Trends in US Medicare Decedents’ Diagnosis of Dementia From 2004 to 2017

This cross-sectional study assesses trends in frequency of diagnosis of Alzheimer disease and related dementias in the last 2 years of life and any associated changes in billing practices, characteristics of the diagnosed population, and intensity of end-of-life care.


Introduction
Alzheimer disease and related dementias (ADRD) for which we have no effective disease-modifying treatment are estimated to cost the US as much as $215 billion annually. 1As the population ages, the importance of ADRD as a cause of or contributor to death continues to increase. 2The recognition of dementia and the burden to affected individuals and their families has increased efforts to address the needs of this population.In addition to the human toll of the disease, there are also high costs associated with both medical care and long-term care as people with dementia progress from mild impairment toward death.The increased attention has led to reassessments of how to account for the presence of dementia in the processes leading up to death.
Recently, contributing factors such as ADRD have been taken into consideration when determining causes of death, 2,3 and ADRD has increased from the eighth leading cause of death in the US in 2000 to the sixth in 2018 4,5 (and the fifth among older adults 6 ).This change reflects changing attitudes and practices for documenting the presence of this disease, which is also likely happening in clinical practice.Dementia has long been noted to be underdiagnosed, 7 but higher levels of public education and more diagnostic and treatment options may be changing how often clinicians document and bill for ADRD.Moreover, several external factors could affect ADRD coding, such as the adoption of electronic health records, the influence of the National Alzheimer Project Act 8 on disease awareness and increased research funding, and changes in the allowable number and coding systems used for diagnoses on Medicare claims.
Understanding whether ADRD identification has increased is important because it may suggest greater opportunity to address burdensome end-of-life care.But it also has health policy importance.
Dementia was added into the risk adjustment strategy Hierarchical Condition Categories (HCC), constructed from algorithms of claims, and used by the Centers for Medicare & Medicaid Services for compensating health plans. 9,10In addition, ADRD coding is important for economic evaluations of how dementia contributes to the costs of care.
Therefore, the objective of this study was to determine whether the likelihood of dying with an ADRD diagnosis, identified from billing data, has changed over time.We examined diagnosis of ADRD among decedents because end-stage disease should be less influenced by temporal changes in diagnostic technologies used in early disease stages. 11,12The hypothesis was that ADRD diagnosis frequency would increase over time coincident with changes in diagnostic billing behavior.We examined whether the potential change in ADRD diagnosis frequency was associated with changes in the characteristics of the ADRD-identified population.Lastly, we examined whether trends toward greater diagnosis frequency were associated with changes in the intensity of care at the end of life.

Study Design
A serial cross-sectional study of Medicare decedents using a 20% national sample of fee-for-service enrollees from 2004 to 2017 was conducted to examine changes in the percentage of older adults receiving an ADRD diagnosis within 2 years of death.This study received approval from the institutional review board of Dartmouth College, waiving informed consent for the use of deidentified data, and abides by Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Study Population
Using the 20% Master Beneficiary Summary Files, we identified all individuals who died between 2004 and 2017.Among decedents, we restricted to those who were continually enrolled in both Medicare Parts A and B, and not enrolled in Medicare Advantage for 24 months before the date of death.With a 2-year look-back period for diagnosis identification, decedents must have been 67 years or older at the time of death to be included.We chose a 2-year look-back period because the

JAMA Health Forum | Original Investigation
Trends in US Medicare Decedents' Diagnosis of Dementia From 2004 to 2017 majority of decedents having an ADRD diagnosis will receive at least 1 ADRD diagnosis on a claim within 2 years of death. 13The final sample consisted of 3 515 329 older decedents.

