At Home Primary Care and Medicare Costs in High-Risk Elders
Medicare claims data were used to identify a case cohort of FFS Medicare beneficiaries who enrolled in the HBPC program from 2004 to 2008. A 3:1 direct matching methodology was used to create a control cohort. Inclusion criteria were aged 65 and older and without health maintenance organization coverage during the month of enrollment and for 3 months before. The data received included Medicare Parts A and B claims. Using the Residential History File, total costs and patterns of use during the study period were determined for cases and controls, and the two groups were compared using univariate analysis and multivariate linear regression models. The follow-up period began in the month after the index month and continued until death, last month of FFS eligibility, long-term nursing home placement, or end of study period in December 2008.
An urban HBPC practice in Washington, District of Columbia, was examined. Outcomes for 722 incident HBPC cases and 2,161 well-matched external controls during 2004 to 2008 were examined.
Cases included individuals newly enrolled in the HBPC program during 2004 to 2008. The month of program enrollment was the index month. Nine hundred nine incident HPBC beneficiaries were identified in this time period; 197 were excluded because they lacked Medicare FFS eligibility, resided in a nursing home, or died during the index month. Controls were excluded for the same reasons. Subjects were eligible if they had a Medicare SNF stay but were not eligible if they were in a nursing home for long-term care. Controls were selected from a large pool of beneficiaries in Washington, District of Columbia, and urban counties of Virginia, Maryland, and Pennsylvania. The total control pool from which 2,161 matched controls were drawn consisted of 1,765,972 Medicare beneficiaries.
Cases and controls were matched at the index month for sex; age bands; race and ethnicity; Medicare buy-in status; long-term nursing home placement status; death in index month; diagnosis of Alzheimer's disease (International Classification of Diseases, Ninth Revision, (ICD-9) code 331.0) or a chronic mental illness (CMI) such as schizophrenia, depression, psychosis, or alcohol abuse (ICD-9 codes 201, 292, 295–98, 300, 303–4, 311); FFS eligibility; and frailty index. There is a close relationship in frail elders between reporting of dementia and behavioral diagnoses. Because the effect of CMI and dementia on care management is intertwined and not easily distinguishable, a combined Alzheimer's disease and CMI category was used. Three controls were available for 718 of 722 cases, and one or two controls were available for the other four cases, for a total of 2,161 controls. The first three matches achieved through random selection were used.
Frailty was measured using an index score developed by JEN Associates. A linear relationship between the JEN Frailty Index (JFI) and the probability of health service usage and future nursing home entry has been found in high-risk populations, including but not limited to homebound persons. The JFI sums the presence (score = 1) or absence (score = 0) of 13 categories of illness linked to need for long-term supportive services. The 13 categories are minor or major ambulatory impairment, mental health diagnosis, mental retardation, dementia, impairment in sensory function or self-care, presence of general symptoms, diagnosis of cancer, presence of major chronic diseases, pneumonia, renal disease, or other medical risks. Summed JFI scores create ranking groups (0–3 low, 4–6 medium, ≥7 high). Major ambulatory impairment was defined according to the presence of certain diagnoses, such as hip fracture, stroke, and falls, which served in this analysis as a claims data proxy for functional impairment.
Baseline characteristics were determined using claims data for diagnoses and use patterns during the 4-month baseline period. Comorbidity flags were yes-or-no indicators of the presence of major selected chronic diseases as primary or secondary diagnoses in the index year.
The HBPC program has served ill elders in Washington, District of Columbia, since 1999 under the auspices of the Geriatrics Division of MedStar Washington Hospital Center (MWHC). The HBPC program is similar to VA HBPC in its use of an interprofessional team of physicians, nurse practitioners (NPs), and mental health staff with an elderly, chronically ill population. The MWHC program differs from some house call programs around the United States because the team physicians follow individuals in the hospital and at home, and the core team has a strong social work component.
HBPC recipients have multiple chronic illnesses such as dementia, CHF, diabetes mellitus, chronic obstructive pulmonary disease (COPD), stroke, and severe arthritis. The HBPC program delivers detailed care coordination at home with a team of geriatricians, NPs, social workers, licensed practical nurses, and office coordinators. The physicians perform an initial visit, visit beneficiaries every 3 to 4 months, provide 24-hour-a-day, 7-day-a week on-call telephone coverage, and perform hospital attending duties. The NPs make frequent visits, ranging from every 8 weeks to several times a week, depending on medical necessity. The social workers provide case management for psychosocial and supportive services. Team members occasionally make joint visits to resolve conflicts in care plan, address staff safety concerns, or resolve ethical questions. Weekly team meetings allow discussion of individuals with unstable conditions and direct communication with home health, mental health, and pharmacy colleagues. The team uses a wireless electronic health record with live access to inpatient and outpatient records and applies home-based diagnostic technology.
