Effects of Medicare Advantage on preventive care use and health behavior

Abstract

Objective

To examine the effects of Medicare Advantage (MA) enrollment on preventive care use and health behavior.

Data Sources

The Medicare Current Beneficiary Survey, the Area Health Resources File, the Geographic Variation Public Use File, and the Centers for Medicare and Medicaid Services annual risk and ratebook files for 2012–2016.

Study Design

Outcomes included 11 measures of preventive care use and six measures of health behavior. My primary independent variable was MA enrollment. For each outcome, I first conducted linear regression analysis while adjusting for individual‐level and county‐level characteristics. Then, I conducted the following alternative analyses to account for differences in observed and/or unobserved characteristics between MA and traditional Medicare (TM) enrollees: propensity score (PS) matching analysis and instrumental variable (IV) analysis.

Data Collection/Extraction Methods

I extracted 9399 MA enrollees and 15,543 TM enrollees.

Findings

Linear regression and PS matching analyses showed that MA enrollment was statistically significantly associated with higher likelihood of having blood pressure measurement, cholesterol measurement, and influenza vaccine, lower likelihood of receiving an HbA1C test, and higher likelihood of currently smoking. However, the magnitude of the associations was small. There were no statistically significant associations in other measures. IV analyses also found no or limited evidence that MA enrollment led to statistically significant changes in preventive care use and health behavior. Specifically, MA enrollment led to statistically significant improvements in the likelihood of doing any physical activities (1.29 [95% CI: 0.51–2.07]) or doing muscle‐strengthening activities (0.72 [95% CI: 0.03–1.41]). No statistically significant changes were observed in other measures.

Conclusions

MA plans may not necessarily increase the use of preventive services and improve health behaviors. As improvements in preventive services and health behavior may have the potential to achieve better outcomes while lowering costs, policy makers should consider developing targeted interventions for MA to achieve those improvements.

What is known on this topic

• Because Medicare Advantage (MA) plans are paid on a capitated basis, there may be an incentive for MA plans to provide more preventive services and encourage enrollees to improve their health behavior.

• There is mixed evidence on the effect of MA enrollment on preventive care use.

• There is limited understanding of the effect of MA enrollment on health behavior.

What this study adds

• I examined the effects of MA enrollment on preventive care use and health behavior.

• My propensity score matching analysis and instrumental variable analysis showed that MA enrollment was associated with no or limited increases in the use of clinically meaningful preventive services.

• There was limited evidence that MA enrollment was associated with improvements in health behavior.

1. INTRODUCTION

Medicare Advantage (MA) is Medicare's managed care alternative to traditional fee‐for‐service Medicare (TM) and provides coverage for inpatient and outpatient care through private health plans. MA enrollment has substantially increased over the past decade, growing from 12 million in 2011 to 26 million in 2021. ¹ Because MA plans are paid on a capitated basis, there may be an incentive for MA plans to deliver care efficiently, provide more preventive services, and encourage enrollees to improve their health behavior. Evidence suggests that MA enrollment improves the efficiency of care delivery. ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ Research has found that healthier individuals tend to enroll in MA than TM, ⁸ , ⁹ , ¹⁰ likely biasing any direct comparison between MA and TM enrollees. Thus, prior studies used an instrumental variable (IV) approach to address selective enrollment into MA and found that MA enrollment led to decreases in health care use overall. ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ However, there were no significant differences in patient satisfaction between MA and TM enrollees, especially access to care. ⁴ , ⁵ This suggests that low health care use among MA enrollees may reflect high efficiency of care rather than under‐provision of care.

MA plans may have the potential to use more preventive services and enhance health behavior in order to increase plan efficiency, but less is known about whether MA enrollment increases preventive care use and improves health behavior. First, preventive care helps detect or prevent serious diseases and medical problems before they can become major. Thus, MA plans may be incentivized to improve primary care access by ensuring a regular source of preventive care. However, there is mixed evidence. While one study found that MA plans encouraged the use of primary care by expanding networks of primary care providers, ¹¹ others found that MA plans offered narrow networks, ¹² particularly for primary care services. ¹³ There is also evidence that MA enrollees used primary care or preventive care at a higher proportion than TM enrollees, ¹⁴ , ¹⁵ but these studies did not fully account for selective enrollment into MA, which possibly led to biased estimates. In addition to preventive care, positive health behavior may also help prevent diseases and promote health. Thus, MA plans may be incentivized to keep enrollees healthy and minimize disease progression. This may be particularly relevant to MA plans as MA plan payments have been adjusted only minimally for clinical characteristics of enrollees, leading to overpayments for healthier enrollees. ⁸ , ⁹ , ¹⁰ Consequently, MA plans may provide educational programs or interventions to encourage enrollees to engage in more positive health behavior such as quitting smoking or drinking less alcohol. For example, 69% of MA enrollees had extra benefits for fitness programs and gym memberships. ¹⁶ To the best of my knowledge, however, there is limited understanding of the effect of MA enrollment on health behavior.

