Abstract
Background:
Hospitalization with heart failure (HF) may signal an increased risk of Alzheimer’s disease and related dementias (ADRD). Nursing homes routinely assess cognition but the association of these results with new ADRD diagnosis in a population at high risk of ADRD is not known.
Objective:
To determine the association between nursing home cognitive assessment results and new diagnosis of dementia after heart failure hospitalization.
Methods:
This retrospective cohort study included Veterans hospitalized for HF and discharged to nursing homes, from 2010 to 2015, without a prior diagnosis of ADRD. We determined mild, moderate, or severe cognitive impairment using multiple items of the nursing home admission assessment. We used Cox regression to determine the association of cognitive impairment with new ADRD diagnosis during 365 days of follow-up.
Results:
The cohort included 7,472 residents, new diagnosis of ADRD occurred in 4,182 (56%). The adjusted hazard ratio of ADRD diagnosis was 4.5 (95% CI 4.2, 4.8) for the mild impairment group, 5.4 (95% CI 4.8, 5.9) for moderate impairment, and 4.0 (95% CI 3.2, 5.0) for severe impairment compared to the cognitively intact group.
Conclusion:
New ADRD diagnoses occurred in more than half of Veterans with HF admitted to nursing homes for post-acute care.
Keywords
INTRODUCTION
Consequences of Alzheimer’s disease and related dementias (ADRD) threaten recovery from heart failure hospitalization, yet clinicians often fail to recognize cognitive impairment during acute care [1]. Timely recognition of cognitive impairment in persons with heart failure may prompt referral to a medical specialist trained in dementia care and enable more caregiver engagement in disease management. Because heart failure management often requires attention to multiple medications, diet, blood pressure, and weight, successful management relies on patients’ or caregivers’ intact cognitive functioning. Veterans Affairs directives require specific indications for cognitive testing [2]; presence of heart failure is not sufficient though cognitive impairment is known to influence outcomes [3]. In contrast, community nursing homes perform cognitive screening regardless of need.
Nursing homes complete Resident Assessment Instruments according to Centers for Medicare and Medicaid Services policy and record results in the Minimum Data Set 3.0 (MDS), including a Brief Interview for Mental Status (BIMS) consisting of three-word recall and questions about orientation [4]. By contrast, neither ambulatory care nor acute care in Veterans Affairs Medical Centers include universal cognitive screening regardless of symptoms [2]. Therefore, nursing home admission of Veterans with heart failure provides a unique opportunity to examine the relationship of cognitive screening results with subsequent diagnosis of ADRD. The MDS 3.0 Cognitive Function Scale (CFS) summarizes the BIMS and other MDS variables related to cognition resulting in a score of intact or mild, moderate, or severe impairment [4]. MDS 3.0 also includes a Confusion Assessment Method, a screening tool for possible delirium. In Veterans discharged to nursing homes after heart failure hospitalization, a CAM assessment suggesting delirium was associated with less improvement in activities of daily living function [5].
Previous studies report the association between nursing home cognitive assessment and diagnosis of dementia, but not with specific attention to heart failure. In a general nursing home population, MDS-CAM suggesting delirium was associated with a higher rate of new ADRD diagnosis [6]. The strength of association varied across levels of CFS, with the strongest association found in mild impairment. Because heart failure increases risk of incident ADRD [7], and because cognitive impairment threatens heart failure self-management [3], the association between nursing home cognitive screening results and new ADRD diagnosis merits further investigation. This study will determine the association between each level of cognitive impairment and a new diagnosis of ADRD in the year following nursing home admission. We hypothesize that increasing incremental impairment measured by CFS will be associated with progressively higher rates of new ADRD diagnosis.
MATERIALS AND METHODS
Study design
This observational retrospective cohort study examined the association between results of cognitive impairment testing upon admission to nursing homes after heart failure hospitalization and clinical recognition of ADRD as represented in the use of diagnostic codes. The institutional review board of the VA Medical Center approved the study as a secondary data analysis. The analysis uses VA health system data linked to data from Centers for Medicare and Medicaid Services (CMS), including the Minimum Data Set 3.0 (MDS).
Sample
This study identified acute heart failure (HF) admissions in VA medical centers from January 1, 2011 to December 31, 2018 using ICD-9 and ICD-10 codes. For Veterans with more than one HF admission during this time period, we randomly selected one index HF admission to avoid selection bias from including the chronologically earliest HF admission. We then used MDS to include all nursing home (NH) admissions within 7 days of discharge from an index HF admission. Next, we excluded Veterans with an ADRD diagnosis during the index HF admission or in the preceding 3 years. We then used MDS and VA Long Term Care files to exclude Veterans with NH care in the 12 months prior to inclusion. Finally, we excluded Veterans who died before NH admission, and due to missing data elements needed for CFS.
