Exploring machine learning strategies for predicting cardiovascular disease risk factors from multi-omic data.

BACKGROUND: Machine learning (ML) classifiers are increasingly used for predicting cardiovascular disease (CVD) and related risk factors using omics data, although these outcomes often exhibit categorical nature and class imbalances. However, little is known about which ML classifier, omics data, or upstream dimension reduction strategy has the strongest influence on prediction quality in such settings. Our study aimed to illustrate and compare different machine learning strategies to predict CVD risk factors under different scenarios. METHODS: We compared the use of six ML classifiers in predicting CVD risk factors using blood-derived metabolomics, epigenetics and transcriptomics data. Upstream omic dimension reduction was performed using either unsupervised or semi-supervised autoencoders, whose downstream ML classifier performance we compared. CVD risk factors included systolic and diastolic blood pressure measurements and ultrasound-based biomarkers of left ventricular diastolic dysfunction (LVDD; E/e' ratio, E/A ratio, LAVI) collected from 1,249 Finnish participants, of which 80% were used for model fitting. We predicted individuals with low, high or average levels of CVD risk factors, the latter class being the most common. We constructed multi-omic predictions using a meta-learner that weighted single-omic predictions. Model performance comparisons were based on the F1 score. Finally, we investigated whether learned omic representations from pre-trained semi-supervised autoencoders could improve outcome prediction in an external cohort using transfer learning. RESULTS: Depending on the ML classifier or omic used, the quality of single-omic predictions varied. Multi-omics predictions outperformed single-omics predictions in most cases, particularly in the prediction of individuals with high or low CVD risk factor levels. Semi-supervised autoencoders improved downstream predictions compared to the use of unsupervised autoencoders. In addition, median gains in Area Under the Curve by transfer learning compared to modelling from scratch ranged from 0.09 to 0.14 and 0.07 to 0.11 units for transcriptomic and metabolomic data, respectively. CONCLUSIONS: By illustrating the use of different machine learning strategies in different scenarios, our study provides a platform for researchers to evaluate how the choice of omics, ML classifiers, and dimension reduction can influence the quality of CVD risk factor predictions.

Determinants of left ventricular diastolic function-The Cardiovascular Risk in Young Finns Study.

Decreased left ventricular (LV) diastolic function is associated with increased all-cause mortality and risk for a heart failure. The determinants of LV diastolic function have been mainly studied in elderly populations; however, the origin of LV heart failure may relate to the lifestyle factors acquired during the life course. Therefore, we examined biochemical, physiological, and lifestyle determinants of LV diastolic function in 34-49-year-old participants of the Cardiovascular Risk in Young Finns Study (Young Finns Study). In 2011, clinical examination and echocardiography were performed for 1928 participants (880 men and 1048 women; aged 34-49 years). LV diastolic function was primarily defined using E/é-ratio (population mean 4.8, range 2.1-9.0). In a multivariate model, systolic blood pressure (P < 0.005), female sex (P < 0.005), age (P < 0.005), waist circumference (P = 0.024), smoking (P = 0.028), serum alanine aminotransferase (P = 0.032) were directly associated with E/é-ratio, while an inverse association was found for height (P < 0.005). Additionally, a higher E/é-ratio was found in participants with concentric hypertrophy compared to normal cardiac geometry (P < 0.005). Other indicators of the LV diastolic function including E/A-ratio and left atrial volume index showed similarly strong associations with systolic blood pressure and age. In conclusion, we identified systolic blood pressure, waist circumference and smoking as modifiable determinants of the LV diastolic function in the 34-49-year-old participants of the Young Finns Study.

Cardiovascular Risk Factors in Childhood and Left Ventricular Diastolic Function in Adulthood.

BACKGROUND AND OBJECTIVES: Cardiovascular risk factors, such as obesity, blood pressure, and physical inactivity, have been identified as modifiable determinants of left ventricular (LV) diastolic function in adulthood. However, the links between childhood cardiovascular risk factor burden and adulthood LV diastolic function are unknown. To address this lack of knowledge, we aimed to identify childhood risk factors associated with LV diastolic function in the participants of the Cardiovascular Risk in Young Finns Study. METHODS: Study participants (N = 1871; 45.9% men; aged 34-49 years) were examined repeatedly between the years 1980 and 2011. We determined the cumulative risk exposure in childhood (age 6-18 years) as the area under the curve for systolic blood pressure, adiposity (defined by using skinfold and waist circumference measurements), physical activity, serum insulin, triglycerides, total cholesterol, and high- and low-density lipoprotein cholesterols. Adulthood LV diastolic function was defined by using E/é ratio. RESULTS: Elevated systolic blood pressure and increased adiposity in childhood were associated with worse adulthood LV diastolic function, whereas higher physical activity level in childhood was associated with better adulthood LV diastolic function (P < .001 for all). The associations of childhood adiposity and physical activity with adulthood LV diastolic function remained significant (both P < .05) but were diluted when the analyses were adjusted for adulthood systolic blood pressure, adiposity, and physical activity. The association between childhood systolic blood pressure and adult LV diastolic function was diluted to nonsignificant (P = .56). CONCLUSIONS: Adiposity status and the level of physical activity in childhood are independently associated with LV diastolic function in adulthood.

