The effect of Medicare Annual Wellness Visits on breast cancer screening and diagnosis

Abstract

Objective:

The Medicare Annual Wellness Visit (AWV)—a prevention-focused annual check-up—has been available to beneficiaries with Part B coverage since 2011. The objective of this study was to estimate the effect of Medicare AWVs on breast cancer screening and diagnosis.

Data Sources and Study Setting:

The National Cancer Institute’s Surveillance, Epidemiology, and End Results cancer registry data linked to Medicare claims (SEER-Medicare), HRSA’s Area Health Resources Files, the FDA’s Mammography Facilities database, and CMS “Mapping Medicare Disparities” utilization data from 2013-2015.

Study Design:

Using an instrumental variables approach, we estimated the effect of AWV utilization on breast cancer screening and diagnosis, using county Welcome to Medicare Visit (WMV) rates as the instrument.

Data Collection/Extraction Methods:

66,088 person-year observations from 49,769 unique female beneficiaries.

Principal Findings:

For every 1-percentage point increase in county WMV rate, probability of AWV increased 1.7-percentage points. Having an AWV was associated with a 22.4-percentage point increase in probability of receiving a screening mammogram within six months (P<0.001). There was no statistically significant increase in probability of breast cancer diagnosis (overall or early stage) within six months of an AWV. Findings were robust to multiple model specifications.

Conclusions:

Performing routine cancer screening is an evidence-based practice for diagnosing earlier-stage, more treatable cancers. The Medicare Annual Wellness Visit effectively increases breast cancer screening and may lead to more timely screening. Continued investment in Annual Wellness Visits supports breast cancer screening completion by women who are most likely to benefit, thus reducing the risk of over-screening and over-diagnosis.

Introduction

The Medicare Annual Wellness Visit (AWV) is a prevention-focused annual check-up available to fee-for-service beneficiaries with Part B coverage since 2011.^1,2 While there is ongoing debate about the utility of annual check-ups,^3,4 Medicare continues to make large investments in the AWV – spending more than $1.1 billion in 2019 alone.⁵ Since its introduction, the AWV has gained popularity among patients and clinicians.^6,7 The AWV represents a rare opportunity in primary care to escape the “tyranny of the urgent”⁸ and focus on prevention, including cancer screening. With a generous reimbursement structure for clinicians, and no patient cost-sharing, the AWV incentivizes clinicians and patients to engage in longer visits focused on preventive rather than acute care. However, utilization of the visit has not been consistent and varies by patient, clinician, and practice characteristics.^9,10

Utilization of screening mammography and diagnosis of earlier stage breast cancer have been proposed to measure the effectiveness of AWVs to achieve their prevention-focused objectives.⁷ The AWV would improve cancer-related outcomes if beneficiaries receiving AWVs faced barriers to care such as lack of awareness of cancer risk or availability of screening and clinicians lacked time to discuss cancer screening during problem-focused visits. Since Medicare eliminated cost-sharing for all preventive services – including the AWV and screening mammograms – beginning in 2011, it is plausible that AWVs not only facilitate referrals for screening, but also make beneficiaries aware of their eligibility for screening and the absence of cost obligations, thus lowering barriers to cancer screening. From the clinician side, AWVs are reimbursed at higher rates than some return patient visits and have a dedicated prevention and planning focus. Clinicians value this unique opportunity to strengthen the relationship with patients outside of problem-oriented visits while also closing gaps in preventive care.⁷

Difficulty studying the causal effects of the AWV arises from the non-random utilization of the visit. Patients who are more likely to take advantage of AWVs are the patients who are also more likely to be healthier or have more interest in prevention services and thus are likely to experience better outcomes regardless of the AWV. Patients in worse health and who present with competing health care demands are less likely to complete an AWV.^6,11 Indeed, when clinicians perceive their patients to be sicker and less likely to benefit from the AWV, they are less likely to offer it. Further, not all clinicians find AWVs useful or have the absorptive capacity to initiate a new visit type, while some clinics with higher-acuity patients are not financially incentivized to adopt the visit.⁷ Thus, there is both patient and clinician selection into which patients receive an AWV.

