Rebecca A. Silliman, MD, PhD

 A false-positive mammogram can result in anxiety, distress, and increased perceptions of breast cancer risk, potentially changing how women utilize healthcare. This study examined whether having an abnormal mammogram, considered a proxy for elevated risk perception, was associated with greater future health service use (outpatient visits and referrals).

 A retrospective cohort study using electronic health record data, spanning 2008 to 2012, from Boston Medical Center, a safety-net hospital.

 We grouped 3920 women aged 40 to 75 years receiving primary care and who had a mammogram between 2010 and 2011 into 3 categories: false-positive mammogram at index date; previous false positive, but normal index mammogram; and no history of false-positive mammograms. We contrasted the longitudinal changes in outpatient visits and provider referrals, before versus after the index mammogram, between women with false-positive mammogram and those without using Poisson regression models with a difference-in-differences specification. Clinical, visit, and demographic data were obtained from the institutional clinical data warehouse.

 Adjusting for baseline differences in sociodemographic characteristics across risk groups and for secular changes between pre- and postindex periods, a current false-positive mammogram was associated with an 18% increase in overall outpatient visits (incidence rate ratio [IRR], 1.18; 95% CI, 1.07-1.51), but no corresponding increase in provider referrals (IRR, 1.15; 95% CI, 0.99&#8209;1.34), relative to never having a false positive. A previous false-positive mammogram had no associated change in outpatient utilization (IRR, 0.99; 95% CI, 0.91-1.07).

 Providers should discuss the implications of mammography findings at the time of screening to help mitigate potential detrimental effects and promote appropriate engagement in health services.

We used a difference-in-differences approach to measure changes in healthcare utilization associated with false-positive mammography. A current false positive was associated with an 18% increase in outpatient visits, but no increase in provider referrals, relative to women with no false positive.

Trends encouraging mammography adherence increase opportunities for false positives, which we have shown to impact short-term health service use.

Counseling about potential false positives and their meaning prior to screening may mitigate the anxiety experienced at the time of an abnormal mammogram.

Interventions to ensure appropriate utilization are needed to address women’s concerns while promoting evidence-based care.

Despite debate about the benefits and harms of screening mammography,

 it remains the best available tool to detect breast cancer early and prevent morbidity and mortality. As such, the majority of women participate in mammography screening. In 2013, the CDC estimated that 69.1% of women older than 50 years underwent mammography

 as recommended by the United States Preventive Services Task Force.

 Identified harms of mammography include a high prevalence of false positives, need for additional imaging and biopsies, potential overdiagnosis of breast cancer, radiation exposure, and masking bias in dense breast tissue.

 These issues have received extensive attention in the literature and are important concepts for women to understand when considering screening.

False-positive mammograms and the resulting psychosocial harms have garnered attention in recent years.

 Over a 10-year screening period, the cumulative probability of a false-positive mammogram is 42% to 61%, depending on screening interval.

 Anxiety and distress are common among those experiencing a false-positive mammogram, with effects persisting in the months following the test.

 The prevalence of false positives and documented negative psychosocial outcomes are, in part, the basis for not routinely screening women younger than 50 years, who are more frequently affected by these outcomes.

 Guidelines instead promote shared decision making for these women, although this ideal is not often realized in practice.

 Research has shown that women are not often aware of the harms of mammography, including the potential for false-positive findings and future testing.

Having a false-positive mammogram is further associated with increases in individuals’ risk perceptions, despite no actual change in risk or health threat,

 that can persist long after the abnormal test and affect health service use.

 Although there is interest in determining the multidimensional benefits and harms associated with screening and risk communication, the effects of designating individuals as “at risk” on healthcare utilization measures beyond the downstream diagnostic workup are understudied. This is a critical gap in knowledge at a time when improving cancer screening rates is emphasized through payment schemes intended to improve quality of care.

 Yet the process of undergoing a mammogram and receiving results may change the way a woman conceptualizes her health and engages with the medical system. Understanding the impact of these tests on how women utilize healthcare services over time is needed to improve the quality of care.

This study examined whether having an abnormal mammogram, considered a proxy for elevated risk perception,

 was associated with greater future health service use. Previous studies have examined breast-specific health service use in relation to other clinical markers of risk status, such as genetic mutations or family history.

 This study expanded the measure of healthcare utilization beyond breast-specific services among a population of women undergoing mammography. We hypothesized that women who experienced a recent or previous false positive would have higher postmammogram visit and referral rates relative to women without false-positive mammograms. Although we were unable to fully characterize the purpose and content of these outpatient visits and referrals, this study sought to deliver an initial assessment of healthcare utilization changes following a false-positive test result, establishing the basis for a detailed exploration of such visits and assessment of whether additional service use represents value-based care.

This retrospective cohort study examined whether experiencing a false-positive mammogram recently or in the past was associated with healthcare utilization by comparing rates of outpatient visits and referrals in the 2 years prior to an index mammogram with 1 year following. This study was determined to be exempt by the Boston University Medical Campus Institutional Review Board.

Utilization data came from Boston Medical Center (BMC), an urban safety-net hospital. Included women: 1) were aged 40 to 75 years, 2) received primary care at BMC, 3) had a screening mammogram at BMC in 2010 or 2011 that did not result in a cancer diagnosis, and 4) had their first mammogram at 40 years or older. Women receiving primary care at BMC were identified by a documented visit with a designated provider in the preceding 2 years. Prior work has demonstrated that these patients use the hospital system as their main source of care for all conditions, including emergency and specialist care.

 The first mammogram during 2010 or 2011 was defined as the index mammogram. Women with a prior cancer or cancer diagnosed during the study period or those who died during the study were excluded.

Administrative and clinical data were obtained through the clinical data warehouse, a comprehensive database that aggregates data from multiple electronic hospital sources. The use of retrospective clinical data provided a measure of actual utilization rates, eliminating bias from self-reported data. Data collected spanned calendar years 2008 to 2012.