Influence of early-life body mass index and systolic blood pressure on left ventricle in adulthood - the Cardiovascular Risk in Young Finns Study.

BACKGROUND: Increased left ventricular mass (LVM) predicts cardiovascular events and mortality. The objective of this study was to determine whether early-life exposures to body mass index (BMI) and systolic blood pressure (SPB) affects the left ventricular structure in adulthood. METHODS: We used longitudinal data from a 31-year follow-up to examine the associations between early-life (between ages 6-18) BMI and SPB on LVM in an adult population (N = 1864, aged 34-49). The burden of early-life BMI and SBP was defined as area under the curve. RESULTS: After accounting for contemporary adult determinants of LVM, early-life BMI burden associated significantly with LVM (3.61 g/SD increase in early-life BMI; [1.94 - 5.28], p < 0.001). Overweight in early-life (age- and sex-specific BMI values corresponding to adult BMI > 25 kg/m2) associated with 4.7% (2.5-6.9%, p < 0.0001) higher LVM regardless of BMI status in adulthood. Overweight in early-life combined with obesity in adulthood (BMI > 30kg/m2) resulted in a 21% (17.3-32.9%, p < 0.0001) increase in LVM. Higher early-life BMI was associated with a risk of developing eccentric hypertrophy. The burden of early-life SPB was not associated with adult LVM or left ventricular remodeling. CONCLUSIONS: High BMI in early-life confers a sustained effect on LVM and the risk for eccentric hypertrophy independently of adulthood risk factors. KEY MESSAGES Excess in BMI in early-life has an independent effect on LVM and the risk of developing eccentric hypertrophy regardless of overweight status in adulthood. Systolic blood pressure levels in early-life did not have an independent effect on LVM or LV remodeling. The clinical implication of this study is that primary prevention of obesity in early-life may prevent the development of high LVM and eccentric hypertrophy.

Fatty liver index and left ventricular mass: prospective associations from two independent cohorts.

OBJECTIVES: Heart disease is the most common cause of death in patients with nonalcoholic fatty liver disease (NAFLD). Emerging data have shown that NAFLD may affect subclinical myocardial remodeling, mainly left ventricular hypertrophy; however, evidence from the prospective studies is still lacking. METHODS: Prospective analyses were performed to investigate the association of fatty liver index (FLI) with left ventricular mass (LVM) among 1962 participants from the Bogalusa Heart Study (BHS, 1995-2010) and 1547 participants from the Cardiovascular Risk in Young Finns Study (YFS, 2001-2011) free of cardiovascular diseases (CVD) at baseline. LVM was assessed by two-dimensional guided M-mode echocardiography and indexed (LVMI) to body height (m2.7). Multivariable regression models were applied after adjustment for traditional CVD risk factors. RESULTS: In both cohorts, we observed significant and positive associations between FLI and LVM (BHS: ß=0.59, P < 0.001; YFS: ß=0.41, P < 0.001) and LVMI (BHS: ß=0.14, P < 0.001; YFS: ß=0.09, P < 0.001). In addition, we found that the relationship between FLI and LVMI was stronger in women than men (BHS: P-interaction = 0.01; YFS: P-interaction < 0.01); and the relationship between FLI and LVM/LVMI was stronger in black than white individuals (LVM: P-interaction = 0.02; LVMI: P-interaction = 0.04). Moreover, we found that the associations of FLI with LVM and LVMI were attenuated by high physical activity, especially in BHS (P-interaction = 0.02). CONCLUSION: Our findings from two independent prospective cohorts indicate that FLI is positively associated with LVM/LVMI, independent of traditional cardiovascular risk factors. Such relationships are more pronounced among women and black individuals and are attenuated by high physical activity.

Aortic sinus diameter in middle age is associated with body size in young adulthood.

OBJECTIVE: Aortic sinus dilatation can lead to aortic valve regurgitation or even aortic dissection. Our objective was to examine the association between body surface area (BSA) measures from childhood to middle age and aortic sinus diameter in middle age. Understanding the relation of these two clarifies how aortic size is normally determined. METHODS: Cardiovascular Risk in Young Finns Study is a longitudinal study with follow-up of over 31 years (1980-2011). The study comprises information of body composition from multiple time points of 1950 subjects with cardiac ultrasound measurements made in 2011. The association between BSA in different ages and aortic sinus diameter in middle age was analysed by linear regression modelling adjusted with age, sex and diastolic blood pressure. Missing BSA values were derived for each life year (ages 3-33 years) from subject-specific curves for body weight and height estimated from longitudinal measurements using mixed model regression splines. RESULTS: BSA estimates in early 20s are most strongly associated with aortic sinus diameter in middle age. Top association was observed at age 23 years with one SD increase in estimated BSA corresponding to 1.04 mm (0.87-1.21 mm) increase in aortic diameter. Increase in body weight beyond early 20s does not associate with aortic sinus diameter, and the association between middle age BSA and aortic size is substantially weaker (0.74 mm increase (0.58-0.89 mm)). These results were confirmed in a subpopulation using only measured data. CONCLUSION: The association between aortic sinus diameter and BSA is stronger when considering BSA in young adulthood compared with BSA in middle age.