Predictors in Youth of Adult Cardiovascular Events

Abstract

BACKGROUND AND OBJECTIVES

Childhood risk factors are associated with cardiovascular events in adulthood. We compared the utility of a risk model based solely on nonlaboratory risk factors in adolescence versus a model that additionally included lipids to predict cardiovascular events in adulthood.

METHODS

The study comprised 11 550 participants from 7 longitudinal cohort studies in the United States, Australia, and Finland with risk factor measurements in adolescence and followed into adulthood. The adolescent risk factors were defined by using clinical standards including overweight or obesity, elevated blood pressure, smoking, and borderline high or high levels of total cholesterol and triglycerides. The main outcomes were medically adjudicated fatal or nonfatal cardiovascular disease events occurring after age 25.

RESULTS

Of 11 550 participants (55.1% female, mean age 50.0 ± 7.7 years), 513 (4.4%) had confirmed cardiovascular events. In a multivariable model (hazard ratio [95% confidence interval]), elevated blood pressure (1.25 [1.03–1.52]), overweight (1.76 [1.42–2.18]), obesity (2.19 [1.62–2.98]), smoking (1.63 [1.37–1.95]), and high total cholesterol (1.79 [1.39–2.31]) were predictors of cardiovascular events (P < .05). The addition of lipids (total cholesterol and triglycerides) into the nonlaboratory model (age, sex, blood pressure, BMI, and smoking) did not improve discrimination in predicting cardiovascular events (C-statistics for the lipid model 0.75 [SD 0.07] and nonlaboratory model 0.75 [0.07], P = .82).

CONCLUSIONS

Nonlaboratory-based risk factors and lipids measured in adolescence independently predicted adult cardiovascular events. The addition of lipid measurements to nonlaboratory risk factors did not improve the prediction of cardiovascular events.

What’s Known on This Subject:

Cardiovascular disease is a multifactorial disease beginning in childhood and primordial prevention would ideally be targeted to children and adolescents with modifiable risk factors. Lipid screening in adolescence could allow for the early identification of dyslipidemia but requires substantial resources.

What This Study Adds:

An approach that used only nonlaboratory risk factors obtained in youth predicted cardiovascular events as accurately as one that additionally contained information on lipids. This nonlaboratory approach offers an alternative to identifying adolescents at risk when laboratory testing is inconvenient.

Atherosclerosis is a multifactorial disease with its roots in childhood.^1,2 Recently, we showed that the childhood cardiovascular risk factors of BMI, systolic blood pressure, total cholesterol level, triglyceride level, and youth smoking, particularly in combination beginning in early childhood, were associated with adult cardiovascular events and death from cardiovascular causes before the age of 60 years.³ These data suggest that the primordial prevention of cardiovascular disease (CVD) should be targeted to children and adolescents.

In 2011, the National Heart, Lung, and Blood Institute (NHLBI) recommended the universal screening of lipid levels in youth.⁴ By contrast, in 2023, the US Preventive Services Task Force did not recommend screening in general risk assessment, arguing that the evidence to demonstrate its effectiveness was insufficient.⁵ In addition, the 2018 multi-society lipid guideline and the European Society of Cardiology/European Atherosclerosis Society lipid guideline recommended lipid screening for children and adolescents with a family history of either early CVD or significant hypercholesterolemia.^6,7 In addition, the International Atherosclerosis Society guidance for familial hypercholesterolemia recently provided recommendations regarding lipid measurements, suggesting that multiple screening strategies (selective, opportunistic, and universal) should ideally be used to detect familial hypercholesterolemia.⁸ Therefore, there exists inconsistency in the messages that pediatric care providers are receiving from expert bodies.

Efforts to create a risk stratification system that could simplify general risk assessment in situations in which laboratory testing is inconvenient or unavailable are needed. A study in adults⁹ revealed promising results supporting this kind of approach, with a nonlaboratory-based risk model predicting CVD events as accurately as one that relied on laboratory-based measurements. We previously reported that the addition of lipid measurements to traditional clinic-based risk factor assessment provided a statistically significant but clinically modest improvement in adolescent prediction of high carotid intima-media thickness in adulthood.¹⁰

In this study, we use data from the International Childhood Cardiovascular Cohort (i3C) Consortium, which includes 7 longitudinal cohort studies (Australia, Finland, and 5 in the United States) of cardiovascular risk factors initiated in childhood that have followed participants into adulthood. Our aim was to compare risk prediction models based on nonlaboratory versus nonlaboratory plus lipid data obtained at ages 12 to 19 years for predicting adult CVD events. The analyses of this study relate to the aspect of general pediatric lipid screening and do not address the purpose related to familial hypercholesterolemia.

