Subclinical cardiovascular disease and frailty risk: the atherosclerosis risk in communities study

This study used data from the Atherosclerosis Risk in Communities (ARIC) Study. Data were obtained from the public database of the National Heart, Lung, and Blood Institute Biologic Specimen and Data Repository Information Coordinating Center. Data usage was approved by the Human Ethical Committee of West China Hospital of Sichuan University. The study design and procedures of the ARIC study were established in 1987 [17]. Briefly, the ARIC is a prospective cohort study of 15,792 adults aged between 45 and 64 years from four American communities (Forsyth County, North Carolina; Jackson, Mississippi; suburban Minneapolis, Minnesota; and Washington County, Maryland). Participants were followed-up every 3 years until Visit 4 (1996–1998) and invited for Visit 5 (2011–2013) after 15 years. Physical examination and interview were carried out at each follow-up. Participants were also contacted by telephone each year to obtain information about their life conditions and hospitalisation. All participants provided written informed consent. The experimental protocol was established according to the ethical guidelines of the Helsinki Declaration and was approved by the institutional review boards at the collaborating medical centers: University of Mississippi Medical Center Institutional Review Board (Jackson Field Center); Wake Forest University Health Sciences Institutional Review Board (Forsyth County Field Center); University of Minnesota Institutional Review Board (Minnesota Field Center); and the Johns Hopkins School of Public Health Institutional Review Board (Washington County Field Center).

This study investigated the relationship of subclinical CVD, assessed by cardiac biomarkers obtained at Visit 2 (1990–1992) and Visit 4, with frailty at Visit 5. For the present study, Visit 2 was set as the baseline and there were 14,348 participants in the ARIC study. For primary analysis, we excluded participants without hs-cTnT or NT-proBNP values (N = 762), those with a history of CVD (N = 1311, including heart failure, myocardial infarction, and coronary heart disease) or chronic renal injury (N = 182) at baseline, those who died during follow up (N = 6179), and those who did not undergo frailty assessment (N = 715). Thus, 5199 participants were included in the primary analysis. Loss to follow-up due to mortality was of particular concern. In order to reduce the impact of mortality and decrease the possibility of reverse causality, the association of cardiac biomarkers at Visit 4 (secondary analysis) and longitudinal patterns of cardiac biomarkers from Visit 2 and Visit 4 (dynamic analysis) with risk of frailty were investigated. Finally, a total of 4727 participants were included in the secondary and dynamic analyses according to a similar exclusion standard (Fig. 1).

The concentration of biomarkers was measured from stored plasma samples collected at Visit 2, using a sandwich immunoassay method on the Roche Elecsys 2010 Analyzer (Roche Diagnostics, Indianapolis, Indiana) at the University of Minnesota in 2012–2013. Samples had been stored since collection at − 70 °C. The detection limit of hs-cTnT was 5 ng/L (Elecsys Troponin T; Roche Diagnostics). The value of 1.5 ng/L was imputed for individuals with undetectable hs-cTnT. The inter-assay coefficient of variation for Hs-cTnT was 6.4% at a mean control of 29 ng/L [18]. The detection limit of NT-proBNP was 5 pg/ml. The value of 2.5 pg/ml was imputed for individuals with undetectable NT-proBNP. The inter-assay coefficient of variation for NT-proBNP was 7.4% at a mean control of 134 pg/ml [19].

An elevated hs-cTnT level was set as ≥14 ng/L, according to the 99th percentile value reported by the manufacturer, and defined as subclinical myocardial damage [20]. An elevated NTproBNP level was set as ≥100 or 300 ng/L according to previous pre-specified levels [21‐23], yielding a 98% negative predictive value to exclude heart failure [24]. The predictive values of these cut-off points for CVD were also validated by previous ARIC studies [22, 25]. Thus, there were three judgment criteria for subclinical CVD in this study: hs-cTnT ≥14 ng/L, NT-proBNP ≥100, or NT-proBNP ≥300 ng/L.

