Abstract
Background
Cancer survivors may have elevated atherosclerotic cardiovascular disease (ASCVD) risk. Therefore, we tested how accurately the American College of Cardiology/American Heart Association 2013 pooled cohort equations (PCEs) predict 10-year ASCVD risk in cancer survivors.
Objectives
To estimate the calibration and discrimination of the PCEs in cancer survivors compared to non-cancer participants in the Atherosclerosis Risk in Communities (ARIC) study.
Methods
We evaluated the PCEs’ performance among 1244 cancer survivors and 3849 cancer-free participants who were free of ASCVD at the start of follow-up. Each cancer survivor was incidence-density matched with up to five controls by age, race, sex, and study center. Follow-up began at the first study visit at least 1 year after the diagnosis date of the cancer survivor and finished at the ASCVD event, death, or end of follow-up. Calibration and discrimination were assessed and compared between cancer survivors and cancer-free participants.
Results
Cancer survivors had higher PCE-predicted risk, at 26.1%, compared with 23.1% for cancer-free participants. There were 110 ASCVD events in cancer survivors and 332 ASCVD events in cancer-free participants. The PCEs overestimated ASCVD risk in cancer survivors and cancer-free participants by 45.6% and 47.4%, respectively, with poor discrimination in both groups (C-statistic for cancer survivors = 0.623; for cancer-free participants, C=0.671).
Conclusions
The PCEs overestimated ASCVD risk in all participants. The performance of the PCEs was similar in cancer survivors and cancer-free participants.
Implications for Cancer Survivors
Our findings suggest that ASCVD risk prediction tools tailored to survivors of adult cancers may not be needed.
Keywords: Cardiovascular disease, Cancer, Risk prediction, Survivorship, Cardio-oncology
Introduction
Seventeen million Americans, including nearly 20% of adults older than 65, are cancer survivors [1, 2]. Cancer survivors and their healthcare providers need to understand how cancer, its causes, and its treatments can degrade health and quality of life [3]. Of particular concern is cardiovascular health [3-6]. Cancer survivors may have elevated risk for cardiovascular disease (CVD), [7] and standard CVD risk calculators such as the pooled cohort equations (PCEs) may not accurately predict cardiovascular outcomes in this group [8].
Multiple pathways may affect CVD risk in cancer survivors. Cancer and CVD share many risk factors, including older age, male sex, and smoking, resulting in higher CVD incidence among survivors [3]. Additionally, many cancer treatments are cardiotoxic [4] and cancer itself may promote a pro-thrombotic state [3]. Furthermore, cancer may lead to behavior changes, such as decreased adherence to statins, blood pressure, and diabetes medication, potentially increasing CVD incidence among otherwise high-risk individuals.
The PCEs were developed and recommended by the American College of Cardiology (ACC) and the American Heart Association (AHA) [10, 11]. The PCEs predict the 10-year risk of ASCVD, which includes non-fatal myocardial infarction, coronary heart disease death, and stroke [10]. PCE-predicted ASCVD risk is calculated from sex, race, blood cholesterol, systolic blood pressure, use of blood pressure medication, diabetes status, and smoking status. The PCEs’ primary purpose is to prevent ASCVD through the prescription of statins [12] and antihypertensives [13].
The hypothesized causal relationships between cancer, PCE scores, and ASCVD incidence are shown in Fig. 1. Cancer, its causes, and treatments are expected to influence both PCE scores and ASCVD incidence. However, it is unclear whether the PCEs are robust to all the different factors by which cancer survivors may differ from otherwise similar individuals.
Fig. 1.
Causal diagram of the associations between ASCVD risk scores, cancer status, and ASCVD events
A CVD score has been developed for survivors of pediatric cancer after common CVD risk predictors were shown to be inadequate in that group [14-17]. In adults, a longitudinal study found the Framingham risk score (another CVD risk predictor) underestimated CVD risk in 206 women aged 33–87 with breast cancer [18]. Additionally, a recent cross-sectional study of 15,096 adults found cancer survivors had 3.42 (95% confidence interval (CI): 2.51–4.66) times the odds of being categorized as high ASCVD risk by the PCEs but did not investigate the longitudinal association of the PCEs with ASCVD incidence or the performance of the PCEs in this population [19].
If the PCEs perform poorly in cancer survivors compared to the general population, there may be a need to incorporate cancer into existing ASCVD risk predictors or develop risk prediction tools specific to this population [8]. We investigated whether the PCEs adequately predict 10-year ASCVD risk in survivors of adult cancers in the Atherosclerosis Risk in Communities (ARIC) study.
Methods
Cohort design
The ARIC study is a prospective cohort of 15,792 mostly Black and White males and females aged 45–64 at visit 1. Participants were recruited beginning in 1987 from four study centers: Forsyth County, NC; Jackson, MS; suburban Minneapolis, MN; and Washington County, MD [20]. Minneapolis and Washington County participants were primarily White, and participants in Jackson were all Black. Participants have attended up to seven study visits: visit 1 (1987–1989, n = 15,641), visit 2 (1990–1992, n = 14,204), visit 3 (1993–1995, n = 12,746), visit 4 (1996–1998, n = 11,505), visit 5 (2011–2013, n = 6461), visit 6 (2016–2017, n = ~4214), and visit 7 (2018–2019, n = ~3792) [21]. Health information, hospitalizations, and medications (for the past 2 weeks) were assessed in annual (1988–2011, response rate 90–99%) then semi-annual (2012–present, response rate 83–90%) follow-up calls [21]. Participants were excluded from this analysis if they belonged to races other than White or Black (N = 48), did not consent to participate in non-CVD studies (N = 139), or had prevalent cancer at visit 1 (N = 910). The ARIC study has been approved by the Institutional Review Boards of all four study centers.
