U.S. Department of Health and Human Services

Assessing Home Health Care Quality for Post-Acute and Chronically Ill Patients: Final Report

Executive Summary

Christopher M. Murtaugh, Ph.D., Timothy R. Peng, Ph.D., Stanley Moore, B.S., and Gill A. Maduro, Ph.D.

Visiting Nurse Service of New York, Center for Home Care Policy and Research

August 2008

This report was prepared under contract #HHS-100-03-0011 between the U.S. Department of Health and Human Services (HHS), Office of Disability, Aging and Long-Term Care Policy (DALTCP) and the Urban Institute. For additional information about this subject, you can visit the DALTCP home page at http://aspe.hhs.gov/_/office_specific/daltcp.cfm or contact the ASPE Project Officer, Hakan Aykan, at HHS/ASPE/DALTCP, Room 424E, H.H. Humphrey Building, 200 Independence Avenue, S.W., Washington, D.C. 20201. His e-mail address is: Hakan.Aykan@hhs.gov.

The opinions and views expressed in this report are those of the authors. They do not necessarily reflect the views of the Department of Health and Human Services, the contractor or any other funding organization.

BACKGROUND AND PURPOSE

Home health agencies serve patients with a range of health care needs including those with relatively short-term post acute care needs as well as individuals who are chronically ill and have more long-term needs. None of the 12 quality measures Centers for Medicare and Medicaid Services (CMS) currently reports as part of the Home Health Quality Initiative (HHQI), however, is assessed and reported separately for clinically important subgroups of patients within the larger home health population. In contrast, all nursing home quality measures that CMS publicly reports are assessed separately for a nursing home’s short-stay and long-stay residents. Because home health quality indicators are not reported separately for important subgroups of home health patients, risk-adjustment of HHQI outcomes takes on more importance. Without adequate risk-adjustment, public reporting could mislead consumers and be unfair to agencies admitting larger shares of patients who tend to do worse on some or all HHQI measures.

This project examines the current approach to public reporting of Medicare home health agency quality with a particular focus on how the current HHQI measures perform as quality indicators for the diverse home health population. Four key analytic questions are addressed: (1) Can clinically meaningful groups of home health patients be identified (e.g., post acute, chronically ill); (2) To what extent do agencies serve different types of patients; (3) Do these patient groups differ in publicly-reported outcomes; and (4) To what extent does risk-adjustment reduce (eliminate) any differences in outcomes. The project builds on an earlier Office of the Assistant Secretary for Planning and Evaluation-funded project where alternative risk-adjustment approaches were developed and tested to assess the quality of home health care. The results of the project are intended to help CMS and other Departmental agencies in their efforts to monitor and improve the quality of home health care.

METHODS

There were two major phases of data analysis: (1) preliminary analyses conducted using files available at the beginning of the project developed for earlier studies; and (2) final analyses conducted using a file constructed and delivered by the CMS Outcome and Assessment Information Set (OASIS) contractor at the University of Colorado to address the research questions specific to this project. A file that included all home health discharges, regardless of when they were admitted, was requested to assure full representation of short and long-stay Medicare and Medicaid home health patients. Long-stay patients are a group of particular interest since they are expected to be disproportionately individuals who are chronically ill and are likely to be underrepresented in HHQI analytic files. Two years of home health discharges were requested to address a second perceived limitation of HHQI. A substantial number of agencies care for relatively few Medicare or Medicaid beneficiaries in a year and do not meet the minimum sample size requirements for reporting some or all of the publicly-reported outcomes. (Currently, outcomes are not publicly-reported when an agency has fewer than 20 home care episodes with the potential to have an outcome.) Two years of data were combined to minimize the number of agencies excluded from project analyses.

The CMS contractor at the University of Colorado drew the data from the OASIS National Repository at CMS. All OASIS discharge assessments in calendar years 2004 and 2005 were identified where there also was an admission (i.e., Start of Care) OASIS assessment. The final analytic file includes a total of 6,493,623 home health episodes with one of three types of OASIS discharges in 2004 and 2005. Medicare and Medicaid provider numbers were used to identify the number discharges from the same agency during the two-year period and to link to the 2006 Provider of Services file to obtain information on agency characteristics.

