Psychosocial profiles and longitudinal achievement of optimal cardiovascular risk factor levels: The Coronary Artery Risk Development in Young Adults (CARDIA) Study

Abstract

Psychosocial factors are associated with the achievement of optimal cardiovascular disease risk factor (CVDRF) levels. To date, little research has examined multiple psychosocial factors simultaneously to identify distinguishing psychosocial profiles among individuals with CVDRF. Further, it is unknown whether profiles are associated with achievement of CVDRF levels longitudinally. Therefore, we characterized psychosocial profiles of individuals with CVDRF and assessed whether they are associated with achievement of optimal CVDRF levels over 15 years. We included 1,148 CARDIA participants with prevalent hypertension, hypercholesterolemia and/or diabetes mellitus in 2000–2001. Eleven psychosocial variables reflecting psychological health, personality traits, and social factors were included. Optimal levels were deemed achieved if: Hemoglobin A1c (HbA1c) < 7.0%, low-density lipoprotein (LDL) cholesterol < 100 mg/dl, and systolic blood pressure (SBP) < 140 mm Hg. Latent profile analysis (LPA) revealed three psychosocial profile groups “Healthy”, “Distressed and Disadvantaged” and “Discriminated Against”. There were no significant differences in achievement of CVDRF levels of the 3 targets combined across profiles. Participants in the “Distressed and Disadvantaged” profile were less likely to meet optimal HbA1c levels compared to individuals in the “Healthy” profile after demographic adjustment. Associations were attenuated after full covariate adjustment. Distinct psychosocial profiles exist among individuals with CVDRF, representing meaningful differences. Implications for CVDRF management are discussed.

Introduction

A growing body of literature has identified a broad range of psychological and social factors that are relevant for predicting cardiovascular disease risk factors (CVDRF) such as hypertension and diabetes (Spruill et al., 2019). For example, individuals with high levels of depressive symptoms, specific subsets of personality traits, including trait anxiety and certain social factors (e.g., racism related vigilance) were more likely to develop hypertension or diabetes than those with lower levels of these psychosocial factors (Baek, Hur, Kim, & Youm, 2016; Cuevas, Williams, & Albert, 2017; Graham et al., 2020; Hackett & Steptoe, 2016; Hicken, Lee, Morenoff, House, & Williams, 2014; Johnson, 2019; Kubzansky et al., 2020; Levine et al., 2021). In addition to conferring higher risks, psychosocial factors have important implications for achieving optimal CVDRF levels for individuals with a risk factor. For instance, in cross-sectional studies, perceived control (Gonzalez, Shreck, Psaros, & Safren, 2015b) and social support were associated with glycated hemoglobin (HbA1c) levels among individuals with diabetes (Misra & Lager, 2009). Taken together, this rapidly developing body of research demonstrates that multiple psychosocial factors, including psychological health (e.g., depressive symptoms), personality traits (e.g., anger-in expression), and social factors (e.g., childhood environment), are contributors to the development of CVDRF and achievement of optimal CVDRF levels (Levine et al., 2021).

Despite increasing evidence that multiple psychosocial factors may be critical for achieving optimal CVDRF levels among individuals with a risk factor, several gaps remain in the literature. Very little research has examined these multiple psychosocial factors simultaneously in relation to achieving optimal CVDRF levels. Rather, most studies focus on one or a few psychosocial factors, such as depression (Carroll, Phillips, Gale, & Batty, 2010; Mulle & Vaccarino, 2013; Simonsick, Wallace, Blazer, & Berkman, 1995). Increasing research and theory argues that multiple psychosocial factors combine to enhance cardiovascular disease risk (Pikhart & Pikhartova, 2015; Sims, Glover, Gebreab, & Spruill, 2020), and scholars have called for new comprehensive empirical approaches that map onto this research (Figueroa, Frakt, & Jha, 2020). Importantly, each of these psychosocial factors do not exist in isolation, and there are often shared relationships among the factors that should be addressed analytically (Clark et al., 2012). In addition, the current body of research demonstrates that some psychosocial factors have not been examined in the context of achieving optimal CVDRF levels. Increasing research suggests that positive psychological factors may be protective, and may be associated with favorable cardiovascular health outcomes (Boehm et al., 2018; DuBois et al., 2015). However, these positive psychological factors have not been extensively studied. Thus, a more complete inclusion of psychosocial factors is needed.