Cancer history
Incident cancer cases were ascertained through December 2015 via linkage to state cancer registries and supplemented by hospital discharge summaries and medical record abstraction [21]. Each participant was considered to become a cancer survivor on the diagnosis date of their first primary cancer, except for non-melanoma skin cancer. Cancer survivors were included in the analysis if they were alive and present at the next study visit at least 1 year after diagnosis. Cancer death was assessed via death certificates through 2015.
Study design
We conducted two main analyses: (1) an unmatched analysis with all participants beginning follow-up at visit 4 and (2) a matched analysis with all participants beginning follow-up at the first visit at least 1 year after their cancer diagnosis or selection as a control (visits 2–6).
Unmatched analysis
For our first analysis, participants with a cancer diagnosis at least 1 year before visit 4 were compared to those without cancer history by visit 4. Visit 4 was chosen as the start of follow-up because participants at visit 4 ranged in age from 54 to 75, which is within the intended age range for the PCEs. In addition, 434 cancer survivors and 9034 cancer-free participants were available to begin follow-up at visit 4, making it the first ARIC visit with a sufficient sample size for this analysis. Participants with less than 1 year between their cancer diagnosis and visit 4 (n = 102) were excluded from the analysis to reduce the effect of current cancer treatment on PCE components. Because the goal was to determine how well the PCEs predict future ASCVD at a single time point (as would be done in a clinical encounter), participants with no cancer history at visit 4 were included as cancer-free participants in this analysis even if they were diagnosed with cancer at some later date.
Matched analyses
In addition, we conducted a matched analysis to maximize the number of cancer survivors included while ensuring similarity between cancer survivors and cancer-free participants. All cancer survivors with a study visit at least 1 year after their cancer diagnosis were compared to matched cancer-free participants (“controls”). Each cancer survivor was matched with up to 5 randomly selected ASCVD-free and cancer-free participants at the time of cancer diagnosis/matching. The matching was conducted using incidencedensity sampling in Stata’s survival time to case–control function, and participants were matched by age (within 5 years), sex, race, and study center. The index date equals the cancer diagnosis date (for cancer survivors) or the corresponding date for the controls. The start of follow-up was the first study visit at least 1 year after the index date. Selected controls who developed cancer between the index date and the beginning of follow-up were excluded from the analysis (n = 823). Selected controls who developed cancer after the start of follow-up were included as controls and may also be included as cancer survivors with follow-up starting at the next visit. Participants who developed ASCVD before the start of follow-up were excluded (n = 427). A schematic demonstrating the study design is shown in Fig. 2. This matched analysis was repeated for male and female participants separately and for prostate and female breast cancer survivors individually.
Fig. 2.
Visualization of study follow-up for a series of hypothetical participants
Measures
PCE components
The PCEs are computed based on eight components: race (Black or White), sex, total cholesterol, HDL cholesterol, systolic blood pressure, blood pressure treatment (yes or no), diabetes (yes or no), and current smoking status (yes or no).
Static participant characteristics, including date of birth, race, and sex, were collected at visit 1. For the unmatched analysis, all other PCE components were measured at visit 4. For matched analyses, the PCE components were measured at the first visit at least 1 year after their index date. All PCE components were time-invariant, with a single measure at the beginning of follow-up. Smoking status (current, former, or never) was collected by participant self-report at each visit. Participants brought medications they had taken in the previous 2 weeks to each study visit, including antihypertensives and statins, for recording [22].
A random-zero sphygmomanometer measured systolic blood pressure two (at visit 4) or three (all other visits) times. The average of the last two measurements was used. Blood was collected at each visit, and total serum cholesterol was measured using standardized enzymatic assays. In addition, HDL cholesterol was measured using a 2-reagent homogeneous assay. The following criteria assessed diabetes: fasting glucose ≥ 126 mg/dL or non-fasting glucose ≥ 200 mg/dL, or self-report of a diabetes diagnosis by a physician or current treatment for diabetes [23].
Incident ASCVD
The PCEs predict the 10-year risk of ASCVD (defined by the ACC and AHA as non-fatal MI, coronary heart disease (CHD) death, or ischemic stroke). All ASCVD outcomes were ascertained through participant report in annual or semi-annual phone calls, review of hospital discharges, and health department death certificate files, with physician adjudication. For our analysis, the individual components of each ASCVD outcome were defined as follows: CHD: definite or probable MI, definite fatal coronary heart disease; stroke: definite or probable ischemic or hemorrhagic stroke. Participants were followed for ASCVD outcomes for up to 10 years, starting at visit 4. Because the PCEs estimate 10-year ASCVD risk, any participant with more than 10 years of follow-up after the measurement of their PCE components was censored at 10 years.
PCE calculation
The PCEs include four separate equations for Black female, Black male, White female, and White male participants. For each, diabetes (yes vs. no) and smoking status (current vs. not current) were dichotomous, and continuous values for age, total cholesterol, HDL cholesterol, and systolic blood pressure were log-transformed. Risk of ASCVD was calculated using the procedure described in the 2013 ACC/AHA Guideline on the Assessment of Cardiovascular Risk [10]. To create scores for each participant, values for each component were multiplied by a coefficient and summed. A constant for each group (male and female, Black and White) was then raised to the power of the calculated score to derive 10-year predicted ASCVD-free survival and risk (= 1 – survival). Finally, ASCVD risk was multiplied by 100 and expressed as a percentage. Calculations for each score are shown in Supplemental Table S1.