Cognitive impairment
For the independent variable, we calculated the MDS 3.0 Cognitive Function Scale (CFS) in the manner described elsewhere [4]. The CFS algorithm combines multiple MDS items to assign a cognitive status of intact, mild, moderate, or severe impairment. CFS derives from the BIMS score in residents who complete that assessment, and from the MDS Cognitive Performance Scale in other residents [8]. We calculated CFS using the first MDS assessment in the nursing home stay which contained the requisite data elements.
Alzheimer’s disease and related dementias
For determining prevalent ADRD as an exclusion criterion and to ascertain the study outcome of incident ADRD, this analysis adapts the 1999–2020 CMS Chronic Conditions Warehouse (CCW) definition of ‘Alzheimer’s Disease and Related Disorders or Senile Dementia’ [9]. This definition includes a list of ICD-10 codes, some of which represent specific dementia subtypes, and some of which represent ADRD non-specifically. We modified this list by removing ICD-10 code F05. F05 represents ‘delirium due to known physiological condition.’ Because delirium, a distinct clinical entity from dementia [10], may confound the association between cognitive impairment and ADRD diagnosis, inclusion of F05 would undermine the goal of accurately modeling this relationship [11]. Our algorithm uses a 3-year reference period to determine prevalent ADRD, and considers diagnoses reported in VA data, CMS claims, and the MDS for both prevalent and incident ADRD. We defined a new ADRD diagnosis as a diagnosis code from our list used in one of these sources during the 1-year follow-up period.
Covariates
Using the VA Corporate Data Warehouse, we determined the baseline demographic, utilization, and clinical characteristics of the cohort overall and according to the admission CFS score. Demographic variables included age, sex, race, and rural residence as recorded in the VA Computerized Patient Record System (CPRS). Utilization variables included total VA and VA fee-basis medical expenses and total number of hospital days in the year prior to index admission. We used the most recent ejection fraction value recorded within 1 year of the hospital admission date. For the Elixhauser comorbidity score, and specific categories of medical comorbidity, we used the year prior to hospital admission as the ascertainment period.
Statistical analysis
For comparison of baseline characteristics across levels of CFS, we used ANOVA for continuous variables and chi-square tests for categorical variables to compute the p value for the group effect of the CFS variable. We used Cox proportional hazards models to compare the rate of new ADRD diagnosis in mild, moderate, and severe impairment to intact CFS. All models adjusted for baseline characteristics which could influence the likelihood of impaired performance on the cognitive screening and subsequent diagnosis of dementia (i.e., age, gender, race, total medical expenses in the prior year, the length of the index HF admission, ejection fraction, use of palliative care in the prior year, rural place of residence, the comorbidities which compose the Elixhauser Comorbidity Index, diagnosis of stroke, and facility effects). The models also controlled for competing risk of mortality, determined through the VA corporate data warehouse. Statistical analyses used Microsoft SQL Server Management Studio version 18 (Microsoft Corporation) and SAS enterprise guide version 7.1 (SAS institute).
Sensitivity analysis
As a robustness check, we considered only VA encounters and CMS claims and not MDS-reported diagnoses when ascertaining the outcome of new ADRD diagnosis.
RESULTS
Study cohort
Beginning with 1,011,395 potentially eligible HF admissions, 7,472 NH admissions met the inclusion criteria (Fig. 1). Residents’ mean (SD) age was 76 (10) years, 218 (3%) were female, 1,102 (15%) were Black, and 575 (8%) were Hispanic (Table 1). Ejection fraction was reduced in 1,628 (22%), mid-range in 1,004 (13%), preserved in 2,013 (27%), and missing in 2,827 (38%). MDS-CAM was positive in 168 (2%), and the mean (SD) Elixhauser Comorbidity Index was 5.2 (2.8). Ejection fraction data was missing in 2,827 (38%) residents. Race data was missing in 15 (0.2%), and data on place of residence was missing in 7 (0.1%).
Study flow diagram.
Baseline characteristics of nursing home residents
Comorbidity frequencies ascertained in year prior to index hospital admission. aMinimum Data Set 3.0 Confusion Assessment Method. bLength of Stay. cTotal inpatient days in the 12 months prior to the index hospital stay. dTotal Veterans Affairs and Veterans Affairs fee-basis medical expenses in the 12 months prior to the index hospital stay.
