Risk of Coronary Artery Disease Associated With Transitions in Metabolic Health in a Clinical Cohort of 69 272

Abstract

Background

The effect of recent onset metabolic dysfunction on coronary artery disease (CAD) risk is poorly understood. We developed a large data set linking metabolic phenotypes and clinical outcomes to quantify CAD risks associated with adverse metabolic transition.

Methods

Clinical parameters, measures of metabolic dysfunction components (diabetes, hypertension, elevated triglycerides, and low high‐density lipoprotein) and incident CAD were curated in a clinical cohort from a single quaternary medical center. Associations between body weight, metabolic abnormalities, and changes in metabolic health status were assessed for prevalent and incident CAD.

Results

Increasing body mass index category, presence of metabolic dysfunction (>2 components), and number of metabolic abnormalities were associated with prevalent CAD among 844 841 individuals. In time‐to‐event analyses (N=69 272), ~38% of subjects initially free of any metabolic dysfunction abnormality developed ≥1 abnormality over a 3‐year run‐in period. All metabolic abnormalities, whether new‐onset or preexisting, significantly increased 10‐year CAD event probability, and there was a progressive increase in 10‐year CAD risk with increasing burden of metabolic abnormalities. In models adjusting for metabolic dysfunction, 10‐year CAD risk increased for individuals with body mass index >25 kg/m² (hazard ratio, 3.8). Compared with individuals with a body mass index of 25 kg/m², any short‐term weight gain or loss of >5% increased or decreased CAD risk, respectively, in individuals with a body mass index of 35 kg/m².

Conclusions

In a large clinical cohort, the transition to metabolic dysfunction is common, occurs rapidly, and significantly increases incident CAD risk. The effect of body weight and weight loss on CAD risk is nonlinear. Interventions to prevent progression to metabolic dysfunction are needed.

Nonstandard Abbreviations and Acronyms

COMMODORE

Cardiovascular and Multiple Metabolic Disease in Obesity Resource

hypertension

hypertension

Research Perspectives.

What Is New?

• The transition from metabolic health to metabolic dysfunction is common in clinical practice, can occur over short time periods, and is associated with a rapid and significant increase in risk for coronary artery disease.

What Question Should Be Addressed Next?

• What patients are at greatest risk for adverse metabolic transition and can hospital‐based interventions to prevent the transition to metabolic dysfunction decrease coronary artery disease risk and improve patient outcomes?

Obesity is an important risk factor for multiple cardiovascular conditions, including coronary artery disease (CAD). ¹ , ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ Risk is highest among individuals with a constellation of biochemical and physiologic abnormalities known as metabolic syndrome, which is thought to be driven largely by visceral adiposity, inflammation, and insulin resistance. ⁸ , ⁹ , ¹⁰ These abnormalities promote dysregulation of glucose and lipid homeostasis and contribute to hypertension. ¹¹ There has been uncertainty regarding the magnitude of risk associated with excess body weight in the absence of hypertension, diabetes, and obesity‐associated dyslipidemia. ¹² , ¹³ However, accumulating evidence suggests that such “metabolically healthy” obesity may be associated with a higher incidence of CAD. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ At least part of this risk derives from the development of metabolic abnormalities over time, ¹⁷ , ¹⁸ , ¹⁹ suggesting that, for many individuals, apparent metabolic health is, in fact, a dynamic, transitional phenotype for eventual metabolic dysfunction and associated health risk. However, studies assessing the effects of body weight, metabolic health, and the transition to metabolic dysfunction on CAD risk have largely been conducted in population‐based and ambulatory cohorts, and their effects in contemporary clinical populations are incompletely defined. ² , ²⁰ , ²¹ , ²² , ²³ Understanding these dynamic effects is critical to the creation and implementation of clinical strategies to identify at‐risk patients and mitigate the untoward effects of obesity and weight‐related metabolic dysregulations.

We created a large, hospital system‐based resource of well‐phenotyped individuals with dense electronic health record (EHR) data. Leveraging local expertise and resources in computational phenotyping, we curated adult individuals for indices of body weight, including directly measured weight and weight indexed to height (body mass index [BMI]), metabolic phenotypes, and cardiovascular outcomes to develop the Vanderbilt Cardiovascular and Multiple Metabolic Disease in Obesity Resource (COMMODORE). We leverage this resource to assess associations between body weight, metabolic abnormalities, and changes in metabolic health status with prevalent and incident CAD.

Methods

The study was approved by the Vanderbilt University Medical Center Institutional Review Board and deemed exempt as the project involves deidentified clinical data and thus is considered nonhuman subjects research. The aggregate data supporting the findings of this study are available from the corresponding author upon reasonable request; however, due to institutional data use agreements, access to individual‐level data is limited.

Study Population

The study population was derived from the synthetic derivative, a deidentified mirror of the Vanderbilt University Medical Center EHR designed to support research. ²⁴ The synthetic derivative contains inpatient and outpatient clinical data for >2.9 million individuals including demographics, vital signs, height and weight measurements, diagnostic and procedure codes, clinical documents, laboratory values, and medications. Demographic variables including sex, race, and ethnicity are based on EHR report, which may reflect self‐report as well as administrative or clinical assignment.

Curation of Clinical Variables

All adult heights and weights were extracted and cleaned. Full details of data cleaning and harmonization procedures are available in Supplemental Methods. Weights were recorded as directly measured values and used, in conjunction with height measures, to calculate BMIs. BMI classification was based on median BMIs and World Health Organization criteria, ²⁵ normal: 18.5 to 24.99 kg/m², overweight: 25.0 to 29.99 kg/m², and obese: ≥30 kg/m². Obesity, was further classified by obesity class, as class I: 30.0 to 34.99 kg/m², class II: BMI 35 to 39.99 kg/m², and class III: ≥40 kg/m². Underweight individuals (BMI <18.5 kg/m²) were excluded because underweight status in clinical populations is frequently associated with comorbidities unrelated to the goals of this analysis.