In this study, I examined the effects of MA enrollment on preventive care use and health behavior. Specifically, I conducted several analyses. First, I conducted linear regression analysis while adjusting for individual‐level and county‐level characteristics. Then, I conducted the following alternative analyses to account for differences in observed and/or unobserved characteristics between MA and TM enrollees: (1) propensity score (PS) matching analysis and (2) IV analysis.

2. METHODS

2.1. Data and sample

I used data from multiple sources for 2012–2016. First, I used the Medicare Current Beneficiary Survey, which offers a nationally representative sample of the Medicare population. The data combines Medicare claims and administrative data with data collected via survey. Second, I used the Area Health Resources File and Geographic Variation Public Use File to obtain county‐level demographic and socioeconomic data. Finally, I used the Centers for Medicare and Medicaid Services (CMS) annual risk and ratebook files to estimate county‐level MA benchmarks.

Using the data, I first identified Medicare enrollees (aged 65 years and older) with 12‐month continuous enrollment in MA or TM. I excluded those whose original Medicare eligibility was attributable to end‐stage renal disease or disability, those who did not have Parts A and B benefits, and those who died within the year. As MA plans are not available in some counties, I also excluded those residing in counties without an MA plan. Some beneficiaries were included in the data over the course of multiple years, but I treated the MCBS data for each year as an independent annual cross‐sectional survey.

2.2. Outcomes

I included two types of (self‐reported) outcomes from survey data. First, I included 11 measures of preventive care use. The first three were the preventive services recommended by the United States Preventive Services Task Force (USPSTF). ¹⁷ Specifically, following the USPSTF guidelines, I constructed binary variables measuring the use of the following preventive services: (1) blood pressure screening, (2) cholesterol screening, and (3) influenza vaccination. For blood pressure screening, there is lack of evidence for ideal blood pressure screening intervals. As the USPSTF recommends blood pressure screening every 1–2 years depending on blood pressure level, however, I assessed whether enrollees had the service within the past two years. The USPSTF recommends cholesterol screening every five years for older adults, and thus I assessed whether enrollees had the service within the past five years. Also, the USPSTF recommends flu vaccinations every year, and therefore I assessed whether enrollees had the services within the past year. The last eight were process measures of diabetes care. Specifically, I constructed binary variables measuring the use of the following process measures of diabetes care: (1) insulin use, (2) medication use, (3) check for sores on feet, (4) blood sugar (glucose test), (5) blood pressure check at home, (6) comprehensive foot exam, (7) hemoglobin A1C (HbA1C) test, and (8) well‐controlled blood sugar. When examining insulin use and medication use, I limited the sample to those with Medicare prescription drug coverage. Second, I included six measures of health behavior. Specifically, I constructed binary variables measuring the following health behavior: (1) currently smoking, (2) drinking alcohol, (3) drinking more than four drinks per day, (4) drinking more than four drinks ten days a month, (5) doing any physical activities (including vigorous or moderate activities), and (6) doing muscle‐strengthening activities. The sample size varied by outcomes as response rates varied by question.

2.3. Independent variables

My primary independent variable was MA enrollment. To adjust for differences in individual‐level characteristics, I included age, sex, race/ethnicity, education, household income, dual eligibility for Medicare and Medicaid, marital status, residence in a metropolitan area, self‐reported comorbidity (hardening of arteries, hypertension, myocardial infarction, stroke, cancer, rheumatoid arthritis, osteoporosis, diabetes, and Alzheimer's disease and related dementias), self‐reported health status (general health status compared with same‐age people and overall health status compared with a year ago), and activities of daily living limitations. To adjust for differences in county‐level characteristics, percentages of those older than 65 years old, percentages of White residents, percentages of Black residents, percentages of Hispanic residents, percentages of dual‐eligible Medicare beneficiaries, county‐level hierarchical condition categories risk score, median household income, and number of hospital beds, medical doctors, and primary care physicians per 1000 residents were included.

