Health care utilization before and after a dementia diagnosis in Medicare Advantage vs Traditional Medicare

Abstract

Background:

Half of all Medicare beneficiaries are enrolled in Medicare Advantage (MA). Many studies document lower care utilization and mortality in MA than traditional Medicare (TM), but evidence for persons with Alzheimer’s disease and related dementias (ADRD) is limited.

Methods:

We conducted a retrospective cohort study of 2015–2018 Medicare claims and encounter data for community-dwelling beneficiaries ages 65 and over in TM and MA with an incident ADRD diagnosis in 2017. We compared monthly hospitalization rates and outpatient visits 12-months before and after diagnosis and mortality one-year from diagnosis. Models adjusted for socio-demographic characteristics and co-morbidities. Sensitivity analyses addressed residual confounding using a control group with incident arthritis/glaucoma or excluding MA-Special Needs Plans, and potential underreporting by restricting to MA plans with high data completeness.

Results:

Among 454,508 beneficiaries diagnosed with ADRD in 2017, 250,837 (55%) were in TM and 203,671 (45%) in MA. Four to 12 months before diagnosis, monthly hospitalizations and outpatient visits were similar in TM and MA. In the diagnosis month, 36.5% of beneficiaries in TM had a hospitalization compared to 25.4% in MA, an adjusted difference of 10.7 percentage points [95% CI: 10.3, 11.1]. Beneficiaries in TM averaged 10.5 outpatient visits in the diagnosis month compared to 8.4 in MA, an adjusted difference of 1.59 visits [95% CI: 1.47–1.70]. Utilization differences narrowed but remained higher in TM for many months. One-year mortality was 27.9% in TM and 22.2% in MA; an adjusted odds ratio of 1.152 [95% CI: 1.135–1.169] for those in TM compared to MA. Controlling for hospitalization in the diagnosis month substantially reduced the mortality difference.

Conclusion:

Hospitalization rates and outpatient visits increased more after an ADRD diagnosis in TM than MA. One-year post-diagnosis, mortality was not higher in MA than TM but comparisons of quality of life and caregiver burden are needed.

INTRODUCTION

In 2023, 50% of Medicare beneficiaries, more than 30 million people, were enrolled in Medicare Advantage (MA) plans, up from 13% in 2005.^1,2 Unlike traditional Medicare (TM), MA plans, which receive capitated per member per month payments from the federal government, have financial incentives to manage the cost and quality of care and encourage preventive care. Many studies suggest that the care provided by MA may be more efficient.³ Use of emergency department care and high-cost imaging are 20% to 30% lower in MA than TM,⁴ and appropriate ambulatory care, such as breast cancer screenings and cholesterol testing, are more common in MA.⁵ In support of better dementia detection, beneficiaries in MA are more likely to report having had structured cognitive assessments than those in TM.⁶

Despite considerable evidence that care patterns differ in MA and TM,^7,8 we know relatively little about differences in care for persons with Alzheimer’s disease and related dementias (ADRD). Analysis of the Medicare Current Beneficiary Survey, a nationally representative survey of the Medicare population, found lower health care use for persons with dementia in MA compared to TM but self-reported use and diagnosis data are subject to recall bias and even more so for persons with dementia.⁹ Studies demonstrated high rates of potentially avoidable hospitalizations (PAH) prior to an ADRD diagnosis for beneficiaries in TM and higher PAH spending after an ADRD diagnosis.^10,11 Multiple studies also demonstrated health care spending spikes leading up to and remains elevated after a diagnosis of ADRD or Mild Cognitive Impairment (MCI) for beneficiaries in TM.^12–14 It is unclear, however, whether care patterns, spending or health outcomes differ for persons with dementia in TM and MA. Projections of continued growth in the number of older persons in MA plans and who are disproportionately from underrepresented minority populations at higher risk of dementia, underscore the urgent need to fill these scientific gaps. The aim of this study was to quantify differences in care – hospitalization rates and outpatient visits – in the months before, during and after an ADRD diagnosis for beneficiaries in TM versus MA and to measure subsequent differences in one-year mortality following diagnosis.

METHODS

Data and Sample

This retrospective cohort study used claims data from 100% of community-dwelling Medicare beneficiaries in traditional Medicare (TM) enrolled in Medicare Parts A, B and D, and Medicare Encounter data on all beneficiaries enrolled in Medicare Advantage plans (MA) between 2015 and 2018. Medicare enrollment information, including MA plan enrollment, was obtained from the Master Beneficiary Summary File.