Clinical phenotypes, including CAD outcomes, were extracted using validated algorithms ²⁶ , ²⁷ and methodologies described in detail in the supplemental methods. Briefly, CAD was identified using a random forest classifier built using 162 CAD‐related International Classification of Diseases (ICD) codes (w.g., 410* “Myocardial Infarction,” 414.* “Ischemic heart disease”), Current Procedural Terminology codes, laboratory values, medication entries, and text mentions, such as “coronary cath/catherization,” “left/right coronary stenosis,” and “CABG [coronary artery bypass graft].” Laboratory values were extracted from structured tables and an assessment of the completeness of laboratory variable completeness data in the entire cohort is available in Table S1. Waist circumference and fasting glucose measures were not widely available in the synthetic derivative and were not used in classification of metabolic syndrome status. Similar to other studies, ¹⁴ we used a modified definition of metabolic syndrome that we subsequently refer to as “metabolic dysfunction,” defined as the presence of ≥2 of the following: (1) diabetes, (2) hypertension, (3) high‐density lipoprotein (HDL) <40 mg/dL (men) or <50 mg/dL (women), or (4) triglycerides >150 mg/dL.

Statistical Analysis

Descriptive and Cross‐Sectional Analyses

Summary statistics for categorical and continuous variables were presented as percentage and median and interquartile range, respectively. Between group comparisons were made using Wilcoxon, Kruskal–Wallis, and chi‐square tests as appropriate. Multivariable logistic regression was used to assess the associations between BMI group and metabolic dysfunction with prevalent CAD. We also performed stratified analyses by 10‐year age group, sex, and race. All regressions were adjusted for age, sex (for nonsex‐stratified analyses), and race (for nonrace‐stratified analyses). When relevant, the Bonferroni adjusted P value for significance is provided throughout the results to account for multiple testing.

Incident CAD Analyses

We used time‐to‐event analysis methodology to estimate the effects of body weight and metabolic status on incident CAD risk (Figure S1). Included subjects were required to have 4 height and weight measures collected during a prespecified 3‐year qualification (ie, “run‐in”) period, with ~1 year (±4 months) between each weight measure (t₀–t₃). Time of entry into the analysis, t₀, was the date of the first qualifying height and weight measure. Individuals with CAD before t₃ were excluded from the analysis. For all included subjects, components of metabolic dysfunction were classified as “preexisting” (present at t₀), “new onset” (diagnosed at t₁ through t₃), or “absent” (absent at t₃). The full medical record after the qualification period (after t₃) constituted the follow‐up time. To allow for flexibility, the effects of height, weight, and age were assumed to be smooth but not linear, using restricted cubic spline functions. These functions employ piecewise cubic polynomials to create smooth curves. The number of knots in the spline was determined based on the effective sample size to achieve a balance between flexibility and the risk of overfitting. This approach was selected to more accurately model weight changes that may have occurred during the qualification period and because of known limitations of BMI as a measure of weight with respect to prediction of cardiovascular outcomes. ²⁸ Observed CAD‐free probability over time was first visualized separately for each component of metabolic dysfunction using Kaplan–Meier plots. Log‐rank tests were then conducted to compare the Kaplan–Meier curves to determine if there were significant differences among these curves. Subsequently, the 10‐year cumulative risk of CAD was estimated using a Cox proportional hazards model that included age, sex, race, log‐transformed height (at t₃), all weights (at t₀–t₃), components of metabolic dysfunction (classified as preexisting, new onset, or absent). Risk was estimated with respect to the population's median age of 51.2 years, White race, and female sex. We also performed Cox proportional hazards to test the association between individual metabolic abnormalities and the overall burden (ie, total number) of metabolic abnormalities and risk of incident CAD. This model was fit by treating the number of metabolic abnormalities (ranging from 0 to 4) at the end of the run‐in period as categorical and the hazard ratios (HRs) were calculated based on each category (0–4 metabolic abnormalities) versus 0 abnormalities. To determine whether the HRs derived from this analysis are significantly different from one another we used a Wald test to compare a model with freely estimated beta coefficients for metabolic abnormalities against a restricted model. In the restricted model beta coefficients for metabolic abnormalities are set as equal, for example 1 versus 0 and 2 versus 0.

Data management was performed with Python 3.6.4 (www.python.org), and statistical analyses and data visualization were performed in R 3.4.3. ²⁹

Results

Clinical Characteristics of COMMODORE

We identified 844 841 individuals with complete demographics (race, sex, date of birth) and at least 1 BMI (Table 1). The cohort was 41.6% male, 73.7% White, and 9.4% Black, with a median age of 47.4 years (31.7–61.5) and median medical record length from first BMI to last entry of 2.8 years (0.5–7.0).

Table 1.