The objective of this study was to estimate the effect of the Medicare AWV on breast cancer screening and diagnosis. To estimate a plausibly causal effect of the AWV on outcomes, we rely on an instrumental variable approach (IV) using county-level utilization rates of Medicare’s Welcome to Medicare Visit (WMV) to measure the local area’s proclivity for prevention. The WMV is a one-time benefit available to all fee-for-service beneficiaries within their first 12 months of Medicare enrollment. We posit that higher utilization of the WMV in a county signals stronger emphasis on prevention and regional practice patterns. Beneficiaries might be more likely to receive an AWV in areas where wellness and prevention are more common. The IV approach helps mitigate the selection bias that makes causal estimates of AWV outcomes difficult.¹²

Methods

Study Sample

We selected Medicare FFS beneficiaries from 2013-2015 in the SEER-Medicare linked cancer registry and claims database (SEER18).¹³ The sample was selected from the cancer and 5% non-cancer cohorts and limited to female beneficiaries aged 66-73 as of January 1, 2014 (index date) because they were age-eligible for breast cancer screening according to USPSTF guidelines.¹⁴ To identify beneficiaries most likely “due” for biennial screening, we excluded observations from beneficiaries with a screening mammogram in the previous year. The sample was further limited to beneficiaries with continuous Medicare Parts A and B coverage from 12 months prior to the index date until the end of the two-year data window, or a breast cancer diagnosis, or death, whichever came first. Thus, the cohort included women who were at risk of a first breast cancer diagnosis, were eligible for routine breast cancer screening, and were eligible for and more likely to benefit from the AWV.

Measures

We included two outcomes: screening mammogram and breast cancer diagnosis (overall and early stage). Outcomes had to follow the AWV within six months to be considered related to the AWV. AWVs, which can occur once per year, were identified using procedure codes (HCPCS: G0438, G0439). Breast cancer screening and dates of service were identified using Current Procedural Terminology (CPT)/HCPCS codes for screening mammography (CPT: 76092, 77052, 77057, 77063, 77067; HCPCS: G0202, G0203). Diagnosis outcomes were identified using the first occurrence of breast cancer from the SEER registry. We considered cancers to be early stage if the SEER summary stage was recorded as in situ, local, or regional by direct extension only.¹⁵ From each AWV date, we followed beneficiaries for up to six months, minimizing the concern that outcomes occurring much later would be attributed as causal effects of the visit. AWVs occurring less than six months before the end of the observation period were excluded due to inadequate follow-up time.

Beneficiaries without an AWV were assigned a simulated, uniformly distributed random index date from which to calculate follow up.¹⁶ Without inclusion of the simulated AWV date, results are biased toward the null because beneficiaries with a screening or cancer diagnosis but no AWV would be “credited” for their outcome regardless of when it occurred, while only outcomes occurring after the AWV are counted for beneficiaries with an AWV. Women who had a screening mammogram within the calendar year were not included as a control observation in the following year. Figure 1 displays the actual and simulated AWV dates.

Figure 1.

Distribution of Observed and Simulated AWV dates

Annual county-level WMV rates in the year preceding the observation year, the instrument, were obtained from the Center for Medicare and Medicaid Services (CMS) “Mapping Medicare Disparities by Population” tool.¹⁷ CMS computes and publishes annual rates calculated as the number of WMVs in the inpatient, outpatient, and carrier files among all eligible FFS beneficiaries in each county. Like the AWV, there is variability in use of the WMV, with notable geographic differences in utilization rates, along with differences by gender, race/ethnicity, and Medicare supplement enrollment status.¹⁸ Our underlying assumption is that county-level utilization of the WMV reflects practice patterns for preventive primary care services in the area, which conditional on other covariates, is not directly associated to the outcomes we study (exclusion restriction).¹⁹ One way this can occur is when clinics conduct outreach to their patients who are eligible for wellness visits.^7,20,21 A woman’s county of residence is plausibly independent from county WMV utilization rates, thus the area-level variation that we employ as the instrument acts to pseudo-randomize women to the AWV.²² For women who are “encouraged” and receive an AWV because they are persuaded by the culture of wellness and prevention in their county of residence, the counterfactual argument is that those same women living in a similar county with low emphasis on prevention would be unlikely to receive an AWV. This empirical strategy has been used to establish the impact of the AWV on dementia diagnosis, for which screening, the intermediate outcome, is not standardized.²³ Geographic variation is commonly employed as an instrument in the oncology literature because it easily passes the relevance test and can be reliably computed from administrative data.¹⁹