Our sample included beneficiaries ages 65 and over who were diagnosed with ADRD in 2017 and were continuously enrolled in Medicare Part D and either TM or MA in 2015 through 2018 or died in 2017 or 2018 and were living in the community. We employed a rigorous technique for measuring diagnosed ADRD in claims that: (1) uses codes for dementia or mild cognitive impairment from inpatient, outpatient, home health care, skilled nursing facility, and carrier claims and the International Classification of Diseases, Ninth and Tenth Revisions (ICD-9-CM, ICD-10-M) diagnoses codes for dementia (Supplemental Table S1); (2) requires a second diagnosis or death in 1 year to exclude potential rule-out diagnoses.^15,16,17 Month of diagnosis was based on the first claim with the documented diagnosis. While our main analytic sample is drawn from 100% of TM claims and MA encounter data, we conducted sensitivity analyses that restrict the MA data to contracts demonstrated to have highly complete data.^18,19 This approach, which has been used previously to compare utilization in MA and TM, addresses concerns about encounter data quality.^19,20

Outcomes

Hospitalizations were identified in the inpatient files and outpatient visits from the carrier (physician) and outpatient files. Death within 1-year of diagnosis was identified in the Master Beneficiary Summary File.

Study Design

We analyzed the trajectory of hospitalization rates and the average number of outpatient visits in each of the 12-months before and after an incident ADRD diagnosis for those in TM versus MA. We also compared mortality rates at 1-year after an ADRD diagnosis for those in TM versus MA and relative to mortality rates at 1-year after an arthritis or glaucoma (but no dementia) diagnosis for persons in TM relative to MA. The control group was introduced to help address potentially unobserved differences in the underlying health of those in TM versus MA. We used diagnosis codes and CCW algorithms to identify beneficiaries with arthritis or glaucoma. This strategy of using persons with incident arthritis or glaucoma has been used previously to address unobserved confounding for Medicare beneficiaries with ADRD.²¹

Statistical Analysis

We used an event study framework and flexible, nonparametric ordinary least squares (OLS) regression models to study hospitalization rates and average outpatient visit counts by month before, during and after an ADRD diagnosis for individuals in TM relative to MA. Specifically, we estimated OLS regressions of the share of beneficiaries who were hospitalized and, separately, the average number of outpatient visits for each month relative to the diagnosis date adjusted for health and socioeconomic characteristics, indicator variables spanning 11 months prior to up through 12 months after a dementia diagnosis, an indicator equal to 1 if the beneficiary was enrolled in TM and 0 if enrolled in MA, and interactions between the month relative to diagnosis indicators and the TM indicator. The indicator for 12^th month prior to diagnosis was excluded such that the other monthly indicators captured changes relative to this initial month. The models adjusted for the following demographic, socioeconomic and health differences in the distribution of beneficiaries in TM compared to MA, measured monthly: age (3-year age bins from 67–90 and then a bin for ages 91+), sex and race/ethnicity (White, Black, Hispanic, Asian, American Indian or Alaska Native, Missing/Other), dual eligibility status, enrollment in Part D low-income subsidy and diagnosis with comorbid conditions of diabetes, hypertension, hyperlipidemia, stroke, acute myocardial infarction, and atrial fibrillation. Comorbid conditions were identified using diagnosis codes and CCW algorithms in each year. We also adjusted for calendar year (2016, 2017 or 2018) to control for annual trends in utilization and calendar month of diagnosis (Jan-Dec) to control for seasonal effects. Estimates were weighted by the monthly number of beneficiaries in TM or MA. The key coefficients of interest were from the interactions between the TM indicator and the month relative to diagnosis indicators; they captured the change in each outcome for beneficiaries in TM relative to MA for each month before and after diagnosis relative to the 12^th month prior to diagnosis.

We estimated logistic regression models of the probability of death within one-year of diagnosis for persons with dementia in TM relative to MA with adjustments for the health and socio-demographic characteristics of beneficiaries as described above. In addition to estimating the difference between TM and MA, we estimated logistic models of one-year mortality post diagnosis of persons with ADRD in TM compared to MA relative to a control group of persons post diagnosis with arthritis or glaucoma (and not dementia) in TM compared to MA.