We used a difference-in-differences (DID) approach, which measures the difference in utilization before and after the index mammogram for women who experienced a false-positive mammogram and contrasts this difference with the corresponding difference for women who experienced no false-positive mammogram.

 This approach adjusts for pre-index differences in utilization between the 2 groups and secular changes in utilization between pre-index and postindex periods. We defined the 2-year period before and 1-year period after the index mammogram visit as the pre-index and postindex periods, respectively. This provided a stable measure of utilization prior to the index mammogram and allowed for assessment of bias that may have threatened internal validity.

An abnormal mammogram was defined using criteria of the 

Breast Imaging Reporting and Data System (BI-RADS) Atlas

 1 (normal) or 2 (benign) classifications. 

 scores of 0 (incomplete), 3 (probably benign), 4 (suspicious), or 5 (highly suggestive of malignancy) were included in the abnormal mammogram group, as they required a follow-up or further testing. False positives were defined as those abnormal findings that did not result in a cancer diagnosis as assessed by the Tumor Registry.

 results were used to create 3 groups of women that reflected levels of risk designation

: 1) current false positive (high risk): women experiencing a false-positive mammogram at the index date and indicative of high risk perception; 2) previous false positive (intermediate risk): women who had a normal mammogram result on the index date with a false-positive result in the past; and 3) no false positives (low risk): women who had a normal index mammogram result and no previous false-positive results. We distinguished current from previous false-positive mammograms to test if changes in utilization following a current false positive would persist over time.

We examined 2 utilization indicators as outcomes: referrals and outpatient visits. The number of referrals measured provider-initiated orders for health services and was totaled for each time period. Referrals were captured as “orders,” and all types (ie, specialty visits, laboratory) were compiled to measure the total count documented in the medical record over each time frame. Counts of all outpatient visits attended for each 12-month period were also measured and included all primary care, specialty, emergency department, laboratory, and procedure visits. Diagnostic tests included in the follow-up for an abnormal mammogram were excluded.

Demographic covariates included age, race/ethnicity, insurance, education, and primary language. Baseline comorbidities were measured by identifiying all unique diagnosis codes (

International Classification of Diseases, Ninth Revision, Clinical Modification

]) in the pre-index period and grouping them into 30 categories using the Elixhauser classification.

 A dichotomous indicator variable for time (pre-index vs post index) was created and later interacted with the 3 groups to assess differences over time by group.

Comparison of Pre-Index Longitudinal Trends Between Comparison Cohorts

A critical assumption underlying the appropriateness of the DID design was that the longitudinal trends in utilization for the 2 comparison cohorts, women with a current or previous false-positive mammogram versus those without, diverged from a parallel trajectory only after the false-positive mammogram event; that is, prior to this event, the longitudinal trends between the 2 groups should have followed similar trends (ie, were parallel).

 We tested this “parallel trends” test by using only pre-index event data, dividing the 2 pre-index years of utilization into 2 single-year utilization periods; estimated a similarly specified DID model limited to only the pre-index data; and tested if the changes in utilization between the 2 years were similar between the 2 comparison groups (current vs never; previous vs never). Confirmation of parallel trends enhanced the ability of the main DID model (with complete data) to attribute postindex utilization changes to false-positive mammograms.

Another potential source of confounding was systematic differences in the prevalence of the observed demographic and socioeconomic covariates and pre-index utilization between the groups of women compared. One approach to adjust for potential imbalance was the use of propensity score weighting known as standardized mortality ratio weighting (SMRW).

 In this 2-step method, we first estimated a logistic regression of the indicator of false-positive mammogram (0/1) on age, race, language, education, insurance, Elixhauser categories, and pre-index visits (or referrals) to obtain the predicted probability of false-positive mammogram (propensity score); in the second step, we estimated the main regression model using SMRW, wherein women with false-positive mammograms were assigned a weight of 1 and women without false-positive mammograms were assigned a weight equal to the ratio of the propensity score to 1 minus the propensity score.

 The propensity score weighting was used to produce comparable groups with parallel pre-visit trends.

We performed descriptive analyses and examined patient characteristics by group. Our core analysis estimated the change in each utilization outcome associated with a current (at index mammogram) or previous false-positive mammogram (prior to index, with normal index mammogram) using a weighted Poisson regression model with a DID specification.

 Multivariable models adjusted for covariates, including comorbidities present at the index mammogram. Two models for each outcome were constructed: one to compare the current versus never false-positive groups and the other to compare the previous versus never false positive-groups. Thus, 4 final DID models were run to account for selection of appropriate comparison groups via propensity score weighting.

The primary independent variable of interest in the DID specification was the interaction between time and having a false-positive mammogram result, which compared the average difference in utilization in the pre- versus postindex mammogram periods for the current and previous false-positive mammogram groups compared with those with no false-positive mammograms. We hypothesized that a current false-positive would result in higher rates of referrals and outpatient visits in the follow-up period, as women might feel that they are at higher risk.

To perform the parallel trends test described above, we estimated an identically specified weighted Poisson DID model, with the data limited to the pre-index period and with utilization in the 2 pre-index years as the 2 periods compared. All analyses were perfomed in SAS version 9.3 (SAS Institute Inc; Cary, North Carolina). Estimates are reported as incidence rate ratios (IRRs) with 95% CIs. Statistical significance was set at α = .05 for all models.

False-positive rates may be confounded by age due to differences in the properties of mammography and breast tissue structure.

 We conducted a sensitivity analysis to assess the influence of age, stratifying the sample into 2 groups (40-49 and 50-75 years). As Poisson models may not capture overdispersion in outcomes, we performed estimation using negative binomial distribution models.