Statistical Analysis

The estimation used a two-stage residual inclusion (2SRI) approach, which provides consistent estimation of binary outcomes.^24,25 In the first stage, we used logistic regression to estimate the probability that a beneficiary received an AWV as a function of the instrument, year and age indicators, and beneficiary- and county-level covariates. In the second stage, we estimated a second logistic regression model for the outcome of interest using an indicator for AWV utilization and the residuals from the first-stage regression:

$$\mathit{logit}({\mathit{AWV}}_{\mathit{it}})={\beta }_0+{\beta }_1{\mathit{WMV}}_{\mathit{it}-1}+{\mathit{X}}_{\mathit{it}}^{\prime }{\beta }_2+{\mathit{Z}}_i^{\prime }{\beta }_3+{\alpha }_t$$ (1)

$$\mathit{logit}(Y_{\mathit{it}})={\gamma }_0+{\gamma }_1{\mathit{AWV}}_{\mathit{it}}+{\gamma }_2{\mathit{AWV}}_{-{\mathit{res}}_{\mathit{it}}}+{\mathit{X}}_{\mathit{it}}^{\prime }{\gamma }_3+{\mathit{Z}}_i^{\prime }{\gamma }_1+{\alpha }_t$$ (2)

where i indexes an individual, t indexes the calendar year, WMV is the county-specific WMV utilization rate, AWV is an indicator for having an AWV claim, AWV_res is the residual from the first stage, X is a vector of time-varying county-level covariates from the Area Health Resources Files and Food & Drug Administration, Z is a vector of time-invariant beneficiary characteristics, Y is an indicator for breast cancer screening or diagnosis (modeled separately), and $\alpha$ are year indicators (fixed effects). The coefficient of interest is ${\gamma }_1$, which captures the local average treatment effect of the AWV on outcomes (Y). We report marginal effects, which are the differences in adjusted average predicted probabilities of the outcome. Standard errors were bootstrapped with 1,000 replications and clustered at the individual level to account for the longitudinal nature of the data.

The IV models estimate a local average treatment effect (LATE), applicable to individuals whose treatment was influenced by the instrument, the so-called “compliers.”^26,27 Some Medicare beneficiaries will always (or never) take advantage of no-cost AWVs, regardless of practice patterns in their area. However, compliers are beneficiaries whose AWV utilization is influenced by the practice patterns in their area and the preference of clinicians to perform or recommend preventive services. Thus, the LATE captured by our IV approach applies only to the individuals who received an AWV because they resided in a county with a stronger likelihood of a WMV, reflecting local norms, related to peer-influence and other local practices and policies. This estimand is of policy interest because wellness visits are intended to motivate screening in asymptomatic women.

In additional analyses we 1) examined shorter follow-up within 3 months post-AWV, 2) estimated un-instrumented models (“naïve” estimates), 3) excluded AWVs occurring in December and January when the distribution of AWVs was less uniform than the rest of the calendar year due to seasonal variations in health care utilization, 4) limited the sample to only the first observation for each beneficiary to avoid using the same women as treated and control observations in different calendar years, 5) analyzed a sub-group of women from the 5% non-cancer cohort only, 6) included state fixed-effects to account for time-invariant differences/unobserved confounders across states, and 7) used current year WMV rates (in place of prior-year WMV rates) as the instrument to insure the population receiving a WMV was distinct from beneficiaries receiving an AWV.