2.4. Statistical analysis

I estimated unadjusted sample characteristics and outcomes between MA and TM enrollees. I used χ² tests for categorical variables and analysis of variance for continuous variables. To estimate differences in preventive care use and health behavior between MA and TM enrollees, I conducted several analyses. First, I conducted linear regression analysis while adjusting for individual‐level and county‐level characteristics. However, there is evidence showing that healthier individuals tend to enroll in MA than TM, ⁸ , ⁹ , ¹⁰ making a direct comparison between MA and TM enrollees potentially biased. Thus, I conducted the following alternative analyses to account for differences in observed and/or unobserved characteristics between MA and TM enrollees: (1) PS matching analysis and (2) IV analysis.

For PS matching analysis, I followed prior research and computed the inverse probability of treatment weighting (IPTW) as a propensity for enrolling in MA based on the variables described. ¹⁴ , ¹⁸ Then, I estimated the difference in outcomes among MA enrollees relative to TM enrollees using a linear regression model after applying the IPTW. We examined whether IPTW‐weighted samples were balanced on sample characteristics between MA and TM enrollees. We also examined whether the propensity score had a sufficient overlap between MA and TM enrollees.

For IV analysis, I followed prior research and decomposed the MA benchmark into exogenous and endogenous components, and used the exogenous component as my instrument. ⁶ , ¹⁹ MA payments have been determined by multiple factors, but the county‐level benchmark rate is likely to be the most dominant. There is evidence that higher benchmarks for MA plans were related to greater availability of the plans, more generous benefits, and higher MA enrollment. ² , ²⁰ , ²¹ Specifically, CMS uses county‐level TM cost data to determine the benchmark rate for MA plans. The benchmark is the average cost of coverage for a TM beneficiary. However, MA plans also submit bids every year and each bid is compared to a payment area's benchmark. If a plan bids below the benchmark, it is paid the base rate plus a rebate. If a plan bids above the benchmark, it is paid the base rate and enrollees pay the difference between the bid and the benchmark in the form of a premium.

The Affordable Care Act (ACA) has made fundamental changes to the MA payment schedule. Prior to the ACA, MA benchmarks were not directly linked with TM spending, and thus the MA benchmark has been used as an exogenous instrument for MA enrollment. As the ACA has linked TM spending to benchmarks more directly, the MA benchmark is no longer a plausibly exogenous instrument for MA enrollment. Starting in 2012, MA benchmarks have been rated based on the following formula that accounts for lagged TM spending and projected TM spending for the current year, and partially relies on the rate used prior to the ACA.

$${\mathit{\text{Benchmark}}}_{\mathit{ct}}=\left\{\begin{array}{c}\left(\mathit{App}{\%}_{\mathit{ct}}\times {\mathit{\text{APCC}}}_{\mathit{ct}}\times {\omega }_{\mathit{ct}}\right)+\left(\left[1-{\omega }_{\mathit{ct}}\right]\times {\mathit{Old}}_{\mathit{ct}}\right)\mathit{\text{if}}{\mathit{Old}}_{\mathit{ct}}\ge {\mathit{\text{Benchmark}}}_{\mathit{ct}}\\ {\mathit{Old}}_{\mathit{ct}}\mathit{\text{if}}{\mathit{Old}}_{\mathit{ct}}<{\mathit{\text{Benchmark}}}_{\mathit{ct}}\end{array}\right.$$

${\mathit{\text{Benchmark}}}_{\mathit{ct}}$ is the MA benchmark for county $c$ in year $t$. $\mathit{App}\%$ is the applicable percentage obtained by county‐specific ranking on TM spending as average (adjusted) per capita cost (APCC). Each county receives an “applicable percentage” based on its quartile of lagged TM spending (115%, 107.5%, 100%, and 95% of projected contemporaneous TM spending for counties in the 1st, 2nd, 3rd, and 4th quartile of lagged TM spending). Counties that change quartiles from one year to the next receive the average of the two applicable percentages for the 1st year as a transition. The applicable percentage is then multiplied by projected contemporaneous TM spending. For counties that change quartiles, the applicable percentage creates discontinuous changes in benchmarks both between counties and within counties over time. The quartile‐based benchmarks support higher payments to MA plans in counties with lower TM spending. $\mathit{\text{APCC}}$ is the projected contemporaneous spending for TM after adjusting for basic demographic factors. $\omega$ is a weight of the old and the new benchmarks as the new benchmarks were phased in over two, four, or six years. $\mathit{Old}$ is the payment rate used prior to the ACA.