Methods

Study Sample

The sample was composed of 11 550 participants aged 12 to 19 years at baseline from the i3C Consortium in 7 childhood cohorts from the United States, Australia, and Finland.^11,12 Data from the first available study visit were used within the age range of 12 to 19 years. All risk factor measurements were derived from the same study visit. All 7 cohorts included in this analysis have been previously described in detail,¹¹ and a brief description of the cohorts is provided in the Supplemental Information. The cohort studies followed protocols approved by local ethics committees, with signed informed consent or assent for participants as children or adolescents. All cohorts had data collected in clinical examinations that obtained the participants’ age, sex, height, weight, and blood pressure. Fasting levels of plasma or serum lipid and lipoprotein measurements were measured using standard methods. Data on youth smoking were based on reports by the participants during childhood augmented by adult recall of the smoking initiation date and analyzed as a dichotomous variable (yes versus no).³

Between 2015 and 2019, the i3C Consortium conducted a coordinated study to locate and survey the now-adult participants.³ US and Australian participants completed a Heart Health Survey to self-report cardiovascular events and medical procedures and update other relevant information; National Death Indices were also searched to determine the cause of death of participants. Cardiovascular event data for the Finnish participants through December 31, 2018 were obtained from Finland’s national medical registries. Altogether, 20 659 participants were personally located or identified as deceased with a coded cause of death. Of these 20 659, this report includes 11 550 participants who had data available between 12 and 19 years of age, the age range when smoking data were collected.

Definition of Risk Factors During Adolescence

To increase the generalizability of these data to the clinical setting and to be consistent with current recommendations, we analyzed the data using categorical adolescent risk factors. Overweight status was defined according to the extended international (International Obesity Task Force) cutoffs for BMI.¹³ Elevated blood pressure was defined according to the Clinical Practice Guideline for Screening and Management of High Blood Pressure in Children and Adolescents.¹⁴ High-risk plasma lipid levels were defined according to the NHLBI guidelines.⁴

Cardiovascular Events

The primary study endpoint was the first instance of fatal or non-fatal myocardial infarction, stroke, transient ischemic attack, ischemic heart failure, angina, peripheral artery disease, carotid intervention, abdominal aortic aneurysm, or coronary revascularization recorded in physician or hospital records or identified through the coded cause of death. Non-fatal events were self-reported in the Heart Health Survey by US and Australian participants, and medical records were requested for confirmation. A physician adjudication committee reviewed the records and classified each event as confirmed CVD, not an event, or not adjudicable. Fatal event diagnoses were adjudicated using the International Classification of Diseases coded causes of death. In Finland, diagnoses of fatal and non-fatal CVD events were based on International Classification of Diseases codes obtained from the national medical or death registries.

Statistical Methods

Statistical analyses were performed with SAS 9.4. Statistical significance was inferred at a 2-tailed value of P ≤ .05. The normality assumptions of the residuals were assessed by examining histograms of the residuals and normal probability plots. To examine the associations of exposures in adolescence and subsequent cardiovascular events, analyses of survival data based on the Cox proportional hazards model were used. All analyses were adjusted for age, sex, and study cohort. The ability of nonlaboratory model (including age, sex, study cohort, blood pressure, BMI, smoking) and lipid model (nonlaboratory model and additionally total cholesterol and triglycerides) data in adolescence to predict CVD events in adulthood was assessed by using Uno’s C-statistics in all study participants and in various subgroups.¹⁵ Additionally, more straightforward yet naïve Harrell’s C-statistics were calculated for similar models to further assess the predictive abilities. Although the Harrel’s C-statistic calculates the percentage of only comparable pairs whose risk has been correctly identified by the model and discards the pairs that are incomparable because of censoring, Uno’s method is independent of the censoring. The differences between the area under the curve functions were calculated by using Uno’s inverse probability of censoring weighting technique. The confidence limits were calculated on the basis of perturbation resampling.¹⁶ We also calculated category-free net reclassification improvement (NRI) and integrated discrimination improvement (IDI) to examine whether, based on these statistics, the addition of lipids enhances the prediction of CVD events compared with the nonlaboratory model.¹⁷ Models were tested for multicollinearity by using the variance inflation statistic, and no sign of multicollinearity was observed (variance inflation statistic <2.0 for all variables).