Primary outcomes were total Medicare costs, mortality, and pattern of use such as hospital admissions, SNF care, ED visits, skilled home health episodes, hospice, and subspecialist or generalist visits. Generalist visits included all home and office visits by NPs or primary care physicians, including internal medicine, family medicine, and geriatrics. All participants were included in follow-up analysis with Medicare claims data; death, long-term nursing home placement, entry into a Medicare Advantage program, or end of study period truncated their follow-up. The team did not provide direct care but coordinated discharge planning for beneficiaries admitted to a Medicare SNF bed.
Medicare FFS claims data were analyzed for all cases and controls during 2004 to 2008. CMS granted the use of Medicare Standard Analytic Files, including claims from all covered services except Part D records. The data release met privacy requirements of the federal government and was approved by the institutional review board of MWHC. Individual-level longitudinal records were constructed, including summaries of payments, patterns of use, and flags for selected diagnoses. Dates of death came from Social Security Administration benefit records.
Univariate analysis was performed using analysis of variance, chi-square tests, and t-tests. Descriptive statistics were used to calculate prevalence of selected major chronic diseases, demographic characteristics, costs, and use patterns for baseline and follow-up periods. Multiple linear regression models were used to measure differences in Medicare costs, mortality, hospital admissions, hospital days, SNF days, ED visits, and specialist and generalist encounters. Covariates' estimated effect on expenditures was derived from a linear regression model, based on stepwise selection of major selected chronic diseases and baseline period use, with separate variables for home health, hospitalization, and SNF care. The premodel matching of participant characteristics, streamlining of factors in the stepwise selection, and the use of score-based complexity covariates allowed for control of interactions.
Presence of selected major chronic diseases, including osteoarthritis, coronary heart disease, CHF, COPD, cerebrovascular disease, and diabetes mellitus (ICD-9 codes 7145, 410–11, 413–14, 427–28, 491–93, 496, 430–438, 250) was controlled for. Covariates were chosen based on external clinical judgment and on evidence from the literature of what affects use. A Cox proportional hazards model was used to assess differences in mortality during follow-up, which ensured that effects of end-of-life events and unequal follow-up time were equally distributed. The Cox model controlled for information censoring in the baseline matching and covariate selection. Death was a proportional risk over time for the aging population.
Log + 1 transformation was applied to improve usage modeling. The parameter estimates with log + 1 transformation can be interpreted as a percentage difference in the model's outcome variable when comparing those with and without predictor variables. All analyses were performed using SAS version 9.1 (SAS Institute, Inc., Cary, NC). The study's funding source played no role in data collection or statistical analysis.
Methods
Design Overview
Medicare claims data were used to identify a case cohort of FFS Medicare beneficiaries who enrolled in the HBPC program from 2004 to 2008. A 3:1 direct matching methodology was used to create a control cohort. Inclusion criteria were aged 65 and older and without health maintenance organization coverage during the month of enrollment and for 3 months before. The data received included Medicare Parts A and B claims. Using the Residential History File, total costs and patterns of use during the study period were determined for cases and controls, and the two groups were compared using univariate analysis and multivariate linear regression models. The follow-up period began in the month after the index month and continued until death, last month of FFS eligibility, long-term nursing home placement, or end of study period in December 2008.
Setting and Participants
An urban HBPC practice in Washington, District of Columbia, was examined. Outcomes for 722 incident HBPC cases and 2,161 well-matched external controls during 2004 to 2008 were examined.
Participant Selection
Cases included individuals newly enrolled in the HBPC program during 2004 to 2008. The month of program enrollment was the index month. Nine hundred nine incident HPBC beneficiaries were identified in this time period; 197 were excluded because they lacked Medicare FFS eligibility, resided in a nursing home, or died during the index month. Controls were excluded for the same reasons. Subjects were eligible if they had a Medicare SNF stay but were not eligible if they were in a nursing home for long-term care. Controls were selected from a large pool of beneficiaries in Washington, District of Columbia, and urban counties of Virginia, Maryland, and Pennsylvania. The total control pool from which 2,161 matched controls were drawn consisted of 1,765,972 Medicare beneficiaries.
Cases and controls were matched at the index month for sex; age bands; race and ethnicity; Medicare buy-in status; long-term nursing home placement status; death in index month; diagnosis of Alzheimer's disease (International Classification of Diseases, Ninth Revision, (ICD-9) code 331.0) or a chronic mental illness (CMI) such as schizophrenia, depression, psychosis, or alcohol abuse (ICD-9 codes 201, 292, 295–98, 300, 303–4, 311); FFS eligibility; and frailty index. There is a close relationship in frail elders between reporting of dementia and behavioral diagnoses. Because the effect of CMI and dementia on care management is intertwined and not easily distinguishable, a combined Alzheimer's disease and CMI category was used. Three controls were available for 718 of 722 cases, and one or two controls were available for the other four cases, for a total of 2,161 controls. The first three matches achieved through random selection were used.