The sample included 3920 women who met inclusion criteria (

). Almost half (42%) were women for whom routine screening was universally recommended (aged 50-64 years), 52% were black, and 61% spoke English as their primary language. Most had public insurance (55%), and 41% had less than a high school education. Seventy-six percent of women had at least 1 Elixhauser-defined comorbid condition. The previous false-positive group had the highest prevalence of 1 or more comorbidities at 80.3% versus 75.6% in the no false-positive group and 71.4% in the current false-positive group (details in 

Associations Between Demographics and Risk Groups

Sample characteristics identified the demographic correlates of false-positive mammograms. The current false-positive group was younger than the other 2 groups (

= .02) was the only other characteristic significantly associated with risk group. Unadjusted visit rates over time by risk group are displayed in 

. Testing differences in these trends across the pre-index period using the SMRW propensity score method resulted in nonsignificant associations, suggesting that the parallel trends assumption was satisfied.

Changes in Utilization Associated With a False-Positive Mammogram

Adjusting for baseline differences in socio&shy;demographic characteristics across groups and for secular changes between the pre- and postindex periods, a current false-positive mammogram was associated with an 18% increase in overall outpatient visits (IRR, 1.18; 95% CI, 1.07-1.51) (

). In contrast, a previous false-positive mammogram had no associated change in utilization (IRR, 0.99; 95% CI, 0.91-1.07). Educational achievement, insurance coverage type, and baseline comorbidities were associated with systematic differences in outpatient visits in the pre- and postindex periods.

Contrary to the hypothesis, women with current false positives saw no increase in referral rates (IRR, 1.15; 95% CI, 0.99-1.34). Those with previous false positives also did not have a significant change in referral rates (IRR, 0.99; 95% CI, 0.83-1.20). Although point estimates for referrals were similar to the visit outcome, differing baseline outcome rates resulted in varying magnitudes, affecting the significance of the association. Nonwhite race, having public insurance, and baseline comorbidities were associated with higher referral rates.

A stratified analysis by age was conducted to account for the lower sensitivity and specificity of mammography for younger women

). The increase in visits associated with a false positive appeared to be driven by women younger than 50 years, with no significant change in utilization for older women (IRR, 1.13; 95% CI, 0.97-1.31); however, the general trend was similar. Stratifying by age, older women (aged 50-74 years) with current false positives received significantly fewer referrals compared with those without false positives (IRR, 0.85; 95% CI, 0.79-0.92), whereas there was no significant difference in referral rates across false-positive groups among younger women (aged 40-49 years). Estimates of change for the current and previous false-positive groups were robust when re-estimating the models using a negative binomial model.

Having a false-positive mammogram was associated with significantly increased rates of outpatient visits in the year following the test, suggesting that utilization may be driven by patients seeking care, but is likely not generated by the health system as there was no corresponding increase in referrals for care. Increased utilization dissipated over time: We found comparable visit and referral rates between those with a previous false positive and a recent normal mammogram and those with no false-positive mammograms.

One striking finding from this study was the marked increase in outpatient visits following an abnormal mammogram. Although increased breast-specific utilization is expected in order to rule out a cancer diagnosis,

 the reasons for increases in other types of care are less obvious. Our findings could indicate that women may conceptualize themselves as less healthy than prior to their false-positive mammogram

 or that they may experience increased anxiety, leading to additional visits that increase reassurance through monitoring. Prior study findings have confirmed that increases in utilization may be related to the negative psychosocial outcomes of having a false-positive mammogram, showing increases in mental health services following the receipt of an abnormal mammogram.

Undergoing imaging and receiving a normal result has been characterized as a “gift of knowing,” providing reassurance to those who engage in screening.

 This reassurance may be driving decreases in utilization in the previous false-positive group. This group was possibly reassured of good health with the normal mammogram result, thus removing the impetus to continue higher rates of utilization. The increase in immediate post—false-positive visits suggests that communication at the time of mammogram or in the immediate postmammogram period may be required to mitigate short-term anxiety and increased utilization. This may be particularly true for underserved populations similar to those studied here: We found that women with a high school education and black women had significantly more utilization associated with a false positive. This finding is similar to others that have noted that minority and low-literacy women experience higher distress and lower understanding when presented with abnormal mammogram results relative to white and high-literacy women.

 Interventions targeted toward these populations, such as patient navigation, have been shown to decrease anxiety following a false positive,

 but navigator effects on utilization beyond the diagnostic follow-up are limited. Prospectively understanding the role of anxiety generated by a false-positive test in subsequent healthcare utilization is required to tailor interventions to promote appropriate, high-quality care to women undergoing mammography screening.

Articles

This study demonstrated that a false-positive mammogram was associated with increases in outpatient visits, but not provider referrals, for 1 year post mammogram.

False-Positive Mammography and Its Association With Health Service Use

 To evaluate cardiovascular disease (CVD) risk factors in older breast cancer survivors compared with a group of women without breast cancer.

The retrospective study included (1) women aged 65 or more years who were initially diagnosed with stage I or II breast cancer from 1990 to 1994 in 6 US health plans and who survived at least 5 years post-diagnosis (cases) and (2) a matched comparison group. They were followed for a maximum of 15 years.

 Data sources included medical charts and electronic health records. Cases (n = 1361) were matched on age, health plan site, and enrollment year to women in the comparison group (n = 1361). Subjects were followed to the first CVD outcome, health plan disenrollment, death, or study end. We compared rates of CVD in these 2 groups and used Cox proportional hazard models to estimate the hazard ratio (HR), considering body mass index, smoking history, diabetes, and hypertension.

 The strongest predictors of CVD were smoking history (HR = 1.29; 95% confidence interval [CI], 1.15-1.46), diabetes (HR = 1.72; 95% CI, 1.48-1.99), and hypertension (HR = 1.48; 95% CI, 1.31-1.67) rather than breast cancer case-comparison status (HR = 0.97; 95% CI, 0.87-1.09).

Results suggest that long-term prognosis in breast cancer patients is affected by management of preexisting conditions. Assessment of comorbid conditions and effective management of diabetes and hypertension in older breast cancer survivors may lead to longer overall survival.