Results

Characteristics of the Study Participants

The risk factors of the study participants at ages 12 to 19 years are described in Table 1. Of 11 550 participants (55.1% female, mean age 50.0 ± 7.7 years when outcome data were obtained during adulthood), 513 (4.4%) had confirmed cardiovascular events. The characteristics of study participants and those not included in the analyses because of insufficient data are compared in Supplemental Table 4.

TABLE 1.

Participant Characteristics

	BHS		CDAH		MN		Muscatinea		NGHS		Princeton		YFS		All
N	2862		426		362		4259		436		423		2782		11 550
Sex, % female	56.9		51.4		45.6		52.5		100.0		54.6		52.0		55.1
Age at baseline, y	13.9	1.5	14.0	1.5	14.5	1.9	14.0	1.6	16.9	2.7	15.1	1.9	14.2	2.4	14.2	2.0
Age at censoring, y	47.2	7.2	46.2	2.2	32.4	2.5	54.9	5.7	40.2	1.3	56.6	5.7	48.8	5.1	50.0	7.7
Total cholesterol, mmol/L	4.16	0.74	4.45	0.75	3.88	0.72	4.02	0.70	4.21	0.83	4.40	0.83	5.14	0.92	4.36	0.90
LDL-C, mmol/L	2.37	0.66	2.68	0.69	2.27	0.62	2.25	0.60	2.53	0.75	2.75	0.76	3.21	0.84	2.69	0.83
HDL-C, mmol/L	1.52	0.49	1.42	0.30	1.13	0.26	1.27	0.26	1.34	0.28	1.33	0.32	1.57	0.32	1.47	0.40
Triglycerides, mmol/L	0.84	0.44	0.75	0.37	1.05	0.61	0.89	0.43	0.98	0.50	0.91	0.42	0.79	0.36	0.86	0.43
BMI	20.9	4.6	19.5	2.8	23.0	5.5	20.8	3.7	24.2	6.8	21.4	4.5	19.3	3.0	20.6	4.2
Systolic blood pressure, mmHg	107	9	113	13	108	9	115	13	108	9	109	12	115	11	112	12
Diastolic blood pressure, mmHg	54	12	67	12	57	13	66	12	67	9	67	10	67	10	63	13
Smoking, %	42.8		34.5		34.0		34.7		39.7		35.9		42.6		38.8
CVD events (%)b	139 (4.9)		4 (0.9)		0 (0)		252 (5.9)		3 (0.7)		31 (7.3)		84 (3.0)		513 (4.4)

BHS, Bogalusa Heart Study; CDAH, Childhood Determinants of Adult Health Study; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; MN, Minnesota Childhood Cardiovascular Cohorts; NGHS, NHLBI Growth and Health Study; YFS, Cardiovascular Risk in Young Finns Study. Values are mean (SD) unless otherwise stated.

^a LDL-C and HDL-C only available in subsample of 812 participants in the Muscatine Study.

^b A total of 213 (35%) of the events were fatal.

Nonlaboratory-Versus Lipid-Based (ie, Nonlaboratory With Lipid Measurements) Risk Assessment in Predicting CVD Events in Adulthood

Univariable analyses (Table 2) assessing relations between risk factors and CVD events revealed significant associations for age, sex, and most of the categorical risk factors (overweight, obesity, elevated blood pressure, smoking, high total cholesterol, high triglycerides, borderline low HDL cholesterol, low HDL cholesterol, and high LDL cholesterol). Table 3 reveals the results for a multivariable model assessing risk ratios for CVD events in adulthood according to nonlaboratory and lipid risk factors measured during adolescence. Among the nonlaboratory risk factors, overweight, obesity, elevated blood pressure, and smoking remained significantly associated with CVD events. When lipids (total cholesterol and triglycerides) were introduced into the model, the significant associations for overweight, obesity, elevated blood pressure, and smoking remained, with high total cholesterol also being significant. When total cholesterol was replaced with low-density lipoprotein (LDL) cholesterol and high-density lipoprotein (HDL) cholesterol in the lipid model (lipid model including LDL cholesterol, HDL cholesterol, and triglycerides; N = 8048), the significant association remained for overweight, obesity, elevated blood pressure, smoking, and high LDL cholesterol (Supplemental Table 5).