Finally, it is necessary to implement empirical approaches that adequately address emerging practices in clinical and research contexts. Scholars have noted that clinicians are screening for multiple psychosocial factors, and this practice and recommendations for it are increasing (K. W. Davidson & McGinn, 2019). Despite the proliferation of data, there is a need for the implementation of modeling techniques to address the complexity associated of how these psychosocial variables are experienced differently across the CVDRF population. Patients with CVDRF are not a homogenous group, and evidence demonstrates the presentation of multiple psychosocial factors across patients can vary (Chirinos, Murdock, Leroy, & Fagundes, 2017; Cornelius et al., 2018). For example, van Montfort and colleagues identified three psychosocial “profiles” of individuals from the general Dutch population. One profile representing women with Type D personality and no experiences of trauma had the highest odds of lipid abnormalities as compared to other profiles of women (van Montfort, Mommersteeg, Spek, & Kupper, 2018). This paper further demonstrates that identifying unique psychosocial profiles may have implications for understanding who is more likely to achieve optimal CVDRF levels. Thus, in the current study, we leverage multiple variables to empirically identify unique “psychosocial profiles” of individuals based on their presentation of these many factors.

In the current study, our first aim was to characterize the psychosocial profiles of individuals with traditional CVDRF of hypertension, hypercholesterolemia, and/or diabetes. By determining the saliency of specific psychosocial variables, we are adding uniquely to the literature on psychosocial factors and CVDRF. We used latent profile analysis (LPA), which is a powerful person-centered statistical technique that allows for the identification of profiles among a group of individuals. Based on the rich literature on psychosocial factors, we hypothesized that distinct profiles would emerge (Howard & Hoffman, 2018). The second aim of the study was to investigate whether these profiles were associated with achieving optimal CVDRF levels over time. To date, few longitudinal studies have examined psychosocial factors in relation to achieving optimal CVDRF levels. While these studies are foundational, the psychosocial profile of individuals with optimal clinical management is still unknown. To achieve these aims, we use the rich longitudinal data available in the Coronary Artery Risk Development in Young Adults (CARDIA).

Materials and Methods:

Study Population

The data are from the Coronary Artery Risk Development in Young Adults (CARDIA) study (Friedman et al., 1988). CARDIA is a longitudinal multi-center cohort study of the development of cardiovascular disease. CARDIA used population-based sampling in Birmingham, AL, Chicago, IL and Minneapolis, MN and sampled from a register of members of a health maintenance plan in Oakland, CA between 1985 and 1986. Ultimately, they enrolled 5,115 white and black men and women aged 18–30 years old. Participants completed follow-up visits in the following exam years: year 2 (1987–1988), year 5 (1990–1991), year 7 (1992–1993), year 10 (1995–1996), year 15 (2000–2001), year 20 (2005–2006), year 25 (2010–2011), and year 30 (2015–2016).The CARDIA study was approved by each field center’s Institutional Review Board at each exam, and all participants provided written informed consent.

This analysis included participants who were aware that they had prevalent hypertension, hypercholesterolemia, and/or diabetes at the year 15 exam in 2000–01 (baseline for this study). This exam year was chosen as baseline because there was an extensive assessment of psychosocial variables available to identify distinct profiles. Participants indicated awareness of their diagnosis at baseline by responding “yes” to questions about each of these three conditions (e.g., “Has a doctor or nurse ever said that you have diabetes?”). Four individuals with 95% missing psychosocial and demographic data and one with approximately 50% missing psychosocial and demographic data were excluded and the final analytic sample included 1,148 participants with any CVDRF, of which 604 had prevalent hypertension, 636 had hypercholesterolemia, and 210 had diabetes. Within this analytic sample, some participants exclusively had one risk factor. Specifically, 95 individuals had diabetes, 366 had hypertension, and 422 had hypercholesterolemia exclusively.