This study was considered exempt research using secondary data by the [OMITTED FOR REVIEW] Institutional Review Board (#19-0424) and approved by the [OMITTED FOR REVIEW] Review and Monitoring System. All analyses were conducted using Stata/MP 17.0 (College Station, TX). We considered statistical significance at the p=0.05 level using 2-sided tests of significance.

Results

Sample Characteristics

The sample included 66,088 person-year observations for 49,769 unique beneficiaries. Table 1 shows baseline characteristics at the person- and county-level, combined and by receipt of AWV. Over 86% of beneficiaries were White. Most beneficiaries had low comorbidity burden (≤ 2 comorbid conditions), measured by the Charlson Comorbidity Index. County characteristics demonstrate that more AWVs occurred where the underlying populations were more highly educated, urban, affluent, and with higher Medicare Advantage enrollment rates. AWVs occurred in counties where physician supply, per 100,000 county residents, was more plentiful; however, there were no underlying differences in the supply of short-term general hospital beds or the number of mammography imaging centers.

Table 1.

Sample Characteristics

	Overall	AWV claim	No AWV claim	P-value
Total N (row %)	49,769 (100)	11,779 (23.67)	37,990 (76.33)
At baseline (Index date January 1, 2014 or first observed)
Age cohort, N (col %)				0.056
66	615 (1.24)	179 (1.51)	436 (1.15)
67	8,173 (16.42)	1,917 (16.27)	6,256 (16.47)
68	7,702 (15.48)	1,843 (15.65)	5,859 (15.42)
69	6,711 (13.48)	1,621 (13.76)	5,090 (13.40)
70	6,694 (13.45)	1,575 (13.37)	5,119 (13.47)
71	7,016 (14.10)	1,656 (14.06)	5,360 (14.11)
72	6,743 (13.55)	1,590 (13.50)	5,153 (13.56)
73	6,115 (12.29)	1,398 (11.87)	4,717 (12.42)
Race/ethnicity, N (col %)				<0.001
White	43,230 (86.86)	10,406 (88.34)	32,824 (86.40)
Black	3,050 (6.13)	626 (5.31)	2,424 (6.38)
Hispanic	366 (0.74)	51 (0.43)	315 (0.83)
Asian/Pacific Islander	1,160 (2.33)	232 (1.97)	928 (2.44)
American Indian/Alaskan Native	139 (0.28)	12 (0.10)	127 (0.33)
Other/Unknown	1,824 (3.66)	452 (3.84)	1,372 (3.61)
Health Status, # Charlson Comorbidities				<0.001
0	34,769 (69.86)	8,596 (72.98)	26,173 (68.89)
1-2	12,974 (26.07)	2,827 (24.00)	10,147 (26.71)
3-4	1,676 (3.37)	306 (2.52)	1,381 (3.64)
5+	350 (0.70)	61 (0.52)	289 (0.76)
At index date (Jan 1, 2014)
WMV utilization rate, mean % (SD)	8.73 (4.05)	9.83 (4.08)	8.38 (3.97)	<0.001
County-level socio-demographic factors, mean (SD)
% age 65+	14.85 (3.65)	14.82 (3.60)	14.86 (3.67)	0.311
% non-English speaking	4.35 (3.80)	4.41 (3.69)	4.33 (3.84)	0.037
% with less than high school education	8.76 (3.66)	8.41 (3.47)	8.86 (3.72)	<0.001
% enrolled in Medicare Advantage	28.10 (13.92)	28.69 (13.30)	27.91 (14.10)	<0.001
% in a rural county	15.65 (36.33)	11.49 (31.89)	16.94 (37.51)	<0.001
% with median household income >$50,000	77.16 (41.98)	81.26 (39.02)	75.89 (42.77)	<0.001
Median home value (1000s)	$297.62 (181.66)	$312.94 (187.32)	$292.87 (179.61)	<0.001
County health care supply factors, mean (SD)
Num. mammography imaging centers	24.91 (42.75)	25.06 (40.71)	24.86 (43.36)	0.658
Physicians, per 100k population	303.95 (172.42)	327.36 (172.69)	296.70 (171.69)	<0.001
Short-term general hospital beds, per 100k population	225.52 (150.96)	224.39 (147.43)	225.87 (152.04)	0.353
Outcomes during follow-up (2014-2015)
Person-year Observations N (row %)	66,088 (100)	14,393 (21.78)	51,695 (78.22)
AWV claim, N (col %)
2014	--	4,237 (16.70)	--
2015	--	10,156 (24.94)	--
Screening Mammogram, N (col%)
2014	8,071 (31.81)	2,237 (52.80)	5,834 (27.61)	<0.001
2015	19,402 (47.65)	6,843 (67.38)	12,559 (41.09)	<0.001
Breast Cancer Diagnosis, N (col %)
2014	2,071 (8.16)	415 (9.79)	1,656 (7.84)	<0.001
2015	3,535 (8.68)	941 (9.27)	2,594 (8.49)	0.016