Following prior research, ⁶ , ¹⁹ I used the MA benchmark to determine the plausibly exogenous variation induced by within‐county changes in quartiles and to use this variation, which is less likely to be related to MA enrollment, to estimate the causal effects of MA enrollment. As the applicable percentage generates discontinuous changes in benchmarks both between counties and within counties over time, these discontinuities idiosyncratically change the relationship between the benchmark and APCC. In other words, the quartile‐based benchmarks can create discontinuities in payment when counties have similar TM spending but are assigned to a different payment category when the ranking of county‐level TM spending changes from year to year. As I controlled for between‐county variation in the effect of the discontinuities on the benchmark, however, my analysis relied on the discontinuous within‐county variation included in the benchmark by a given county moving to different quartiles, generating plausibly exogenous variation in MA enrollment. Using the CMS annual risk and ratebook files, I ran the following regression to decompose the MA benchmark into exogenous and endogenous components:

$${\mathit{\text{Benchmark}}}_{\mathit{ct}}=\alpha \left(\mathit{App}{\%}_{\mathit{ct}}\times {\mathit{\text{APCC}}}_{\mathit{ct}}\times {\omega }_{\mathit{ct}}\right)+\beta \left(\left[1-{\omega }_{\mathit{ct}}\right]\times {\mathit{Old}}_{\mathit{ct}}\right)+\gamma {\mathit{Cap}}_{\mathit{ct}}+\delta {\omega }_{\mathit{ct}}+\theta {\mathit{\text{APCC}}}_{\mathit{ct}}+C_t$$

$\mathit{Cap}$ is an indicator for whether the benchmark is capped and C indicates county fixed effects.

To precisely estimate the plausibly exogenous variation induced by within‐county changes in quartiles, I included county fixed effects, making it possible to remove between‐county variations in the effect of the discontinuities on the MA benchmark. I also controlled for the independent effects of either APCC or the phase‐in factor on the MA benchmark, and thus I did not require them to be independently exogenous. Then, I used the coefficient $\alpha$ on the triple interaction to estimate the exogenous component of the MA benchmark (Exog) and on the remaining coefficients to estimate the endogenous component of the MA benchmark (Endog). Finally, I used the exogenous component as my instrument.

A valid instrument must satisfy two main requirements. First, the instrument must be strongly correlated with the treatment variable, in my case, with MA enrollment. To assess whether my instrument was strong, I tested its association with MA enrollment and then examined F statistics, where a value greater than 10 traditionally indicates a strong instrument. Also, I examined whether sample characteristics were balanced across values of the instrument. Second, the instrument must not influence the outcome except through its association with the treatment variable. There are no established methods to empirically confirm this second requirement. Following prior research, ¹⁹ however, I mechanically extracted the instrument's relationship with projected contemporaneous TM spending and other potentially endogenous variables, ensuring the exogeneity of the instrument.

I conducted two‐stage IV probit regression. In the first stage, I obtained the estimated likelihood of MA enrollment while adjusting for individual‐ and county‐level characteristics and accounting for selective enrollment into MA. In the second stage, I estimated the relationship between estimated enrollment in MA plans from the first stage and the outcomes of interest. Both stages adjusted for the aforementioned variables and adjusted the SEs for clustering within the county. My IV estimates represent the local average treatment effect (LATE) of MA enrollment, which is the average causal effect for those whose enrollment status is sensitive to the instrument (residing in counties with high exogenous MA benchmarks), known as “compliers.” As a sensitivity analysis, I followed prior research and included the number of MA plans at the county level as an additional instrument. ²²

For all analyses, I included state and year‐fixed effects. I used survey weights to adjust sample characteristics to be representative of the Medicare population.

3. RESULTS

I included 9399 MA enrollees and 15,543 TM enrollees (Table 1). There were statistically significant differences in individual‐level characteristics between MA and TM enrollees in terms of age, race/ethnicity, education, household income, dual eligibility for Medicare and Medicaid, residence in a metropolitan area, comorbidities (hardening of arteries, cancer, and diabetes), overall health status compared with a year ago, and activities of daily living limitations.

TABLE 1.