Defining and Describing Different Types of Home Health Patients

The types of patients served by home health agencies are not well defined but include individuals with short-term post acute care needs as well as chronically ill persons with long-term health care needs. We relied primarily on the literature and advice of clinical experts to develop an algorithm for identifying subgroups of the home health population that differ in: (1) whether they are post acute; and (2) whether they are chronically ill. Differences on these dimensions plausibly affect home health care needs and, more importantly for this project, home health outcomes. A small Technical Advisory Group reviewed the preliminary algorithm and distributional data, and suggested refinements in our approach.

Differences in the sociodemographic, payer and clinical characteristics of the patient groups were examined as part of the final analyses. No statistical tests of the significance of differences were estimated because we examined a census of all discharges during 2004 and 2005. In addition, given the very large number of observations, even trivial differences would be statistically significant. We focused instead on the magnitude of observed differences among the groups.

Analysis of Types of Patients Served by Home Health Agencies

The agency is the unit of analysis when examining agency characteristics and the extent to which agencies serve different types of patients. There were a total of 8,094 agencies with at least one OASIS discharge in 2004 or 2005. All agencies were included in descriptive analyses of agency characteristics. We excluded agencies with too few discharges during the two-year study period to estimate reliable statistics (i.e., agencies with fewer than 100 discharges) when examining the extent to which agencies serve patients classified into the final five patient groups. The 6,113 agencies with 100 or more OASIS discharges in 2004 or 2005 represent agencies included in analyses of whether agencies serve different types of patients.

Whether agency characteristics differ by size was examined first. For this analysis we grouped all agencies (i.e., all 8,094 agencies with at least one OASIS discharge) into quintiles by size as measured by the number of OASIS discharges in 2004 and 2005 at each agency. Then, to determine whether agencies serve different types of patients, we examined the relative distribution of the five patient groups among agencies with at least 100 discharges (6,113 agencies) that differ in size, ownership and geographic location.

Analysis of Performance of Publicly-Reported Quality Indicators

The unit of analysis when examining the performance of the 12 HHQI measures was the home health episode. All discharges in the two-year period were included in these analyses. However, when determining the proportion of episodes improving in an HHQI quality measure, CMS excludes episodes that are not eligible to have a particular outcome. The criteria we used to evaluate each HHQI measure include: (1) the share of each patient group with the potential to have the outcome; (2) the extent to which each unadjusted outcome varied among the patient groups; and (3) the skewness of the quality indicator (i.e., whether a very high proportion of eligible episodes either had or did not have the outcome regardless of patient group).

The extent to which risk-adjustment reduced differences in unadjusted outcomes across the five patient groups then was examined. Logistic regression, the statistical method currently used by CMS to risk-adjust HHQI outcomes, was used to estimate risk-adjustment models for all 12 publicly-reported measures. The theory and evidence-based logistic regression models developed for risk-adjusting outcomes in the earlier project were employed here. For each outcome we estimated a core model with a common set of risk factors used to risk-adjust each HHQI quality measure, and then a full model where the common set of risk factors is supplemented by additional risk factors specific to each outcome. We also estimated a full risk-adjustment model for all outcomes that included as explanatory variables binary indicators defining the five patient groups. The magnitudes of the parameters for the patient group indicators are measures of potential bias in the risk-adjustment models.

RESULTS

Can Clinically Meaningful Groups of Patients Be Identified?

The final algorithm classifies all home health episodes into five mutually exclusive and exhaustive groups based on OASIS data recorded at home health admission. The algorithm first divides episodes into those where patients have chronic conditions that are not well controlled in two or more body systems (i.e., “clinically complex” patients) versus all others. The algorithm then divides each of these two groups into: (1) “post acute care” admissions (i.e., patients with an inpatient stay in a hospital, inpatient rehabilitation facility, or skilled nursing facility (SNF) in the 14 days prior to home health admission); and (2) all others (i.e., “community” admissions). The post acute care group is further divided into: (1) “restorative care” admissions (i.e., patients with a surgical wound, a diagnosis of injury or trauma, or a surgical or orthopedic aftercare diagnosis); and (2) all others.

