Individual-level trends in mammography screening over 14 years: results from an all-payer statewide claims database

Abstract

Background:

Our objective was to characterize individual-level screening patterns during a fourteen-year period spanning changing breast cancer screening guidelines and the COVID-19 pandemic.

Methods:

We conducted a retrospective analysis of mammography screening in women ages 40 and older using 2010–2023 data from the Vermont Health Care Uniform Reporting and Evaluation System, a statewide all-payer insurance claims database.

Results:

Biennial mammography screening adherence declined between 2010–2023 by 20.7 and 6.7 percentage points for women aged 40–49 and 50–74 years, respectively. Younger birth cohorts consistently had lower screening rates at any given age compared to older cohorts. Cohorts under age 50 during the COVID-19 pandemic have had the largest post-pandemic recoveries, with post-pandemic screening rates exceeding pre-pandemic screening adherence.

Conclusions:

While mammography screening rates have been declining both between and within birth-cohorts, pandemic disruptions have narrowed the gaps resulting from these trends.

Impact:

Sustaining post-pandemic screening momentum may be critical to reversing long-term declines in adherence.

INTRODUCTION

Utilization of mammography screening in the US has fluctuated over the past two decades in response to influences such as the COVID-19 pandemic¹ and changes to breast cancer screening guidelines.² While survey-based studies generally report high adherence (70–80%) to biennial mammography among women aged 50–74 years throughout the entire period,^3–5 more rigorous studies relying on medical records data or insurance claims data have reported lower adherence (50–65%),^2,6–8 with more dramatic and sustained declines in response to the COVID-19 pandemic^1,9–14 and the 2009 changes to the US Preventive Services Task Force (USPSTF) breast cancer screening recommendations.^2,8,15–21 While not uniformly observed,⁶ a number of such studies have indicated a long-term decline in mammography screening adherence in the US since 2009^7,13,15,21 and reported contemporary screening rates that remain lower than pre-pandemic rates.^12–14 The USPSTF updated mammography screening guidelines in 2024 to recommend biennial screening for women ages 40–74,²² meaning that routine screening is recommended for women ages 40–49 for the first time since 2009. Understanding past declines in screening may be important for developing strategies for re-establishing screening in women ages 40–49.

Our understanding of screening utilization in the US is generally informed by cross-sectional studies of population mammography adherence, which report age-specific screening utilization rates by calendar year. Cross-sectional analyses may mask cohort effects and cannot capture within-woman adherence patterns (i.e., how individual women’s screening practices change over time). Few studies to date have investigated within-woman longitudinal mammography screening patterns beyond short-term screening behavior (e.g., 2–3 consecutive rounds).^13,23–25 A few studies have used Medicaid and/or Medicare insurance claims databases to characterize the frequency of adherence over longer term periods (e.g., 10 years) and identified predictors of maintaining mammography adherence.^26,27 Others have characterized maintenance of screening adherence within specific intervention programs.^28–30 We are aware of no studies that have examined variation in adherence patterns over time in the US within birth cohorts.

We previously reported a sustained decline in age-specific mammography screening adherence rates in Vermont over the fourteen years using medical records data from a statewide registry.¹³ We sought to generate new insights regarding mammography adherence by characterizing individual-level screening patterns, investigating variation in screening patterns across age cohorts, and evaluating predictors of adherence such as primary care engagement, health insurance coverage, and rural residence within a longitudinal cohort. We used data from the Vermont statewide all-payer claims database from a fourteen-year period (2010–2023) spanning changing breast cancer screening guidelines and the COVID-19 pandemic.

MATERIALS AND METHODS

Study setting and data sources

Our analysis uses data from the Vermont Health Care Uniform Reporting and Evaluation System (VHCURES), Vermont’s all-payer claims database. VHCURES includes medical and pharmacy claims from all of the state’s Medicare and Medicaid enrollees, as well as from 60 percent of commercial enrollees.³¹ Self-insured employer plans governed by the Employee Retirement Income Security Act (ERISA) are, as of 2016, not required to contribute claims to the database;³² however, some large employers contribute voluntarily. The study was approved by the University of Vermont IRB on August 14, 2023.