Sample characteristics of TM and MA enrollees

Characteristics	% or mean (SD)		p‐value
	TM enrollees (N = 15,543)	MA enrollees (N = 9399)
Individual‐level characteristics, %
Age			<0.001
65–69	16.5	17.1
70–74	22.6	24.4
75–79	20.5	20.9
≥80	40.4	37.6
Female	56.6	56.3	0.585
Race/ethnicity			<0.001
Non‐Latino White	84.1	76.9
Non‐Latino Black	7.6	9.9
Non‐Latino Asian	1.4	1.8
Latino	5.6	9.7
Others	3.0	3.0
Education			<0.001
Less than high school	18.8	22.0
High school completion	34.2	36.1
Some college or associate's degree	20.9	19.2
Bachelor's degree	13.7	11.9
Advanced degree	12.1	10.4
Household income			<0.001
Less than $25,000	35.3	41.3
$25,000–$40,000	45.2	41.0
More than $40,000	16.7	15.8
Dual eligibility	11.2	12.8	<0.001
Married	53.8	54.2	0.537
Metro area	69.8	83.3	<0.001
Comorbidity
Hardening of arteries	11.2	10.1	0.015
Hypertension	72.1	72.5	0.431
Myocardial infarction	13.3	13.1	0.648
Stroke	11.2	11.0	0.751
Cancer	42.4	36.8	<0.001
Rheumatoid arthritis	16.4	17.1	0.353
Osteoporosis	23.8	23.5	0.628
Diabetes	28.5	31.6	<0.001
ADRD	6.4	6.4	0.934
Good overall health status compared with same‐age people, %	82.5	83.1	0.186
Good health compared to prior year	79.6	80.8	0.026
Number of ADL limitations			<0.001
0	50.5	54.5
1–2	20.2	19.8
3+	29.3	25.7
County‐level factors, mean (SD)
Percent of 65 years and older	15.6 (4.6)	15.4 (4.1)	<0.001
Percent of White residents	69.0 (19.9)	66.3 (21.2)	<0.001
Percent of Black residents	7.1 (7.5)	6.8 (7.0)	0.008
Percent of Hispanic residents	1.3 (1.7)	1.7 (2.1)	<0.001
Percent of dual‐eligible Medicare beneficiaries	20.8 (8.4)	23.2 (9.9)	<0.001
Average HCC risk score	15.3 (5.6)	15.2 (4.9)	<0.001
Percent of residents with incomes below poverty	1.0 (0.1)	1.0 (0.1)	<0.001
Number of hospital beds per 1000 residents	2232.5 (4047.8)	3152.6 (4987.0)	<0.001
Number of medical doctors per 1000 residents	2.4 (1.6)	2.6 (1.5)	<0.001
Number of primary care physicians per 1000 residents	0.8 (0.4)	0.9 (0.4)	<0.001

Abbreviations: ADL, activities of daily living; ADRD, Alzheimer's disease and related dementias; HCC, hierarchical condition category; MA, Medicare advantage; TM, traditional fee‐for‐service Medicare.

My unadjusted analysis showed that there were no or small differences in preventive care use and health behavior between MA and TM enrollees (Table 2). Specifically, MA enrollees were more likely than TM enrollees to have blood pressure measurements (77.4% vs. 76.5%), cholesterol measurements (99.2% vs. 98.7%), and comprehensive food exams for diabetes care (66.7% vs. 65.7%). However, MA enrollees were more likely than TM enrollees to currently smoke (15.9% vs. 13.6%). There were no statistically significant differences in other outcomes.

TABLE 2.

Unadjusted outcomes between TM and MA enrollees

Outcomes	TM enrollees		MA enrollees		p‐value
	N	% of “yes” values	N	% of “yes” values
Preventive care use
Diagnostic and preventive testing
Blood pressure measurement	15,261	76.5	9225	77.4	0.004
Cholesterol measurement	15,438	98.7	9361	99.2	0.003
Influenza vaccine	14,935	97.7	9062	98.2	0.059
Diabetes care
Insulin use	1243	25.2	1258	22.8	0.061
Medication use	1238	66.0	1256	63.7	0.347
Check for sores on feet	2075	73.9	1346	72.5	0.193
Blood for sugar (glucose) test	2073	68.6	1343	66.8	0.139
Blood pressure check at home	2074	53.8	1346	51.9	0.247
Comprehensive food exam	2059	65.7	1342	66.7	0.001
HbA1C test	7383	97.8	4412	96.7	0.882
Blood sugar well control	1978	85.2	1281	85.6	0.982
Health behavior
Currently smoking	8885	13.6	5233	15.9	<0.001
Drinking alcohol	7890	41.6	4846	41.9	0.781
Drinking alcohol more than ten days a month	7898	18.1	4835	16.7	0.115
Drinking more than four drinks per day	7891	6.0	4840	6.4	0.420
Doing any physical activities	15,543	60.5	9399	58.2	0.356
Doing muscle‐strengthening activities	15,543	38.4	9399	34.2	0.680