Psychosocial variables

We selected 11 psychosocial variables for the LPA. The variables were selected if they were available at study baseline, and if they represented three areas of interest, namely psychological health, personality traits, and social factors. As previously stated, research has demonstrated that these facets are associated with CVDRF (Baek et al., 2016; Boehm et al., 2018; K. Davidson, Jonas, Dixon, & Markovitz, 2000; Gonzalez, Shreck, Psaros, & Safren, 2015a; Hackett & Steptoe, 2016; Hicken et al., 2014; Jonas & Lando, 2000; Lewis, Williams, Tamene, & Clark, 2014; Momtaz et al., 2012; Ostir, Berges, Markides, & Ottenbacher, 2006; Schmitz et al., 2012; Trudel-Fitzgerald, Boehm, Kivimaki, & Kubzansky, 2014). Further, each individual variable was selected based on existing research demonstrating evidence of an association with cardiovascular health. Below, we detail the variables included.

The Center for Epidemiologic Studies Depression Scale (CES-D) (Radloff, 1977) is a widely used measure for depressive symptoms. We computed a score by summing across all 20 items where higher scores indicate greater depressive symptoms. The Chronic Burden scale (Bromberger & Matthews, 1996; Mujahid, Roux, Cooper, Shea, & Williams, 2011) assesses ongoing chronic stressors across five domains of an individual’s life (e.g., self-health, loved one-health, job, relationship, and financial problems). We computed a sum score of all five items, where higher values indicate greater chronic burden.

Perceived control (also known as mastery) (Pearlin, Menaghan, Lieberman, & Mullan, 1981; Pearlin & Schooler, 1979) was assessed with the Sense of Mastery Scale. This scale was used to assess the amount of control that a participant perceived to have over their life (e.g., “I have no control over things happening to me.”). We computed a sum score across all seven items. Higher scores indicate greater perceived control. Optimism was measured with a six-item version of the revised Life Orientation Test (LOT-R) (Scheier, Carver, & Bridges, 1994). Higher scores indicate greater optimism. The Reactive Responding scale was used to assess each individual’s self-regulation across three sub dimensions, namely emotional reactivity, goal orientation, and vigilance (Taylor & Seeman, 1999). We computed a sum across the three items in each sub dimension. Higher scores indicate greater emotional control, greater goal orientation, and less vigilance, respectively. Finally, we included the Anger-In Expression scale (Spielberger, 1985) to assess individual differences in tendencies to keep anger “inside”. We computed a sum across all eight items in the scale. Higher scores indicate greater expressed anger.

Neighborhood cohesion, in which participants indicated the interpersonal nature of their neighborhood environment (Sampson, Raudenbush, & Earls, 1997), was computed as a sum across five items (e.g., people around here are willing to help their neighbors). Higher scores indicate greater social cohesion. The Childhood Family Risk scale was included to assess the nature of an individual’s childhood home environment. We computed a sum across all seven items, higher scores indicate a more positive environment (Felitti et al., 1998; Taylor, Lerner, Sage, Lehman, & Seeman, 2004). The Family Social Support and Strain scale was used to assess the nature of an individual’s family relationships. We computed a sum across the four items of the social support subscale, where higher scores indicate more social support. Likewise, we computed a sum across the four items of the strain scale, where higher scores represent a more stressful family environment (Schuster, Kessler, & Aseltine, 1990). We included an abbreviated version of the Lubben Social Network scale. Participants indicated the size of their social network, on three items. We computed a sum across the three items, and higher scores indicate a more expansive social network (Lubben, 1988). Finally, we included the Experiences of Discrimination scale to assess participant’s experiences with racial discrimination, gender discrimination, and socioeconomic-status (SES) discrimination. Participants indicated “Yes” or “No” to questions that asked if participants have experienced discrimination, if participants have been prevented from doing something or been hassled, or if participants have felt inferior across seven sectors (e.g., school, job, home, etc.). We computed the sum of these seven items for the three sub dimensions (i.e., racial-, gender-, and SES- ) of discrimination (Krieger, Smith, Naishadham, Hartman, & Barbeau, 2005).

Achievement of optimal cardiovascular disease risk factor (CVDRF) levels

Achievement of optimal CVDRF levels was assessed using data available in CARDIA at the year 20, 25, and 30 (2015–16) examinations. Based on treatment guidelines and previous research, we considered optimal CVDRF levels to be achieved at a given exam if participants met the following optimal targets: Hemoglobin A1c (HbA1c) < 7.0%, low-density lipoprotein (LDL) cholesterol < 100 mg/dl, and systolic blood pressure (SBP) < 140 mm Hg. Achievement of all 3 optimal targets was the primary outcome. In addition, we examined HbA1c, LDL, and SBP as independent outcomes. Missing data for these outcome variables were not imputed.