Participants

Our analysis considered two populations: an overall population and a population continuously enrolled in insurance reporting to VHCURES. The overall population included observations for women in years between 2010 and 2023 in which they were at least 40 years old (including if they turned 40 during the year), and were enrolled for all 12 months in the year. Our continuously enrolled cohort was a subset of the overall population who met an additional criterion: they were continuously enrolled in an insurance plan reporting to VHCURES every year 2010–2023 or turned 40 after 2010 and were continuously enrolled for every year after that. By definition, women who died before 2023 were not included in the continuously enrolled cohort as they would not appear in VHCURES as insurance plan enrollees for the entire 2010–2023 period. Women with zip codes outside of Vermont were excluded from both populations.

Measures, Definitions, Outcomes

Our primary outcome was adherence with biennial mammography screening, which we defined as having a screening mammogram in the current or prior calendar year. We did not explicitly consider screening interval in our definition of screening adherence; women were considered adherent to biennial screening each calendar year in which they were screened as well as the subsequent calendar year. We did not distinguish between women screened at annual vs. biennial intervals in our analysis, as our analysis was framed around USPSTF screening guidelines.

Screening mammograms were identified using the Current Procedural Terminology (CPT) codes: 77063, 77067, 76092, 77057, G0202 and revenue code 0403. Our analyses examined the relationship between biennial screening and age, age category, year, time period relative to the COVID-19 pandemic, rurality, insurance, and primary care engagement. Age categories were 40–49, 50–74, and 75 and older. We considered three time periods in our analysis: 2010–2019 was the pre-pandemic period, 2020–2021 was the pandemic period (reflecting the years of most intense COVID-19 social distancing), and 2022–2023 was the post-pandemic period. We defined rurality using the Rural-Urban Commuting Area codes,³³ with categories collapsed to create a dichotomous rural-urban variable. We constructed mutually exclusive insurance enrollment categories with the following hierarchy for individuals with more than one type of insurance in a 12-month period: Medicare, commercial, Medicaid. We applied the same hierarchy both when individuals had certain forms of insurance for only part of the year and in when individuals had multiple forms of insurance at the same time. Note that only individuals with some form of insurance for all 12 months of the year could included in our study populations as uninsured individuals are not included in VHCURES.

Primary care engagement was defined as having a preventive, annual wellness, or outpatient evaluation and management visit in a given year. Procedure codes used to identify primary care were 99495, 99496, 99490, 99439, 99201–99215, 99381–99389, G0402, G0438, and G0439.

Statistical Analysis

We first examined trends in biennial mammography screening by age group among our entire population and in our continuously enrolled cohort. We considered the age groups 40–49, 50–74, and 75+ based on USPSTF breast cancer screening guidelines and guideline changes. Routine screening has consistently been recommended for women ages 50–74 since 2002.^22,34–36 Routine screening was not recommended for women ages 40–49 between 2009 and 2023 and routine screening has not been recommended for women ages 75+ since 2009^22,35,36 Next, we examined screening rate trends by age in our continuously enrolled population, sub-divided by age cohort. Age cohorts were defined by 5-year birth year groups with the youngest cohort born in 1979–1983 and the oldest born in 1934–1938. We used logistic regression with individual fixed effects to model the within-individual relationship between biennial breast cancer screening and age, year, insurance, rurality, and primary care engagement. The relationship between screening and age was modeled as a piecewise linear function that allowed for both discontinuities and changes in slope as age category changed. Analyses were conducted in R (version 4.5.0).

RESULTS

The study included 244,075 women in the full cohort and 71,640 women in the continuously enrolled cohort (Table 1). The distribution of age at cohort entry was slightly older for the full cohort compared to the continuously enrolled cohort (mean, 56.7 vs. 55.6 years old). Individuals in our full cohort appear in the data for an average span of 8.7 years with 0.5 years of missing data within that span. About a third of women in both the full and continuously enrolled cohorts had Medicare at cohort entry. Medicaid was more common in the full cohort (16.3 vs. 11.3 percent at entry) likely because Medicaid enrollees are more likely to experience loss or change in health insurance, which would disrupt enrollment in the VHCURES database. Those in the continuously enrolled cohort were more likely to have had a primary care visit in the entry year. Rurality at cohort entry was similar in both cohorts, with just over two thirds of each classified as rural, similar to estimates from the US Census.³⁷

Table 1.

