Association of antecedent cardiovascular risk factor levels and trajectories with cardiovascular magnetic resonance-derived cardiac function and structure

The KORA S4 study (“Cooperative Health Research in the Region of Augsburg”) is a population-based cohort conducted in the city of Augsburg (Southern Germany) and two surrounding counties between 1999 and 2001 (baseline examination). Of all 4261 participants of the baseline examination cycle (Exam 1), 3080 individuals also participated in the 7-year follow-up examination (2006–2008; Exam 2) and 2279 subjects also participated in the 14-year follow-up examination (2013–2014; Exam 3) [13]. At Exam 3, a total of 400 participants without a history of stroke, myocardial infarction, and arterial vessel occlusion were examined by 3 T CMR in the KORA CMR sub-study that was planned with sufficient power for several analyses of subclinical cardiovascular disease differences in different risk groups [14]. Participants with missing data for cardiac imaging (n = 20), non-participation in Exam 2 (n = 18) and with missing data on risk factors and potential confounders at any of the Exams (n = 13) were excluded from the present analysis, yielding an analytical sample of 349 participants (143 women; aged 39 to 73 years) (Fig. 1).

CMR examinations were performed on a 3 T system (Magnetom Skyra, Siemens Healthineers, Erlangen, Germany) using an 18 channel body array coil in combination with the table-mounted spine matrix coil [14]. Individuals were scanned in a supine position.

The aim of the study was to focus on 5 selected main cardiovascular risk factors (systolic BP, diastolic BP, waist circumference, HbA1c, LDL-C) that were modifiable, clinically measureable, lowly inter-correlated, and of a continuous measurement character. At each examination cycle, systolic and diastolic BP were measured three times at the right arm of seated participants after a minimum resting period of 5 min and a time interval of 3 min between measurements. An oscillometric digital BP monitor (HEM-705CP, Omron Corporation, Tokyo, Japan) and a cuff of appropriate size were used. The mean of the second and third BP measurements was used for the present analyses [19].

Waist circumference was measured with an inelastic multi-colored measuring tape (Fa Hoechstmass) in cm to the closest 0.1 cm. Waist circumference was measured at the level midway between the lower rib margin and the iliac crest while the participants breathed out gently.

Laboratory measurements including HbA1c, total cholesterol, high- and low-density lipoprotein cholesterol (HDL-C, LDL-C) were performed as described elsewhere [13]. Briefly, for the assessment of total cholesterol, LDL-C and HDL-C, enzymatic, photometric assays were used at the baseline examination and enzymatic, colorimetric Flex assays were used at Exam 2 and Exam 3. HbA1c was measured by a turbidimetric inhibition immunoassay at baseline (S4) and by cation-exchange high performance liquid chromatographic, photometric assays at Exam 2 and 3.

Further health-related variables were measured in a comparable fashion in all three KORA examination cycles by standardized interview and in a comprehensive physical examination. Participants were classified as never-smoker, ex-smoker or current smoker. Pack-years were calculated by multiplying the number of packs of cigarettes smoked per day by the number of years the participant has smoked. Participants were classified as being physically active if they did regular sports in summer and winter and were active for ≥ 1 h per week in at least one season or as physically inactive if they did less sports [20]. Alcohol intake was assessed using a validated recall method, calculating alcohol intake in grams per day from participants’ self-reported intake of beer, wine, sparkling wine or distilled spirits over the previous weekend and workday [21].

Diabetes was self-reported or defined by the use of glucose-lowering medication. Hypertension was defined as systolic BP ≥ 140 mmHg or diastolic BP ≥ 90 mmHg [22] or the use of antihypertensive medication under the awareness of having hypertension. Medication intake within the last seven days was recorded during a medical interview by computer-based software, and participants were also asked to bring their medication packages with them.

Weight and height were determined by Seca's measuring systems (Seca GmbH & Co, KG, Hamburg, Germany; [weight: Exam 1: SECA 709, Exam 2: SECA 709, Exam 3: SECA 635 or SECA 877 or SECA measuring station 285; height: Exam 1: SECA 221, Exam 2: SECA 242, Exam 3: SECA 242]) with either calibrated steelyards or digital scales allowing accurate measurements up to 0.1 kg and 0.1 cm, respectively. BMI was calculated as weight divided by squared height (kg/m²).

waist circumference was divided by hip circumference to get waist-hip-ratio (WHR) and by height to get waist-height-ratio (WHtR). Hip circumference was measured at the widest protrusion of the gluteal region between the superior border of the iliac crest and crotch. Body surface area (BSA) was calculated according to the Du Bois formula (BSA = 0.007184 × (height in cm)^0.725 × (weight in kg)^0.425) [23].