The frailty phenotype was classified according to the definition by the Cardiovascular Health Study and previous ARIC studies [8, 26, 27]. The frailty definition includes five component criteria: (1) Shrinking: defined as 10% unintentional weight loss between two visits or body mass index < 18.5 kg/m² at the last visit. (2) Weakness: defined as sex- and BMI-specific grip strength in the lowest 20th percentile. (3) Poor endurance and energy: defined as participants who answered ‘some of the time’ or ‘most of the time’ to the statements ‘I felt everything I did was an effort’ and ‘I could not get going’ from the Centre for Epidemiological Research Depression scale. (4) Slowness: defined as the slowest 20% of the population, based on time to walk 4 m. (5) Low physical activity level: defined as the lowest 20th percentile based on the Modified Baecke questionnaire. Participants who met three or more criteria were classified as frail (even if there were missing variables); one or two criteria, pre-frail; and without these characteristics, robust. To minimize sample loss, this study used the definition of the second edition, which considers items that cannot be completed due to physiological function limitations as positive. The frailty phenotype assessment was limited to Visit 5, and frailty outcomes were defined as ≥3 criteria vs. < 3 criteria in this study.

Covariates were selected based on existing literature on frailty in previous ARIC studies [28‐31]. The potential factors affecting frailty included age, sex, race, drinking status (never, former, current), smoking status (never, former, current), education status (less than high school, high school, college), total cholesterol (mg/dl), high-density lipoprotein (mg/dl), triglyceride (mg/dl), blood glucose (mmol/l), creatinine (mg/ml), cognitive scores, hypertension, diabetes, cancer, and chronic obstructive pulmonary disease (COPD) disease.

Alcohol use was ascertained using an interviewer-administered questionnaire. Participants were asked whether they currently consume alcohol or whether they had done previously. We defined “never drank” as participants who reported never drinking, “current drinker” as participants who reported active drinking at Visit 2, and “former drinker” as participants who had previously consumed alcohol but did not report active drinking at Visit 2 [32]. Smoking was defined in a similar manner [33]. The average systolic blood pressure and diastolic blood pressure in the sitting position was calculated with an automatic oscillographic sphygmomanometer (OMRONHEM-907XL) after sitting for 5 min. The body mass index was calculated based on height and weight measurements. Hypertension was defined as self-reported antihypertensive drug use or blood pressure ≥ 140/90 mmHg. Diabetes was defined as fasting blood glucose ≥126 mg/dl, non-fasting blood glucose ≥200 mg/dl, or self-reported antidiabetic drug use. History of CVD was defined as electrocardiogram abnormalities, self-reported history of previous coronary procedures at baseline, or any definite CVD events observed during the follow-up. The presence of COPD was based on self-reported physician diagnosis or the obstructive vital capacity measurement model, which was defined as the ratio of forced expiratory volume in 1 s /forced vital capacity < 0.70. Cancer was determined by extracting medical records and discharge codes for self-reported cases. The cognitive test included the Delayed Word Recall Test, the Word Fluency Test, and the Digit Symbol Substitution Test. Cognitive Z scores were calculated by averaging individual Z scores from three tests [34, 35].

Parametric continuous variables were reported as the mean ± standard deviation and compared using analysis of variance. Meanwhile, non-parametric continuous variables were reported as the median (25th, 75th percentiles) and compared using the Mann-Whitney U test. Categorical variables were reported as frequencies and percentages and compared using the chi-square test.