The five mutually exclusive and exhaustive groups are listed below with the percent of all episodes they represent and the average home health length of stay (LOS) for each group.

1. Community Admission: Clinically Complex -- 8.8 percent of all admissions and average LOS of 90.0 days.

2. Community Admission: Other -- 21.7 percent of all admissions and average LOS of 66.0 days.

3. Post Acute: Clinically Complex -- 17.6 percent of all admissions and average LOS of 58.7 days.

4. Post Acute: Restorative Care -- 41.2 percent of all admissions and average LOS of 40.1 days.

5. Post Acute: Other -- 10.7 percent of all admissions and average LOS of 55.4 days.

The five groups differed in the relative distribution of sociodemographic and clinical characteristics on admission as well as LOS and publicly-reported outcomes. The starkest difference among the five groups was between the “clinically complex community admission” and the “post acute restorative care” groups. Patients in the former group were much more likely to have sensory and communication impairments as well as cognitive deficits on admission relative to the latter group. The diagnostic profile of these two groups also differed markedly. Diabetes and hypertension were common problems among clinically complex community admissions and a significant minority had Alzheimer’s disease or other types of dementia. These diagnoses were reported far less often among patients in the post acute restorative care group. Instead, 42.2 percent had an orthopedic primary diagnosis that affected the score on the clinical dimension of the original Medicare prospective payment system (in place beginning in October 2000) including 29.1 percent with “abnormal gait or other symptoms involving nervous and musculoskeletal systems.” Given differences in clinical condition on admission it is not surprising that the mean LOS for clinically complex community admissions was over twice that of the post acute restorative care group (90.0 versus 40.1 days, respectively).

To What Extent Do Agencies Serve Different Types of Patients?

We found considerable variation in patients served by agencies that differ in size, ownership, control and geographic location. Perhaps the most surprising result is the strong relationship between agency size and the relative share of two of the five types of patients served. Specifically, as agency size increased, the proportion of clinically complex community admissions typically decreased and the proportion of post acute restorative care admissions increased. We also found that hospital-based agencies were more likely to care for post acute home health patients -- in particular, post acute restorative care patients -- relative to other agencies, and that for-profit agencies were more likely to serve community admissions compared to other agencies.

Large differences also were found in the relative distribution of the five types of patients among the nine Census divisions. It is possible that some of this variation reflects differences in the Medicare and Medicaid populations in different areas of the country. However, the magnitude of some of the differences (e.g., the three-fold difference between the New England and West South Central Divisions in the share of clinically complex community admissions) suggests other factors may contribute to these differences. These could include differences in home health agency characteristics, variation in the supply of other types of providers that are potential substitutes for home health care (e.g., SNFs), differences in physician practice patterns, and differences in how OASIS fields used to define patient groups (e.g., admission diagnoses and their severity) are recorded.

Do the Five Patient Groups Differ in Publicly-Reported Outcomes?

The share of each group’s discharges that had the potential to have the 12 HHQI outcomes varied substantially; however, at least half of the discharges in each group had the potential to have most outcomes. The magnitude of differences in unadjusted health status outcomes among the five patient groups was more than 20 percentage points in some cases. The post acute restorative care group had the best outcome on all eight of the health status quality indicators (i.e., the largest percent improving among those eligible to improve) while the two community admission groups had the worst outcomes. The generally worse clinical condition on admission of the individuals in the two community admission groups, relative to the post acute restorative care group, clearly appears to affect their rate of improvement over the course of the home health episode.

The percent of episodes with the three utilization outcomes, as well as the one adverse event outcome were highly skewed across all groups. Less than 10 percent of all admissions were hospitalized, received emergent care for any reason, or emergent care for a wound infection; while over 95 percent were discharge to the community. This result partly reflects the approach we chose to defining home health episodes relative to the approach used to by CMS to define episodes for public reporting of home health quality. In any case, while the magnitude of the utilization outcomes that we report is not large for hospitalization or emergent care, the relative differences among the groups were substantial. The highest rates of hospitalization and emergent care were found among two of the three post acute care groups (i.e., the clinically complex and the “other” groups) while the lowest and second lowest rates of hospitalization and emergent care, respectively, were among post acute restorative care admissions.