HbA1c was determined at exam years 20–30 using Tosoh G7 (variant mode) high-performance liquid chromatography (national Glycohemoglobin Standardization Program-certified assays). Optimal HbA1c was defined as (HbA1c < 7.0), While an HbA1c of < 7% is appropriate for many nonpregnant adults, the ADA 2020 guidelines also state that less stringent goals (e.g., <8%) may be better for individuals with certain pre-existing conditions (e.g., history of severe hypoglycemia or macrovascular complications). To address possible concerns regarding differences in optimal target HbA1c levels, we conducted a sensitivity analysis with HbA1c < 8.0. For exam year 30, HbA1c was only available women and not available for men.

LDL cholesterol was estimated from plasma blood samples using the Friedewald equation for individuals with fasting triglyceride values < 100 mg/dL (Friedewald, Levy, & Fredrickson, 1972). Optimal LDL was defined as (< 100 mg/dL) based on treatment guidelines (Grundy et al., 2004).

Resting SBP was measured at each exam by trained technicians three times at 1-minute intervals using a random-zero sphygmomanometer (year 20) or an oscillometer (years 25–30), and the second and third measurements were averaged together. Blood pressure measures were calibrated to account for the changes in this method. Optimal SBP was defined as (<140 mm Hg) because this was the clinical standard at the time of the year 20–30 exams (Chobanian et al., 2003).

Covariates

Demographic factors included participant’s age in years, sex, race (i.e., non-Hispanic black or non-Hispanic white), highest grade of education measured in years, marital status, field center, and barriers to assessing medical care. Participants indicated their sex as male, female, or as having undergone a sex change. Two participants indicated a female to male sex change and were classified as male in our analysis. Marital status was categorized as “single”, “married/partnered”, or “divorced/separated/widowed”. We computed barriers to accessing medical care, by taking a sum of three variables (1= No; 0 = Yes) measured at baseline that reflect barriers to accessing medical care, (i.e., “In the past two years, have you always had health insurance or other coverage for medical care?”, “Was there ever anytime during the past two years when you did not seek medical care because it was too expensive or health insurance did not cover it? (Reverse coded)” and “Do you have a usual source of medical care?”). Higher scores indicate greater barriers to accessing medical care.

Behavioral covariates assessed at each exam year included physical activity, smoking status, and alcohol consumption levels. Physical activity, measured from the physical activity questionnaire, was computed by multiplying the intensity of various activities and the frequency of engaging in those activities, resulting in “exercise units” (EU). This instrument is comparable in its reliability and validity to other physical activity measures (Jacobs, Ainsworth, Hartman, & Leon, 1993; Jacobs Jr, 1989). Participants were classified as “non-smoker” if they reported they have never used any tobacco product, or if they indicated using any tobacco product but have never smoked cigarettes regularly. Participants were classified as an “ex-smoker” if they indicated ever using any tobacco product, reported ever smoking cigarettes regularly, but do not currently smoke cigarettes regularly. Participants were classified as a “current smoker” if they indicated using any tobacco product, reporting smoking cigarettes regularly, and if they are currently smoking cigarettes regularly. Alcohol consumption was calculated for each individual, and then categorized as “none”, “moderate” or “heavy”. Individuals were classified as “none” if they indicated having never had a drink within the past year or if the number of drinks participants indicated they usually have per week was zero. The category of moderate is defined as up to one drink per day for women and up to two drinks per day for men. The category of heavy is defined as fifteen drinks or more for men, and eight drinks or more for women per week. We used the Dietary Guidelines of drink-equivalents to calculate participant’s consumption of wine, beer, and spirits (Committee, 2015). Finally, we included waist circumference at baseline as our anthropometric factor. Waist circumference was measured in centimeters (cm) and rounded to the nearest 0.5 cm.