Characteristics of the study population.

	Full cohort (N=244,075)	Continuously enrolled cohort† (N=71,640)
Age at entry, mean (SD)	56.7 (13.4)	55.6 (10.8)
Age group at entry
40–49, percent (N)	35.8 (87,333)	32.2 (23,037)
50–74, percent (N)	52.3 (127,710)	64.4 (46,118)
75+, percent (N)	11.9 (29,032)	3.5 (2,485)
Span of years, mean (SD)	8.7 (4.8)	12.7 (3.3)
Missing years, mean (SD)	0.5 (1.4)	0.0 (0.0)
Entry period, percent (N)
2010–2013	71.2 (173,681)	87.1 (62,391)
2014–2018	14.2 (34,709)	4.6 (3,320)
2019–2023	14.6 (35,685)	8.3 (5,929)
Exit period, percent (N)
2010–2013	10.3 (25,198)	0 (0)
2014–2018	15.7 (38,383)	0 (0)
2019–2023	74.0 (180,494)	100.0 (71,640)
Insurance at entry, percent (N)
Medicare	32.9 (80,485)	33.4 (23,971)
Medicaid	16.3 (39,853)	11.3 (8,146)
Commercial	50.7 (123,737)	55.2 (39,523)
Primary care visit in entry year, percent (N)	79.8 (194,849)	86.8 (62,181)
Rurality in entry year, rural percent (N)	68.4 (167,114)	67.4 (48,280)

^†enrolled every year 2010–2023 OR turned 40 after 2010 and continuously enrolled for all subsequent years.

Figure 1 shows biennial mammography screening trends by age group for the full and continuously enrolled cohorts. All groups saw declining screening rates from 2010–2023 as well as a drop in screening in 2020–2021 during the COVID-19 pandemic with a post-pandemic rebound. For women 40–49, screening rates dropped 16.5 and 20.7 percentage points from 2010–2023 in the full and continuously enrolled cohorts, respectively. For women 50–74 these declines were 8.7 and 6.4 percentage points. As expected, screening rates were higher in the continuously enrolled cohort, as the continuously enrolled cohort excludes women who lack stable health insurance.

Figure 1.

Population level trends in biennial mammography screening adherence by year and age group. Pandemic years (2020–2021) are circled.

Screening trends were approximately parallel between the full cohort and continuously enrolled cohort for the 40–49 and 50–74 year old populations. Screening rates fell more quickly in the continuously enrolled 75+ population compared to the full 75+ population because the continuously enrolled population excluded individuals who died during the 2010–2023 time period. Thus, the age distribution of the continuously enrolled population was younger at study entry and increased annually, whereas the age distribution of the full population was relatively steady throughout the study period. Because screening rates decline with age after 75, screening rates declined markedly for the continuously enrolled population.

Figure 2 shows biennial screening rates by age for 10 cohorts based on birth year (5-year bins). In general, younger cohorts have lower screening rates at any given age compared to older cohorts. In addition, the dominant trend across all cohorts was to have decreasing rates of biennial screening over time until the COVID-19 pandemic. Cohorts under the age of 50 at the time of the COVID-19 pandemic appeared to have the largest post-pandemic recoveries with their highest rates of screening rates occurring post-pandemic. Those ages 50–74 at the time of the pandemic had screening rates that returned to pre-pandemic levels. For cohorts over the age of 75 at the time of the COVID-19 pandemic, screening rates continued to decline post-pandemic.

Figure 2.

Screening trends by 5-year age cohort for the continuously enrolled population. Screening rates are plotted as a function of each cohort’s mean age in each year; years are represented by the points’ colors. Pandemic years (2020 and 2021) are circled.