Our results suggest that older long-term breast cancer survivors (initially diagnosed with early-stage disease) have a cardiovascular disease (CVD) risk similar to that of otherwise healthy women of comparable ages.

The established risk factors (very old age, smoking history, diabetes, and hypertension) were more predictive of CVD risk than breast cancer history status.

Long-term prognosis in older breast cancer patients is affected by management of preexisting conditions, and these may be best managed by primary care providers.

Management of comorbidities in survivors should not be different from that in the general population of older patients.

More than half of the 2.6 million breast cancer survivors living in the United States are over age 65 years,

and the fraction of older people with cancer is growing, partly attributable to the success of cancer screening and treatment. Consequently, the number of older cancer survivors at risk of developing other age-related conditions such as cardiovascular disease (CVD) is increasing. Many of these older breast cancer patients also have comorbid conditions or other CVD risk factors.

 Moreover, CVD is the leading cause of death in breast cancer survivors.

 As comorbidities impact prognosis and cardiovascular outcomes in breast cancer patients, the role of primary care physicians in the care of survivors is expanding to manage these preexisting conditions.

Despite CVD being the leading cause of morbidity in older breast cancer survivors, very few studies have examined CVD risk factors in such women, whether these factors differ from those in women in the general population, and the long-term impact of these risk factors on CVD outcomes.

 For example, prior studies on CVD risk in older cancer survivors were limited by cross-sectional designs; few included information on health status prior to cancer diagnosis; and even fewer included data from comparison subjects without a cancer history.

Given these limitations, it is unclear whether there is excess risk of CVD among breast cancer survivors.

Examining CVD risk poses a challenge, as long-term observation periods are required. Further, CVD is more common in older adults in general and especially in those who have established risk factors other than cancer treatments. Because few older breast cancer survivors are treated with chemotherapy,

 particularly those agents known to be cardiotoxic, examining the impact of comorbidities on CVD risk is crucial. Therefore, a well-characterized comparison group with long follow-up is essential to determine whether there truly is excess morbidity in older women treated for breast cancer.

The purpose of this investigation was to determine whether incident CVD was greater in a group of older breast cancer survivors versus a cancer-free comparison group, and if the excess risk could be attributed to differences in comorbid conditions. To this end, we compared incident CVD in the 2 groups over a 15-year follow-up period, incorporating baseline risk factors such as race/ethnicity, body mass index (BMI), smoking history, diabetes, and hypertension.

We identified women 65 years or older who were diagnosed with early-stage breast cancer (American Joint Commission on Cancer  TNM stage I, IIA, or IIB) from January 1, 1990, through December 31, 1994, who survived at least 5 years after the initial breast cancer diagnosis. We selected 5-year survivors because this time period is most often used as a benchmark to define recovery.

 These women were participants in the BOWI study.

 Briefly, the BOWI multisite cohort study is a 10-year longitudinal study focusing on the effectiveness of treatment for breast cancer. Women in the BOWI cohort were identified through Cancer Research Network (CRN) managed care systems: Group Health Cooperative, Seattle, Washington; Kaiser Permanente, Southern California;

Lovelace Health System, New Mexico; Henry Ford Hospital and Health System, Detroit, Michigan; Health Partners, Minnesota; and Fallon Community Health Plan, Massachusetts. 

 These CRN sites were selected to achieve diversity in geography, system size, and patient populations.

The BOWII study extended data collection through 5 additional years of follow-up on the BOWI cohort and added a comparison group. The eligible BOWII case group for this analysis consisted of 1361 five-year breast cancer survivors. Comparison women were selected from the source population of each health plan. Comparisons included women who were cancer free at the time of the case’s year of diagnosis, and frequency matched (1:1) on age, health plan site, and enrollment year. These potential confounders were selected as matching variables because they are strongly associated either with survival or with treatments. To be eligible, comparison women also had to survive 5 years after enrollment into the study cohort. The final cohort consisted of 2722 women (1361 matched pairs). Women were followed from 5 years after the index date (breast cancer diagnosis date or matched enrollment  date) until first CVD event, disenrollment from the health plan, loss to clinical follow-up, death, or completion of 15 years of follow-up  (up to December 31, 2009), whichever occurred first. The protocol for this study was reviewed and approved by the institutional review board at each participating CRN site.

Data on CVD outcomes, demographics, comorbidities, and other covariates were ascertained primarily from the women’s medical  records and were supplemented with electronic health records. Standardized medical record reviews were conducted at each site by trained medical record abstractors. A detailed description of the data collection system and the training procedures implemented to standardize data collection across sites has been published elsewhere.

 Mortality data (date of death and whether  the cause was related to breast cancer) were collected using the National Death Index. We used mortality information for censoring in the analysis.

 Study outcomes included the following CVD events and were examined as a single binary composite outcome (presence or absence of any event): myocardial infarction, congestive heart failure, coronary artery disease, arrhythmias, and cerebrovascular disease. The CVD events were identified by the first occurrence of a diagnosis in the women’s medical charts during the study follow-up period. Women with a diagnosis for multiple events on the same day were assigned a single outcome using a priority scale based on importance per the recommendation of one of the study clinicians (RAS): (1) myocardial infarction, (2) coronary artery disease, (3) cerebrovascular disease, (4) arrhythmia, and (5) congestive heart failure.

Demographics and Cardiovascular Disease Risk Factors. 

We gathered information on date of birth, race, and ethnicity. We also collected information on diabetes, hypertension, smoking status, and BMI for both groups. Information closest to time of entry into the cohort was used in the analyses.

Differences in demographic characteristics and CVD risk factors between case and comparison women were first examined by comparing frequency distributions (

 or Fisher exact tests). Cox proportional hazards models were used to estimate the hazard ratio (HR) of the association of CVD composite outcome with case-comparison status. Because we examined  only the first CVD outcome (and not multiple occurrences), and as this model is the standard approach for analyzing cohort data, we used the Cox model.