Statistical analysis
Descriptive statistics were calculated for participants with and without cancer history at visit 4 (for the unmatched analysis) as well as cancer survivors and matched controls (for the matched analysis). We calculated means and standard deviations (SD) for continuous variables; and frequencies and percentages (for categorical variables) of participant characteristics (age, study center, race, and sex) in each group. Mean PCE risk and means (SD) or frequencies (percentages) for each PCE component were calculated. PCE risk was also categorized by risk thresholds: low/intermediate risk (0 to less than 7.5%), high risk (7.5 or greater). The crude ASCVD incidence rate was calculated for cancer survivors and those with no cancer history at visit 4 (for the unmatched analysis) and at the start of their follow-up (for the matched analysis).
The performance of the PCEs was evaluated (1) for all participants with follow-up beginning at visit 4 and (2) in matched analyses starting at least 1 year after a cancer diagnosis. In matched analysis, we evaluated calibration and discrimination separately for male and female participants and prostate and female breast cancer survivors. There was insufficient sample size to stratify by gender or cancer site in the unmatched analysis.
We calculated two measures of score performance: calibration and discrimination. Calibration describes how accurately the score predicts risk. In population studies, calibration is estimated by comparing predicted to actual case counts within categories of predicted risk [24]. Participants were divided into quintiles of 10-year PCE-predicted ASCVD risk. To evaluate calibration, PCE-predicted and estimated “actual” ASCVD incidence rates were compared in cancer survivors and cancer-free participants.
Loss to follow-up and censoring prevented us from observing all ASCVD events that occurred in the 10 years following score calculation. We estimated the number of ASCVD events in our sample within 10 years by multiplying the Kaplan–Meier event rate in each PCE-predicted risk quintile by the number of participants. Overall agreement between predicted and Kaplan–Meier-estimated events was summarized using a Hosmer–Lemeshow goodness-of-fit test separately in participants with and without a history of cancer.
Discrimination is a relative measure of performance, evaluating whether participants with high model-predicted risk have an increased risk of the outcome (in this analysis, ASCVD) and vice versa [24]. To assess discrimination, we calculated Harrell’s C statistic for a Cox proportional hazards model with PCE-predicted risk as the exposure and 10-year ASCVD incidence as the outcome [25]. For each model, a C-statistic greater than or equal to 0.7 indicated adequate discrimination [26]. To compare discrimination of the PCE scores, bootstrapping with 1000 iterations was used to calculate confidence intervals (CI) for the C-statistics separately in cancer survivors and cancerfree participants, as well as the difference between the two C-statistics and a p value for the null hypothesis that there was no difference between the two C-statistics. All analyses were conducted in Stata version 16. (StataCorp. 2019. Stata Statistical Software: Release 16. College Station, TX: StataCorp LLC.)
Sensitivity analyses
We conducted six sensitivity analyses described below. All sensitivity analyses were matched as described under “Matched analyses”.
Cancer severity is one source of heterogeneity in the ASCVD risk profile of cancer survivors as a population since people with higher stage and more aggressive cancers might receive different or more cardiotoxic treatments and might also die of cancer before ASCVD events can occur. Therefore, we excluded all participants who eventually died of cancer to capture ASCVD risk in long-term survivors of non-fatal cancers.
In some analyses, participants could have an extended period between their cancer diagnosis and baseline, resulting in measurements that may have more in common with participants with no history of cancer than recent cancer survivors. Therefore, we excluded all participants who had more than 3 years between their cancer diagnosis and PCE measurements to account for this.
Participants with fewer than 3 years between their cancer diagnosis and baseline study visit were excluded to ensure all participants in acute cancer treatment were excluded.
The PCEs have been shown to overestimate ASCVD risk in older adults, but many ARIC participants were older than the recommended age range when their follow-up began in our primary analysis. Thus, we excluded all participants who were 76 or older at baseline because the AHA and ACC recommend using the PCEs to evaluate risk in participants aged 40–75 years [11].
Previous analyses have found that excluding statin users improves the calibration of the PCEs because statins reduce ASCVD risk dramatically but are not included in the PCEs. Hence, we performed an analysis excluding statin users.
Most prostate cancer patients do not receive cardiotoxic treatments. Thus, they are not subject to many of the pathways that may increase or decrease ASCVD risk in cancer survivors. As the most common cancer site in our matched analysis, prostate cancer survivors’ similarity to participants without cancer might hide a real difference between survivors of other cancers and cancer-free individuals. To investigate this, we excluded prostate cancer survivors from our analysis.
Our main matched analyses were matched on the date of cancer diagnosis date, and some controls were lost to follow-up, died, or were diagnosed with cancer between their selection as controls and the start of follow-up. To assess any bias introduced by this matching process, we repeated our main matched analysis with cancer survivors matched with controls at the first visit after their cancer diagnosis.
Results
The unmatched analysis included 434 cancer survivors and 9034 individuals with no history of cancer at visit 4. Cancer survivors were older on average than those with no history of cancer and more likely to be White and male. Of cancer survivors, most had breast (n = 114) or prostate (n = 112) cancer (Table 1). The median time from cancer diagnosis to PCE measurement was 3.98 years (range: 1.02–10.1 years). Cancer survivors had fewer current smokers, less prevalent diabetes, lower total cholesterol, lower systolic blood pressure, and higher use of blood pressure medications than those without cancer (Table 1), but had higher PCE-predicted risk (14.5% vs. 12.3%) (Table 2). Cancer survivors had lower statin use on average than cancer-free participants (Table 1), although they were more likely to have risk above the threshold for statin prescription (69.6% versus 61.6%). In crude analyses, cancer survivors had 14.6 more ASCVD events per 100,000 person-years than survivors (Table 2).