While some work suggests favorable selection into MA has declined considerably,^22,23 concerns about residual confounding may remain. We performed numerous sensitivity analyses to address concerns that differential morbidity or mortality of those in MA versus TM could bias our results. First, we differenced out care and mortality patterns for a control group of persons diagnosed with arthritis or glaucoma in TM compared to MA. Second, we tested the sensitivity or our results to the exclusion of beneficiaries enrolled in MA Special Needs Plans (SNPs), which in 2018 accounted for much of the observable differences between TM and MA.²⁴ Finally, we tested the sensitivity of the mortality results to endogenous plan switching, relaxing the restriction on continuous enrollment and including plan switchers in the analysis. We assigned “plan switchers” to TM or MA based on their enrollment status in 2017.

This study complied with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for cohort studies (STROBE Statement S1). This research was approved by the Institutional Review Board of the University of Southern California, which waived the requirement for informed consent because there was no direct contact with patients and only deidentified claims data were used in the analyses. All analysis was conducted using SAS Enterprise Guide 7.1.

RESULTS

Among 454,508 beneficiaries with an incident ADRD diagnosis in 2017, 250,837 (55%) were in TM and 203,671 (45%) were in MA (Table 1). Females accounted for 151,737 (60.5%) beneficiaries in TM and 119,226 (58.5%) in MA. 204,253 (81.4%) of beneficiaries with an incident dementia diagnosis were non-Hispanic white, 20,190 (8%) non-Hispanic Black and 14,498 (5.8%) Hispanic in TM compared to 134,513 (66%) white, 24,335 (11.9%) Black, and 35,033 (17.2%) Hispanic in MA. The mean (SD) age at diagnosis was 81.5 (7.61) years in TM and 80.8 (7.35) in MA. 47,952 (19.1%) beneficiaries were dual eligible in TM compared to 33,768 (16.6%) in MA. Beneficiaries with an ADRD diagnosis in TM relative to MA had lower rates of diabetes (35.5% vs. 38.4%) and hyperlipidemia (56.9% vs. 60.7%) but higher rates of stroke (6.7% vs. 5.7%) and atrial fibrillation (15.7% vs. 12.3%). Beneficiaries with incident ADRD diagnoses represent 2.5% of all beneficiaries in TM and 2.2% of those in Medicare Advantage (Table S2). Beneficiaries with an incident arthritis/glaucoma diagnosis were slightly younger (mean age of 75.35 in TM vs. 75.09 in MA) but racial/ethnic composition and co-morbid conditions between those in TM versus MA were similar to that for the ADRD sample (Table S2).

Table 1.

Characteristics of Beneficiaries Diagnosed with ADRD in 2017 in Traditional Medicare vs. Medicare Advantage

	Traditional Medicare	Medicare Advantage
Total Sample	250,837	203,671
Female N(%)	151,737 (60.5)	119,226 (58.5)
Race/Ethnicity N(%)
White	204,253 (81.4)	134,513 (66.0)
Black	20,190 (8.0)	24,335 (11.9)
Hispanic	14,498 (5.8)	35,033 (17.2)
Asian	8,062 (3.2)	7,090 (3.5)
American Indian/Alaska Native	971 (0.4)	335 (0.2)
Missing/Other	2,863 (1.1)	2,365 (1.2)
Age at Dx
Mean (sd)	81.45 (7.61)	80.75 (7.35)
<75	57,881 (23.1)	51,287 (25.2)
75-84	108,129 (43.1)	92,108 (45.2)
85+	84,827 (33.8)	60,276 (29.6)
Income-related Assistance
Dual	47,952 (19.1)	33,768 (16.6)
LIS	8,926 (3.6)	8,462 (4.2)
Comorbid Conditions N (%)
Diabetes	88,959 (35.5)	78,169 (38.4)
Hypertension	184,173 (73.4)	149,282 (73.3)
Hyperlipidemia	142,799 (56.9)	123,546 (60.7)
Stroke	16,875 (6.7)	11,567 (5.7)
AMI	2,882 (1.1)	1,945 (1.0)
Atrial fibrillation	39,429 (15.7)	25,035 (12.3)
Health Care Utilization in the 12th Month Prior to Diagnosis
Inpatient Visits N (%)	6,688 (2.7)	3,801 (1.9)
Avg. # of Monthly Outpatient Visits	2.89 (3.97)	2.33 (3.57)

Notes: Sample is continuously enrolled in either TM and Part D or MA and Part D in 2015 through 2018 or death in 2017 or 2018. Incident ADRD diagnosis is measured in 2017, and is defined as no diagnoses in 2015 or 2016, verified by a second diagnosis or death in one year.