We tested the proportional hazards assumption using Schoenfeld residuals.

 The subjects’ entry into the analysis corresponded to 5 years after the initial breast cancer diagnosis for the cases and matched enrollment date for the comparisons. We examined the association with 2 models: a parsimonious model adjusted for the matching factors, age, and site, and another model that included hypertension, diabetes, smoking history, BMI, and race/ethnicity. Life table analysis was used to compare CVD incidence by case-comparison status.

Demographic characteristics, comorbidities, and CVD risk factors for cases and comparisons are displayed in 

. Although the majority of women in both groups were white non-Hispanic (81.9% and 84.3% for cases and comparisons, respectively), the race/ethnicity distribution was similar in both groups (

 = .34). Case patients were more likely to be obese (BMI >30 kg/m

 .005) than comparison women. There were no significant differences observed in smoking history and diabetes between the 2 groups.

displays follow-up characteristics of the case and comparison groups. The mean follow-up time among cases was 5.0 years (1828 days) after entry into the cohort compared with 5.3 years (1942 days) among comparisons. Nearly 20% of women in both groups completed 15 years of follow-up (17.4% and 19.7% for cases and comparisons, respectively). Nearly half of the women in both groups experienced a CVD event during the follow-up period, with a slightly higher proportion of comparisons experiencing a CVD event (47.7%) than cases (45.3%). The fraction of deaths was nearly 2- greater in cases (21.7%) than in comparisons (12.1%). As expected, there was a greater risk of death due to breast cancer among cases. Nearly one-third died of breast cancer in the  case group (97/295) versus 1% in the comparison group (2/165).

Of the 2722 total women in the entire cohort, 1266 (46.5%) experienced a CVD event. Of the 1266 women, 740 experienced 1 of the 5 CVD events constituting the composite outcome and 526 experienced 2 or more CVD events during follow-up (data not shown). 

shows the incidence of first CVD event during the follow-up period (maximum 10 years) for the composite outcome as well  as the individual conditions. No overall differences were found with the composite CVD outcome (

= .40) nor with the individual outcomes in terms of risk. The rates of the composite CVD outcome and the individual outcomes were also similar in cases and comparisons (83.3/1000 person-years vs 82.90/1000 person years, respectively). We also examined the risk of CVD in the 2 groups  over time. The life table curves for CVD events by case-comparison status showed no difference in overall survival probabilities  between the 2 groups (data not shown). 

 presents the adjusted HR for the association between CVD and case-comparison status. Multivariable models were initially adjusted for matching factors (model 1 included age at diagnosis and health plan site). Cases were no more likely to experience a CVD event than comparisons (HR = 1.00; 95% confidence interval [CI], 0.90-1.12). In model 2, we further examined race/ethnicity, BMI, smoking history, diabetes, and hypertension. Interestingly, the strongest predictors of CVD were smoking history, diabetes, and hypertension rather than breast cancer case-comparison status (HR = 0.97; 95% CI, 0.87-1.09). For example, smoking history  increased CVD risk by nearly 30% (HR = 1.29; 95% CI, 1.15-1.46). Women with diabetes were 72% more likely to develop CVD (HR = 1.72; 95% CI, 1.48-1.99), and those with hypertension were 48% more likely to develop CVD after accounting for case-comparison status (HR = 1.48; 95% CI, 1.31-1.67). Also, white women were 50% (HR = 1.51; 95% CI, 1.28-1.77) more likely to develop CVD than minority women. As expected, CVD risk increased with older age (

for trend <.10). We did not find a trend with BMI categories, possibly due to missing values. We repeated the Cox regression analysis excluding the 121 survivors exposed to chemotherapy (and their matched comparisons). As the subset results were similar, we reported the HRs of the full cohort.

Our goal was to determine whether older cancer survivors are at increased risk of CVD events compared with the general population. In this population-based study of more than 2700 breast cancer survivors 65 years and older and age-matched comparison women, we found that the risk of CVD was similar in the 2 groups. The incidences of the composite CVD outcome were the same, as were the incidences of the individual conditions.

Very few studies have compared CVD risk factors and outcomes of cancer survivors with those of a general population of women to determine whether there truly is an excess risk among cancer survivors. Our results were consistent with the few studies that have included a comparison group. For example, a Dutch study examined more than 4000 breast cancer survivors and compared their CVD incidence with that of a general female population, and determined that radiation to the breast alone did not increase CVD risk.

 Similar to our present study, breast cancer survivors followed in Ontario, Canada, also did not have an increased risk of  myocardial infarction compared with age-matched women from the general population (other CVD outcomes were not studied).

 However, in these 2 studies, comorbid conditions were not examined. The remaining studies examined CVD outcomes within a group of breast cancer survivors and focused on cancer treatments, whereas our aim was to examine risk factors common to both groups. Further, few of these studies included a cancer-free comparison group and, again, did not examine the impact of comorbid conditions or lacked information on lifestyle factors (smoking history, BMI).

Our study has a number of strengths. A major advantage of this study is that data elements were mainly obtained from medical record reviews, not solely from electronic clinical data.

 We carefully identified a group of comparison women (without breast cancer) to determine whether there truly was excess CVD morbidity in the cancer case group. In addition, mean follow-up time was similar in both groups, thereby reducing the potential for selection bias. Importantly, the case and comparison groups were followed a maximum of 15 years; manifestation of CVD requires more than a decade of follow-up.