TABLE 2.

HRs for CVD Events in Adulthood According to Nonlaboratory Risk Factors and Lipids

Adolescent Risk Factor	No. of Participants With Event/Total No.	HR	95% CI
Age, y	513/11 550	1.10	1.05−1.16
Sex, male	513/11 550	2.12	1.77−2.54
Blood pressurea
Normotensive	267/7383	Ref
Elevated blood pressure	246/4167	1.45	1.20−1.75
BMIb
Normal weight	314/9098	Ref
Overweight	118/1754	1.96	1.59−2.42
Obese	54/698	2.73	2.04−3.65
Smoking
Nonsmoker	237/7070	Ref
Smoker	276 /4480	1.67	1.40−1.99
Total cholesterol
Normal (<4.40 mmol/L)	287/6660	Ref.
Borderline high (≥4.40−5.17)	122/2940	1.17	0.95−1.46
High (≥5.18)	104/1950	2.03	1.58−2.61
Triglycerides
Normal (<1.02 mmol/L)	339/8599	Ref
Borderline high (≥1.02−1.46 mmol/L)	107/2072	1.19	0.95−1.48
High (≥1.46 mmol/L)	67/879	1.82	1.39−2.37
LDL cholesterol
Normal (<2.85 mmol/L)	169/5093	Ref.
Borderline high (≥2.85−3.36 mmol/L)	43/1456	1.16	0.82−1.66
High (≥3.37 mmol/L)	68/1511	1.93	1.39−2.68
HDL cholesterol
Normal (>1.16 mmol/L)	205/6419	Ref.
Borderline low (1.16-1.03 mmol/L)	32/785	1.53	1.05−2.24
Low (<1.03 mmol/L)	42/872	1.74	1.23−2.44

Analyses were adjusted additionally for age in adolescence, sex, and study cohort.

^a Age- and sex-specific values defined according to the Clinical Practice Guideline for Screening and Management of High Blood Pressure in Children and Adolescents.

^b Age- and sex-specific values defined according to the Cole classification.

TABLE 3.

Multivariable HRs for CVD Event According to a Nonlaboratory Model and a Lipid Model Including Additionally Total Cholesterol and Triglycerides

		Nonlaboratory Model		Lipid Model
No. of Participants With Event/Total No.		513/11 550		513/11 550
Adolescent risk factor		HR	95 % CI	HR	95 % CI
Blood pressurea	Normotensive	Ref.		Ref.
	Elevated blood pressure	1.27	1.04−1.54	1.25	1.03−1.52
BMIb	Normal weight	Ref.		Ref.
	Overweight	1.83	1.48−2.27	1.76	1.42−2.18
	Obese	2.48	1.84−3.33	2.19	1.62−2.98
Smoking	Nonsmoker	Ref.		Ref.
	Smoker	1.63	1.37−1.94	1.63	1.37−1.95
Total cholesterol	Normal (<4.40 mmol/L)	N/A		Ref.
	Borderline high (≥4.40−5.17)			1.10	0.88−1.37
	High (≥5.18)			1.79	1.39−2.31
Triglycerides	Normal (<1.02 mmol/L)	N/A		Ref.
	Borderline high (≥1.02−1.46 mmol/L)			1.03	0.82−1.28
	High (≥1.46 mmol/L)			1.27	0.96−1.69

Analyses were adjusted additionally for age in adolescence, sex, and study cohort.

^a Age- and sex-specific values defined according to Clinical Practice Guideline for Screening and Management of High Blood Pressure in Children and Adolescents.

^b Age- and sex-specific values defined according to the Cole classification.

As shown in Fig 1, both in nonlaboratory and lipid (nonlaboratory plus lipids) models, the number of adolescent risk factors was associated with subsequent cardiovascular events.

FIGURE 1.

HRs for CVD event according to the number of risk factors stratified by nonlaboratory and lipid (nonlaboratory plus lipids) models. If a risk factor was above the normal cut points derived from the recommendations of the Expert Panel on Integrated Guidelines for Cardiovascular Health and Risk Reduction in Children and Adolescents or participant reported smoking, the risk factor was considered positive.