Unadjusted Differences in Hospitalizations and Outpatient Visits

In the 12^th month prior to an ADRD diagnosis, 6,688 (2.7%) beneficiaries in TM had an inpatient stay versus 3,801(1.9%) in MA (Table 1). In the 6 months preceding an ADRD diagnosis, monthly hospitalization increased from less than 3% in both groups to about 6% in MA and 9% in TM in the month before diagnosis (Figure 1). Hospitalization rates spiked in the month of diagnosis (time 0) to 36.5% in TM and 25% in MA. While hospitalization rates declined precipitously thereafter, they remained elevated for several months after diagnosis.

Figure 1. Trends in Health Care Utilization in the 12 Months Before and After an ADRD Diagnosis for Beneficiaries Enrolled in Traditional Medicare and Medicare Advantage.

Hospitalization rates (A) and average outpatient visit counts (B) by month relative to the month of an ADRD diagnosis for beneficiaries continuously enrolled in traditional Medicare (TM) or Medicare Advantage (MA). Based on Medicare claims and Encounter data.

The average number (s.d.) of outpatient visits in the 12^th month prior to an ADRD diagnosis, was 2.89 (3.97) in TM and 2.33 (3.57) in MA (Table 1). Monthly outpatient visits evolved in a similar way to hospitalization rates, increasing from less than 3 visits per month in both TM and MA to 3 to 5 visits per month in the 1 to 2 months prior to diagnosis and sharply increasing to 10.6 visits in TM and 8.4 visits in MA in the month of diagnosis. The number of visits declined precipitously thereafter for beneficiaries in both TM and MA but remained elevated for several months after diagnosis.

Regression Model of Hospitalizations and Outpatient Visits

Based on event study models adjusted for socio-demographic characteristics and co-morbid conditions, hospitalization rates were statistically indistinguishable in TM and MA eleven to four months prior to diagnosis relative to 12 months prior. In the 3, 2 and 1 months prior to diagnosis, hospitalization rates were 0.7 (p<0.01), 1.3 (p<0.01), and 2.4 (p<0.01) percentage points higher in TM relative to MA. (Figure 2 and Table S3). During the month of diagnosis, hospitalization rates were 10.7 (p<0.01) percentage points higher. Rates remained higher in TM relative to MA in the 12 months following diagnosis, although the difference narrowed over time from 3.7 (p<0.01) percentage points in first month after diagnosis to 0.8 (p<0.01) in the 12^th month after diagnosis.

Figure 2. Changes in Health Care Before and After an ADRD Diagnosis for Beneficiaries in Traditional Medicare and Medicare Advantage relative to 12 Months Prior to Diagnosis.

Hospitalization rates (A) and average outpatient visit counts (B) by month relative to the month of an ADRD diagnosis (month 0) for beneficiaries continuously enrolled in traditional Medicare relative to Medicare Advantage. All models adjust for age, race and sex (Age) and sequentially add controls for dual eligibility, low-income subsidy status, comorbid conditions (Age, SES, Health), and month and year indicators (Age, SES, Health, Month & Year FE). Based on Medicare claims and Encounter data. Dotted lines indicate 95% confidence intervals from the “Age, SES, Health, Month & Year FE” models.

A similar pattern was found for average outpatient visit counts. Eleven to three months prior to diagnosis, outpatient visits were statistically indistinguishable in TM and MA relative to 12 months prior. Three, 2 and 1 month prior to diagnosis, beneficiaries in TM had 0.18 (p<0.01), 0.34 (p<0.01), 0.74 (p<0.01) more outpatient visits in TM relative to MA. During the month of diagnosis, beneficiaries in TM have 1.6 (p<0.01) more outpatient visits. Six months to a year later, however, the outpatient visit differential goes to zero (Figure 2).