Table 1.
Characteristics of cancer survivors and cancer-free participants
| Unmatched analysis (beginning follow-up at visit 4) |
Matched analysis (beginning follow-up at visits 2–7, depending on the timing of cancer diagnosis)a |
|||
|---|---|---|---|---|
| Cancer survivors | Cancer-free participants | Cancer survivors | Cancer-free participants | |
| N | 434 | 9034 | 1244 | 3849 |
| Cancer site (N, %) | ||||
| Breast | 114 (26.3%) | n/a | 366 (29.4%) | n/a |
| Prostate | 112 (25.8%) | n/a | 267 (21.5%) | n/a |
| Colorectal | 51 (11.7%) | n/a | 125 (10.0%) | n/a |
| Other | 157 (36.2%) | n/a | 486 (39.1%) | n/a |
| Pooled cohort equation (PCE) components | ||||
| Race (N, %) | ||||
| White | 363 (83.6%) | 6948 (76.9%) | 993 (79.8%) | 3191 (82.9%) |
| Black | 71 (16.4%) | 2086 (23.1%) | 251 (20.2%) | 658 (17.1%) |
| Sex (N, %) | ||||
| Female | 212 (48.9%) | 5098 (56.4%) | 573 (46.1%) | 1891 (49.3%) |
| Male | 222 (51.2%) | 3936 (43.6%) | 671 (53.9%) | 1958 (50.9%) |
| Age, years (mean, SD) | 65.2 (5.6) | 62.5 (5.6) | 71.9 (8.6) | 69.9 (8.6) |
| Current smoker, yes (N, %) | 53 (12.2%) | 1327 (14.7%) | 115 (9.2%) | 383 (10.0%) |
| Diabetes, yes (N, %) | 55 (12.7%) | 1267 (14.0%) | 309 (24.8%) | 844 (21.9%) |
| Systolic blood pressure, mmHg (mean, SD) | 126.8 (18.4) | 127.5 (18.9) | 128.5 (17.6) | 128.6 (18.5) |
| Total cholesterol, mg/dL (mean, SD) | 199.2 (34.1) | 201.1 (36.5) | 185.3 (40.2) | 190.0 (41.7) |
| HDL cholesterol, mg/dL (mean, SD) | 48.7 (15.9) | 50.3 (16.6) | 50.0 (14.7) | 50.0 (14.7) |
| Blood pressure medication, yes (N, %) | 187 (43.1%) | 3807 (42.1%) | 785 (63.1%) | 2269 (59.0%) |
| Statin use, yes (N, %) | 38 (8.8%) | 956 (10.6%) | 437 (35.2%) | 1250 (32.6%) |
| Study center (N, %) | ||||
| Forsyth County, NC | 116 (26.7%) | 2191 (24.3%) | 295 (23.7%) | 927 (24.1%) |
| Jackson, MS | 59 (13.6%) | 1859 (20.6%) | 223 (17.9%) | 589 (15.3%) |
| Suburban Minneapolis, MN | 132 (30.4%) | 2512 (27.8%) | 385 (31.0%) | 1271 (33.0%) |
| Washington County, MD | 127 (29.3%) | 2472 (27.4%) | 341 (27.4%) | 1062 (27.6%) |
For the matched analysis, participants were matched by age (within 5 years), race, and gender, resulting in similar measures across groups in this table. Differences in these categories are explained by censoring between control selection and the start of follow-up
Table 2.
Pooled cohort equation (PCE) predicted ASCVD risk and crude ASCVD incidence in cancer survivors and cancer-free individuals
| Unmatched analysis (beginning follow-up at visit 4) |
Matched analysis (beginning follow-up at visits 2–7, depending on the timing of cancer diagnosis)a |
|||
|---|---|---|---|---|
| Cancer survivors | Cancer-free participants | Cancer survivors | Cancer-free participants | |
| N | 434 | 9034 | 1244 | 3849 |
| Pooled cohort equation (PCE) predicted risk (95% CI) | 14.5% (13.6–15.5) | 12.3% (12.1–12.5) | 26.1% (25.2–27.1) | 23.1% (22.5–23.6) |
| PCE risk thresholds (N, %) | ||||
| Low/intermediate risk (0 to less than 7.5%) | 132 (30.4%) | 3464 (38.3%) | 172 (13.8%) | 690 (17.9%) |
| High risk (7.5% or greater%) | 302 (69.6%) | 5570 (61.7%) | 1072 (86.2%) | 3159 (82.1%) |
| ASCVD incidence | ||||
| ASCVD events (crude) | 38 | 733 | 110 | 332 |
| ASCVD events (estimated over 10 years) | 47.1 | 820.1 | 176.5 | 466.3 |
| Person-years of follow-up | 3710.93 | 83,496.02 | 8224.70 | 29,034.04 |
| ASCVD events per 100,000 person-years (crude) | 102.4 | 87.8 | 1337.4 | 1143.5 |
For the matched analysis, participants were matched by age (within 5 years), race, and gender, resulting in similar measures across groups in this table. Differences in these categories are explained by censoring between control selection and the start of follow-up
The matched analysis included 1244 cancer survivors and 3849 cancer-free controls. The median time from cancer diagnosis to PCE measurement was 3.70 years (range: 1.01 to 17.22 years). By design, matched cancer survivors and controls were similar across demographic characteristics and PCE components. However, cancer survivors were slightly older and more likely be to Black and male than cancer-free participants because some participants were excluded when they developed cancer after their selection as controls (Table 1). Cancer survivors had higher PCE-predicted risk than controls (26.1% vs. 23.1%) and higher incidence of ASCVD (Table 2).