Certain limitations of this study must be considered. Although we captured low-density lipoprotein (LDL) cholesterol and glycated  hemoglobin (A1C) levels from health plan laboratory databases, very few women had these values (overall, about 70% of women had missing LDL levels and 85% had missing A1C levels; data not shown). Therefore, we did not include these variables in the analyses. However, the percentage of missing values was nondifferential (71% of the cases had missing LDL cholesterol levels vs 67% of  comparisons; 83% of cases had missing A1C levels vs 85% of comparisons), so any bias resulting from this exclusion was likely to be minimal. Similarly, many women were missing BMI, but again, the percentage of missing values was nondifferential (roughly 20% in both groups). In addition, all women in this study were selected  from integrated healthcare delivery systems, so their experiences may not reflect those of others in the general population. However, we selected a nationally representative group of health plans, and the race/ethnic distribution of these subjects was similar to the source population at each health plan.

Because our main objective was to examine factors common to both cases and comparisons, and because very few breast cancer  patients in our study underwent chemotherapy (121 of 1361 cases), we did not examine the impact of chemotherapy within the cancer case group. In a sensitivity analysis, when we excluded breast cancer survivors exposed to chemotherapy, the results were similar to those with the full cohort. In general, as older breast cancer survivors are less likely to receive chemotherapy, our results are still generalizable to a broader group of older breast cancer survivors.

 However, our results may only be generalizable to breast cancer survivors cared for in other integrated healthcare delivery systems. For similar reasons, we did not examine the  impact of aromatase inhibitors in the breast cancer survivor group. Few survivors used such drugs (98 of 1361 cases), as the majority were diagnosed before the availability of aromatase inhibitors in the mid 2000s. However, given the long latency of CVD (which may be 10 or more years), we were able to capitalize on the long observation period in this study (maximum of 15 years  postdiagnosis).

 Before risk/benefit ratios can be determined, data on the long-term toxicity profiles of aromatase inhibitors need to  be available, and these data are still pending. In addition, while a few studies suggest that aromatase inhibitors may slightly  increase cardiotoxicity,

 Therefore, given that nearly 900 of the 1361 breast cancer patients used tamoxifen, it is possible that the lack of difference in CVD outcomes between cases and comparisons could be  partially attributed to the cardioprotective effect of tamoxifen.

Overall, our results suggest that older long-term, early-stage breast cancer survivors have a risk of CVD similar to that of otherwise healthy women of comparable ages. The established risk factors (very old age, smoking history, diabetes, and hypertension) were more predictive of CVD risk than breast cancer history status. For long-term survivors, comorbid conditions are a greater health threat than the initial cancer.

 This study demonstrates that long-term prognosis in older breast cancer patients is affected by  management of preexisting conditions, and these may be best managed by primary care providers.

Our results also suggest that  management of comorbidities in survivors should not be different from that in the general population of older patients. The transfer of  cancer survivorship care to primary care settings is a challenge; however, careful attention to follow-up care of other chronic diseases  in survivors may best be facilitated in coordinated healthcare systems. 

 From Kaiser Permanente (RH, CA, VQ), Pasadena, CA; Boston Medical Center, (RAS), Boston, MA; Boston University, School of Public Health (MP, ST), Boston, MA; Wake Forest School of Medicine (AMG), Winston-Salem, NC; Group Health Research Institute (AK), Seattle, WA; VA-Healthcare System (ST), Boston, MA; Henry Ford Health Systems (MEUY), Detroit, MI.

Research reported in this publication was supported by the National Cancer Institute, National Institutes of Health,  under award numbers R01CA093772 (RAS), R01CA136743 (RH), and U19CA079689 (EW).

 Dr Haque reports that she has received grants from California Breast Cancer Research Program. The other  authors (MP, AMG, AK, SST, CA, RAS, VQ, MUY) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (RH, SST, RAS, MUY); acquisition of data (RH, AK, VQ, MUY); analysis and  nterpretation of data (RH, SST, RAS, MUY); drafting of the manuscript (RH, SST, RAS, and MUY); critical revision of the manuscript for important intellectual content (RH, SST, RAS, MUY); statistical analysis (RH, SST); provision of study materials or patients (RH, AK, VQ, MUY); obtaining funding (RAS); administrative, technical, or logistic support (RH, CA, RAS, VQ); and supervision (RH, RAS).

Reina Haque, PhD, MPH, Kaiser Permanente Southern California, Department of Research & Evaluation,  100 South Los Robles Ave, 2nd Floor, Pasadena, CA 91101. E-mail: reina.haque@kp.org.1. McCaskill-Stevens W, Abrams JS. Comorbidities in the aging breast cancer population: are current assessments leading to improved outcomes? J Natl Cancer Inst. 2011;103(14):1072-1073.

2. Parry C, Kent EE, Mariotto AB, Alfano CM, Rowland JH. Cancer survivors: a booming population. Cancer Epidemiol Biomarkers Prev. 2011;20(10):1996-2005.

3. Patnaik JL, Byers T, Diguiseppi C, Denberg TD, Dabelea D. The influence of comorbidities on overall survival among older women diagnosed with breast cancer. J Natl Cancer Inst. 2011;103(14):1101-1111.

4. Patnaik JL, Byers T, DiGuiseppi C, Dabelea D, Denberg TD. Cardiovascular disease competes with breast cancer as the leading cause of death for older females diagnosed with breast cancer: a retrospective cohort study. Breast Cancer Res. 2011;13(3):R64.

5. Roychoudhuri R, Robinson D, Putcha V, Cuzick J, Darby S, Møller H. Increased cardiovascular mortality more than fifteen years after radiotherapy for breast cancer: a population-based study. BMC Cancer. 2007;7:9.

6. Correa CR, Litt HI, Hwang WT, Ferrari VA, Solin LJ, Harris EE. Coronary artery findings after left-sided compared with right-sided radiation treatment for early-stage breast cancer. J Clin Oncol. 2007;25(21): 3031-3037.

7. Jagsi R, Griffith KA, Koelling T, Roberts R, Pierce LJ. Stroke rates and risk factors in patients treated with radiation therapy for early-stage breast cancer. J Clin Oncol. 2006;24(18):2779-2785.