Clinical Characteristics and Demographics of the COMMODORE Data Set in All Individuals and Across Body Mass Index Groups

	Overall	Normal	Overweight	Class I obesity	Class II obesity	Class III obesity
Number of individuals	844 841	275 525	270 925	161 606	76 804	59 981
Lifetime median body mass index, kg/m2	27.5 (23.9–32.3)	22.6 (21.1–23.8)	27.3 (26.1–28.6)	32.1 (31.0–33.4)	37.0 (35.9–38.3)	44.1 (41.7–48.2)
Median age, y	47.4 (31.7–61.5)	39.7 (25.6–58.8)	50.1 (34.3–63.9)	51.3 (37.4–62.9)	49.7 (36.9–60.7)	46.6 (35.1–57.4)
Male, sex	351 290 (41.6)	96 005 (34.8)	133 772 (49.4)	74 858 (46.3)	29 215 (38.0)	17 440 (29.1)
Race
Asian	14 001 (1.7)	7802 (2.8)	4363 (1.6)	1292 (0.8)	362 (0.5)	182 (0.3)
Black	79 652 (9.4)	18 479 (6.7)	22 445 (8.3)	17 732 (11.0)	10 436 (13.6)	10 560 (17.6)
Native American/Alaskan	1549 (0.2)	460 (0.2)	567 (0.2)	306 (0.2)	116 (0.2)	100 (0.2)
Other**	3326 (0.4)	1265 (0.5)	1148 (0.4)	574 (0.4)	211 (0.3)	128 (0.2)
White	622 691 (73.7)	205 658 (74.6)	203 031 (74.9)	118 411 (73.3)	54 885 (71.5)	40 706 (67.9)
Unknown*	108 406 (12.8)	37 954 (13.8)	34 138 (12.6)	19 768 (12.2)	9210 (12.0)	7336 (12.2)
Ethnicity
Hispanic	20 091 (2.4)	5632 (2.0)	6948 (2.6)	4403 (2.7)	1911 (2.5)	1197 (2.0)
Length of record, y	2.8 (0.5–7.0)	2.6 (0.5–6.4)	2.9 (0.5–7.2)	2.9 (0.5–7.3)	3.0 (0.5–7.3)	2.7 (0.4–6.9)
Coronary artery disease	42 541 (5.0)	9132 (3.3)	15 784 (5.8)	10 493 (6.5)	4601 (6.0)	2531 (4.2)
Hypertension	302 227 (35.8)	63 147 (22.9)	96 361 (35.6)	71 795 (44.4)	38 202 (49.7)	32 722 (54.6)
Diabetes	91 953 (10.9)	14 129 (5.1)	25 420 (9.4)	23 193 (14.4)	14 887 (19.4)	14 324 (23.9)
Metabolic dysfunction	156 822 (18.6)	25 350 (9.2)	49 438 (18.2)	40 235 (24.9)	22 352 (29.1)	19 447 (32.4)
Estimated glomerular filtration rate, mL/min	89.0 (73.7–104.7)	93.0 (77.9–109.0)	87.4 (72.7–102.5)	86.0 (71.0–101.2)	86.7 (70.8–102.9)	89.6 (72.5–106.5)
Triglycerides, mg/dL	113.0 (79.0–166.0)	88.0 (64.0–126.0)	114.0 (80.5–165.0)	133.0 (94.0–192.0)	139.0 (98.0–200.0)	134.0 (95.0–189.5)
Total cholesterol, mg/dL	182.0 (156.5–208.0)	180.0 (155.0–205.0)	184.0 (158.0–210.0)	183.0 (157.0–210.0)	181.0 (156.0–207.5)	178.0 (154.0–203.0)
Triglycerides/HDL ratio	2.3 (1.4–3.9)	1.5 (1.0–2.5)	2.3 (1.4–3.8)	3.0 (1.9–4.8)	3.2 (2.0–5.2)	3.1 (2.0–4.8)
Hemoglobin A1c, mmol/mol	5.7 (5.3–6.6)	5.5 (5.1–6.0)	5.7 (5.2–6.4)	5.8 (5.3–6.8)	5.9 (5.4–7.0)	6.0 (5.5–7.1)
Brain natriuretic peptide, ng/L	111.0 (39.5–331.0)	152.8 (50.0–495.0)	126.0 (44.5–372.0)	100.0 (36.5–281.0)	86.0 (32.0–238.0)	73.0 (30.0–194.0)
HDL, mg/dL	49.0 (40.0–61.0)	58.0 (46.0–71.0)	49.0 (40.0–60.0)	45.0 (37.0–55.0)	43.5 (36.0–53.0)	43.5 (36.0–52.0)
Low‐density lipoprotein, mg/dL	103.0 (82.0–125.0)	99.0 (80.0–120.5)	105.0 (84.0–128.0)	105.0 (83.0–128.0)	103.5 (82.0–126.0)	102.5 (82.0–124.0)
C‐reactive protein, mg/L	5.7 (1.7–24.5)	3.6 (1.0–24.2)	4.8 (1.6–22.1)	6.3 (2.4–22.4)	8.1 (3.5–25.1)	11.8 (5.2–32.8)

Median (lower; upper quartile) of individual lifetime medians are shown for continuous phenotypes. N (%) are displayed for binary phenotypes. Laboratory values are present for a subset of patients with information in their records. COMMODORE indicates Cardiovascular and Multiple Metabolic Disease in Obesity Resource; and HDL, high‐density lipoprotein.

^*The Unknown race category includes individuals with an electronic health record reported race that is missing, unknown, or reported as belonging to >1 race group.

^**The Other race category includes individuals who reported other on the clinical intake forms.

The median BMI of the overall cohort was 27.5 kg/m² (23.9–32.3) and 298 391 (35.3%) individuals had a median BMI >30 kg/m². Metabolic dysfunction was present in 156 822 individuals (18.6%) (Table S2). Individuals with metabolic dysfunction had higher median BMI, were older, and were more commonly male and Black (P<0.001 for all). Differences in the median BMI, metabolic dysregulation, and other clinical characteristics based on race and sex are presented in Tables S3 through S6.