In the main analysis beginning at visit 4, calibration was poor in both groups, overestimating risk by 25.4% in cancer survivors and 26.0% for cancer-free participants. C-statistics were 0.685 (95% CI: 0.615–0.755) and 0.692 (95% CI: 0.672–0.713) for cancer survivors and cancer-free participants, respectively (Table 3).
Table 3.
Comparison of calibration and discrimination of the pooled cohort equations (PCEs) in unmatched and matched analyses among cancer survivors and cancer-free persons, ARIC
| Model description | Unmatched analysis Unmatched, follow-up starting at visit 4 |
Matched analyses | ||||
|---|---|---|---|---|---|---|
| Overall | Females | Males | Breast cancer survivors |
Prostate cancer survivors |
||
| Starting years of follow-up (range) | 1996–1998 | 1990–2017 | 1990–2017 | 1990–2017 | 1990–2017 | 1990–2017 |
| Cancer survivors | ||||||
| N | 434 | 1244 | 573 | 671 | 263 | 366 |
| Median age at the start of follow-up | 65 | 73 | 73 | 73 | 71 | 73 |
| Predicted ASCVDa eventsb | 63.05 | 324.77 | 124.83 | 199.94 | 51.54 | 110.96 |
| Observed ASCVD eventsc | 47.05 | 176.53 | 86.04 | 92.26 | 30.50 | 53.49 |
| % over-estimation | 25.4% | 45.6% | 31.1% | 53.8% | 40.8% | 51.8% |
| C-statistic (95% confidence interval) | 0.685 (0.611–0.759) | 0.623 (0.571–0.673) | 0.688 (0.620–0.754) | 0.578 (0.500–0.657) | 0.645 (0.508–0.782) | 0.636 (0.542–0.729) |
| Cancer-free participants | ||||||
| N | 9034 | 3849 | 1825 | 1934 | 837 | 977 |
| Median age at the start of follow-up | 62 | 73 | 70 | 73 | 69 | 71 |
| Predicted ASCVDa eventsb | 1107.86 | 887.21 | 331.00 | 516 | 130.15 | 261.22 |
| Observed ASCVD eventsc | 820.09 | 466.26 | 157.22 | 261.35 | 73.23 | 140.86 |
| % over-estimation | 26.0% | 47.4% | 52.5% | 49.4% | 43.7% | 46.0% |
| C-statistic (95% confidence interval) | 0.692 (0.673–0.711) | 0.671 (0.645–0.687) | 0.689 (0.641–0.737) | 0.601 (0.562–0.640) | 0.683 (0.617–0.750) | 0.624 (0.567–0.681) |
| Difference between cancer survivors and controls | ||||||
| Difference between C-statistics | – 0.007 | – 0.049 | – 0.001 | – 0.023 | – 0.038 | 0.011 |
| p value | 0.85 | 0.097 | 0.98 | 0.61 | 0.62 | 0.84 |
ASCVD atherosclerotic cardiovascular disease
Predicted events are the product of the number of participants in each quintile of PCE-predicted risk multiplied by the mean predicted risk in the quintile
Observed events are the product of the number of participants multiplied by the Kaplan–Meier annual event rate multiplied by 10 years
In matched analysis, calibration was similar between the two groups, overestimating risk by 45.6% for cancer survivors and 47.4% for controls. Discrimination was poor in both groups (C (cancer survivors)=0.623 (95% CI: 0.571–0.673); C (controls): 0.671 (95% CI: 0.645–0.687), p for the difference between C-statistics: 0.097). Among female participants, ASCVD risk was overestimated by 31.1% in cancer survivors, compared to 53.2% in matched controls. However, discrimination was nearly identical: Harrell’s C statistics were 0.688 (95% CI: 0.620–0.754) for cancer survivors and 0.689 (95% CI: 0.645–0.687) for controls. In male participants, calibration was similar in both groups: 53.8% for cancer survivors and 49.4% for matched controls. Discrimination was similar across groups: 0.578 (95% CI: 0.500–0.657) for cancer survivors and 0.601 (95% CI: 0.562–0.640) for matched controls. Likewise, there was no meaningful difference between calibration or discrimination in breast or prostate cancer survivors and matched controls (Table 3).
The PCE scores overestimated risk in cancer survivors and cancer-free participants in all matched sensitivity analyses by 27.2–54.3%. Discrimination was also poor for all sensitivity analyses for cancer survivors and cancer-free participants. In most sensitivity analyses, calibration was similar for cancer survivors and cancer-free participants (within five percentage points). Conversely, the PCEs overestimated ASCVD risk by 8–10% more than cancer-free participants in analyses (1) excluding participants with more than 3 years between their index date and the start of follow-up and (2) excluding participants who were more than 75 years old or older at baseline (Supplemental tables S2-S15).
Discussion
This analysis demonstrates the complex relationship among cancer survivorship, PCE-predicted ASCVD risk, and true ASCVD risk. In crude analyses, cancer survivors had higher PCE-predicted risk and ASCVD incidence rates than participants with no history of cancer.