Kaplan–Meier plots for the cumulative incidence of frailty according to elevated and low cardiac biomarker levels at Visit 2 were conducted and compared using the log-rank chi-square test. The Cox proportional hazards regression models were used to assess the relationship of the cardiac biomarkers at Visits 2 and 4 with the time to the frailty event. To further determine whether these relationships were independent of risk factors, the model was adjusted according to demographic variables, physiological variables, laboratory test results, and chronic medical conditions. To reduce sample loss caused by partly missing covariates (such as education, income, smoking, and drinking) when the Cox regression models were run, we conducted multiple imputations to simulate the missing information. Missing covariates were observed in 0.3% of participants for education and lifestyle variables, in 3.3% of participants for laboratory examination, and in 0.4% of participants for chronic conditions. The missing information was imputed simultaneously by conducting regression-based multiple imputations according to Rubin’s rules [36].

As loss to follow up due to mortality could introduce a major selective bias, additional analyses were conducted. First, Cox regression models were used to analyse the relationship between elevated cardiac biomarkers and composite events (frailty plus all-cause mortality). Further, a competing risk analysis was conducted to compare the association of elevated cardiac biomarkers with frailty and all-cause mortality, where each outcome was simultaneously modelled as a different event, and the Wald test was used to compare the parameter estimates between frailty and all-cause mortality [37].

Correlations between continuous cardiac biomarkers levels at Visit 2 and adjusted hazard ratios (HRs) for frailty events were analyzed using restricted cubic linear splines of the cardiac biomarkers with three evenly spaced knots, determined by Harrell et al.’s method [38]. In addition, after adjusting for the same confounding factors, Cox regression analysis was performed to evaluate the association of the cardiac biomarkers at Visit 2 and 4 with frailty in different subgroups of age, sex, race, hypertension, and diabetes, and their interactions were tested.

To examine the association of longitudinal patterns of midlife cardiac biomarkers (representing the change in subclinical CVD from Visit 2 to Visit 4) with frailty, the Cox regression models were conducted. Participants were divided into four groups according to two dichotomies: elevated or normal biomarkers and their change over time. The groups were as follows: (1) stable low: low cardiac biomarker at both Visits 2 and 4; (2) ascending: low cardiac biomarkers at Visit 2 and elevated cardiac biomarkers at Visit 4; (3) descending: elevated cardiac biomarkers at Visit 2 and low cardiac biomarkers at Visit 4; and (4) stable elevated: high cardiac biomarkers at Visits 2 and 4.

Considering that the two cardiac biomarkers represent different types of subclinical CVD, joint association of hs-cTnT (cut-off of 14 ng/l) and NT-proBNP (cut-off of 100 pg/ml) was evaluated by analysing their cross-categories. The population was divided into four groups: (1) low hs-cTnT plus low NT-proBNP; (2) elevated hs-cTnT plus low NT-proBNP; (3) low hs-cTnT plus elevated NT-proBNP; and (4) elevated hs-cTnT plus elevated NT-proBNP. The NT-proBNP cut-off of 300 pg/ml was not included in this analysis because only three participants with elevated hs-cTnT plus elevated NT-proBNP.

Among the 14,348 participants at Visit 2, 5199 participants aged 55.1 ± 5.1 years were included in this study. Among them, 360 (6.9%) and 2417 (46.5%) participants developed frailty and pre-frailty, respectively, at Visit 5, at the age of 75.7 ± 5.2 years. The baseline (1990–1992) characteristics are shown and compared in Table 1. Compared with the robust group, the frailty and pre-frailty groups were older, more likely to be female, African-Americans, have hypertension and diabetes, have lower education, and have higher concentrations of cardiovascular risk markers in midlife (Visit 2).

To assess whether the 5199 participants could represent the entire population at Visit 5 (N = 6538), we compared their baseline characteristics. The study population had similar baseline characteristics, such as age, race, and cardiovascular risk marker levels to the 1339 individuals who were excluded (P > 0.05); however, the study population had a slight female predominance and higher cholesterol levels (P < 0.05, Supplementary Table S1).