Statistical Analyses

Latent Profile Analysis (LPA)

Mplus version 6.0 was used for the LPA analyses. LPA was used to characterize profiles among all psychosocial variables. LPA uses continuous variables to compute profiles of individuals, with the goal of maximizing homogeneity within a profile and heterogeneity between profiles. LPA is considered an individual-based approach, because it identifies similarities between individuals rather than associations among variables. The optimal number of profiles was determined after examination of the following model-fit indices: the Akaike information criteria (AIC), the Bayesian information criteria (BIC), the sample-size adjusted BIC (ABIC), log-likelihood (LL), entropy, the Lo–Mendell–Rubin adjusted likelihood ratio test (ALRT), and the parametric bootstrapped likelihood ratio test (BLRT). Better fitting models were determined by smaller AIC, BIC, ABIC and LL values. Entropy values closer to 1.0 indicate better model fit, with values over 0.80 being considered noteworthy (Roesch, Villodas, & Villodas, 2010). The The ALRT and BLRT provide a p-value for each model solution indicating that a model with one less profile is rejected in favor of the estimated model. See table 1 for model fit indices, and descriptive statistics are presented for all psychosocial variables and within each profile group in table 2.

Table 1.

Fit indices for the latent profile analysis

No. of clusters	No. of parameters	AIC	BIC	aBIC	LL	Entropy	ALRT (p)	BLRT (p)
1	32	90765.419	90926.884	90825.242	−45350.710	-	-	-
2	49	88595.139	88842.382	88686.742	−44248.569	0.863	<0.001	<0.001
3	66	87467.545	87800.567	87590.930	−43667.773	0.895	<0.001	<0.001
4	83	86965.061	87383.861	87120.227	−43399.531	0.908	0.083	<0.001
5	100	86579.882	87084.460	86766.829	−43189.941	0.849	0.086	<0.001
6	117	86321.158	86911.514	86539.885	−43043.579	0.846	0.6461	<0.001
7	134	86084.500	86760.634	86335.008	−42908.250	0.852	0.3246	<0.001
8	151	85877.057	86638.969	86159.346	−42787.528	0.87	0.1697	<0.001

Note. AIC, Akaike information criterion; BIC, Bayesian information criterion; aBIC, Adjusted BIC; LL, log-likelihood; ALRT, Lo-Mendell-Rubin Adjusted likelihood ratio test; BLRT, bootstrapped likelihood ratio test.

Table 2.

Descriptive statistics of psychosocial variables in the LPA

	Overall Sample		Discriminated Against		Distressed and Disadvantaged		Healthy
	n= 1,148		n= 169 (14.72%)		n= 240 (20.91%)		n= 739 (64.37%)
	M	SD	M	SE	M	SE	M	SE
CES-D	10.08	8.41	14.38	0.96	18.23	1.18	6.37	0.3
Neighborhood Cohesion (NC)	17.79	3.7	16.09	0.3	16.04	0.27	18.77	0.15
Perceived Control (PC)	21.28	4.48	19.74	0.46	17.75	0.44	22.82	0.2
Emotional Reactivity (ER)	6.38	2.19	6.25	0.19	5.14	0.2	6.82	0.09
Goal Orientation (GO)	8.6	2.07	8.95	0.17	7.66	0.17	8.84	0.09
Vigilance (VIG)	6.6	2.08	5.61	0.18	5.84	0.18	7.08	0.08
Optimism (OP)	16.63	3.62	15.75	0.35	13.25	0.37	17.97	0.16
Anger-In Expression (ANG)	6.03	3.46	6.62	0.34	7.7	0.39	5.34	0.12
Childhood Environment (CE)	16.19	4.09	15.05	0.4	14.09	0.39	17.15	0.17
Family Social Support (SS)	13.9	2.42	13.1	0.28	12.11	0.29	14.68	0.09
Family Social Strain (ST)	8.54	2.61	9.91	0.23	10.4	0.24	7.61	0.11
Social Network (SN)	5.98	2.57	5.72	0.21	4.65	0.19	6.48	0.11
Chronic Burden (CB)	4.34	3.44	5.64	0.32	6.32	0.44	3.37	0.12
Discrimination-Gender (DG)	1.44	1.79	4.05	0.2	1.4	0.14	0.85	0.05
Discrimination-Race (DR)	1.44	1.85	4.21	0.18	1.15	0.15	0.9	0.06
Discrimination -SES (DS)	0.95	1.62	4.17	0.16	0.56	0.09	0.34	0.04