However, the performance of the PCEs was similar for cancer survivors compared to those with no cancer history. The PCEs overestimated ASCVD risk in both cases and cancer-free participants by at least 25%, and discrimination was poor in both groups. In female participants, ASCVD risk was overestimated nearly twice as much in cancer-free participants as cancer survivors, but discrimination was almost identical between these two groups. Conversely, the calibration and discrimination of the scores in men did not differ meaningfully by cancer status. In all other analyses, the performance of the PCEs was similar for cancer survivors compared to those with no cancer history.
Because the ARIC study was among the cohorts pooled for the creation of the PCEs, the poor calibration and discrimination observed in this analysis are somewhat surprising. However, participants of our study began follow-up at an older age (47–94 in matched analyses) than participants included in the initial development of the PCEs (40–79 years old across all cohorts). Validation studies of the PCEs find nearly universal overestimation, particularly when including statin users and older adults [27], as most of our analyses do. In addition, a growing body of evidence has shown that race-stratified risk scores can cause severe damage to Black patients by over- or underestimating risk and leading providers to provide either unnecessary risky treatments or insufficient treatment to Black patients [28, 29]. Although these concerns are well known and there are continuing efforts to recalibrate and improve the PCEs, they are still widely used in clinical practice and recommended by the AHA and ACC [11, 27]. Thus, it is still important to determine how their performance differs for cancer survivors.
Two small studies have investigated the relationship between specific cancers, CVD risk scores, and CVD risk. A cross-sectional analysis of 787 testicular cancer survivors and matched cancer-free participants found Framingham risk scores (a similar score to the PCEs) were strongly associated with shared cancer and CVD risk factors, such as smoking status and physical activity, but not with cardiotoxic chemotherapy history. As with our analysis, they found that after matching on shared predictors (in this case, age, race, and education), the Framingham risk scores did not differ between cancer survivors and cancer-free participants [30]. On the other hand, in longitudinal analysis of 152 HER-2 positive breast cancer patients who had been referred to a cardiology-oncology clinic, 14% had a cardiovascular event within 40 months of follow-up, despite predicted 10-year risk of 11.2% [18].
ASCVD risk in cancer survivors varies by cancer site, treatment, stage at diagnosis, and severity. For example, colorectal, breast, and lung cancer incidence is strongly associated with ASCVD risk factors such as smoking and obesity, while prostate cancer has a weaker association with these risk factors.
The better performance (due to higher incidence relative to PCE scores) of the PCEs in female cancer survivors in our analysis compared to female participants without cancer may be partially explained by higher ASCVD risk in some female cancer survivors. A cross-sectional study of 15,095 adults found that, although PCE-predicted risk was higher for cancer survivors than people without cancer history overall, elevated risk varied dramatically by cancer site, where survivors of testicular and bladder/kidney cancers had 11.47 and 7.27 times the odds of being characterized “high risk” (PCE risk ≥ 7.5%) than cancer-free participants. In contrast, cervical cancer survivors had lower odds than cancer-free participants (OR: 0.81, 95% CI: 0.29–2.24) [19]. These findings may indicate that cancer-specific ASCVD prediction tools are needed for breast cancer survivors or patients in acute cancer treatment but traditional calculators such as the PCEs are no worse for long-term survivors of most cancers than the general population.
Limitations
Although we investigated the performance of the PCEs in breast and prostate cancer survivors, the sample size of these analyses was small, and the findings were imprecise. We did not have sufficient power to investigate the performance of the PCEs by any other cancer site or treatment.
Another limitation is the high average participant age in our analysis. The PCEs are recommended for use among people aged 40–75 [11], and multiple studies show the scores overestimate risk [27, 31, 32], particularly for older adults [33]. Because there were at least 14 years between visit 4 (1996–1998, median participant age 63) and visit 5 (2011–2013, median participant age 76), many participants began their follow-up at visit 5 when they were older than the intended age for the PCEs. Analyses with younger age at baseline all had better calibration and discrimination than our main models, but performance was poor.
Additionally, with a median time from cancer diagnosis to PCE measurement of 3.9 years, most participants included in this analysis were long-term cancer survivors, who may be generally healthier and at lower risk for ASCVD than participants who did not survive until their PCE measurements. Further research is needed to determine whether the PCEs should be modified for patients in active or recent cancer treatment. Conversely, since CVD outcomes develop over a long period of time, the impact of cancer on the performance of the PCEs may only be apparent in longer-term cancer survivors.
Furthermore, certain cardiotoxic treatments are only used for specific cancer sites and stages of cancer diagnosis, and many, such as anthracyclines, lead to heart failure, rather than atherosclerotic outcomes. Because treatment information was not available for most participants, we were not able investigate whether performance varied by treatments. Additionally, scores predicting broader CVD definitions (including heart failure) may be more impacted by cancer status.
In addition, because two-thirds of participants in our matched analyses began follow-up between 2011 and 2019 and outcome surveillance was available through 2019, some participants were followed for less than 10 years, so ASCVD events after censoring were not captured. We attempted to address this limitation by estimating the 10-year incidence from the Kaplan–Meier event rate; however, estimates may still be imprecise.
Our unmatched analysis and some sensitivity analyses had few ASCVD events and may have been underpowered. However, our main finding was robust across analyses.
Strengths and conclusions
Despite its limitations, the ARIC study’s large sample size and longitudinal measurement of each PCE component, cancer, and ASCVD outcomes provide a unique opportunity to evaluate the performance of the PCEs.