Obesity, Metabolic Dysfunction, and Prevalent Coronary Artery Disease

Consistent with prior cross‐sectional studies, we replicated many known associations between obesity, metabolic dysfunction, and CAD prevalence, supporting the validity of the data set. For example, increasing BMI category was associated with higher prevalence of metabolic dysfunction (P<0.001; Table 1 and Figure S2A), all individual metabolic dysfunction components (P<0.001 for all; Table 1 and Figure S2B through S2E), and a higher prevalence of CAD (P<0.001) (Table 1). The presence of ≥1 metabolic dysfunction components was strongly associated with prevalent CAD in all BMI categories in the overall population (P<2.2×10⁻¹⁶ for all). In the absence of metabolic abnormalities, the association between BMI group on CAD risk was small and generally similar across race and sex (Tables S4 and S6, respectively).

Preexisting Metabolic Abnormalities and Incident Coronary Artery Disease Risk

There were 69 272 individuals who met inclusion criteria for time‐to‐event analysis. During a median follow‐up of 6.58 years (IQR: 4.70‐9.49), 15 074 (21.8%) subjects developed CAD (Table 2). Individuals who developed CAD were significantly older, heavier, predominantly male, more often Black, and had a higher burden of preexisting metabolic dysfunction components (Table 2). In adjusted Cox proportional hazards regression analysis, age, male sex, and preexisting diabetes, hypertension, and elevated triglycerides were independent predictors of incident CAD. In adjusted models there was no association between body weight and CAD risk (Figure 1). A comparison of CAD risk based on the number of metabolic abnormalities at the end of the run‐in period as a continuous variable demonstrated a dose‐dependent relationship between CAD risk and number of metabolic dysfunction components (Figure 2). We also assessed the risk for CAD across the various groups by treating the number of abnormalities as categories (Table S7). All categories including a metabolic abnormality at t₃ demonstrated a significant increase in CAD hazards. We then performed a Wald test to explicitly examine whether the HRs between the groups differ and found that all HRs were different with the exception of the categories including 2 or 3 abnormalities (P=0.24). The greatest hazard existed for individuals with 4 metabolic abnormalities compared with those with none (HR, 2.32 [95% CI, 1.86–2.88]).

Table 2.

Clinical Characteristics of the Population With Incident CAD

	Incident coronary artery disease		P value
	No (n=54 198)	Yes (n=15 074)
Minimum age, y	44.3 (30.3–56.9)	53.6 (42.7–63.6)	<0.001
Sex, male	17 290 (31.9)	5852 (38.8)	<0.001
Race			<0.001
Asian	1175 (2.2)	228 (1.5)
Black	5411 (10.0)	1804 (12)
Native American/Alaskan	99 (0.2)	27 (0.2)
Other*	161 (0.3)	43 (0.3)
White	46 417 (85.6)	12 917 (85.7)
Unknown†	935 (1.7)	55 (0.4)
Height, cm	167.6 (162.5–175.3)	167.6 (162.6–177.8)
Weight, kg
Weight t0, kg	76.7 (64.0–92.0)	82.1 (68.6–97.5)	<0.001
Weight t1, kg	77.2 (64.6–92.5)	82.4 (69.0–97.5)	<0.001
Weight t2, kg	77.7 (65.0–93.0)	82.5 (69.0–98.2)	<0.001
Weight t3, kg	78.0 (65.3–93.4)	82.6 (69.0–98.4)	<0.001
Weight change t3–t0, kg	1.3 (−2.3–5.0)	0.7 (−2.8–4.5)	<0.001
Record length, y	6.9 (4.9–9.9)	5.8 (4.1–8.2)	<0.001
Time to outcome, y		5.8 (4.1–8.2)
Metabolic syndrome components
Diabetes			<0.001
Absent	48 259 (89.0)	12 154 (80.6)
Preexisting	1860 (3.4)	1167 (7.7)
New onset	4079 (7.6)	1753 (11.6)
Hypertension			<0.001
Absent	36 912 (68.1)	6989 (46.4)
Pre‐existing	3812 (7.0)	3012 (20.0)
New onset	13 474 (24.9)	5073 (33.6)
Low high‐density lipoprotein‡			<0.001
Absent	48 240 (89.0)	12 506 (83.0)
Preexisting	935 (1.7)	523 (3.4)
New onset	5023 (9.3)	2045 (13.6)
Elevated triglycerides§			<0.001
Absent	48 689 (89.9)	12 381 (82.2)
Preexisting	833 (1.5)	593 (3.9)
New onset	4676 (8.6)	2100 (13.9)

Data are presented as the median (lower; upper quartile) or number (%).

^*The Other race category includes individuals who reported other on the clinical intake forms.

^† The Unknown race category includes individuals with electronic health record reported race that is missing, unknown, or reported as belonging to >1 race category.

^‡ <40 mg/dL in men and <50 mg/dL in women.

^§ >150 mg/dL.

Figure 1. Independent effects on the risk of developing coronary artery disease.

The presented data represent adjusted hazard ratios of age (61.1 vs 35.8), height (177 vs 163 cm), sex, and race, as well as individual components of metabolic dysregulation. The variable race unknown includes individuals where the electronic health record‐reported race is missing, unknown, or reported as >1 race or ethnic group. HDL indicates high‐density lipoprotein; and HR, hazard ratio.

Figure 2. Burden of multiple metabolic abnormalities.

Cox proportional hazards analyses results for the association between increasing burden of metabolic abnormalities (hypertension, diabetes, triglycerides >150 mg/dL, HDL <40 mg/dL [men] or <50 mg/dL [women]) and coronary artery disease risk. HDL indicates high‐density lipoprotein; and HR, hazard ratio.