There were 168 (3.2%), 909 (17.5%), and 122 (2.4%) participants with hs-cTnT ≥14 ng/L, NT-proBNP ≥100 pg/ml, and NT-proBNP ≥300 pg/ml at Visit 2, respectively. During the median follow-up of 21.7 (interquartile range: 20.0–22.3) years from Visit 2, participants with hs-cTnT ≥14 ng/L, NT-proBNP ≥100 pg/ml, or NT-proBNP ≥300 pg/ml had significant higher incidence of frailty (17.9% vs. 6.7%; 11.8% vs. 5.9%; 19.7% vs. 6.8%; P < 0.001 for all, respectively, Supplementary Fig. S1). Similar results were found in participants at Visit 4 (Supplementary Fig. S2).

In the Kaplan–Meier analysis, participants with hs-cTnT ≥14 ng/L, NT-proBNP ≥100 pg/ml, or NT-proBNP ≥300 pg/ml had a significantly higher cumulative incidence of frailty compared to participants with normal cardiac biomakers (67.5% vs. 35.1%; 56.1% vs. 30.0%; 70.1% vs. 37.8%; P < 0.001 for all, respectively, Fig. 2).

In Cox regression analysis, compared with participants with hs-cTnT < 14 ng/L at Visit 2, those with hs-cTnT ≥14 ng/L had an HR for frailty of 2.127 (95% confidence interval (CI): 1.129–4.010, P = 0.020), even after adjusting for demographic variables, physiological variables, laboratory examination, and chronic medical conditions (Table 2). Similarly, NT-proBNP ≥100 or 300 pg/ml was an independent risk factor for frailty (cut-off = 100 pg/ml, HR = 1.33, 95% CI: 1.01–1.76, P = 0.045; cut-off = 300 pg/ml, HR = 2.61, 95% CI: 1.28–5.33, P = 0.008). When models were run without multiple imputations, the results were consistent (data not shown). However, NT-proBNP ≥300 pg/ml at Visit 4 was not independently associated with incident frailty (P > 0.05, Supplementary Table S4).

Participants with elevated hs-cTnT (≥14 ng/L) or NT-proBNP (≥300 pg/ml) levels had more than a five-fold increased risk of frailty combined with mortality (P < 0.001, Supplementary Table S2). However, in the competing risk analysis, the association of elevated hs-cTnT (≥14 ng/L) or NT-proBNP (≥300 pg/ml) at Visit 2 with frailty showed no change for all-cause mortality (P = 0.647; P = 0.781; respectively).

In the restricted cubic spline models, the adjusted HR for incident frailty increased linearly with an increase in hs-cTnT and NT-proBNP levels at Visit 2 (Fig. 3). The relationship between hs-cTnT and frailty was curvilinear, with the appearance of a plateau for values over 14 ng/L. In contrast, the relationship between NT-proBNP and frailty was rectilinear. Regardless of the cut-off point of 100 or 300 pg/ml, the curve had a high slope, indicating that there was no unique threshold for NT-proBNP (Fig. 3).

At Visit 2 (Supplementary Table S3) and Visit 4 (Supplementary Table S4), subgroup analysis showed that the hs-cTnT and NT-proBNP levels were also associated with frailty events in different subgroups of age, sex, race, and chronic condition (P for interaction > 0.05).

Data from Visit 2 were entered into a multivariate Cox regression model for the correlation analysis between joint association of cardiac biomarkers and frailty. The results showed that individuals with elevation of both hs-cTnT (≥14 ng/L) and NT-proBNP (≥100 pg/ml) had a higher risk for frailty than those with low cardiac biomarker levels (HR: 4.15, 95% CI: 1.50–11.48, P = 0.006). In addition, the HR for elevated NT-proBNP alone was 1.42 (95% CI: 1.11–1.76, P = 0.041). However, elevated hs-cTnT alone did not evidently confer an increased frailty risk (Fig. 4). In Visit 4, participants with any elevation in hs-cTnT (≥14 ng/L) or NT-proBNP (≥100 pg/ml) categories had a significantly higher risk of frailty than individuals with low cardiac biomarker levels (Fig. 4).