Figure 2 reveals the receiver operating characteristic curve values (Uno’s C-statistics) for the nonlaboratory (blood pressure, BMI, smoking) and lipid model (nonlaboratory plus total cholesterol and triglycerides) prediction of adult CVD events. The addition of lipids to the nonlaboratory model did not lead to higher C-statistics (C-statistics for the lipid model 0.75 [SD 0.07] and nonlaboratory model 0.75 [0.07], P = .82). When male and female participants, smokers and nonsmokers, participants with normal weight and overweight, participants within US cohorts, and participants within non-US cohorts were analyzed separately, the results remained essentially similar (Fig 2). When Harrell’s C-statistics were calculated for similar models (Supplemental Table 6), the results were essentially similar to those shown in Fig 2. The overall improvement in category-free NRI was 0.11 (95% confidence interval [CI] 0.008–0.21); for cases, it was −0.03, and for controls, it was 0.14, indicating that the addition of lipids improves the prediction of nonevents rather than events. However, IDI was 0.002 (P = .89), indicating that the difference in average predicted risks between the individuals with and without the outcome did not increase significantly when lipids were included in the prediction model.

FIGURE 2.

Receiver operating characteristic curve values for a model only including age and sex and comparisons of the nonlaboratory (age, sex, blood pressure, BMI, smoking) with the lipid models (additionally total cholesterol and triglycerides) in adolescents for the prediction of adult cardiovascular events.

When only fatal CVD events were used as an outcome, the results remained essentially similar. The number of adolescent risk factors was associated with fatal CVD events in both nonlaboratory and lipid models (Supplemental Fig 3). In C-statistics, no significant difference was observed when lipids were added to the nonlaboratory model (C-statistics for the lipid model 0.75 [SD 0.05] and nonlaboratory model 0.72 [0.06], P = .14).

Discussion

The findings from 7 international longitudinal cohorts examined in this study reveal that the risk of CVD events can be predicted by nonlaboratory adolescent risk factors (overweight, obesity, hypertension, and smoking). Lipid values were also associated with adult CVD events; however, the addition of lipids did not improve the models of prediction for CVD events.

In our recent report, we showed that childhood cardiovascular risk factors, BMI, systolic blood pressure, smoking, total cholesterol, and triglycerides were significantly associated with the incidence of adult CVD events by midlife.³ Previously, data from the Coronary Artery Risk Development in Young Adults study revealed a relation between the Framingham risk score and cardiovascular events in a young adult cohort (ages 22–36) followed for 20 years.¹⁸ It is noteworthy to mention that the original Framingham risk score does not include BMI or obesity, and when obesity is included, the overall prediction does not change.¹⁹ However, BMI in childhood and overweight or obesity in late adolescence have been associated with increased cardiovascular morbidity and mortality in adulthood.^20,21 The findings of the present study are consistent with the earlier observations revealing the relation of early risk factor profiling to CVD events.

In the NHANES I adult population (baseline, 25–74 years of age, N = 14 407), Gaziano et al⁹ examined whether a risk prediction model that did not require any laboratory tests could be as accurate as one requiring laboratory information. They observed that a model with nonlaboratory-based risk factors predicted CVD events as accurately as one that relied on laboratory-based values. The nonlaboratory model included age, blood pressure, smoking, BMI, history of diabetes mellitus, and history of blood pressure treatment, whereas in the laboratory-based model, BMI was replaced with total cholesterol level. In addition, the Fuster-BEWAT Score (blood pressure, exercise, weight, alimentation, and tobacco) that requires no laboratory tests has been shown to predict the presence and extent of subclinical atherosclerosis in adults with similar accuracy than the ideal cardiovascular health score that is recommended for use in primary prevention.²² Accordingly, we previously reported that the addition of adolescent lipid measurements to traditional clinic-based risk factor assessment (sex, blood pressure status, BMI status) provided a statistically significant but clinically modest improvement on prediction of high carotid intima-media thickness in adulthood.¹⁰ In the present study, our findings from 7 cohort studies among youth 12 to 19 years of age were essentially similar. Although, in line with our previous analyses,³ adolescent lipid levels in this sub-study were related to CV events, nonlaboratory-based risk factors predicted CVD events as accurately as an approach that additionally considered lipids. Our results were consistent in several subgroup analyses based on sex, study cohort location, smoking, and adiposity status. In addition, we observed statistically significant improvement in the category-free NRI. However, there was no improvement in IDI in the prediction of CVD events when the laboratory model was compared with the nonlaboratory model. Thus, it is likely that such small movement noticed in the NRI is not clinically relevant. There are several possible explanations for this finding. First, within the baseline age range of our study cohort, there are substantial changes in the lipid profile related especially to pubertal development and growth.²³ These physiologic changes may reduce the ability of the lipid profile to add to the risk prediction. Second, the variables in the nonlaboratory model, especially BMI and blood pressure, contain information about the lipid levels,²⁴ so this model partly, implicitly, includes lipids. Third, especially compared with BMI measurements, there is more physiologic (long-term) and analytical variation (day-to-day) in lipid values, and thus, the use of repeated lipid measures instead of a single measure could improve the predictive performance of lipids.²⁵