Lability of Metabolic Health and Incident Coronary Artery Disease Risk

A substantial number of subjects developed metabolic abnormalities during the 3‐year run‐in qualification period. Of the 59 770 patients with no metabolic abnormalities at the start of the run‐in period, 22 717 (38.0%) developed ≥1 abnormalities over 3 years. Conversely, no patients went from having a metabolic abnormality during 1 of the 3 years to not having an abnormality in a subsequent year. Which abnormality developed varied across the population as did the associated risk for CAD. For example, 8.4% of the overall population (4097/54 198 without CAD events and 1753/15 074 with CAD events) developed diabetes and 26.8% (13 474/54 198 with CAD events and 5073/15 074 with CAD) were diagnosed with hypertension. For all metabolic abnormalities, development during the qualification period was independently association with CAD and the magnitude of risk conferred by the new diagnosis was, in general, similar to that associated with preexisting abnormalities (Figure 3). In the case of hypertension, however, patients with a preexisting diagnosis were at significantly greater risk for incident CAD than those with new‐onset hypertension during the qualification period, P<0.001 (Figures 1 and 3C). The largest independent hazards for CAD were due to increasing age (HR, 1.91 [95% CI, 1.81–2.02]; P<0.001) and a diagnosis of hypertension (pre‐existing, HR, 1.63 [95% CI, 1.55–1.71], and new‐onset, HR, 1.48 [95% CI, 1.42–1.54]; P<0.001 for both).

Figure 3. Kaplan–Meier plots of metabolic abnormalities on the risk of coronary artery disease.

Coronary artery disease‐free probability for (A) diabetes, (B) high‐density lipoprotein <40 mg/dL (men) or <50 mg/dL (women), (C) hypertension, and (D) triglycerides >150 mg/dL. The line color denotes absence (red); development during the landmark period, t₀–t₃ (blue); or presence at the start of the landmark period, t₀ (green) of the tested metabolic abnormality. CAD indicates coronary artery disease; d, diabetes; hdl, high‐density lipoprotein; htn, hypertension; and tri, triglycerides.

The development of metabolic abnormalities during the run‐in period varied by differential patterns of baseline characteristics across the population (Table 3). Individuals without metabolic abnormalities, that is, absent, were younger and less often male than those who had a chronic or developed abnormality. Individuals who had a preexisting abnormality, for the most part, were more likely to have another metabolic abnormality at baseline. Interestingly, individuals who developed low high‐density lipoprotein almost uniformly had hypertension at baseline (99.1%) compared with those who had preexisting (40.7%) or absent (8.9%) low high‐density lipoprotein. Both increasing age and weight were associated with a greater percentage of conversion to metabolic unhealth, for example, development of ≥1 metabolic abnormalities, during the 3‐year run‐in period (Figure 4).

Table 3.

Baseline Characteristics of the Population at t₀ Stratified by the Presence, Absence, or Development of Individual Components of Metabolic Dysregulation During the Run‐In Period

	Diabetes				High‐density lipoprotein				Hypertension				Triglycerides
	Absent (n=60 413)	Chronic (n=3027)	Developed (n=5832)	P value	Absent (60 746)	Chronic (1458)	Developed (7068)	P value	Absent (43 901)	Chronic (6824)	Developed (18 547)	P value	Absent (n=61 070)	Chronic (1426)	Developed (6776)	P value
Age, y	45.6 (31.7–58.9)	52.0 (35.3–63.6)	53.5 (42.8–63.2)	<0.001	46.6 (33.1–58.7)	47.3 (33.1–58.7)	47.2 (36.0–58.6)	<0.001	40.7 (27.9–53.3)	57.9 (47.8–67.2)	54.1 (43.5–64.2)	<0.001	46.0 (31.6–58.5)	52.2 (40.7–63.3)	50.1 (39.9–60.4)	<0.001
Sex, male	19 236 (31.8)	1335 (44.1)	2571 (44.1)	<0.001	19 679 (32.4)	537 (36.8)	2926 (41.4)	<0.001	12 828 (29.2)	2715 (39.8)	7599 (41.0)	<0.001	19 041 (31.2)	653 (45.8)	3448 (50.9)	<0.001
Race				<0.001				<0.001				<0.001				<0.001
Asian	1224 (2.0)	46 (1.5)	133 (2.3)		1181 (1.9)	37 (2.5)	185 (2.6)		1013 (2.3)	81 (1.2)	309 (1.7)		1178 (1.9)	33 (2.3)	192 (0.3)
Black	5777 (9.6)	569 (18.8)	869 (14.9)		6070 (10.0)	227 (15.6)	918 (13.0)		3430 (7.8)	1244 (18.2)	2541 (13.7)		6637 (10.9)	114 (8.0)	464 (6.9)
Indian	98 (0.2)	6 (0.2)	22 (0.4)		94 (0.2)	4 (0.3)	28 (0.4)		81 (0.2)	15 (0.2)	30 (0.2)		104 (0.2)	2 (0.1)	20 (0.3)
Native American	168 (0.3)	10 (0.3)	26 (0.5)		164 (0.3)	6 (0.4)	34 (0.5)		139 (0.3)	18 (0.3)	47 (0.3)		157 (0.3)	8 (0.6)	39 (0.6)
White	52 239 (86.5)	2384 (78.8)	4711 (80.8)		52 324 (86.1)	1169 (80.2)	5823 (82.4)		38 477 (87.7)	5439 (79.7)	15 418 (83.1)		52 101 (85.3)	1255 (88.0)	5978 (88.2)
Other/unknown*	907 (1.5)	12 (0.4)	71 (1.2)		895 (1.5)	15 (1.0)	80 (1.1)		761 (1.7)	27 (0.4)	202 (1.1)		893 (1.5)	14 (1.0)	83 (1.2)
Height, cm	168 (163–175)	169 (163–178)	170 (163–178)	<0.001	168 (162–175)	168 (161–176)	170 (163–178)	<0.001	168 (163–175)	168 (161–178)	168 (163–178)	<0.001	168 (162–175)	170 (162–178)	170 (163–180)	<0.001
Weight, kg	76.2 (64.0–90.9)	86.2 (72.0–103.9)	91.7 (77.1–108.9)	<0.001	76.2 (64.0–91.2)	88.5 (74.4–104.5)	90.5 (75.9–105.9)	<0.001	73.3 (62.2–88.0)	85.2 (71.7–101.2)	86.2 (72.2– 101.6)	<0.001	76.2 (64.0–91.3)	87.8 (73.6–103.4)	89.4 (75.9–103.9)	<0.001
Medical history
Diabetes	0 (0.0)	3027 (100.0)	0 (0.0)	<0.001	2206 (3.6)	376 (25.8)	445 (6.3)	<0.001	1019 (2.3)	1353 (19.8)	655 (3.5)	<0.001	2231 (3.7)	399 (28.0)	397 (5.9)	<0.001
Low high‐density lipoprotein†	952 (1.6)	376 (12.4)	130 (2.2)	<0.001	0 (0.0)	1458 (100.0)	0 (0.0)	<0.001	612 (1.4)	594 (8.7)	252 (1.4)	<0.001	699 (1.1)	671 (47.1)	88 (1.3)	<0.001
Hypertension	4951 (8.2)	1353 (44.7)	520 (8.9)	<0.001	5400 (8.9)	594 (40.7)	7006 (99.1)	<0.001	0 (0.0)	6824 (100.0)	0 (0.0)	<0.001	5223 (8.6)	724 (50.8)	877 (12.9)	<0.001
High triglycerides‡	894 (1.5)	399 (13.2)	133 (2.3)	<0.001	693 (1.1)	671 (46.0)	62 (0.9)	<0.001	456 (1.0)	724 (10.6)	246 (1.3)	<0.001	0 (0.0)	1426 (100.0)	0 (0.0)	<0.001
Metabolic components				<0.001				<0.001				<0.001				<0.001
0	54 555 (90.3)	0 (0.0)	5215 (89.4)		53 778 (88.5)	0 (0.0)	5992 (84.8)		42 168 (96.1)	0 (0.0)	17 602 (94.9)		54 125 (88.6)	0 (0.0)	5645 (83.3)
1	5081 (8.4)	1449 (47.9)	488 (8.4)		5739 (9.5)	452 (31.0)	827 (11.7)		1426 (3.2)	4816 (70.6)	776 (4.2)		5803 (9.5)	301 (21.1)	914 (13.5)
2	615 (1.0)	1163 (38.4)	92 (1.6)		1127 (1.8)	506 (34.7)	237 (3.4)		260 (0.6)	1480 (21.6)	130 (0.7)		1076 (1.8)	591 (41.4)	203 (3.0)
3	162 (0.3)	280 (9.3)	37 (0.6)		102 (0.2)	365 (25.0)	12 (0.2)		47 (0.1)	393 (5.8)	39 (0.2)		66 (0.1)	399 (28.0)	14 (0.2)
4	0 (0.0)	135 (4.5)	0 (0.0)		0 (0.0)	135 (9.3)	0 (0.0)		0 (0.0)	135 (2.0)	0 (0.0)		0 (0.0)	135 (9.5)	0 (0.0)