Generalized Estimating Equations

Statistical analyses were done using SAS version 9.4 (SAS Institute, Cary, NC). Generalized estimating equations assuming independence correlation structure were used to examine the associations of psychosocial profiles with optimal CVDRF levels while accounting for correlated observations among participants over time. We computed the odds ratio (OR) and 95% confidence interval (CI) for optimal CVDRF levels as a composite and for each of its components. Three models with varying levels of adjustments were used to examine the associations of psychosocial profiles with the optimal CVDRF levels. Model 1 was unadjusted. Model 2 was adjusted for demographic factors including age, sex, race, education, field center, marital status, and barriers to accessing medical care. Model 3 was also adjusted for behavioral and anthropometric factors that may be on the causal pathway between psychosocial profiles and optimal CVDRF levels, including alcohol consumption, smoking status, physical activity, and waist circumference. We used multiple imputation by chained equations (MICE) to impute data for missing covariates (Sterne et al., 2009).

Results

Descriptive Statistics

At baseline, the mean age of participants was 40.61 years old (SD = 3.62), 58.4% were female, and 51.6% were non-Hispanic black (Table 3). Most participants had some college education (M = 14.7 years; SD = 0.08), and 60.6% were married. In addition, most participants had no barriers to medical care (81.5%). No participants were pregnant. In year 15, 25.0% had optimal LDL cholesterol and 87.20% of participants had optimal SBP. Since HbA1c data were not available at baseline, the percentage of participants with optimal HbA1c or optimal CVDRF levels could not be determined. In year 20, 90.20% of participants had optimal HbA1c, 33.89% had optimal LDL cholesterol, 87.41% had optimal SBP, and 25.03% had optimal CVDRF levels. In year 25, 87.99% of participants had optimal HbA1c, 35.71% had optimal LDL cholesterol, 85.11% had optimal SBP, and 25.64% had optimal CVDRF levels. In year 30, 97.75% of participants had optimal HbA1c, 39.90% had optimal LDL cholesterol, 84.30% had optimal SBP, and 25.89% had optimal CVDRF levels.

Table 3.

Participant Characteristics

Baseline (Y15) Characteristics	Overall	Healthy	Distressed & Disadvantaged	Discriminated Against
N(%)	1148(100.00)	739(64.37)	240(20.91)	169(14.72)
Field Center, %
Birmingham, AL	26.57	25.98	27.50	27.81
Chicago, IL	21.25	22.60	19.58	17.75
Minneapolis, MN	23.26	22.46	22.92	27.22
Oakland, CA	28.92	28.96	30.00	27.22
Age, Years (SD)	40.61(3.62)	40.57(3.59)	40.27(3.66)	41.25(3.63)
Gender, %
Female	58.36	55.21	66.67	60.36
Race, %
Non-Hispanic Black	51.57	44.52	52.08	81.66
Education, Highest Grade (years)	14.68(0.08)	14.98(0.09)	13.95(0.17)	14.34(0.16)
Marital Status, %
Single	20.70	17.77	25.83	26.21
Married	60.56	67.21	53.75	41.12
Divorced or Separated	18.74	15.02	20.42	32.66
Barriers to Medical Care, %
0 Barriers	81.51	87.94	69.17	70.95
1 Barrier	11.95	8.67	17.92	17.81
2 Barriers	5.41	2.85	10.83	8.87
3 Barriers	1.13	0.54	2.08	2.37
Physical Activity Intensity Score	306.20(7.70)	322.74(9.83)	244.17(14.08)	321.95(21.27)
Alcohol Consumption Level, %
None/Light	51.80	49.58	53.46	59.17
Moderate	35.90	39.45	29.83	28.99
Heavy	12.30	10.97	16.71	11.83
Smoking Status, %
Non-smoker	58.62	63.06	52.08	48.52
Ex-smoker	19.51	21.11	16.25	17.16
Smoker	21.87	15.83	31.67	34.32
Waist Circumference, cm	94.40(0.48)	93.05(0.59)	97.87(1.18)	95.42(1.04)
History of High Blood Pressure, %	52.94	48.98	60.59	59.52
History of High Cholesterol, %	57.14	59.37	51.12	55.49
History of Diabetes Mellitus, %	18.49	15.96	24.68	20.83
Family History of HBP, %	71.00	68.20	76.10	76.26
Mean Lipids LDL, mm/dl	124.30(1.08)	125.13(1.35)	122.92(2.23)	122.65(2.77)
Mean Hemoglobin A1c, g/dl	--	--	--	--
Mean Systolic BP, mmHg	119.15(0.53)	117.59(0.65)	121.73(1.16)	122.29(1.34)

Note. We used multiple imputation by chained equations (MICE) to impute data for missing covariates. The demographic variables included in the table include MICE imputation. Hemoglobin A1c was not available at year 15.