Identification of Decedents With ADRD Diagnosis
The presence of an ADRD diagnosis was identified from inpatient records (Medicare Provider Analysis and Review [MedPAR]), professional service claims (carrier file and selected outpatient hospital claims for clinician visits in underserved settings), home health records, and hospice claims.We applied a standard, previously validated claims algorithm that requires the presence of a single claim with an ADRD diagnosis 14,15 rather than 2 claims, as used for other diseases, owing to concerns for underdiagnosis.We also applied a more stringent approach by requiring 2 or more claims with an ADRD diagnosis for professional services at least 7 days apart.[18] Changes in clinical practice or volume of care delivered in each setting could influence frequency of diagnosis documentation or setting where diagnosis occurred across the inpatient, professional service, home health, or hospice settings.In addition to identifying how many decedents had received an ADRD diagnosis, we examined whether changes had occurred in the setting where the diagnosis was billed.Trends were examined in the occurrence of an ADRD diagnosis in each claim type (ie, nonmutually exclusive), as well as when the diagnosis appeared in only 1 claim type.For descriptive purposes, we superimposed the external events (adoption of electronic health records, 19,20 billing rule changes, and enactment of the National Alzheimer Project Act) that we hypothesized could have been associated with the likelihood of ADRD diagnosis.

Demographic Characteristics
Changes in the likelihood of receiving an ADRD diagnosis could lead to differences in the composition of the ADRD decedent population.We examined demographic characteristics and selected comorbidities among ADRD decedents relative to changes in decedents who did not receive an ADRD diagnosis.We examined characteristics in 2004 vs 2017 and in 2011 when the expansion in the number of diagnoses on health care claims went into effect. 21Demographic characteristics at the time of death included age, sex, and race and ethnicity.Race and ethnicity was categorized as Black vs non-Black because the percentage of Black and non-Black individuals was stable over time, unlike categorization of White race, which declined as coding of Hispanic ethnicity increased. 22Non-Black race and ethnicity included American Indian or Alaska Native, Asian, Hispanic, White, and those of other races and ethnicities.The Elixhauser Comorbidity Index was used to identify clinically significant comorbidities in the last year of life, 23 including cancer, diabetes, congestive heart failure, chronic pulmonary disease, and kidney failure.

Volume of Health Service Use
We also determined whether health service use temporally increased because previous research shows that more services would lead to a higher likelihood of diagnosis. 24Number of outpatient services (evaluation and management visits, 25 emergency department visits, and home health days) and inpatient service use (inpatient evaluation and management visits, intensive care unit [ICU]   stays, inpatient days, and skilled nursing facility days) were assessed 1 year before death.

End-of-Life Treatments
Based on the MedPAR file, we report place of death as occurred in a (1) hospital without an ICU stay, (2) hospital with an ICU stay, or (3) some other type of Medicare-reimbursed inpatient stay (primarily a skilled nursing facility).We classified all other decedents with no MedPAR services on the date of

Trends in ADRD Diagnosis Among Decedents
Among the 20% sample of Medicare fee-for-service decedents from 2004 to 2017, when adjusted for age and sex the percentage of decedents who received an ADRD diagnosis increased from 34.7% in 2004 to 47.2% in 2017, which equates to a 36.0%increase over the 14 years (P < .001; Figure 1 19,20 2).Decedents aged

ADRD Diagnosis by Service Setting
Trends in the service types for which an ADRD diagnosis was billed have changed since 2004 in 2 main ways.The likelihood of receiving an ADRD diagnosis has increased considerably among the inpatient, hospice, and home health settings (Figure 3A).3B).

Characteristics and Service Use Trends Among Decedents
In comparison with decedents without ADRD, the changes in demographics and comorbidities over time for the decedents with ADRD were very similar (Table 1).The age, sex, and race and ethnicity changes were small among decedents with ADRD and paralleled the changes in the non-ADRD group.
The percentage-point change in each comorbidity was the same for decedents with ADRD compared with decedents without ADRD, although the level of comorbidities differed between the groups in expected ways owing to the nature of ADRD.
Between 2004 and 2017, there were changes in demographic characteristics within the ADRD decedent population worth noting (Table 1 We found that the number and type of health service contacts increased for some types of care but not for the services from which most ADRD diagnosis claims are occurring.The number of outpatient evaluation and management visits, which are face-to-face visits with a clinician, increased