Note: 66,088 person-year observations for 49,769 fee-for-service (FFS) Medicare beneficiaries who were female, aged 66-73 years in 2014 and had no history of cancer (of any type) as of January 1, 2014. Utilization and health status measures came from SEER-Medicare cancer denominator and claims files. County-level socio-demographic factors came from the Area Health Resources Files. Chi-square tests were used to compare categorical variables, two-sample T tests were used to compare proportions/means of continuous variables.

Abbreviations: WMV = Welcome to Medicare Visit; AWV = Annual Wellness Visit; SD= standard deviation; col % = column percent

First Stage Model

County-level WMV visit rate was between 0-30% (mean: 8.06%; standard deviation [SD]: 3.80). The county-level WMV rate in the previous year was a strong, linear predictor of individual-level AWV utilization, supporting the monotonicity assumption (i.e., “no defiers”) of IVs (Figure 2). For every 1-percentage point increase in county WMV rate, the probability of an individual receiving an AWV in the next calendar year increased by 1.7% (95% confidence interval [CI]: 1.6 – 1.8). The first stage F-statistic when estimating a linear probability model, an indicator of the strength of the instrument,²⁸ showed the WMV to be a very strong instrument for AWV utilization (F(1, 44935)=690.20; P<0.001). The corresponding Chi-square test from the first-stage logistic model in the 2SRI approach was 2353.07 (P<0.001).

Figure 2.

Predicted probability of an AWV, by county-level WMV rate in the previous calendar year

Breast Cancer Screening and Diagnosis

Our 2SRI model predicted a 37.6-percent probability of a screening mammogram within 6 months for beneficiaries with an AWV, compared to a 15.2-percent probability among beneficiaries without an AWV. The corresponding marginal effect was a 22.4-percentage point increase in the probability of screening mammograms among those with an AWV relative to those without an AWV (P<0.001) (Table 2). We found no difference in the probability of new breast cancer diagnosis within 6 months for those with an AWV compared to those without an AWV, the marginal effect was a 0.6-percentage point difference (P=0.793). Similarly, we found no difference in the probability of a new early-stage breast cancer diagnosis, the marginal effect was a 0.5-percentage point difference (P=0.759). The coefficients on the residuals from the first stage models (AWV_res) were not statistically significant (P=0.870; P=0.557; P=0.766), suggesting that the naïve models (reduced form) would yield similar findings.

Table 2.

Two-Stage Residual Inclusion (2SRI) Models of Screening Mammogram and Breast Cancer Diagnosis

	First stage AWV-utilization model			Second-stage outcomes model
	Marginal Effect	95% CI	P	Marginal Effect	95% CI	P
WMV (N = 65,973)	0.017***	(0.016, 0.018)	<0.001
Outcomes measured within 6 months of AWV (N = 44,963)
Screening				0.224***	(0.126, 0.322)	<0.001
Any Diagnosis				0.006	(−0.038, 0.049)	0.793
Early-Stage Diagnosis				0.005	(−0.027, 0.037)	0.759
Outcomes measured within 3 months of AWV (N = 55,160)
Screening				0.317***	(0.226, 0.408)	<0.001
Any Diagnosis				0.039	(−0.016, 0.094)	0.166
Early-Stage Diagnosis				0.026	(−0.023, 0.075)	0.306

^*P<0.05

^***P<0.001

Note: Models are adjusted for beneficiary and county-level factors listed in Table 1. The F-statistic measuring instrument strength is estimated using a linear probability model in the first stage: F = 649.07; P<0.001. The corresponding Chi-square test from the first-stage logistic model in the 2SRI approach was 2353.07 (P<0.001). Standard errors were bootstrapped (1,000 replications), clustering at the individual level to account for the longitudinal nature of the data.