Calibration and discrimination were poor for cancer survivors and cancer-free participants. The overall misestimation we found is in keeping with most recent validation studies, which show an average 60–90% overestimation.
Although more research is needed to investigate how different cancer sites and treatments impact the performance of the PCEs and whether cancer status impacts the prediction of heart failure, especially in middle-aged and recent cancer survivors, we found little evidence that the PCEs performed worse in cancer survivors than cancer-free adults.
Supplementary Material
Funding
The Atherosclerosis Risk in Communities study has been funded in whole or in part with Federal funds from the National Heart, Lung, and Blood Institute, National Institutes of Health, Department of Health and Human Services, under Contract nos. 75N92022D00001, 75N92022D00002, 75N92022D00003, 75N92022D00004, and 75N92022D00005. Studies on cancer in ARIC are also supported by the National Cancer Institute (U01 CA164975). The authors thank the staff and participants of the ARIC study for their important contributions. The content of this work is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Dr. Lutsey was partially supported by NIH NHLBI K24 HL159246. Dr. Joshu was supported in part by the American Cancer Society (Rsg-18–147-01-Cce).
Other.
Cancer data was provided by the Maryland Cancer Registry, Center for Cancer Prevention and Control, Maryland Department of Health, with funding from the State of Maryland and the Maryland Cigarette Restitution Fund. The collection and availability of cancer registry data is also supported by the Cooperative Agreement NU58DP006333, funded by the Centers for Disease Control and Prevention. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the Centers for Disease Control and Prevention or the Department of Health and Human Services.
Abbreviations
- PCE
Pooled cohort equations
- ASCVD
Atherosclerotic cardiovascular disease
- CVD
Cardiovascular disease
- ACC
American College of Cardiology
- AHA
American Heart Association
- CI
Confidence interval
- ARIC
The Atherosclerosis Risk in Communities study
Footnotes
Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s11764-023-01379-0.
Competing interests The authors declare no competing interests.
Ethics approval The ARIC study was approved by the institutional review boards of each study center.
Consent to participate Informed consent was obtained from all individual participants included in the study.
Conflict of interest The authors declare no competing interests.
Data availability
The data generated in this study may be made available upon reasonable request. Consistent with a prespecified policy for access of ARIC data, requests may be submitted to the ARIC steering committees for review. The analytic methods and study materials may be made available to other researchers for purposes of reproducing the results or replicating the procedure upon reasonable request.
References
- 1.Miller KD, Nogueira L, Mariotto AB, et al. Cancer treatment and survivorship statistics, 2019. CA: Cancer J Clin. 2019;69(5):363–385. 10.3322/caac.21565 [DOI] [PubMed] [Google Scholar]
- 2.Annual Estimates of the Resident Population for Selected Age Groups by Sex: April 1, 2010 to July 1, 2019. In: National Population by Characteristics: 2010-2019. 1999. https://www2.census.gov/programs-surveys/popest/tables/2010-2019/national/asrh/nc-est2019-agesex.xlsx. Accessed 23 Aug 2022. [Google Scholar]
- 3.Koene RJ, Prizment AE, Blaes A, Konety SH. Shared risk factors in cardiovascular disease and cancer. Review Circulation. 2016;133(11):1104–14. 10.1161/circulationaha.115.020406. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Alexandre J, Cautela J, Ederhy S, et al. Cardiovascular toxicity related to cancer treatment: a pragmatic approach to the American and European cardio-oncology guidelines. J Am Heart Asso. 2020;9(18): 10.1161/JAHA.120.018403 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Armenian SH, Xu LF, Ky B, et al. Cardiovascular disease among survivors of adult-onset cancer: a community-based retrospective cohort study Article. J Clin Oncol. 2016;34(10):1122. 10.1200/jco.2015.64.0409. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Bradshaw PT, Stevens J, Khankari N, Teitelbaum SL, Neugut AI, Gammon MD. Cardiovascular disease mortality among breast cancer survivors Article. Epidemiology. 2016;27(1):6–13. 10.1097/ede.0000000000000394. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Florido R, Daya NR, Ndumele CE, et al. Cardiovascular disease risk among cancer survivors. J Am College Cardiol. 2022;80(1):22–32. 10.1016/j.jacc.2022.04.042. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Blaes AH, Shenoy C. Is it time to include cancer in cardiovascular risk prediction tools? Editorial material. Lancet. 2019;94(10203):986–8. 10.1016/s0140-6736(19)31886-0. [DOI] [PubMed] [Google Scholar]
- 9.Tseng T-S, Lin H-Y, Moody-Thomas S, Martin M, Chen T. Who tended to continue smoking after cancer diagnosis: the national health and nutrition examination survey 1999–2008. BMC Public Health. 2012;12(1):784. 10.1186/1471-2458-12-784. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Goff DC, Lloyd-Jones DM, Bennett G, et al. ACC/AHA guideline on the assessment of cardiovascular risk: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. Circulation. 2013;129(25):S49–73. 10.1161/01.cir.0000437741.48606.98. [DOI] [PubMed] [Google Scholar]
- 11.Arnett DK, Blumenthal RS, Albert MA, et al. 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines (vol 140, e596 Correction. Circulation. 2019;140(11):E649–50. 10.1161/cir.0000000000000725. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Andrus B, Lacaille D. 2013 ACC/AHA guideline on the assessment of cardiovascular risk. 2014 [DOI] [PubMed] [Google Scholar]