Intensity of Care at the End of Life
From 2004 to 2017, there was a shift toward lower ICU use during a terminal hospitalization among decedents with ADRD (Table 2).Those who died in the hospital without an ICU stay also decreased from 12.3% in 2004 to 5.4% in 2017, whereas death in the community or other non-Medicare reimbursed setting (ie, the unknown category) increased from 65.6% to 78.9% during the same time period.Hospice use increased considerably from 36.4% in 2004 to 62.7% in 2017.Changes from 2004 to 2017 in life-prolonging treatment in the last 6 months of life were mixed: use of ventilation increased from 6.8% to 7.6%, whereas the use of feeding tubes decreased from 5.7% to 3.7%.

Discussion
This study documents that ADRD diagnosis at the end of life is common and has increased from 2004 to 2017 among Medicare beneficiaries such that nearly half of fee-for-service Medicare beneficiaries die with a documented ADRD diagnosis.This is despite few changes in the characteristics of decedents with an ADRD diagnosis relative to the changes in the non-ADRD population, suggesting greater identification of previously undiagnosed cases.Most of the increase  recognition of dementia in the last years of life, the use of life-prolonging services such as mechanical ventilation and dialysis have not declined even while more people with dementia are receiving hospice and fewer dying in hospitals or with feeding tubes.

JAMA Health Forum | Original Investigation
To our knowledge, these analyses are among the first to examine trends in ADRD diagnosis among Medicare decedents.Prior cross-sectional studies estimate 40% of decedents with ADRD using Medicare claims, 13 37.7% with ADRD using claims and survey report, 27 and 21.7% with dementia and 43.1% with cognitive impairment without dementia using cognitive testing. 28We find higher rates-47.2% of decedents with ADRD when using the standard single-claim algorithm and still 39.2% when we increase stringency by requiring 2 claims.The high and increasing rate of ADRD diagnosis, particularly for services often used in late-stage disease (ie, hospice, home health, hospital) and without the identified ADRD population characteristics changing differentially from the general population suggest that rates of underdiagnosis for late-stage disease may be declining.That said, identification within the last 2 years of life is a minor improvement considering the disease course for ADRD spans 10 years. 29e potential benefit of greater disease identification in late-stage disease is the opportunity to engage in end-of-life care discussions.The present results do not support an association between higher ADRD diagnosis frequency and less aggressive end-of-life care.[34] From a health policy perspective, the implications of changes in ADRD diagnosis leading up to death extend beyond the implications related to quality of care.Diagnoses of ADRD were added to  35 This study demonstrates that coding was increasing in fee-forservice (which is not subject to HCC adjustment) prior to this announcement.Further studies need to address the degree that these secular trends, along with new incentives to code ADRD by health plans, will have on costs to Medicare.Medicare claims have recently been shown to have a positive predictive value between 60% and 70%, which may make it challenging to disentangle upcoding, correction of historical underdiagnosis, and misdiagnosis. 36Additionally, 2 recent articles 36,37 and the present analysis demonstrate that underdiagnosis may no longer warrant using 1 claim as opposed to the standard approach of using 2 in ADRD-diagnostic criteria.Thus, changes in ADRD diagnosis over time among Medicare decedents could affect payment and comparison of quality-ofcare outcomes.Changes in the identification of ADRD at the end of life also likely affects both epidemiological measures of disease burden (eg, years of potential life lost) and estimates of the cost burden of ADRD to the US economy.

Limitations
Several limitations must be acknowledged.First, while the present analyses demonstrate changes in ADRD diagnosis at the end of life among Medicare fee-for-service beneficiaries, these results may not extend to other groups (eg, beneficiaries enrolled in Medicare Advantage plans) or early stage disease.Second, inaccuracy of ADRD diagnoses in clinical practice and by extension in claims data likely leads to misclassification; nevertheless, these are the data used in most policy studies.Third, as an observational study we cannot rule out potential residual confounding affecting the results.Lastly, the analyses were not designed to determine whether the changes we observed are related to underlying population disease prevalence or lead to differences in clinical outcomes.