Abbreviations: WMV = Welcome to Medicare Visit; AWV = Annual Wellness Visit

The naïve model is shown in the Supplemental Digital Content, Table S1. Without the IV, we find a 21.6-percentage point increase in the probability of screening mammograms for those with an AWV compared to those without (P<0.001). In the naïve model not accounting for selection, we found a 1.9-percentage point increase in the probability of breast cancer diagnosis and a 0.9-percentage point increase in the probability of breast cancer diagnosed at an early stage for beneficiaries with an AWV (both p<0.001).

When we shortened the follow-up time, allowing for outcomes that occurred only within three months of an AWV or simulated AWV date (Table 2), we find a 31.7-percentage point increase in the probability of breast cancer screening (P<0.001). We find no increase in the probability of breast cancer diagnosis overall (marginal effect: 3.9-percentage points; P=0.166), or in probability of cancer diagnosed at an early stage (marginal effect: 2.6-percentage points; P=0.306) among beneficiaries with an AWV.

Sensitivity Analysis

We conducted multiple analyses to test the robustness of the main findings. When we excluded AWVs occurring in December and January, we find an 18.2-percentage point increase in breast cancer screening (P=0.003), and no difference in breast cancer diagnosis overall (marginal effect: 0.2-percentage points; P=0.959) or at an early stage (marginal effect: 0.6-percentage points; P=0.775) within six months among those with AWVs. Analyses using the first observation only, applying state fixed-effects, and using the current year WMV rate as the instrument all found significant increases in screening mammography use (19.0-pp, P<0.001, 34.3-pp, P<0.001, and 22.4-pp, P<0.001, respectively) and no change in diagnosis of breast cancer overall or at an early stage. Limiting to the 5% non-cancer cohort, we find a sustained increase in screening mammography within 6 and 3 months of an AWV (20.0-pp, P<0.001 abd 27.6-pp P<0.001). Results from all sensitivity analyses are displayed in the Supplemental Digital Content, Tables S2 and S3.

Discussion

In this study, we leveraged the variability in local-area emphasis on wellness and prevention to control for selection bias related to receiving and offering AWVs to assess the causal effect of Medicare’s AWV on breast cancer screening and diagnosis. Our research design allows us to obtain causal effects of the AWV for women who were encouraged to receive an AWV because they lived in a county with a stronger emphasis on prevention. We found that the AWV increased probability of undergoing a screening mammogram within six months following the visit. The effect was larger when we shortened the follow-up period, suggesting that women are referred for mammograms during their AWV and undergo screening shortly after. We found no statistically significant effect of the AWV on new breast cancer diagnoses overall or at an early stage.

Previous studies of the AWV’s role in promoting preventive services in general, which include breast cancer screening, found mixed results.²⁹ A study of AWV outcomes in the first year the visit was available found no difference in preventive service utilization in the traditional Medicare group compared to beneficiaries with other types of coverage (supplemental private insurance, Medicare managed care, or supplemental public coverage).³⁰ However, in the first year, AWV uptake was very low (approximately 7.5% of eligible beneficiaries had a visit)⁶ and there was little patient and clinician awareness of the new benefit. Public awareness and utilization have increased steadily over time as patients have become more aware and clinicians and health care systems have addressed implementation challenges, allowing them to deliver more AWVs. An observational study in a single health care system found AWV utilization associated with increased mammography utilization, but not with the utilization of colorectal cancer screening.³¹ An additional study focusing on the Annual Wellness Visit found that it was not associated with a meaningful change in age-appropriate cancer screening utilization, including mammograms, among patients in clinics that adopted the AWV.³² Yet, another study that accounted for patient self-selection using a propensity-score adjusted difference-in-difference analysis found higher rates of breast cancer screening among beneficiaries with an AWV (81-percent vs. 66-percent).³³