Table 2 shows the results from the fixed effect logistic regression model predicting biennial screening. There was no significant annual change in screening for women between the ages 40–49, followed by a discontinuous increase in screening rates as women entered the 50–74 year old age category (OR=1.19, 95% CI= [0.14–1.24]). Within this age group, screening rates declined modestly each year (OR=0.90, 95% CI= [0.90–0.91]). The rate of decline then increased substantially once women reached age 75 and older (annual change OR=0.50, 95% CI=[0.49–0.50]).

Table 2.

Factors associated with biennial mammography screening adherence in fixed-effects logistic regression.

	Odds Ratio (95% CI)	p-value
Annual change within age groups (slope)
40–49	1.00 (0.99–1.01)	0.83
50–74	0.90 (0.90–0.91)	<0.001
75+	0.50 (0.49–0.50)	<0.001
Age-group discontinuities (jumps)
50–74	1.19 (1.14–1.24)	<0.001
75+	1.06 (1.00–1.14)	0.06
Insurance category (ref. commercial)
Medicaid	0.29 (0.28–0.30)	<0.001
Medicare	1.05 (1.01–1.09)	0.006
Time period (ref. pre-pandemic)
Pandemic year (2020–2021)	0.90 (0.88–0.92)	<0.001
Post pandemic year (2022–2023)	1.71 (1.67–1.76)	<0.001
Rural	1.16 (1.07–1.26)	<0.001
Has PCP	7.77 (7.58–7.97)	<0.001

Women were substantially less likely to have a mammogram when they were on Medicaid compared to when they had commercial insurance (OR=0.29; CI=[0.28–0.30]). In contrast, Medicare was associated with higher screening rates compared to commercial insurance.

The pandemic was associated with lower screening rates compared to the pre-pandemic period and the post-pandemic period was associated with substantially higher screening rates (OR=1.71, 95%CI= [1.67–1.76]). Individuals living in rural areas were modestly more likely to be screened OR=1.16, 95% CI=[1.07,1.26]) and those with primary care visits were substantially more likely to be screened.

DISCUSSION

Our results provide new insights into population-level declines in mammography screening rates by characterizing trends within age-specific cohorts and within-individual screening changes among those with continuous health insurance during 2010–2023. We found that trends in age-specific mammography screening adherence were similar in the continuously enrolled cohort and the full study population, suggesting that analyses of the continuously enrolled cohort are likely representative of patterns in the broader population. The declining pattern of screening in both groups since 2010 was also consistent with that reported in our prior work evaluating mammography screening in Vermont using medical records data from a statewide breast imaging registry,¹³ and similar to declines reported in other study populations.^1,2,6–21

We found a strong birth cohort effect in which more recent cohorts were less likely to be adherent to screening at any given age compared to prior cohorts. Within birth cohorts, we observed a modest increase in screening adherence upon turning age 50 but a steady decline in screening with advancing age from 50 to 74. For example, the average woman born in 1953 had a screening adherence rate of 68% in 2023 at age 70, down from 71% in 2013 at age 60. The pandemic disrupted these trends with declines in screening in all cohorts in 2020–2021. Post pandemic recoveries in screening adherence varied across birth cohorts with younger cohorts reaching screening rates that exceed those from pre-pandemic levels.

The difference in screening rates by birth cohort may indicate a cultural shift across successive generations of women in attitudes towards mammography screening amid increasing media emphasis on the limitations of mammography and inconsistent screening recommendations issued by various professional group and organizations. Shortages of primary care providers may also contribute to the observed trends, as having a PCP was identified as a strong determinant of screening adherence. Other factors including insurance coverage and other economic factors may also be contributing to the observed trends. More research is needed to better understand these trends.

Women who turned age 40 during the pre-2010 era in which routine screening beginning at age 40 was recommended by all national guideline makers likely experienced more uniform promotion of screening. Women who turned age 40 after 2010 faced inconsistent recommendations across guideline makers and were likely exposed to more information regarding screening’s limitations, including false-positive rates and overdiagnosis. The observed birth cohort effects suggest that age-specific screening rates for older women in particular are likely to continue to decline in Vermont. Approximately 71% of 55-year old women born in 1957 were adherent to biennial screening, compared to only 63% of 55-year old women born in 1967. Given our finding that screening adherence also tends to decline with age after age 50, it is likely that younger birth cohorts will continue to have low screening rates in the absence of impactful interventions.