To What Extent Does Risk-Adjustment Reduce Differences in Outcomes?

Models developed in an earlier project were employed in this project to estimate risk-adjustment models for the 12 HHQI outcomes. We found that the risk-adjusted outcomes were remarkably similar to the unadjusted outcomes regardless of group suggesting that risk-adjustment at the aggregate level is good. This is not surprising for the HHQI outcomes where the summary statistics for the risk-adjustment models indeed were good. It is more surprising for outcomes where the summary statistics for the risk-adjustment models were mediocre to poor. The unadjusted outcomes for the latter group, however, generally do not vary as much across the five patient groups; weak risk-adjustment is likely to result in risk-adjusted values close to the overall mean which will not differ that much from unadjusted values.

The potential for bias in the risk-adjustment models, at the same time, was found to be substantial although the actual bias at the aggregate level does not appear to be large. Nevertheless, the direction of the bias is of concern. In particular, agencies admitting a relatively large share of community admission could have publicly-reported health status outcomes that are too low (i.e., the adjusted proportion improving is under-reported), while agencies that admit a relatively large share of post acute restorative care patients could have publicly-reported outcomes that are too high. Risk-adjusted utilization outcomes also appear to favor agencies admitting a relatively large share of post acute restorative care patients. An analysis of individual agency outcomes, which is beyond the scope of this project, is required to better understand the extent of bias and impact on actual agency outcomes and rankings.

CONCLUSIONS AND IMPLICATIONS

The results raise the possibility that HHQI is unfair to agencies admitting a relatively large share of patients who tend to have worse outcomes. In particular, agencies with a relatively large share of clinically complex community admissions could be disadvantaged relative to agencies serving a relatively large share of post acute restorative care patients. Because there are differences in the types of patients served by agencies that vary in size, ownership and geographic location, some of the differences in outcomes observed across the five patient groups could be caused by systematic differences in the quality of care provided by different types of agencies. It seems more likely, however, that important risk-adjusters are omitted from current models.

There are two straightforward approaches to improving current risk-adjustment, each having important drawbacks. One approach is to separately risk-adjust and report outcomes for the patient groups developed here or other groupings of chronically ill and post acute home health patients. This would permit comparison of the quality of care provided to important subgroups of patients across different agencies. As noted above, CMS currently assesses publicly-reported nursing home quality indicators separately for a nursing home’s short-stay and long-stay residents although there is only limited risk-adjustment that mainly is operationalized through exclusion rules (Mukamel et al., 2008). The major drawback of separately risk-adjusting and reporting outcomes for subgroups of home health patients is that far fewer agencies than currently is the case would have the minimum number of episodes with the potential to have HHQI outcomes (i.e., 20 episodes).

The other approach is to estimate models that allow risk-adjusters to vary in their effect on outcomes depending on key measures defining patient groups (e.g., community versus post acute admission). The drawback to developing risk-adjustment models that include interaction terms is that the substantial increase in the complexity of the models limits the ability of providers and consumers to understand risk-adjustment. This may undermine trust in and support for risk-adjusting patient outcomes.

Before exploring alternative approaches to risk-adjustment, however, we strongly recommend that analyses be conducted at the agency level. More work is needed to understand the extent and impact of bias in risk-adjusted outcomes at the agency level as well as the relationship among agency and geographic factors, the types of individuals served, and patient outcomes. While the risk-adjustment models currently used by CMS are likely to produce results similar to those reported here, this also should be evaluated at the agency level.

Project results provide important insights into the types of patients served by home health agencies and differences in their outcomes on publicly-reported measures. Multiple chronic conditions, as in other health care settings, were found to be important contributors to outcomes. Given the limitations of current reporting of medical conditions on OASIS, new ways of insuring better information about chronic conditions need to be developed to improve patient care and the risk-adjustment of outcomes. While many questions remain to be answered, the findings from this project in several areas provide critical information needed by the Department in its efforts to assess and improve the quality of care provided to the diverse home health population.