Data are presented as the median (lower; upper quartile) or number (%).

^*In this table, Other/Unknown includes individuals of racial groups where the percentage of the population is <5% of the total. This also includes individuals with electronic health record‐reported race as Asian, Native American/Alaskan, Other, or where >1 race group is reported or missing.

^† <40 mg/dL in men and <50 mg/dL in women.

^‡ >150 mg/dL.

Figure 4. Effect of baseline age and weight on risk of conversion to metabolic dysfunction.

A 3‐dimensional bar graph demonstrating how both increasing age and weight at the start of the run‐in period, t₀, is associated with a greater percent conversion to metabolic dysregulation, for example, development of ≥1 metabolic syndrome components during the 3‐y run‐in period.

Model‐Based Exploration of Body Weight and Short‐Term Weight Change on Coronary Artery Disease Risk

Although we did not detect an independent effect of body weight in the time‐to‐event analysis, this could have been due to nonlinear effects of weight on CAD risk. Additionally, we also specifically sought to evaluate the impact of short‐term weight change with respect to risk for incident CAD. Therefore, we conducted additional analyses to define the independent contribution of weight and weight change to CAD risk across the full distribution of body weight. We used a Cox regression model to estimate the 10‐year risk of incident CAD across a range of body weights with invariant age, sex, race, and metabolic status. We detected a significant nonlinear relationship between CAD risk and increasing weight (Figure 5, P<0.001). This analysis identified an inflection point at 70 kg, equivalent to a BMI of 25.1 kg/m². Therefore, we performed a cumulative risk analysis for patients with BMI >25 kg/m² and identified a cumulative risk of 10‐year CAD as 3.8.

Figure 5. 10‐y cumulative coronary artery disease risk by weight at t₃ of the run‐in period.

The presented data represents the 10‐y risk for coronary artery disease development based on a patient's final weight during the run‐in period, t₃. The analysis was adjusted to height of 167 cm, age of 49.7 y, female sex, White race, and absence of hypertension, diabetes, high‐density lipoprotein <50 mg/dL, and triglycerides >150 mg/dL before or at t₃. AF indicates atrial fibrillation.

In order to explore the potential impact of short‐term weight change on CAD risk, we modeled the impact of a 5%, 10%, or 30% change in weight (either increase or decrease) during the 3‐year run‐in period on 10‐year CAD risk independent of change in other variables. The model was based on BMI separated into 2 categories: obese (BMI=35 kg/m²) and nonobese (BMI=25 kg/m²) at baseline (Figure 5). Consistent with the observed nonlinear relationship between weight and CAD risk (Figure 4), the effects of weight change differed by baseline weight. Among individuals with Class I, II, or III obesity, any weight change significantly affected incident CAD risk in an additive fashion, with weight gain associated with increased CAD risk and weight loss associated with decreased risk respectively (Figure 6). Conversely, among subjects without obesity only a 30% short‐term weight gain significantly affected CAD risk (HR, 1.13 [95% CI, 1.03–1.24]) (Figure 6).