8. Nilsson G, Holmberg L, Garmo H, Terent A, Blomqvist C. Radiation to supraclavicular and internal mammary lymph nodes in breast cancer increases the risk of stroke. Br J Cancer. 2009;100(5):811-816.

9. Nilsson G, Holmberg L, Garmo H, Terent A, Blomqvist C. Increased  incidence of stroke in women with breast cancer. Eur J Cancer. 2005;41(3):423-429.

10. Patt DA, Goodwin JS, Kuo YF, et al. Cardiac morbidity of adjuvant radiotherapy for breast cancer. J Clin Oncol. 2005;23(30):7475-7482.

11. Giordano SH, Kuo YF, Freeman JL, Buchholz TA, Hortobagyi GN, Goodwin JS. Risk of cardiac death after adjuvant radiotherapy for breast cancer. J Natl Cancer Inst. 2005;97(6):419-424.

12. Doyle JJ, Neugut AI, Jacobson JS, et al. Radiation therapy, cardiac risk factors, and cardiac toxicity in early-stage breast cancer patients. Int J Radiat Oncol Biol Phys. 2007;68(1):82-93.

Effective management of the comorbidities of diabetes and hypertension may increase survival in older breast cancer survivors.

Comorbidities and Cardiovascular Disease Risk in Older Breast Cancer Survivors

To identify factors associated with delayed radiotherapy (RT) in older women with early-stage breast cancer.

We studied 541 women age ≥65 years diagnosed with early-stage breast cancer in 1990-1994 at 5 integrated healthcare delivery systems and treated with breast-conserving surgery and RT, but not chemotherapy. We examined whether demographic, tumor, or treatment characteristics were associated with RT delays of >8 weeks postsurgery using χ2 tests and multivariable logistic regression.

Seventy-six women (14%) had delayed RT, with a median delay of 14 weeks. Even though they had insurance and access to care, nonwhite and Hispanic women were much more likely than white women to have delayed RT (odds ratio = 3.3; 95% confidence interval = 1.7, 10) in multivariable analyses that controlled for demographic and clinical variables.

Timely RT should be facilitated through physician and patient education, navigation, and notification programs to improve quality of care. Queues for RT appointments should be evaluated on an ongoing basis to ensure adequate access. Future research should examine modifiable barriers to RT timeliness and whether delays impact long-term outcomes.

This study identified factors associated with delayed radiotherapy (RT) for women age ≥65 years with breast cancer in 5 integrated healthcare delivery systems.

Delayed RT was experienced by 14% of the study subjects, even though they had insurance and access to care.

Women receiving delayed care were more likely to be nonwhite or Hispanic.

Queues for RT appointments should be evaluated on an ongoing basis to ensure adequate access.

Breast cancer, diagnosed in 180,000 women in the United States in 2008,

 usually is treated with mastectomy or breast-conserving surgery (BCS) with or without radiotherapy (RT).

 Treatment characteristics vary by race/ethnicity, socioeconomic status, geographic region, healthcare access, and insurance status.

Quality of care, including timing and completeness of RT and chemotherapy, can affect recurrence and survival rates.

 The start of RT is time sensitive, yet many factors can affect whether a patient experiences a delayed start. If she has positive nodes, she may be more likely to have longer discussions about treatment options because more options are available.

 As breast cancer treatment requires extensive coordination of care across multiple providers and specialties, patients may experience difficulties navigating the healthcare system.

 Typical treatment starts with surgery and lymph node assessment, followed by recovery time after surgery, subsequent RT to the whole breast for 5 to 6 weeks, and a 1-week RT boost to the tumor bed.

 Positive lymph nodes are an indication for chemotherapy to be part of the treatment.

There is conflicting evidence regarding the optimal timing of RT following BCS.

 One study suggests that RT delay does not negatively affect breast cancer—related health outcomes,

 A systematic review concluded that RT delay was associated with higher rates of local recurrence.

 Delay also may be associated with psychological distress and anxiety, which can negatively impact the care process and outcomes.

This study assesses patient-level predictors of delayed RT in an insured population of older women with early-stage breast cancer treated in integrated healthcare delivery systems, a group that has not been studied for timeliness in receiving RT.

This study is nested in the Breast Cancer Treatment Effectiveness in Older Women Study cohort.

 We included 541 women age ≥65 years diagnosed with unilateral, early-stage breast cancer (American Joint Commission on Cancer stage I or II22) enrolled in 5 integrated delivery systems in the Cancer Research Network, a  collaborative group of research programs, enrolled populations, and data systems of 14 health systems nationwide.

 Women diagnosed from 1990 through 1994 were followed for 10 years. All subjects in this nested cohort received BCS with a full course of whole breast RT, but no chemotherapy. Subjects from the original cohort receiving BCS without RT (23%) or mastectomy (53%) were excluded.

 We restricted the sample to women not receiving chemotherapy to eliminate variations in RT timing related to chemotherapy sequencing. We used automated administrative databases, medical record review, and tumor registries to ascertain age, race/ethnicity, Charlson Comorbidity Index score,

 lymph node status, tumor size, histology, estrogen receptor (ER) or progesterone receptor (PR) status, and any use of tamoxifen. To calculate the Charlson Comorbidity Index score, information on comorbid conditions was collected from medical records for all healthcare services provided in the year before diagnosis. This index includes 18 conditions weighted to predict mortality. We excluded cancer diagnoses from the index and categorized the remaining score as 0, 1, or 2+. Because others have shown that an 8-week delay can impact health outcomes, we defined RT delay as starting RT more than 8 weeks following BCS, based on medical record data.

All research was conducted following approval from local institutional review boards and in accordance with the ethical standards of those boards.

We calculated the percentage of women with delayed RT and the percentage with timely RT by demographic and clinical characteristics. We used 

 tests and fit multivariable logistic regression models to identify factors associated with delayed RT, controlling for integrated delivery system site.