I found evidence of the validity of using PS matching analysis and IV analysis. For PS matching analysis, sample characteristics were similar between MA and TM enrollees after applying IPTW. Also, there was a substantial overlap in propensity scores between MA and TM enrollees. For IV analysis, I found that my instrument appears to be strong except for some process measures of diabetes care. The exogenous component of the benchmark led to a significant decrease in MA enrollment (−0.0005 [95% CI: −0.0007, −0.0004] to −0.007 [95% CI: −0.009, −0.005]) (Table 3). The F‐statistics were generally higher than 20, but the F‐statistics were lower than 10 for 7 process measures of diabetes care. I found that a large number of counties changed quartiles over time (ranging from 25.1% to 33.7% each year). Specifically, about 10%–15% of counties had an increase or a decrease in quartiles, respectively. Furthermore, the exogenous component of the MA benchmark was larger than the endogenous component of the MA benchmark. Also, the exogenous component of the MA benchmark accounted for a large proportion of variations in the MA benchmark (nearly 50%).

TABLE 3.

Results from first stage regression

Outcomes	IV using the exogenous component
	Coefficient (95% CI)	F‐statistics
Preventive care use
Diagnostic and preventive testing
Blood pressure measurement	−0.0005 (−0.0007 to −0.0004)	40.22
Cholesterol measurement	−0.0005 (−0.0007 to −0.0004)	41.23
Influenza vaccine	−0.0006 (−0.0007 to −0.0004)	40.02
Diabetes care
Insulin use	−0.0008 (−0.0012 to −0.0003)	7.62
Medication use	−0.0007 (−0.0012 to −0.0002)	7.69
Check for sores on feet	−0.0008 (−0.0013 to −0.0004)	9.27
Blood for sugar (glucose) test	−0.0009 (−0.0013 to −0.0004)	9.24
Blood pressure check at home	−0.0006 (−0.0007 to −0.0004)	9.27
Comprehensive food exam	−0.0009 (−0.0013 to −0.0005)	9.37
HbA1C test	−0.0006 (−0.0009 to −0.0004)	22.14
Blood sugar well control	−0.0008 (−0.0012 to −0.0004)	8.85
Health behavior
Currently smoking	−0.0006 (−0.0008 to −0.0004)	24.04
Drinking alcohol	−0.0006 (−0.0008 to −0.0004)	20.73
Drinking alcohol more than ten days a month	−0.0006 (−0.0008 to −0.0004)	20.81
Drinking more than four drinks per day	−0.0006 (−0.0008 to −0.0004)	20.85
Doing any physical activities	−0.0007 (−0.0009 to −0.0005)	28.68
Doing muscle‐strengthening activities	−0.0007 (−0.0009 to −0.0005)	28.97

I showed that MA enrollment was statistically significantly associated with no or small increases in the use of preventive services (Table 4). Specifically, linear regression and PS matching analyses showed that MA enrollment was statistically significantly associated with higher likelihood of having blood pressure measurement (0.67 [95% CI: 0.28–1.06] and 0.48 [95% CI: 0.19–0.76] percentage point increases, respectively), cholesterol measurement (0.83 [95% CI: 0.32–1.33] and 0.61 [95% CI: 0.25–0.96] percentage point increases, respectively), and influenza vaccine (2.66 [95% CI: 1.32–4.00] and 1.94 [95% CI: 0.94–2.95] percentage point increases, respectively) and lower likelihood of receiving HbA1C test (0.94 [95% CI: −1.66 to −0.21] and 0.76 [95% CI: −1.28 to −0.25] percentage point decreases, respectively). However, the magnitude of the associations was small. There were no statistically significant associations in other measures. IV analysis also found that MA enrollment did not lead to statistically significant changes in preventive care use.

TABLE 4.