Logistic Models of One-Year Mortality Post Dementia Diagnosis

The mortality rate one year after an ADRD diagnosis was high for both groups – 27.9% in TM and 22.2% in MA. The adjusted odds ratio of death within one year was also higher for those in TM relative to MA, OR 1.229 (95% CI: 1.212–1.247) (Table 2). Controlling for diagnosis in the hospital, a proxy for diagnosis in a sicker state, modestly reduced the odds ratio, OR 1.152 (95% CI: 1.135–1.169). Including a control group of persons diagnosed with arthritis or glaucoma yielded a similar overall odds ratio of death in TM relative to MA. Specifically, the adjusted odds ratio of death within one year was of similar magnitude to the ADRD-only analysis - OR 1.155 [95% CI: 1.136–1.174] for those in TM relative to MA. The adjusted odds ratio of mortality for those with dementia in TM relative to MA was also higher relative to the odds ratio of mortality for the control group of beneficiaries diagnosed with arthritis or glaucoma in TM relative to MA (OR 1.058 [95% CI: 1.035–1.081]). Controlling for diagnosis in the hospital further reduced the magnitude and precision of the ratio of odds ratios of one-year mortality, OR 1.021 (95% CI: 0.999–1.044).

Table 2.

Odds Ratios for 1-Year Mortality for Individuals Diagnosed with Dementia

	TM Relative to MA		DD Model: TM Relative to MA and Relative to Arthritis/Glaucoma
	Main controls	Control for Dx in Hospital	Main Controls	Control for Dx in Hospital
Beneficiaries	454,508	454,508	2,134,665	2,134,665
	1.229***	1.152***	1.155***	1.117***
TM	(1.212-1.247)	(1.135-1.169)	(1.136-1.174)	(1.098-1.136)
			6.040***	5.490***
PLWD			(5.938-6.144)	(5.396-5.586)
			1.058***	1.021
TM*PLWD			(1.035-1.081)	(0.999-1.044)

Notes: The one-year mortality rate was 27.9% in TM and 22.2% in MA. The table shows differential odds of death within one-year in TM relative to MA based on logistic regression models. All models adjust for age in 3-year age bins from 67-90 and then a bin for ages 91+, sex, race/ethnicity (Black, Hispanic, Asian, American Indian or Alaska Native, Missing/Other), dual eligibility status, enrollment in Part D low-income subsidy and diagnosis with the following comorbid conditions: diabetes, hypertension, hyperlipidemia, stroke, acute myocardial infarction, and atrial fibrillation.

Sensitivity Analyses

To address concerns about residual confounding, we first analyzed care patterns for the control group of persons diagnosed with arthritis or glaucoma in TM relative to MA (Figure S1 shows care patterns separately for TM and MA and Figure S2 shows the difference between them). Differencing out monthly inpatient stays or the average outpatient visit counts for this control group only modestly narrowed the gap in care patterns (Figure S3).

Second, we excluded individuals enrolled in MA-SNPs, whose members contributed to much of the observable differences between beneficiaries in TM and MA in 2018.²⁴ Excluding beneficiaries in MA-SNPs had no meaningful impact on either the gap in care patterns (Figure S4) or mortality (Table S4).

Third, we relaxed the continuous enrollment requirement. Including plan switchers in the mortality analysis yielded similar results: OR 1.064 (95% CI: 1.042–1.087) for the ratio of odds ratios of mortality for those with dementia in TM relative to MA and relative to the control group of those with arthritis or glaucoma (Table S5). Controlling for hospitalization in the month of diagnosis reduced the magnitude of the ratio of odds ratios but did not eliminate it, OR 1.024 (95% CI: 1.003–1.046).

We also sought to address concerns about the completeness of MA data, which risks artificially contributing to lower health care utilization in MA relative to TM. Restricting our analyses to MA contracts demonstrated to have highly complete data. only modestly narrowed the differences in utilization in MA and TM ( Figure S5). During the month of diagnosis, those in TM had a 9.79 (p<0.01) percentage point higher hospitalization rate and 1.41 higher outpatient visits than beneficiaries in MA plans with complete contracts (Table S6). Restricting to complete contract data plans had no impact on the mortality results (Table S7), suggesting that beneficiaries in excluded contracts are not selected on health status.

DISCUSSION

This study is the first to use administrative Medicare claims and Encounter data and 100% of Medicare beneficiaries to quantify trends in health care utilization and one-year mortality in community-dwelling beneficiaries diagnosed with ADRD in TM and MA. The estimates address selection into plans with model adjustments for observable differences in the socio-demographic and health of beneficiaries diagnosed with ADRD in TM and MA. The role of unobserved differences in mortality was further assessed by including a control group with incident arthritis/glaucoma, controlling for diagnosis in the hospital and estimating models that include plan switchers.