A small number of studies examine the effect of Medicare’s prevention benefit expansion — including the AWV, cost-sharing elimination, and a temporary physician fee bump — on cancer-related outcomes. One study found no impact of these policies on breast cancer diagnosis rates,³⁴ while another found a modest but statistically significant effect on breast cancer diagnoses in the first three years the policies were in place.³⁵ Our study adds to the literature on the effects of AWVs on breast cancer screening and diagnosis among a policy-relevant population. One important distinction between our study and others is that we account for the non-random utilization of the AWV using an instrument, which may explain why we find no effect on breast cancer diagnosis. Our estimates apply to asymptomatic women who would not have received an AWV if they had resided in a different county. Our naïve model finds an effect on diagnosis, but the effect disappears once non-random selection is considered. Importantly, breast cancer screening rates were relatively high at baseline,³⁶ and women who are screened after an AWV likely reflect a relatively healthy, low- to average-risk population given the IV LATE interpretation. The AWV may have a larger impact on screening and diagnosis of other cancers, such as lung cancer, for which screening is available with no cost-sharing but remains significantly underutilized.^37,38

The AWV presents an opportunity to discuss personalized recommendations for prevention and care planning based on patient preferences and the likelihood of benefit and de-escalation of care once it is deemed no longer appropriate or beneficial for a patient. Excess screening could identify indolent cancers unlikely to cause harm over the lifetime, resulting in over-treatment and treatment-induced morbidity.^39,40 However, AWVs are often longer in duration than other return patient and chronic disease management visits and Medicare is prescriptive about the prevention-related elements that must be addressed. This opportunity for patients and their providers to discuss the individual pros and cons of screening may mean that screening following an AWV is more concordant with patient preferences and the likelihood of offering a favorable risk/benefit profile for that patient.⁴¹ Nonetheless, some features of the AWV are barriers to implementation in some clinical settings, which may perpetuate existing disparities.^7,10 Using lessons from AWV adopters, such as standardizing patient-facing messaging, redistributing clinical workload, and promoting the clinical, financial and interpersonal value may enhance adoption and improvement in preventive service utilization.

A limitation of this study is that treatment effects estimated by IV are LATE and cannot be interpreted as the effect of the visit on all patients, although the LATE is of interest in this study because the population more likely to benefit from a visit focused on prevention are women who would not have been screened otherwise (i.e., compliers). As with other studies that rely on IVs, the validity of our study hinges on the area-level WMV rate not being directly associated with an individual screening and diagnosis outcomes after controlling for other factors. Although using area-level variation is common in IV studies, this assumption cannot be empirically assessed. There may be other unmeasured confounders outside of the individual- and area-level covariates we included. For instance, clinician awareness and ability have been identified as barriers to preventive service delivery.⁴² Finally, because our sample is limited to an age-restricted population of women with FFS Medicare, these findings may not generalize to younger or older women, or those with Medicare Advantage, whose plans may offer other or similar tools to improve screening.

Annual wellness visits increase utilization of preventive services and may be a critical strategy for increasing underutilized services, such as lung cancer screening. Nonetheless, the cost-effectiveness and long-term effects of these visits warrant further study as Medicare continues to grow while facing capacity (and financial) constraints.

Conclusion

Performing routine cancer screening is an evidence-based practice for diagnosing earlier-stage, more treatable cancers. The Medicare AWV effectively increases breast cancer screening and may lead to more timely screening. Continued investment in the AWV supports breast cancer screening completion by women who are most likely to benefit, thus reducing the risk of over-screening and over-diagnosis.

Supplementary Material

Supplemental Data File (.doc, .tif, pdf, etc.)

Table S1. Logistic Regression Model for Outcomes Measured within 6 Months of AWV

Table S2. Outcomes for Sensitivity Analyses, Alternative Sample Specifications

Table S3. Outcomes for Sensitivity Analyses, Alternative Model Specifications