For each KORA examination cycle, descriptive characteristics of participants are provided as means with standard deviations for continuous measurements and as absolute numbers and proportions for categorical measurements. Differences in participant characteristics between different KORA examination cycles were evaluated by one-way repeated measures ANOVA (continuous data) or Cochran's Q test (categorical data) [24]. MRI parameters from Exam 3 were also summarized as means with standard deviations.

The associations of 5 main CVD risk factors (systolic and diastolic BP, waist circumference, HbA1c and LDL-C) from three examination cycles with various CMR outcome variables are displayed in Additional file 1: Table S1. In essence, diastolic BP, waist circumference and LDL-C displayed consistent inverse associations with end-diastolic volumes and stroke volumes of the LV and the RV, respectively, across the three examination cycles. The strength of association was slightly stronger for BP traits and waist circumference from Exam 2 and 3 (contemporaneous with the CMR) than with the risk factors from Exam 1. WC and, to a certain extent also diastolic BP, were positively associated with epicardial and/or pericardial fat (Additional file 1: Table S1).

Systolic BP was most consistently associated with LV stroke volume and myocardial mass. These associations were slightly stronger for systolic BP measurements obtained at Exam 2 and 3, as compared to systolic BP obtained at Exam 1 (Additional file 1: Table S1).

By contrast, HbA1c did not provide evidence for a consistent association with CMR traits of cardiac structure and function or with cardial fat, even though we observed a statistically significant association of HbA1c from the Exam 3 with stroke volume and myocardial mass. However, these associations could not be confirmed with HbA1c levels assessed at the Exam 1 and 2 (Additional file 1: Table S1).

In aggregate, the risk factor profiles from the Exam 3 (contemporaneous with the CMR measurements) explained slightly more variation in the different CMR outcomes than the risk factor levels obtained at baseline or at Exam 2: R²-values for the regression models increased slightly but consistently for all cardiac function and structure parameters from the first to the third examination cycle (Additional file 1: Table S1).

For each risk factor, we defined three trajectory clusters based on mean risk factor levels and the change of the risk factors over time (Fig. 2). Individuals with similar mean values and comparable longitudinal trajectories were grouped together, as detailed in the methods section. In the high and low-level clusters, mean values for systolic and diastolic BP decreased whereas mean values for waist circumference increased over the 14 year follow-up period (Fig. 2). Mean HbA1c values increased substantially in the high level cluster (reflecting a higher proportion of incident diabetes cases; n = 15), but remained essentially unchanged in the middle (incident diabetes cases n = 9) and low-level cluster (n = 0). Mean LDL-C levels slightly decreased in the high-level cluster, whereas LDL-C levels in the middle and low-level cluster slightly increased.

The association of the different individual CVD risk factor trajectory clusters with CMR measures is displayed in Table 3. Consistent with the associations of individual risk factors measurements (Additional file 1: Table S1), the clusters representing high cumulative exposure (high mean levels) of waist circumference, diastolic BP and LDL-C displayed the most consistent evidence for association with end-diastolic diameters and stroke volumes of the LV and the RV (inverse associations) as well as with epi- and pericardial fat (positive associations for waist circumference and diastolic BP) (Table 3). For waist circumference, the mid-level trajectory cluster displayed a comparable pattern of association (same directionality, lower strength of association) as the high level trajectory (Table 3).

We also estimated multivariable trajectory clusters, taking into account longitudinal information from all five CVD risk factors (systolic and diastolic BP, waist circumference, HbA1c, LDL-C). These multivariable risk factor trajectories were clearly and consistently associated with nearly all CMR traits, except for LV mass (which was specifically related to high systolic BP), and with epicardial and pericardial fat. Figure 3 displays the adjusted means of all CMR traits according to the three multivariable risk factor trajectory clusters. In essence, we observed a graded relation between the multivariable trajectory clusters and most CMR traits in the sense that higher exposure to multiple risk factors over time (the middle risk and the high risk trajectories) was associated with lower end-diastolic volumes of the LV and RV, lower stroke volumes of the LV and RV, lower LV diastolic filling rate, as well as with higher levels of epicardial and pericardial fat.

Model A in Table 4 denotes the cross-sectional association of multivariable risk factor clusters from Exam 3 with CMR traits. Once longitudinal information from the multivariable risk factor trajectory clusters is added to the cross-sectional multivariable risk clusters from Exam 3 (Table 4; Model B), the model fit improved [LR tests were statistically significant (Table 4)], and the risk factor clusters from Exam 3 only were no longer statistically significant whereas the longitudinal multivariable risk factor trajectory clusters were statistically significantly associated with various CMR traits (Table 4; Model B).