Figure 6. Impact of short‐term weight change on the 10‐year risk of coronary disease stratified by baseline body mass index.

The presented data represent the modeled 10‐year risk for coronary artery disease development based on a 5%, 10%, or 30% increase or decrease in weight over the 3‐year landmark period. The Cox proportional hazards models are based on a 54‐year‐old male with a height of 177.8 cm. The weight stratification is based on a baseline BMI and separated into categories: (1) normal, corresponding to a BMI of 25 (weight 79 kg) and (2) obese, corresponding to a BMI of 35 (111 kg). BMI indicates body mass index; CAD, coronary artery disease; and HR, hazard ratio.

Discussion

This work demonstrates the first use of the COMMODORE data set to disentangle the impacts of the individual components of metabolic syndrome and their development on CAD. Specifically, this article succinctly validates the data set through confirmation of known associations between prevalent CAD and metabolic dysregulation then expands on its utility. The longitudinally and data density of COMMODORE is a unique strength upon which we capitalized to estimate the 10‐year risk for CAD based on individual metabolic dysregulation components as well as the transition from metabolically healthy to unhealthy metabolic status. Further, because the data set is derived from a clinically relevant real‐world health system, including primary outpatient visit records, the results suggest potential changes that can be made to clinical practice to affect CAD risk.

Although this work focuses on the impacts of obesity and metabolic dysregulation on CAD risk, an important aspect to the work is that the resource and methodologies may extend beyond CAD, and the data structure and related pipelines for cross‐phenotype analyses are a fundamental component of COMMODORE's structure. Although there may be some biases related to hospital‐based ascertainment, as part of a large, regional health care system, COMMODORE largely reflects the health of the broader regional population (eg, the southern United States) and directly represents those seeking care. ³⁰ Importantly, COMMODORE phenotypes go beyond simple billing code‐based definitions and use published algorithms (eg, Phenotype Knowledge Base ³¹ ) or expertly validated multimodality novel phenotypes. Moreover, confidence in the veracity of COMMODORE is supported by the robust replication of known relationships between obesity, metabolic dysregulation, and risk for prevalent CAD across various population strata, the novel findings relating to the transition to metabolic unhealth and CAD incidence support its use to disentangle complex disease relationships. Thus, we have established a large phenotypically rich resource for the evaluation of metabolic dysregulation, obesity, and disease risk in a real‐world setting that also allows for the execution of cross‐sectional analyses as well as studies assessing temporal relationships.

Prior studies have provided conflicting results regarding the CAD risk conferred by obesity in the absence of metabolic abnormalities. Discrepancies may be partly related to definitions used for metabolic health and duration of follow‐up, with metabolically healthy obesity generally showing a more benign risk profile with more stringent criteria for metabolic health (ie, no metabolic abnormalities) and shorter follow‐up duration. ¹² , ¹³ In >90 000 women from the NHS (Nurses' Health Study) followed from 1980 to 2010, ¹⁷ increased CAD risk was seen among women with metabolically healthy obesity at baseline compared with metabolically healthy women with normal weight. Importantly, most initially metabolically healthy women, regardless of obesity status, acquired metabolic abnormalities during follow‐up. Increased CAD risk among metabolically healthy populations with obesity was also observed in a large cohort of patients in primary care in the United Kingdom. ¹⁴ Again, the presence of metabolic abnormalities, regardless of BMI, was associated with progressively increased risk of incident disease. In this relatively young cohort with short follow‐up, ~10% of individuals developed diabetes, hypertension, or hyperlipidemia. The dynamic nature of metabolically healthy obesity and the consequences of transitioning to an unhealthy state are highlighted by a recent publication from MESA (Multi‐Ethnic Study of Atherosclerosis). ¹⁸ In this study, those with metabolically healthy obesity at baseline did not experience higher rates of subsequent cardiovascular disease. However, approximately half of the individuals developed metabolic syndrome during 12 years of follow up, and the conversion to unhealthy obesity was associated with increased cardiovascular disease risk.

In the small subset of patients metabolically healthy obesity, defined as having obesity without metabolic dysregulation, we observed a marginal increase in CAD risk in the cross‐sectional analyses. One potential explanation for the increase in CAD risk is undiagnosed or nascent metabolic dysregulation. This would be supported by the findings of the MESA study and others that have shown a large proportion of patients with obesity eventually develop metabolic syndrome. This increase in CAD risk due to excess weight that may eventually lead to metabolic dysregulation is also supported by our independent effects analysis; where the components of metabolic dysregulation were independently associated with an increase in CAD risk while BMI was not. Alternatively, this may reflect a true effect based on the greater severity of obesity in the COMMODORE cohort. This interpretation would agree with our analysis showing that the independent effect of weight on CAD risk is most evident at higher body weights. We may also have had improved ability to detect effects related to our independent modeling of weight and height; we have previously shown that modeling the continuous relationship between weight and height opposed to the categorical representation using BMI improved prediction of various cardiometabolic traits including CAD. ²⁸ The impact of short‐term weight change on 10‐year CAD risk is a novel finding with potentially important implications for health‐system based weight management strategies to reduce CAD risk. We observed that any change in weight >5% resulted in a corresponding increase or decrease in CAD risk for patients with excess body weight, but increased risk was seen with marked increases in weight only among those with normal BMI. Therefore, weight loss as a means to decrease CAD risk may operate through 2 mechanisms: (1) decreasing risk of conversion to metabolic unhealth and (2) reducing the risk for CAD associated with weight alone.