Characterization of Psychosocial Profiles

Several LPA models were fitted with the number of profiles ranging from 1 through 8 (Table 1). Entropy values ranged between 0.846 and 0.908, indicating great fit of the data across all profile solutions. While the BLRT was significant across all models, the ALRT indicated that a 2-profile solution was significantly better than a 1-profile solution (p <.001), and a 3-profile solution was significantly better than a 2-profile solution (p <.001). The values for AIC, BIC, ABIC, and LL all decreased as the number of profiles increased across models. The largest decreases were observed between the models with a 1–3 profile solution, while the smallest decreases were observed for the 4–8 profile solutions. We selected a 3-profile solution after observing patterns of all model-fit indices and theoretical considerations.

The three profiles were labeled as “Healthy”, “Discriminated Against”, and “Distressed and Disadvantaged” based on the model means. Most participants (64.4%) belonged to the “Healthy” profile, while 20.9% belonged to the “Distressed and Disadvantaged” profile, and 14.7% belonged to the “Discriminated Against” profile. The graphical representation of the 3-profile solution is presented in Fig. 1, in which z-scores for each psychosocial variable are presented across profile groups. The “Healthy” psychosocial profile was marked by lowest levels of CES-D and chronic burden, better social factors including the best childhood environment, largest social network, and neighborhood cohesion. In addition, this group had the greatest perceived control, optimism, lowest anger, and reported the fewest discriminatory experiences. The “Distressed and Disadvantaged” psychosocial profile was marked by highest levels of CES-D, high chronic burden, and lowest optimism and perceived control. This profile group also indicated “disadvantaged” social contexts including worst childhood environment, and the smallest social network. The “Discriminated Against” psychosocial profile was primarily defined by the strikingly high number of gender-, race-, and SES- based discrimination experiences. However, this group also indicated equally high chronic burden as the “Distressed and Discriminated” profile, and equally poor neighborhood cohesion. In addition, this profile indicated the second worst social contexts.

Fig. 1:

In this fgure, all psychosocial variables are z-scored, and higher values indicate a more “negative” score on that variable. CES-D=Center for epidemiologic studies depression scale; NC=Neighborhood cohesion; PC=Perceived control; ER=Emotional reactivity; GO=Goal orientation; VIG=Vigilance; OP=Optimism; ANG=AngerIn expression; CE=Childhood environment; SS=Family social support; ST=Family social strain; SN=Social network; CB=Chronic burden; DG=Discrimination-gender; DR=Discrimination-race; DS=Discrimination-SES

Generalized Estimating Equations

Based on the findings of the LPA, we assigned the “Healthy” profile group as the reference category for all GEE analyses (Table 4). Further, we anticipated that the “Healthy” profile group would be more likely to achieve optimal CVDRF levels as compared to the “Distressed and Disadvantaged” and “Discriminated Against” profiles. The OR for the GEE analyses represent the odds of having achieved optimal CVDRF levels over the past 15 years in the comparison group as compared to the “Healthy profile”.

Table 4.

Association of Psychosocial Profiles with Achievement of Optimal Cardiovascular Disease Risk Factor Levels