Thus, the (longitudinal) multivariable trajectory clusters added significant information in explaining variation of the various CMR traits beyond the multivariable risk profile obtained at Exam 3 (contemporaneous with the CMR exam, Table 4, all p values ≤ 0.011, except for myocardial mass). Results of further risk factor clusters and CMR parameters are available in Additional file 1: Tables S2, S3 and S4, respectively. Sensitivity analyses with adjustment for covariates from Exam 1 (Additional file 1: Table S5) did not reveal substantially different results.

The results can be summarized as follows: First, individual measurements of diastolic BP, waist circumference and LDL-C were consistently, inversely associated with LV and RV stroke volumes and end-diastolic volumes. The association of these risk factors with CMR traits was slightly stronger for risk factors from Exam 2 and 3 (closer to the CMR exam conducted at Exam 3), compared to Exam 1. Second, mainly waist circumference, but also diastolic BP were associated with cardiac fat depots. Third, systolic BP was most significantly associated with myocardial mass, and—to some extent—with LV stroke volume. Fourth, HbA1c did not display a consistent pattern of association with CMR traits or epi- and pericardial fat. Fifth, long-term exposure to high levels of diastolic BP, waist circumference and LDL-C (high risk factor levels trajectories) displayed statistically significant associations with end-diastolic volumes and stroke volumes of both ventricles. Sixth, long-term exposure to high and medium levels of several risk factors (multivariable trajectory clusters) displayed strong and consistent associations with multiple CMR traits and fat depots and added additional information for the explanation of CMR outcomes beyond multivariable risk measurement from Exam 3 (concurrent to the CMR exam).

We observed consistent associations of diastolic BP, waist circumference and LDL-C with LV and RV end-diastolic and stroke volume. While the associations of these CVD risk factors from Exam 2 and 3 were slightly stronger than with risk factors measured at Exam 1, defined risk factors from Exam 1 also displayed significant associations with CMR traits obtained 14 years later at Exam 3. In essence, Exam 1 BP traits and LDL-C were associated with LV mass and stroke volume and Exam 1 waist circumference was related to RV parameters and to cardiac fat depots.

The inverse association of diastolic BP, waist circumference and LDL-C with filling rates, end-diastolic volume and stroke volume could be consistent with these risk factors contributing to smaller stiffer ventricles which are potential precursors of heart failure with preserved ejection fraction later in life [30].

Prior studies provided cross-sectional evidence for associations of standard cardiovascular risk factors with subclinical cardiac remodeling [9‐11, 31]. In a sample of the UK Biobank, the impact of several modifiable risk factors on CMR-measured subclinical alterations of all four cardiac chambers was analyzed [10]. Similarly to our study, higher systolic BP was related to higher LV mass and volumes while higher diastolic BP and higher cholesterol levels were related to lower LV mass and volumes.

The association of systolic BP with LV mass was a prominent results in our analyses and agrees well with another study [31]. In fact, high LV mass is considered an integral measure of long-time exposure to high BP and it has previously been demonstrated in data from the Framingham Heart Study that systolic BP is an important correlate of LV mass progression over 4 years as well as over the adult life course [31]. However, a bit surprisingly, systolic BP was not significantly associated with other CMR traits which is in part in contrast with prior observations [9‐11].

In addition, higher BMI was previously associated with higher LV mass and volumes not indexed by BSA [10] as compared to our study that demonstrated an inverse association of waist circumference with LV mass and volumes. Our results regarding the associations of BP and cholesterol levels with cardiac parameters are also in line with results of a CMR examination of the RV conducted within the community-based MESA study [11]. In the latter study, total cholesterol levels were inversely associated with mass and end-diastolic volume of the LV and RV while HDL-C was positively associated with end-diastolic volume in both ventricles. In contrast, Framingham Heart Study investigators did not reveal independent associations between lipid concentrations and echocardiographic measurements of LV structure in individuals free of CVD [32].

While clinically overt diabetes is associated with an increased risk for adverse cardiac remodeling [10, 11] and incident heart failure [33], it is less clear, whether HbA1c level in a general population sample (including many non-diabetic individuals) confer an increased risk for adverse cardiac remodeling. Analyses in the likewise community-based SHIP study displayed a lack of association of HbA1c levels with LV mass index and fractional shortening, key ultrasonographic measures of cardiac structure and function, respectively [34].

