multivariate models predicting these outcome variables (Section 3.9, Tables 3-24 to 3-27).
In this work, we compare several measures of severity including APR-DRG, MS-DRG and HCCs. The additional contribution of the HCC indicators to the multivariate models flagging comorbid conditions proved quite useful to improving the explanatory power of the models. Greater severity was associated with longer length stay, as expected, regardless of measure used. Severity was also important for explaining the probability of PAC use and the type of PAC service used. Patients with higher severity scores were more likely to use LTCHs, followed by SNFs, then IRFs, and last home health services relative to outpatient therapy services.
Organizational relationships were also important for predicting use. We also found greater likelihood of using a type of PAC if the hospital had a subprovider or co-located PAC provider of that type. For example, multivariate models showed a greater likelihood of using IRFs if the hospital has a subprovider or co-located IRF and a lower likelihood if the hospital had a SNF subprovider, all else equal. Similarly, having a co-located LTCH increased the likelihood of LTCH use while the presence of a SNF or HHA reduced the likelihood of LTCH use, all else equal. And the same is true for the presence of a SNF.
Both these factors (severity and organizational relationships) were also important for predicting readmission rates and average episode payments (Section 3.9, Table 3-27). The probability of readmission increased as severity increased and having a subprovider was negatively associated with readmission rates. Both factors were also statistically significantly associated with episode payments; as severity increased, so did the average payment per episode. Similarly, average episode payments were higher for beneficiaries treated in hospitals with PAC subproviders. This may reflect different resource mixes of the hospitals or reflect higher likelihood of using subproviders where they exist, all else equal.
This work provides an important starting point for predicting beneficiary costliness and outcome variations. Understanding the contributions of better severity and medical complexity measures allows us to refine payment and outcome models. During the coming year, we will be adding data from the Chronic Care Warehouse (CCW) dataset to identify beneficiaries in our 2006 episode file with chronic conditions. Similar to some of the analyses presented in this report, we will look at the patterns of use and expenditures associated with having one or more chronic conditions. This will further allow us to refine the information describing a beneficiary's medical complications and is more comprehensive than our limited application of the HCCs to the index acute admission claims. Second, we will also examine alternative episode definitions including fixed and variable length episodes and episodes initiating in IRF, LTCH, HHA, or outpatient therapy without an index acute hospital admission-so-called community entrants to Medicare post-acute care services. This work will serve as the basis of exploring potential episode-based payment or bundling options and will build on some of the episode composition work presented here.