Cognitive Function Scale and baseline characteristics
According to the MDS 3.0 Cognitive Function Scale, 5,033 (67%) of residents were cognitively intact, 1,829 (25%) had mild cognitive impairment, 524 (7%) had moderate cognitive impairment, and 86 (1%) had severe cognitive impairment. Residents rated as cognitively impaired tended to be older, were more often Black, and were more often Hispanic than residents rated as cognitively intact. Obesity was present in 393 (24%) residents rated cognitively intact, 277 (15%) mildly impaired, 62 (12%) moderately impaired, and 9 (11%) residents rated as having severe cognitive impairment. Mean (SD) length of the index hospital stay was 11.8 (10.3) days in residents rated as cognitively intact, 11.8 (10.4) in mildly impaired, 13.9 (12.4) in moderately impaired, and 16.5 (14.1) in residents rated as having severe cognitive impairment.
Cognitive Function Scale and diagnosis of ADRD
A new diagnosis of ADRD was given to 4,182 (56%) residents during the 365 days following their NH admission. Of those assessed as cognitively intact, or mildly, moderately, or severely impaired, 41%, 85%, 91%, and 71% were diagnosed with ADRD in the year following NH admission, respectively (Fig. 1). Adjusted hazard ratios of new ADRD diagnosis were 4.5 (4.2–4.8) for mild impairment, 5.4 (4.8–5.9) for moderate impairment, and 4.0 (3.2–5.0) for severe impairment (Table 2).
Hazard ratios for new diagnosis of Alzheimer’s disease and related dementias after NH admission following HF hospitalization; comparing mild, moderate, and severe cognitive impairment to cognitively intact
aMinimum Data Set 3.0 Cognitive Function Scale. bAdjustment variables: age, sex, race and ethnicity, total Veterans Affairs and Veterans Affairs fee-basis medical expenses in previous 12 months, total inpatient days in the 12 months prior to the index hospital stay, length of the index hospital stay, ejection fraction, palliative care in the 12 months prior to the index hospital stay, rural place of residence, individual comorbidities in the Elixhauser Comorbidity Index, stroke, VA medical center.
Sensitivity analysis
When we did not consider MDS-reported diagnoses for the outcome of new ADRD diagnosis, a new ADRD diagnosis occurred in 4179 (56%). The adjusted hazard ratio (95% CI) of new ADRD diagnosis was 1.4 (1.3–1.5) in mild impairment, 2.0 (1.8–2.3) in moderate impairment, and 2.2 (2.2.– 3.6) in severe impairment compared with cognitively intact.
DISCUSSION
This observational cohort study found that newly-recorded ADRD diagnosis occurred in 56% of Veterans admitted to NH after HF hospitalization and without a previous diagnosis of ADRD. Furthermore, we found that mild, moderate, and severe impairment were each associated with increased rates of newly-recorded ADRD diagnosis when adjusting for measured sources of confounding. The strength of this association increases in moderate impairment compared to mild impairment but not in severe impairment. Our lower estimated hazard ratio for severe impairment relative to moderate and mild impairment must be interpreted considering our study’s exclusion of residents with previous ADRD diagnosis. We expect that Veterans with ADRD causing severe impairment would have had a previously-recorded ADRD diagnosis at the beginning of follow-up and would not have been included in this study. Empirically, we included a relatively low number (%) of residents with severe impairment: 86 (1%).
Our findings demonstrate that incident ADRD diagnosis occurs very commonly in persons admitted to NH after HF hospitalization. Although we found a strong association between CFS and new ADRD diagnosis, this outcome still occurred in 41% of individuals scoring ‘intact.’ Therefore, the cognitive assessment items on MDS may lack sensitivity to detect incipient ADRD in populations at very high risk. Clinicians may defer diagnosis of dementia beyond the mild-to-moderate stage for numerous reasons. These may include diagnostic discomfort or patient reluctance to share information about symptoms. Clinicians may also lack time to evaluate cognitive impairment, especially in the context of chronic heart failure management, which requires detailed clinical assessment and patient or caregiver education [1]. Clinical recognition and formal diagnosis of ADRD requires a clinician to devote attention to the possibility of a neurocognitive disorder. Such a focus may occur in response to expression of concern by a patient or caregiver or following an acute clinical event such as a heart failure exacerbation. Additionally, the post-acute care setting’s emphasis on discharge planning and multidisciplinary assessment may draw the clinician’s attention to an unexpected CFS finding or other evidence of cognitive dysfunction. With this information, clinicians can consider the possibility of undiagnosed ADRD and its implications for post-discharge management.