We found higher hospitalization for beneficiaries in TM compared to MA in the three months leading up to a dementia diagnosis – about 2.5 percentage points higher in the month before and 11 percentage points higher in the month of diagnosis. Hospitalization rates remained higher in TM for several months thereafter, although the difference narrowed over time. The number of outpatient visits was also higher in the months just before and after a dementia diagnosis among beneficiaries in TM compared to MA, peaking at 1.6 more visits in the month of ADRD diagnosis and declining thereafter.

The odds of dying within one year of an ADRD diagnosis was 1.23 times higher for those in TM compared to those in MA. Much of this difference likely reflected health factors that were not accounted for in our simplest adjusted models. However, a sizeable difference in mortality remained even after introducing a control group of persons with incident arthritis or glaucoma in TM and MA, including plan switchers, excluding those in MA-SNPs, or restricting to those in complete MA contract plans.

The patterns of lower health care use in MA relative to TM may reflect more restricted access to care for beneficiaries in MA. However, we did not find evidence that this led to worse outcomes – that is we found lower mortality after a dementia diagnosis for beneficiaries in MA compared to TM. A recent review article of studies that compared health care use in MA to TM reported fewer hospital admissions but more outpatient visits.³ Results on mortality revealed both higher, lower and no difference. The systematic review revealed knowledge gaps. Most studies examined average effects with few studies on the effects for beneficiaries with specific conditions and none for persons with dementia. With no therapeutics to prevent and slow dementia progression, quality of care for persons living with dementia will continue to be a focus among policymakers, clinicians and persons living with the disease and their families.

Limitations

This study has limitations. First, claims data provide an incomplete record of health and likely do not fully account for health status before, during or after a dementia diagnosis. Since providers in TM rely on claims for payment, this strengthens the record of health for those in TM relative to MA. However, capitated plan payments in MA are adjusted for health status thus MA plans typically do more screening to diagnose underlying health conditions.²⁵ Our analysis adjusted for observable differences between those in TM relative to MA but unobserved differences, including in the timing and severity of diagnosis that prior work has shown to bias costs of care estimates for those with versus without ADRD,²⁶ may remain. For these unobserved differences to bias the reported effects they would need to be different for beneficiaries in TM and MA and in ways that affect health care use and outcomes. Sensitivity analyses that addressed residual confounding by introducing a control group of persons diagnosed with arthritis or glaucoma in TM relative to MA or by excluding MA-SNPs had little impact on either the gap in care patterns or mortality in TM relative to MA. Together these analyses suggest that unobservable characteristics are unlikely to explain the full pattern of utilization or mortality results.

Second, and related to confounding, we do not explicitly measure the severity of disease at diagnosis. If, for example, individuals in TM are diagnosed later in the disease trajectory, this could contribute to the higher rate of health care utilization and mortality after an ADRD diagnosis in TM relative to MA, meaning we overstate the gaps attributable to TM. We attempted to proxy for disease severity in our mortality analysis by controlling for hospitalization during the month of diagnosis. As expected, this control substantially narrowed but did not eliminate the higher odds of one-year mortality in TM relative to MA.

Third, the MA encounter data files are relatively new to researchers and have not been subject to the same scrutiny as TM data. Analysis by the Medicare Payment Advisory Commission shows that MA encounter data quality has improved over time.²⁷ We have assessed and found high quality of dementia diagnosis data in Encounter data and prevalence and incidence rates similar to those in claims data after adjustments for demographic and health differences. Moreover, sizeable gaps in utilization remained even after we restricted the analysis to MA plans with complete contracts, an approach recommended to address MA data quality.^18,19 This restriction also had virtually no impact on our mortality results.

Fourth, beneficiaries may switch between TM and MA in response to a dementia diagnosis. We tested robustness of estimates to endogenous switching of health plans and estimated models that did not require continuous enrollment in MA or TM. Mortality results that included plan switchers were similar to our main results (Table S4).

Finally, while we analyzed differences in hospitalization, we did not specifically assess measures of preventable hospitalizations. This is an important area for future research.