Data on the association of antecedent risk factor levels with clinical CVD events have been published in different cohort studies [35‐37]. Lee and colleagues investigated in the Framingham Heart Study the associations of remote (− 30 years), recent (− 20 years) and current (− 10 years) BP and BMI levels with incident heart failure and revealed that all three measurements of systolic BP, pulse pressure, and BMI were associated with incident heart failure independently of other covariates [36]. This study could not detect a significant impact of current, recent and remote diastolic BP on heart failure risk. Framingham Heart Study investigators have also reported a positive association between antecedent levels of systolic BP, diastolic BP and pulse pressure with the 10-year risk of overall CVD events, including death, myocardial infarction, stroke, claudication, heart failure and others in subgroups of sex and age [37]. A more recent study reported that antecedent systolic BP was a stronger predictor of the 20-year risk of CVD events than current systolic BP, whereas—in the same sample—, antecedent hypertension was not a significant predictor for incident CVD [35]. A further study observed the relation between lipid concentrations and incident heart failure over a long time period (mean = 26 years) and revealed a higher heart failure risk for higher non-HDL-C levels and a lower heart failure risk for higher HDL-C levels [38].

This is the first study to assess associations of repeated risk factor measurements and their trajectories over a 14 year time period with CMR-based measures of cardiac structure and function as well as peri- and epicardial fat. The evidence linking risk factor trajectories to subclinical CVD is not yet available in contrast to the evidence for overt CVD.

In a community-based sample of elderly (≥ 65 years) individuals, BP measurements were recorded at least 4 times over a period of 7 years. The relation of BP trajectory clusters (cluster 1: increasing systolic and diastolic BP from low levels; cluster 2: constant systolic BP and decreasing diastolic BP; cluster 3: decreasing systolic and diastolic BP from high levels) to all-cause mortality, incident CVD (fatal and nonfatal cerebrovascular accidents and myocardial infarctions) and incident heart failure resulted in cluster 3 to be exposed to highest risk for all three outcomes compared to cluster 1 [39]. Attard and colleagues reported 10 different longitudinal BMI trajectories from adolescence to adulthood (mean age 17–29 years) and their association with CVD risk [40] over a period of 12 years. These different BMI trajectories were associated variably with higher risk of diabetes, hypertension and inflammation (as main risk factors for cardiac remodeling) [41] compared to the referent cluster of constant low BMI over time in both, in women and men.

A further study compared metabolic and biochemical risk factor trajectories between a group of incident CVD cases and an age- and sex-matched control group over a time period up to 15–20 years and revealed that systolic BP and waist circumference trajectories were more unfavorable in the CVD group compared to the control group [42]. However, trajectories of BMI, diastolic BP, total and HDL-C levels were not different between both groups of that latter study. This is partly in line with the results from our analysis, where we observed that only some single risk factor trajectory clusters (diastolic BP, waist circumference, LDL-C) were associated with selected cardiac parameters.

However, in our sample the multivariable clusters incorporating all risk factors simultaneously were significantly associated to all cardiac outcomes except LV mass, which underscores the importance of taking the conjoint evolution of risk factors into account.

Strengths of our study include the well-characterized sub-sample of the population-based KORA study, a large prospective cohort study with detailed and highly standardized measurements of a wide range of established cardiovascular risk factors and other phenotypes. Furthermore, we used advanced CMR techniques and quality assured reading procedures to characterize subclinical parameters of cardiac structure and function. To assess the associations between risk factors and CMR parameters we used not only risk factor measurements of the same examination cycle as the CMR examination (Exam 3) but also the examination cycles 7-years and 14-years prior the CMR examination.

The following limitations merit consideration. We conducted the CMR examination only at Exam 3. Thus, we were not able to investigate the associations of risk factor levels from prior exams or risk factor trajectories with changes in subclinical CMR parameters over time. However, we were able to relate longitudinal trajectories of multiple risk factors—in isolation and in combination—over 14 years to various CMR traits. Second, CMR measurements were obtained only in a subsample of the KORA cohort and the representativeness of this study sample for the initial cohort sample is limited. However, an analysis of cardiac parameters using statistical sampling weights to attain representativeness of the CMR sample to the full underlying cohort revealed no major differences to an unweighted analysis [43]. Third, we only identified three clusters for each risk factor and focussed on the most prevalent risk factors in our study sample. Fourth, except for the main analyses the results were not adjusted for multiple comparisons, thus weak associations remain questionable. And fifth, our sample was of European ancestry with unknown generalizability to other ethnic samples.

We observed strong associations between risk factors levels measured over a 14-years’ time horizon with subclinical cardiac remodeling traits as assessed by CMR. These associations suggest that prevention of adverse cardiac remodelling should start early in life, since long term exposure to higher (multiple) risk factor levels demonstrated consistent and graded associations with adverse subclinical alterations of cardiac function and structure. Prevention strategies should include optimal control of modifiable risk factors, including BP, waist circumference, and LDL-C. Further studies should evaluate whether longitudinal risk factor information can be used to improve CVD risk prediction models for adverse cardiovascular remodelling and clinically overt CVD events.