Frailty was measured using an index score developed by JEN Associates. A linear relationship between the JEN Frailty Index (JFI) and the probability of health service usage and future nursing home entry has been found in high-risk populations, including but not limited to homebound persons. The JFI sums the presence (score = 1) or absence (score = 0) of 13 categories of illness linked to need for long-term supportive services. The 13 categories are minor or major ambulatory impairment, mental health diagnosis, mental retardation, dementia, impairment in sensory function or self-care, presence of general symptoms, diagnosis of cancer, presence of major chronic diseases, pneumonia, renal disease, or other medical risks. Summed JFI scores create ranking groups (0–3 low, 4–6 medium, ≥7 high). Major ambulatory impairment was defined according to the presence of certain diagnoses, such as hip fracture, stroke, and falls, which served in this analysis as a claims data proxy for functional impairment.
Baseline characteristics were determined using claims data for diagnoses and use patterns during the 4-month baseline period. Comorbidity flags were yes-or-no indicators of the presence of major selected chronic diseases as primary or secondary diagnoses in the index year.
Intervention
The HBPC program has served ill elders in Washington, District of Columbia, since 1999 under the auspices of the Geriatrics Division of MedStar Washington Hospital Center (MWHC). The HBPC program is similar to VA HBPC in its use of an interprofessional team of physicians, nurse practitioners (NPs), and mental health staff with an elderly, chronically ill population. The MWHC program differs from some house call programs around the United States because the team physicians follow individuals in the hospital and at home, and the core team has a strong social work component.
HBPC recipients have multiple chronic illnesses such as dementia, CHF, diabetes mellitus, chronic obstructive pulmonary disease (COPD), stroke, and severe arthritis. The HBPC program delivers detailed care coordination at home with a team of geriatricians, NPs, social workers, licensed practical nurses, and office coordinators. The physicians perform an initial visit, visit beneficiaries every 3 to 4 months, provide 24-hour-a-day, 7-day-a week on-call telephone coverage, and perform hospital attending duties. The NPs make frequent visits, ranging from every 8 weeks to several times a week, depending on medical necessity. The social workers provide case management for psychosocial and supportive services. Team members occasionally make joint visits to resolve conflicts in care plan, address staff safety concerns, or resolve ethical questions. Weekly team meetings allow discussion of individuals with unstable conditions and direct communication with home health, mental health, and pharmacy colleagues. The team uses a wireless electronic health record with live access to inpatient and outpatient records and applies home-based diagnostic technology.
Outcomes and Follow-up
Primary outcomes were total Medicare costs, mortality, and pattern of use such as hospital admissions, SNF care, ED visits, skilled home health episodes, hospice, and subspecialist or generalist visits. Generalist visits included all home and office visits by NPs or primary care physicians, including internal medicine, family medicine, and geriatrics. All participants were included in follow-up analysis with Medicare claims data; death, long-term nursing home placement, entry into a Medicare Advantage program, or end of study period truncated their follow-up. The team did not provide direct care but coordinated discharge planning for beneficiaries admitted to a Medicare SNF bed.
Statistical Analysis
Medicare FFS claims data were analyzed for all cases and controls during 2004 to 2008. CMS granted the use of Medicare Standard Analytic Files, including claims from all covered services except Part D records. The data release met privacy requirements of the federal government and was approved by the institutional review board of MWHC. Individual-level longitudinal records were constructed, including summaries of payments, patterns of use, and flags for selected diagnoses. Dates of death came from Social Security Administration benefit records.
Univariate analysis was performed using analysis of variance, chi-square tests, and t-tests. Descriptive statistics were used to calculate prevalence of selected major chronic diseases, demographic characteristics, costs, and use patterns for baseline and follow-up periods. Multiple linear regression models were used to measure differences in Medicare costs, mortality, hospital admissions, hospital days, SNF days, ED visits, and specialist and generalist encounters. Covariates' estimated effect on expenditures was derived from a linear regression model, based on stepwise selection of major selected chronic diseases and baseline period use, with separate variables for home health, hospitalization, and SNF care. The premodel matching of participant characteristics, streamlining of factors in the stepwise selection, and the use of score-based complexity covariates allowed for control of interactions.
Presence of selected major chronic diseases, including osteoarthritis, coronary heart disease, CHF, COPD, cerebrovascular disease, and diabetes mellitus (ICD-9 codes 7145, 410–11, 413–14, 427–28, 491–93, 496, 430–438, 250) was controlled for. Covariates were chosen based on external clinical judgment and on evidence from the literature of what affects use. A Cox proportional hazards model was used to assess differences in mortality during follow-up, which ensured that effects of end-of-life events and unequal follow-up time were equally distributed. The Cox model controlled for information censoring in the baseline matching and covariate selection. Death was a proportional risk over time for the aging population.
Log + 1 transformation was applied to improve usage modeling. The parameter estimates with log + 1 transformation can be interpreted as a percentage difference in the model's outcome variable when comparing those with and without predictor variables. All analyses were performed using SAS version 9.1 (SAS Institute, Inc., Cary, NC). The study's funding source played no role in data collection or statistical analysis.
SHARE