Effects of MA enrollment on preventive care use

Outcomes	Change in outcome associated with MA enrollment
	Naïve regression, percentage points (95% CI)	Regression with propensity scores, percentage points (95% CI)	IV probit regression using the exogenous component, coefficient (95% CI)	IV probit regression using the exogenous component and the number of MA plans, coefficient (95% CI)
Diagnostic and preventive testing
Blood pressure measurement	0.67 (0.28 to 1.06)	0.48 (0.19 to 0.76)	−2.08 (−4.8 to 0.64)	−0.05 (−1.07 to 0.97)
Cholesterol measurement	0.83 (0.32 to 1.33)	0.61 (0.25 to 0.96)	0.59 (−1.25 to 2.43)	0.24 (−0.54 to 1.01)
Influenza vaccine	2.66 (1.32 to 4.00)	1.94 (0.94 to 2.95)	0.64 (−0.14 to 1.43)	−0.27 (−0.60 to 0.07)
Diabetes care
Insulin use	−3.93 (−7.87 to 0.01)	−3.16 (−6.69 to 0.38)	0.01 (−1.82 to 1.83)	0.50 (−0.49 to 1.49)
Medication use	−3.54 (−7.80 to 0.72)	−4.10 (−8.22 to 0.02)	−2.18 (−4.43 to 0.06)	−0.48 (−1.48 to 0.52)
Check for sores on feet	0.68 (−3.10 to 4.46)	−0.02 (−3.05 to 3.02)	0.54 (−0.74 to 1.82)	0.24 (−0.48 to 0.96)
Blood for sugar (glucose) test	−1.48 (−5.05 to 2.09)	−1.72 (−4.61 to 1.16)	0.34 (−0.97 to 1.66)	0.25 (−0.49 to 0.99)
Blood pressure check at home	−0.30 (−4.17 to 3.57)	−0.69 (−4.01 to 2.63)	−0.78 (−2.10 to 0.54)	−0.12 (−0.84 to 0.59)
Comprehensive food exam	−1.28 (−4.75 to 2.19)	−1.41 (−4.15 to 1.34)	−0.62 (−2.04 to 0.80)	0.21 (−0.56 to 0.98)
HbA1C test	−0.94 (−1.66 to −0.21)	−0.76 (−1.28 to −0.25)	0.45 (−1.52 to 2.41)	0.35 (−0.64 to 1.34)
Blood sugar well control	0.16 (−2.80 to 3.12)	−0.17 (−2.41 to 2.06)	1.67 (−0.12 to 3.46)	0.19 (−0.71 to 1.08)

Abbreviation: MA, Medicare advantage.

I found that MA enrollment was statistically significantly associated with no or limited improvements in health behavior (Table 5). Specifically, linear regression and PS matching analyses showed that MA enrollment was statistically significantly associated with higher likelihood of currently smoking (2.47 [95% CI: 0.92–4.02] and 2.94 [95% CI: 1.06–4.82] percentage point increases, respectively). However, there were no statistically significant associations in other measures. IV analysis using the exogenous component as an instrument found that MA enrollment led to statistically significant improvements in the likelihood of doing any physical activities (1.29 [95% CI: 0.51–2.07]) or doing muscle‐strengthening activities (coefficient: 0.72 [95% CI: 0.03–1.41]). IV analysis using the exogenous component and the number of MA plans as an instrument found that MA enrollment led to statistically significant increases in the likelihood of drinking alcohol (coefficient: 0.74 [95% CI: 0.30–1.18]) or drinking more than four drinks per day (coefficient: 1.12 [95% CI: 0.41–1.84]). No statistically significant changes were observed in other measures.

TABLE 5.

Effects of MA enrollment on health behavior

Outcomes	Change in outcome associated with MA enrollment
	Naïve regression, percentage points (95% CI)	Regression with propensity scores, percentage points (95% CI)	IV probit regression using the exogenous component, coefficient (95% CI)	IV probit regression using the exogenous component and the number of MA plans, coefficient (95% CI)
Currently smoking	2.47 (0.92 to 4.02)	2.94 (1.06 to 4.82)	−0.21 (−1.23 to 0.81)	0.01 (−0.52 to 0.53)
Drinking alcohol	0.67 (−1.34 to 2.69)	0.84 (−1.03 to 2.71)	0.45 (−0.49 to 1.40)	0.74 (0.30 to 1.18)
Drinking alcohol more than ten days a month	−0.32 (−1.96 to 1.32)	−0.40 (−2.33 to 1.53)	−0.26 (−1.40 to 0.87)	0.28 (−0.23 to 0.78)
Drinking more than four drinks per day	0.42 (−0.74 to 1.58)	0.46 (−1.14 to 2.06)	0.78 (−0.83 to 2.38)	1.12 (0.41 to 1.84)
Doing any physical activities	1.40 (0.07 to 2.88)	1.07 (−0.01 to 2.25)	1.29 (0.51 to 2.07)	0.32 (−0.40 to 0.68)
Doing muscle‐strengthening activities	1.39 (−0.33 to 3.11)	1.32 (−0.35 to 2.99)	0.72 (0.03 to 1.41)	0.19 (−0.14 to 0.51)

Abbreviation: MA, Medicare advantage.