Strong evidence suggests that CVD has its origins in childhood.^1,2 Prevention strategies conducted in children have provided evidence of the benefits of lifestyle counseling on risk markers and remain the cornerstones for promoting cardiovascular health in children at the population level.² Recently, the Special Turku Coronary Risk Factor Intervention Project for Children study revealed favorable effects on risk factors over a period of 26 years after dietary counseling was initiated in infancy and continued throughout childhood.²⁶ In addition to the population strategy, the identification of children who are at a high risk of atherosclerosis could be effective in allowing personalized interventions. However, there is no widely accepted childhood or adolescent risk prediction method that uses risk factor data obtained from apparently healthy youths. From the Pathobiological Determinants of Atherosclerosis in Youth data, a risk score has been developed estimating the probability for coronary artery lesions observed at autopsy, but it is only applicable for individuals 15 to 34 years of age.^27,28

The results from the present study suggest that a risk prediction method based on BMI, blood pressure, and smoking status that does not require any laboratory tests could be noninferior to one requiring laboratory information. This nonlaboratory approach would allow an easy and simple initial evaluation or identification of those youth who might benefit from a further investigation of major modifiable cardiovascular risk factors, such as lipids, as well as therapeutic lifestyle and possible medical interventions. Obvious benefits include avoiding the need to subject a child or adolescent to a blood draw, as well as reduced financial costs. Earlier (from 2011) pediatric guidelines recommend lipid screening for identifying familial hypercholesterolemia and predicting atherosclerosis.⁴ We acknowledge that our findings do not provide any evidence for or against screening for familial hypercholesterolemia, which is an important reason to measure lipid levels in childhood as a clearly defined risk factor for atherosclerotic CVD occurring in 1:250 individuals.²⁹ In addition, lipid measures might enhance the discrimination of individuals already in the high-risk group because participants with all 5 risk factors of the lipid model had notably higher risk for subsequent cardiovascular event (hazard ratio [HR] 6.20; 95% CI 3.59–10.73), compared with participants with all 3 risk factors of the nonlaboratory model (HR 4.05; 95% CI 2.92–5.64).

The strengths of this study include the large sample size, longitudinal study design, extensive follow-up of the participants who were well-phenotyped in adolescence among 7 cohorts, and the adjudication of medical records. However, it was not possible to locate half of the original adolescent participants, and medical records were not available for some participants self-reporting a cardiovascular event (included in the nonevent group). Second, we were unable to consider family history, for example, for familial hypercholesterolemia or pubertal stage, both of which may influence CVD risk factors, because data were not available for all cohorts. Third, lipid measurement methods differed between study cohorts. However, to take this into account, the analyses were adjusted for the cohorts. Finally, residual confounding is a possibility in all observational studies.

Conclusions

Our data from 7 international cohort studies reveal that both nonlaboratory risk factors and lipids in adolescence independently predict adult CVD events. The predictive value of an approach that additionally considered lipids using current clinical standards was not superior to a model that only included nonlaboratory factors.

Glossary

CI

confidence interval

CVD

cardiovascular disease

HDL

high-density lipoprotein

HR

hazard ratio

i3C

International Childhood Cardiovascular Cohort

IDI

integrated discrimination improvement

LDL

low-density lipoprotein

NHLBI

National Heart, Lung, and Blood Institute

NRI

net reclassification improvement