The sample included 541 women, of whom 76 (14%) had a delay of more than 8 weeks (

). Among women with RT delayed more than 8 weeks, the median delay was 14 weeks, and 21 subjects (4%) had delays of more than 12 weeks. More than two-thirds of the sample was younger than age 75 years, and few (3.3%) scored more than 1 on the Charlson Comorbidity Index. Almost 7% were ER−/PR−, and 13% of subjects had positive nodes (

), women of nonwhite race or Hispanic ethnicity had significantly higher odds of delayed RT (odds ratio [OR] = 3.3, 95% confidence interval [CI] = 1.7, 10). We had limited precision in general, as indicated by some wide CIs (eg, tumor size >2 cm [OR = 1.9, 95% CI = 0.9, 4.1] and ER−/PR− status [OR = 1.8, 95% CI = 0.7, 4.9]).

In this fully insured population of older women with breast cancer who had access to integrated delivery systems, about 1 in 7 women experienced delayed RT following BCS, a rate similar to that reported in other studies.

 Surprisingly, in this population, the proportion with RT delays of more than 8 weeks was about the same as that found in a fee-for-service Medicare population (14% vs 16%, respectively).

 One might expect better coordinated, and therefore less delayed, care in integrated healthcare delivery systems, yet delays of more than 8 weeks may be due to healthcare access issues such as transportation problems, language concordance or communication issues, or limited patient knowledge about the importance of timely RT—all barriers that are not mediated by insurance status. Alternatively, delays may be caused by poor care coordination by clinicians and the healthcare system or patient-related barriers, concerns, or anxiety regarding decision making and treatment

The only predictor of delayed RT was nonwhite or Hispanic race/ethnicity, consistent with previous research.11,18 Indeed, delayed RT, which may represent uncoordinated care (among other possible factors), was not affected significantly by other included clinical or demographic factors. Whether these delays in turn affect mortality for patients in integrated delivery systems could not be shown because of limited statistical power.

All eligible women were included in our study population, and we had no loss to follow-up in the treatment period. This design advantage substantially limited the potential for selection bias. Another strength of this study is that all subjects in the cohort had equal access to complete treatment by having access to care in an integrated healthcare delivery system. In addition to sociodemographic information (beyond age), we had important information about the tumors themselves that helped explain treatment receipt and reduced confounding in our analyses. Nonetheless, we did not have information on characteristics such as patient educational attainment, income, transportation availability, radiation oncologist availability, language barriers, or prefer ences for care; doctor-patient communication; or healthcare system programs such as care coordination—all of which may be greater barriers to or facilitators of care for older women diagnosed with breast cancer.

 These characteristics may directly or indirectly influence care-seeking behavior and care receipt among nonwhite and Hispanic women, who are more likely to be of lower socioeconomic status.

 All subjects had Medicare coverage in these healthcare systems with the same benefits for RT and Medicare paying for RT in inpatient, outpatient, and freestanding clinics. Outpatient RT requires a set copayment, which was usually lower than fee-for-service copayments, thereby reducing barriers to care even more. It is unlikely that coverage levels affected timeliness of RT.

We have shown that access to healthcare, as reflected by a patient’s membership in an integrated healthcare delivery system, does not in itself ensure timely therapy. Rates of RT delay were similar between integrated delivery systems and the fee-for-service sector11 and appear to be affected by race/ethnicity. Healthcare delivery systems and providers should consider targeting care coordination efforts to traditionally underserved patients to improve timeliness of RT.

: From the Department of Public Health (HTG), Weill Cornell Medical College, New York, NY; the Department of Geriatrics (SST, RAS) and the Department of Epidemiology (TLL), Boston University Medical Center, Boston, MA; Group Health Research Institute (DSMB), Group Health Cooperative, Seattle, WA; Meyers Primary Care Institute (TSF), Worcester, MA; HealthPartners Research Foundation (FW), Bloomington, MN; Department of Epidemiology & Public Health (MUY), Yale University, New Haven, CT; Henry Ford Health System (MUY), Detroit, MI; Research & Evaluation (VPQ), Kaiser Permanente Southern California, Pasadena, CA; and the Department of Social Services and Health Policy (AMG), Wake Forest University, Winston-Salem, NC.

: Funded by the National Cancer Institute (R01 CA093772; Dr Silliman, principal investigator) and in part by the American Cancer Society (MRSGT-04-002-01-CPHPS; Dr Gold, principal investigator).

: The authors (HTG, SST, DSMB, TSF, FW, MUY, TLL, VPQ, AMG, RAS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

: Concept and design (HTG, SST, DSMB, TSF, FW, MUY, TLL, RAS); acquisition of data (SST, DSMB, TSF, FW, MUY, VPQ, AMG, RAS); analysis and interpretation of data (HTG, SST, DSMB, TLL, VPQ, AMG, RAS); drafting of the manuscript (HTG, DSMB, VPQ); critical revision of the manuscript for important intellectual content (HTG, SST, DSMB, TSF, FW, MUY, TLL, AMG, RAS); statistical analysis (HTG, SST); provision of study materials or patients (DSMB, MUY, VPQ, RAS); obtaining funding (HTG, MUY, VPQ, AMG, RAS); administrative, technical, or logistic support (DSMB, RAS); and supervision (DSMB, FW).

: Heather Taffet Gold, PhD, Department of Public Health, Weill Cornell Medical College, 402 E 67th St, New York, NY 10065-6304. E-mail: heg2001@med.cornell.edu.

Among older women in 5 integrated delivery systems, 14% experienced delayed radiotherapy for breast cancer, even though they had insurance and access to care.


Rebecca A. Silliman, MD, PhD

Articles

False-Positive Mammography and Its Association With Health Service Use

Comorbidities and Cardiovascular Disease Risk in Older Breast Cancer Survivors

Delayed Radiotherapy for Breast Cancer Patients in Integrated Delivery Systems