4. DISCUSSION

I found two primary results. First, I showed that MA enrollment was associated with no or limited increases in the use of clinically meaningful preventive services. Second, there was limited evidence that MA enrollment was associated with improvements in health behavior.

MA plans may have the potential to improve access to preventive care overall, but my findings suggest that there were limited increases in the use of clinically meaningful services. This finding may appear to contradict results from prior research that MA enrollees were more likely to have preventive services than TM enrollees. ¹⁵ However, there is also evidence that use of preventive care was not necessarily higher among MA enrollees than TM enrollees. ²³ There may be multiple explanations. First, use of preventive services was already high among MA enrollees, possibly leading to limited improvement. Second, the ACA eliminated cost‐sharing for many preventive services, which lowered barriers to enrollees' use of preventive services, especially for TM enrollees who often face high‐cost‐sharing requirements. Finally, the prior study only accounted for observable differences between MA and TM enrollees, and thus this finding may be vulnerable to bias attributable to unobservable differences.

I also found evidence that MA plans may be limited in effectively encouraging enrollees to change their health behavior. Prior research found that MA plans may be motivated to improve the efficiency of care delivery, as the potential for cost savings from unnecessary services may be substantial. ²⁴ It is notable that my IV analysis using the exogenous component as an instrument found that MA enrollment led to improvements in the likelihood of doing any physical activities or muscle‐strengthening activities. As MA plans cover the costs of fitness programs and gym memberships, ¹⁶ this may explain the mechanism through which MA enrollees are more likely to engage in physical activities than TM enrollees. However, this finding was not consistently observed across all analyses, suggesting that MA may not be incentivized to promote targeted investment in health behavior. Because improvements in health behavior may have the potential to achieve better outcomes while lowering costs, policy makers should consider developing targeted interventions for MA to achieve those improvements.

My findings should be interpreted with caution. First, my findings may not be generalizable in several ways. Specifically, I only examined a limited number of outcomes, and thus my findings may not be applicable to other outcomes. Moreover, I found my sample covered only about 20% of counties. Thus, the pattern of MA enrollment in these counties may not be applicable to other counties. Furthermore, my IV analysis estimated the LATE of MA enrollment for those whose MA enrollment had been changed due solely to residing in counties with high exogenous MA benchmarks. Therefore, these findings apply only to the complier. Second, there may be some concerns about the validity of my findings. Specifically, I accounted for differences in sample characteristics between MA and TM enrollees, but unobserved differences in individual‐level characteristics may have remained. This concern is likely to be more profound for linear regression and PS matching analyses. Additionally, IV analysis allows me to account for observed and unobserved differences in sample characteristics, but there may be another issue. For example, I assumed that outcomes of interest were independent of the instrument. However, this could not be proven conclusively. Also, I found that my instrument appeared to be weak for several process measures of diabetes care, possibly raising concerns about the validity of my findings. Moreover, I assumed that the instrument did not affect the outcome except through its association with the treatment variable. However, this might not be satisfied if MA plans selectively enroll healthier individuals and or if there are spillover effects from MA to TM. Third, outcome variables were self‐reported and thus findings may be subject to self‐reporting errors. Finally, I did not detect significant differences in some outcomes, which could be attributable to the sample size.

5. CONCLUSIONS

I found that MA enrollment was associated with no or limited increases in the use of clinically meaningful preventive services. There was limited evidence that MA enrollment was associated with improvements in health behavior. These findings suggest that there may still be misaligned incentives for MA plans to use more preventive services and enhance health behavior in order to increase plan efficiency. Thus, policy makers should consider how to better design financial incentives to improve primary care access and encourage enrollees to engage in more positive health behavior.

FUNDING INFORMATION

None.

CONFLICT OF INTEREST

The authors declare that there is no conflict of interest.

Park S. Effects of Medicare Advantage on preventive care use and health behavior. Health Serv Res. 2023;58(3):569‐578. doi: 10.1111/1475-6773.14089