Conclusions

MA plan enrollment continues to grow but our understanding of care patterns in MA remains quite limited, particularly relative to TM and for persons with ADRD. While TM claims still provide a good picture of care for a large segment of the Medicare population and offer researchers the advantage of decades of data, these data are no longer sufficient for characterizing the care experiences of all Medicare beneficiaries. Differences in the benefits available in MA and TM are growing, particularly given dramatic growth in MA Special Needs Plans (MA-SNPs).²⁸ Differences in care may be larger for individuals enrolled in MA-SNPs compared to those in TM. Moreover, the 2018 Creating High-Quality Results and Outcomes Necessary to Improve Chronic Care (CHRONIC) Act removed benefit uniformity requirements between TM and MA and, beginning in 2020, allows MA plans to offer Special Supplemental Benefits for the Chronically ill (SSBCI) or “non-medical” disease-specific benefits such as long-term services and supports.²⁹ While initial data suggest MA plans offered limited nonmedical benefits,³⁰ their offerings are likely to grow over time.

Our work finds that hospitalization rates and outpatient visits increase prior to an ADRD diagnosis for beneficiaries in both plan types but more so for those in TM than MA. The spike in care before an ADRD diagnosis suggests there may be opportunities to improve identification of mild cognitive impairment and dementia at earlier stages in both Medicare Advantage and traditional Medicare. As measured by mortality, outcomes appear if anything worse in TM relative to MA. A possible explanation, although one for which we have no direct evidence, is that care is better managed in MA plans. Evidence is needed, however, on other patient- and caregiver-centered outcomes such as quality of life, care satisfaction and caregiver burden for beneficiaries with an ADRD diagnosis in TM relative to MA. Given the dearth of effective treatments for patients with ADRD, well-managed care may be critical to maintaining patient quality of life and delaying institutionalization or death for individuals with ADRD.

Supplementary Material

Supinfo

Supplementary Text S1.

Figure S1. Trends in Health Care Utilization in the 12 Months Before and After an Arthritis/Glaucoma Diagnosis for Beneficiaries Enrolled in Traditional Medicare and Medicare Advantage.

Figure S2. Changes in Health Care Before and After an Arthritis/Glaucoma Diagnosis for Beneficiaries in Traditional Medicare and Medicare Advantage relative to 12 Months Prior to Diagnosis

Figure S3. Changes in Health Care Before and After an ADRD Diagnosis relative to Before and After an Arthritis/Glaucoma Diagnosis for Beneficiaries in Traditional Medicare and Medicare Advantage and relative to 12 Months Prior to Diagnosis

Figure S4. Changes in Health Care Before and After an ADRD Diagnosis for Beneficiaries in Traditional Medicare and Medicare Advantage Plans other than Special Needs Plans and relative to 12 Months Prior to Diagnosis

Figure S5. Changes in Health Care Before and After an ADRD Diagnosis for Beneficiaries in Traditional Medicare and Complete Contract Medicare Advantage Plans and relative to 12 Months Prior to Diagnosis

Table S1. ICD-9-CM and ICD-10-M Codes Used to Identify Dementia

Table S2. Characteristics of All Beneficiaries and Beneficiaries Diagnosed with Arthritis or Glaucoma in 2017 in Traditional Medicare vs. Medicare Advantage

Table S3. Event Study Model Estimates for Incident Dementia Diagnosis

Table S4. Differential Odds of 1-Year Mortality for Individuals Diagnosed with Dementia Excluding Beneficiaries in MA-SNPs

Table S5. Differential Odds of 1-Year Mortality for Individuals Diagnosed with Dementia Including MA/FFS Switchers

Table S6. Event Study Model Estimates for Incident Dementia Diagnosis Restring to Plans with Complete Contracts

Table S7. Odds of 1-Year Mortality for Individuals Diagnosed with Dementia Restricting to Plans with Complete Contracts

STROBE Statement S1. Observational studies checklist

Key Points:

• Health care use and hospitalizations spike leading up to a dementia diagnosis in both traditional Medicare and Medicare Advantage.

• The increase in care use leading up to a dementia diagnosis is higher among beneficiaries in traditional Medicare than in Advantage plans and remains higher after diagnosis.

• Lower health care use among Medicare Advantage beneficiaries is not associated with higher one-year mortality after a dementia diagnosis.

Why does this paper matter?

Half of all Medicare beneficiaries are now enrolled in Medicare Advantage plans but we know relatively little about differences in care for persons with dementia in Medicare Advantage versus Traditional Medicare despite differences in benefits and provider networks.