The within-individual decline in screening adherence with increasing age during ages 50–74 has not previously been reported as population level analyses do not distinguish between declining screening rates within and between cohorts. The decline in screening with age, observed here even in a population with continuous healthcare insurance coverage, suggests that efforts are needed to encourage women to continue adherence with mammography screening.

The COVID-19 pandemic period effect is evident in our population-level results and the birth year cohort trends in the observation of lower screening rates in 2020–2021. Our population-level analysis demonstrated that screening rates rebounded by 2023 to pre-pandemic levels. However, this varied across birth year cohorts. Younger cohorts had the strongest rebound in screening, well exceeding their screening rates prior to the pandemic, whereas older cohorts had incomplete rebounds. This suggests that efforts to increase initiation of screening or resume screening among younger women after the pandemic may have been more successful than efforts to resume screening among older women previously screened. Nevertheless, trend of lower screening rates in younger birth cohorts persists post-pandemic years, with the most recent cohort having substantially lower screening rates at age 40 than all prior cohorts.

We found that insurance type, urban/rural residence, and primary care provider engagement were associated with screening adherence. Our results are consistent with prior studies demonstrating that mammography screening rates are lower among women covered by Medicaid (or uninsured) compared to women with private insurance or Medicare coverage.³⁸ Prior studies have also demonstrated that attendance at annual wellness visits is associated with increased mammography screening utilization³⁹ and that having a personal health care provider is associated with increased mammography screening rates.⁴⁰ We observed that women with a primary care provider healthcare encounter were more than 7 times more likely to be adherent with mammography screening compared to women without an encounter. In the context of low population-level mammography screening rates, efforts to promote access to primary care and utilization of primary care services are potential strategies to bolster mammography screening.

Prior studies have demonstrated that 50.6% of women living in rural areas of the USA have limited access to mammography (<=20 minute drive time),⁴¹ and that geographic access to mammography is positively associated with mammography screening utilization.^41,42 However, we found that rural residence was associated with a 16% increase in mammography adherence compared to urban residence. This fixed effect estimate represents a within-individual effect; women who moved during the study period were more likely to be adherent to screening when living in a rural residence versus when they lived in an urban residence. Women who transition from urban to rural residence or vice-versa may be experiencing particular socioeconomic conditions (e.g., gain or loss of employment, wealth attainment) that influence their screening behaviors. This may explain why our results differ from cross-sectional studies that compare screening rates between urban and rural residents (the majority of whom were likely stable in their residence). Further research is needed to explore patterns in mammography screening according to rural/urban residence.

Limitations

Our analysis and conclusions are based on a population enrolled in insurance that reports to VHCURES, which excludes uninsured, self-pay, and many self-insured employer plans. As a result, analyses may not be representative of the entire state. Some of our analyses focused on a population continuously enrolled in insurance plans that report to VHCURES. This population is likely to have more stable insurance than others in the state. In addition, Vermont has a largely rural, predominantly white population, and VHCURES does not include data on race and ethnicity; therefore, conclusions may not generalize to the rest of the United States population. In our regression analysis, we examined the association of mammography adherence with having a primary care provider. With VHUCRES, we are unable to determine whether the provider recommended or discussed mammography; analyses using data that capture provider recommendations are needed to better understand the potential effects of primary care providers on mammography uptake.

Conclusions

Our findings contribute to the research on adherence to mammography screening in the context of changing guidelines and COVID-19 public health emergency. We observed both that women born more recently have lower tendency to receive biennial screenings than prior cohorts, and that screenings decline rapidly among older age women. At the same time, receipt of primary care is strongly associated with mammogram adherence in that year. While the COVID-19 pandemic led to immediate declines in mammography screening, post-pandemic rebounds were largest in younger cohorts, potentially disrupting the decade-long trend of declining screening rates among those newly eligible. These findings can be used to target outreach and prevention strategies to engage women more effectively in mammogram screening in the wake of recently expanded USPSTF screening eligibility criteria.