Conclusions
In summary, in this serial cross-sectional study of Medicare decedents, the number of decedents dying with an ADRD diagnosis in the US increased by 36.0% between 2004 and 2017.Inpatient, hospice, and home health settings had the most statistically significant increases in ADRD diagnosis documentation.Individual characteristics and use of services, however, were relatively stable over time.These findings indicate that diagnosis of ADRD has become more common among older US decedents.Although these analyses are unable to fully explain such increases, it may be owing to increased awareness among patients, families, and/or health care professionals, as well as temporal changes in billing practices.
An established list of International Classification of Diseases, Ninth Revision, Clinical Modification and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes were used (eTable in the Supplement).

Figure 1 .
Figure 1.Age-and Sex-Adjusted Trends in the Percentage of Older Adult Medicare Decedents With an ADRD Diagnosis Within the Last 2 Years of Life, 2004-2017

Figure 3 .B
Figure 3. Trends in Older Adult Medicare Decedents Who Received an ADRD Diagnosis by Health Care Service and Single Health Care Service in the Last 2 Years of Life, 2004-2017 26ends in US Medicare Decedents' Diagnosis ofDementia From 2004to 2017 owing to limits of Medicare data, but they were most likely to have died in a community setting or long-term care nursing home.Any claim for hospice services in the last 6 months of life was used to determine the percentage of decedents who used hospice.Lastly, lifeprolonging treatment used in the last 6 months of life included gastrostomy feeding tubes, mechanical ventilation, and hemodialysis.26Overalltrends from 2004 to 2017 in the percentage of decedents who received an ADRD diagnosis were age-and sex-adjusted using direct rate adjustment (with 2011 as the reference year).We used univariate statistical tests to compare the characteristics of decedents with an ADRD diagnosis over time (2004 vs 2017) and with decedents who did not receive an ADRD diagnosis in 2017.Statistical significance for trends in ADRD diagnosis across time was assessed by linear regression using calendar year as a continuous variable.All analyses were completed using SAS, version 9.4 (SAS Institute Inc).We set the critical α level to a 2-sided .05.Owing to the large sample size, even small differences can reach statistical significance.

Table 1 .
Trends in US Medicare Decedents' Diagnosis of Dementia From 2004 to 2017 in ADRD diagnoses occurred in 2011 through 2013 when there was a change in the number of diagnoses allowed on Medicare claims.Of note, the transition from International Classification of Diseases, Ninth Revision, Clinical Modification to International Statistical Classification of Diseases and Related Health Problems, Tenth Revision diagnostic codes in 2015 was not associated with much change in frequency of ADRD diagnosis.There is some evidence that the settings in which ADRD was diagnosed have changed.In more recent years, decedents were more likely to receive an ADRD diagnosis from an inpatient, hospice, or home health service.As a corollary, we found a drop in diagnosis only in professional services claims, decreasing from 40.6% to 15.3%.Despite greater Characteristics of Medicare Decedents With ADRD Diagnosis vs Those Without a Diagnosis in 2004, 2011, and 2017 Test was used for continuous measures and χ 2 test for proportions.Race and ethnicity was categorized as Black vs non-Black because the percentage of Black and non-Black individuals was stable over time.Non-Black race and ethnicity includes American Indian or Alaska Native, Asian, Hispanic, White, and those of other races and ethnicities.

Table 2 .
Use of Health Care Services in the Last Year of Life Among Medicare Decedents With an ADRD Diagnosis in 2004, 2011, and 2017 Trends in US Medicare Decedents' Diagnosis of Dementia From 2004 to 2017 the HCC used for risk adjustment by the Centers for Medicare & Medicaid in 2020, though initially announced in December 2018.