An earlier report on Medicare enrollees entering nursing homes for post-acute care regardless of hospital diagnosis found new ADRD diagnoses to occur in 6.3% of residents over the following year. In this sample, 7% had severe impairment, 9% moderate, 18% mild, and 66% were cognitively intact according to CFS [6]. This study used MDS-CAM as the primary predictor variable and found that positive MDS-CAM was associated with a new ADRD diagnosis with a hazard ratio (95% CI) of 2.6 (1.5–4.6) compared with negative MDS-CAM. In contrast, we found new ADRD diagnoses in 56% of residents following hospitalization for heart failure. This contrast offers compounding evidence for the potential public health impact of cognitive screening in older adults with heart failure. First, the rate of new ADRD diagnosis in residents admitted to post-acute care for heart failure greatly exceeds that in a general post-acute care population. Second, lapses in self-management due to unrecognized ADRD threaten heart failure patients with exacerbation, hospitalization [1], and rehospitalization [3]. Finally, health systems and provider networks may be able to mitigate this threat through timely diagnoses that result in the introduction of appropriate long-term services such as chronic care management [12, 13]. The tendency for Medicare beneficiaries with ADRD to begin accruing increased medical costs in the year before diagnosis offers further support for this proposition [14]. Available services include remotely monitored vital signs and weight monitoring equipment, telephonic, virtual, or in-person care management, and home health services. Indeed, heart failure patients with significant enough cognitive impairment to fail the MiniCog have high risk for rehospitalization once discharged from NH care [3]. Our findings suggest an opportunity to address impairment in heart failure self-management due to ADRD earlier through long-term services and support interventions [15].
Our study benefits from several strengths: it draws upon data linking electronic health records from the largest integrated health system in the United States to CMS and MDS data. It therefore includes a geographically and racially diverse population. There are several limitations to this study: the cohort included very few women. The outcome, defined as clinical use of a diagnostic code representing ADRD, occurs due to the confluence of multiple factors including the severity of cognitive symptoms and the practice patterns of post-acute care providers in coding for an ADRD diagnosis. The CCW algorithm for ADRD’s inclusion of specific and nonspecific diagnostic codes introduces additional heterogeneity into the outcome definition. The study does not allow for comparisons between VA and non-VA care nor does it allow for comparison of post-acute care in community nursing homes to care in other settings such as home health care or VA community living centers. There may be unmeasured sources of confounding due to the observational design of the study.
Conclusion
This retrospective cohort study measured new ADRD diagnosis in Veterans admitted to nursing homes after HF hospitalization, and its association with results of routine cognitive assessment. New ADRD diagnoses occurred in more than half of Veterans with HF admitted to nursing homes for post-acute care, and cognitive impairment was associated with subsequent ADRD diagnosis. Our results demonstrate that persons admitted to nursing homes have after heart failure hospitalization without a previously-recorded ADRD diagnosis have a very high risk of underlying ADRD. Clinicians should be sensitive to signs of cognitive impairment in such patients and provide appropriate assessment and follow-up care. Abnormal results on MDS cognitive assessments were associated with new ADRD diagnosis, but new ADRD diagnosis was still common when performance on these assessments was normal. Therefore, it appears reasonable to carefully assess for ADRD in any individual admitted to a nursing home after heart failure hospitalization, even when MDS cognitive assessments appear normal. Future studies should examine the effects of cognitive screening and appropriate services in Veterans hospitalized with heart failure on patient-centered outcomes.
Footnotes
ACKNOWLEDGMENTS
The views and opinions expressed do not represent the official policies and protocols of the VA or the United States Government.
FUNDING
Drs. Bayer, Erqou, Kunicki, Singh, Bozzay, McGeary, Zullo, Wu, Gravenstein, and Rudolph and Ms. Jiang are employees of the US Department of Veterans Affairs. Drs. Rudolph, Gravenstein, Zullo, and Eqou and Ms. Jiang are funded by the VA Health Services Research and Development Center of Innovation in Long Term Services and Supports (CIN-13-419) and Drs. Erqou and Rudolph are funded by the Providence Evidence Synthesis Program (ESP-22-116). Dr. Zullo was funded, in part, by National Institute on Aging awards R01AG077620, R01AG079295, and R01AG065722-03S1.
CONFLICT OF INTEREST
The authors have no conflict of interest to report.
DATA AVAILABILITY
The participants of this study did not give written consent for their data to be shared publicly, so due to the sensitive nature of the research supporting data is not available.