A major finding of this work is that adverse transition of metabolic health is a central driver of CAD risk. Regardless of the chronicity, presence of any component of metabolic dysregulation significantly increased an individual's risk for CAD with the greatest individual risk being conferred by hypertension. These results have several clinical implications. First, because we observed risk related to metabolic derangement independent of an individual's BMI our findings emphasize the importance of clinical management of metabolic health in all individuals to minimize the risk of CAD. Second, metabolic abnormalities are increasingly common with higher BMI categories. Therefore, interventions targeted toward maintaining metabolic health, including weight maintenance (even for patients with established obesity), would likely be impactful strategies for preventing CAD. Finally, our results highlight that early detection and intervention of hypertension is critical, especially among patients with excess body weight and lend further support to recent hypertension guidelines that emphasize both detection and effective control of hypertension, especially in patients with metabolic dysfunction, as a high priority to reduce risk of cardiovascular disease, including CAD. ³²

Many previous studies have demonstrated that hypertension is one of several common modifiable risk factor for CAD; however, few have attempted to rigorously estimate their relative effect sizes in diverse populations including both sexes, a unique aspect of this research. Although hypertension is consistently among the most important risk factors in published analyses, findings have been variable by population studied and there does not appear to be consensus in the literature regarding the relative effects size of each risk factor. However, because hypertension is highly prevalent, underdiagnosed, and often poorly controlled it consistently has among the highest population attributable risk of conventional risk factors. ³³ , ³⁴ , ³⁵ The rigorous estimation of risk factor relative effect sizes and the prominent effect of hypertension is a unique aspect of this research. There are several potential mechanisms that may explain the significant risk conferred by hypertension in our study. One possibility is demographics of our population. For example, there is a relatively high proportion of women in the cohort (~60% in cases, ~68% in controls). It is known that women have higher cardiovascular disease risk at any given blood pressure and higher incremental risk per increase in blood pressure. ³⁶ , ³⁷ Second, subjects in our analysis are largely middle aged (54 years in cases, 44 years in controls). The effect size of risk factors, including hypertension, is known to vary by age, with higher values typically found in middle age. ³⁸ , ³⁹ It is possible that these specific demographic features resulted in a cohort particularly sensitive to the effects of hypertension. Finally, another possible contributor could be that, compared with other risk factors, hypertension is underdiagnosed and less well controlled compared with other risk factors. The differential effect of preexisting and newly diagnosed hypertension may reflect differences in cumulative exposure.

Our study also has several limitations. BMI has limitations as a measure of obesity, ⁴⁰ , ⁴¹ and we were unable to further characterize the population by body composition or fat distribution using the preferred measures of adiposity. ⁴² , ⁴³ Additionally, some variables of interest, such as waist circumference and measures of physical activity/physical fitness, were unavailable for the vast majority of individuals as these factors are not routinely collected in clinical practice and as such are not part of the EHR. In an attempt to acknowledge the limitations of BMI, we conducted a time‐to‐event analysis using a more flexible height–weight interaction model that does not enforce a fixed relationship between height and weight and improves outcome prediction across various metabolic abnormalities and cardiovascular outcomes. ²⁸ Data were acquired and documented as part of clinical care at a single institution. Consequently, exposures and outcomes for some individuals may be inaccurate and underascertained. There are also limitations in the use of EHR‐reported race descriptors. These limitations include data missingness, inaccurate report, and inadequate precision to wholly reflect all social determinants of health. Despite these limitations we include these variables in our analyses as they are reflective of variables that may influence clinical practice. The impact of these potential biases, however, are significantly reduced because of the size of the cohort, the diversity in clinic sites, and length of the EHR.

Conclusions

Despite limitations, we demonstrate, at scale in a clinically relevant population, COMMODORE, that relationships between obesity, metabolic dysfunction, and CAD risk vary across clinical and demographic variables. Overall, our findings from the cross‐sectional analyses robustly replicate existing relationships between obesity and metabolic dysfunction, and cardiovascular and metabolic phenotypes. Our time‐to‐event analyses provide novel insights into the significant impact that transition to unhealth by means of metabolic dysregulation or weight has on an individual's 10‐year CAD risk. We also demonstrate for the first time that exposure time for hypertension is an important predictor for CAD risk but not for diabetes, low high‐density lipoprotein, or elevated triglycerides. Finally, our results suggest that, with respect to CAD, the most effective health system‐based interventions to mitigate weight‐related complications would be those that prevent the development of metabolic abnormalities rather than target weight‐loss alone.

Sources of Funding

This work was supported through 17SFRN33520017, a Strategically Focused Research Network grant from the American Heart Association. Megan M. Shuey and Rebecca T. Levinson were funded as fellows on 17SFRN33520017. Megan M. Shuey was also supported by National Institutes of Health (K12HD043483). Megan E. Vogel was funded by National Institutes of Research Institutional National Research Service Award 5T32DK007061‐44. Alyssa H. Hasty is supported by a Merit Award from the Veterans Affairs (5I01BX002195) and an Innovative Basic Science award from the American Diabetes Association (1‐17‐IBS‐140). Quinn S. Wells and Rebecca T. Levinson were funded by National Institutes of Research 1R01HL140074‐01. The data set used for the clinical analyses was obtained from the Vanderbilt University Medical Center Synthetic Derivative, which is supported by institutional funding, the 1S10RR025141‐01 instrumentation award, and by the Clinical and Translational Science Award grant UL1TR000445 from National Center for Advancing Translational Sciences/National Institutes of Health.

Disclosures

None.

Supporting information

Data S1

Tables S1–S7

Figures S1–S2

Reference 44