- 13.Whelton PK, Carey RM, Aronow WS, et al. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline for the prevention, detection, evaluation, and management of high blood pressure in adults: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension. 2018;71(6):e13–e115. 10.1161/HYP.0000000000000065 [DOI] [PubMed] [Google Scholar]
- 14.Mertens AC, Liu Q, Neglia JP, et al. 2008. Cause-specific late mortality among 5-year survivors of childhood cancer: the childhood cancer survivor study. Journal of the National Cancer Institute 100(19): 10.1093/jnci/djn310 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Oeffinger KC, Mertens AC, Sklar CA, et al. Chronic health conditions in adult survivors of childhood cancer. New England J Med. 2006;355(15): 10.1056/nejmsa060185 [DOI] [PubMed] [Google Scholar]
- 16.Chow EJ, Chen Y, Kremer LC, et al. Individual prediction of heart failure among childhood cancer survivors Article. J Clin Oncol. 2015;33(5):394–U26. 10.1200/jco.2014.56.1373. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Reulen RC, Winter DL, Frobisher C, et al. Long-term cause-specific mortality among survivors of childhood cancer. JAMA - J Am Med Asso. 2010;304(2): 10.1001/jama.2010.923 [DOI] [PubMed] [Google Scholar]
- 18.Law W, Johnson C, Rushton M, Dent S. The Framingham risk score underestimates the risk of cardiovascular events in the HER2-positive breast cancer population Article. Curr Oncol. 2017;24(5):E348–53. 10.3747/co.24.3684. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Zhang X, Pawlikowski M, Olivo-Marston S, Williams KP, Bower JK, Felix AS. Ten-year cardiovascular risk among cancer survivors: the National Health and Nutrition Examination Survey. PLoS One. 2021;16(3):e0247919. 10.1371/journal.pone.0247919. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Wright JD, Folsom AR, Coresh J, et al. The ARIC (Atherosclerosis Risk in Communities) study: JACC focus seminar 3/8. J Am Coll Cardiol. 2021;77(23):2939–59. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Joshu CE, Barber JR, Coresh J, et al. Enhancing the infrastructure of the Atherosclerosis Risk in Communities (ARIC) study for cancer epidemiology research: ARIC cancer. Cancer Epidemiol Biomarkers Prevention. 2018;27(3):295–305. 10.1158/1055-9965.Epi-17-0696. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Mondul AM, Joshu CE, Barber JR, et al. Longer-term lipidlowering drug use and risk of incident and fatal prostate cancer in black and white men in the ARIC study. Cancer Prev Res. 2018;11(12):779–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.DeBarmore B, Longchamps RJ, Zhang Y, et al. Mitochondrial DNA copy number and diabetes: the Atherosclerosis Risk in Communities (ARIC) study. BMJ Open Diabetes Res Care. 2020;8(1):e001204. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Alba AC, Agoritsas T, Walsh M, et al. Discrimination and calibration of clinical prediction models: users’ guides to the medical literature. JAMA. 2017;318(14):1377–84. 10.1001/jama.2017.12126. [DOI] [PubMed] [Google Scholar]
- 25.Harrell FE, Califf RM, Pryor DB, Lee KL, Rosati RA. Evaluating the yield of medical tests. Jama-J Am Med Assoc. 1982;247(18):2543–6. 10.1001/jama.247.18.2543. [DOI] [PubMed] [Google Scholar]
- 26.Lemeshow Stanley, Sturdivant Rodney, Hosmer David Jr. Applied logistic regression. Hoboken, New Jersey: John Wiley & Sons; 2013. [Google Scholar]
- 27.Cook NR, Ridker PM. Calibration of the pooled cohort equations for atherosclerotic cardiovascular disease: an update. Ann Intern Med. 2016;165(11):786–94. [DOI] [PubMed] [Google Scholar]
- 28.Ahmed S, Nutt CT, Eneanya ND, et al. Examining the potential impact of race multiplier utilization in estimated glomerular filtration rate calculation on African-American care outcomes. J General Inter Med. 2021;36(2):464–71. 10.1007/s11606-020-06280-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Grobman WA, Sandoval G, Rice MM, et al. Prediction of vaginal birth after cesarean delivery in term gestations: a calculator without race and ethnicity. Am J Obstet Gynecol. 2021;225(6):664. e1–664.e7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Feldman DR, Ardeshir-Rouhani-Fard S, Monahan P, et al. Predicting cardiovascular disease among testicular cancer survivors after modern cisplatin-based chemotherapy: application of the Framingham risk score. Clin Genitourinary Cancer. 2018;16(4):e761–9. 10.1016/jxlgc.2018.01.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Emdin CA, Khera AV, Natarajan P, et al. Evaluation of the pooled cohort equations for prediction of cardiovascular risk in a contemporary prospective cohort. Am J Cardiol. 2017;119(6):881–5. 10.1016/j.amjcard.2016.11.042. [DOI] [PubMed] [Google Scholar]
- 32.Ko DT, Sivaswamy A, Sud M, et al. Calibration and discrimination of the Framingham risk score and the pooled cohort equations. Can Med Assoc J. 2020;192(17):E442–9. 10.1503/cmaj.190848. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Nanna MG, Peterson ED, Wojdyla D, Navar AM. The accuracy of cardiovascular pooled cohort risk estimates in US older adults. J General Int Med. 2020;35(6):1701–8. 10.1007/s11606-019-05361-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data generated in this study may be made available upon reasonable request. Consistent with a prespecified policy for access of ARIC data, requests may be submitted to the ARIC steering committees for review. The analytic methods and study materials may be made available to other researchers for purposes of reproducing the results or replicating the procedure upon reasonable request.