Categoriesa	Model 1b OR(95% CI)	Model 2c OR(95% CI)	Model 3d OR(95% CI)
HbA1C < 7.0%
Healthy	Reference	Reference	Reference
Distressed & Disadvantaged	0.64(0.41,0.99)	0.69(0.48,0.99)	0.93(0.63,1.39)
Discriminated Against	0.57(0.33,0.95)	0.71(0.47,1.09)	0.74(0.48,1.14)
SBP < 140 mm Hg
Healthy	Reference	Reference	Reference
Distressed & Disadvantaged	0.66(0.50,0.88)	0.80(0.61,1.06)	0.91(0.68,1.20)
Discriminated Against	0.61(0.45,0.83)	0.92(0.68,1.25)	1.01(0.74,1.37)
LDL < 100 mm/dl
Healthy	Reference	Reference	Reference
Distressed & Disadvantaged	1.06(0.83,1.36)	1.08(0.88,1.34)	1.04(0.84,1.29)
Discriminated Against	1.21(0.90,1.62)	1.24(0.97,1.58)	1.22(0.96,1.56)
CVDRF
Healthy	Reference	Reference	Reference
Distressed & Disadvantaged	0.88(0.64,1.21)	0.89(0.69,1.15)	0.95(0.73,1.24)
Discriminated Against	0.83(0.57,1.21)	0.97(0.71,1.34)	1.02(0.74,1.40)

^aLatent Profile Analysis was run using the following psychosocial variables: CES-Depressive Symptoms, Neighborhood Cohesion, Perceived Control, Reactive Responding-Emotions, Reactive Responding-Goals, Reactive Responding- Vigilance, Optimism, Anger-In Expression, Childhood Family Environments, Positive Social Support, Negative Social Conflict, Social Network, Chronic Burden, Gender Discrimination, Race Discrimination, and Socioeconomic Status Discrimination. Optimal solution included three profiles: “Healthy” (64% of sample), “Distressed and Disadvantaged” (20%), and “Discriminated Against” (14%). The “Healthy” profile was defined by better mental health, social health, and least amount of discrimination. The “Distressed and Disadvantaged” profile was marked by high levels of depressive symptoms and disadvantaged social contexts. The “Discrimination” profile was marked primarily by high levels of discriminatory experiences.

^bModel 1 was unadjusted

^cModel 2 was adjusted for age, sex, race, education, field center, marital status, and barriers to medical care.

^dModel 3 was adjusted for age, sex, race, education, field center, marital status, barriers to medical care, alcohol consumption levels, smoking status, physical activity, and waist circumference.

Distressed and Disadvantaged vs Healthy Profiles.

In model 2, individuals in the “Distressed and Disadvantaged” profile were significantly less likely (OR 0.69, 95% CI: 0.48, 0.99) to have HbA1c < 7.0% compared to individuals in the “Healthy” profile. There was no significant association in model 3. To examine which factor attenuated the association between the “Distressed and Disadvantaged” profile and HbA1c in model 2 and model 3, we examined each behavior and anthropometric factor individually in model 3. The attenuation was primarily driven by waist circumference. For SBP, we observed a marginal effect in the predicted direction in model 2. In model 3, the marginal effect for SBP was removed after fully adjusting for demographic, anthropometric, and behavioral variables. In model 3, we found there were no differences in likelihood of achieving optimal CVDRF levels of all 3 targets combined for the “Distressed and Disadvantaged” profile as compared to the “Healthy” profile.

Discriminated Against vs Healthy Profiles.

In model 2, marginal effects were observed for the “Discriminated Against” profile in the predicted direction for HbA1c. We also observed a marginal effect in model 2 for LDL. In model 3, marginal effects were observed for the “Discriminated Against” profile for HbA1c, and LDL. In model 3, we found there were no differences in likelihood of achieving optimal CVDRF levels of all 3 targets combined for the “Discriminated Against” profile as compared to the “Healthy” profile. See table 4 for the full set of results.

Sensitivity Analysis.

As stated previously, we conducted a sensitivity analysis with HbA1c < 8.0. The analysis revealed HbA1c estimates similar in magnitude, but several significant effects become non-significant when HbA1c was < 8.0. We also aimed to differentiate between individuals with type 1 and type 2 diabetes. However, there was no indication of individuals diagnosed with diabetes at an earlier study year (year 7) and who also met the inclusion criteria for the current study (hypertension and/or hypercholesterolemia at study baseline). Therefore, no sensitivity analysis differentiating individuals with type 1 and type 2 diabetes was conducted.

Discussion

The current study is among the first to characterize psychosocial profiles of individuals with traditional CVDRF, including hypertension, hypercholesterolemia, and/or diabetes. The innovative implementation of LPA moves the literature beyond well-studied single factors, to focus on modeling the psychosocial conditions of life in which individuals with CVDRF live. LPA revealed three distinct, novel psychosocial profiles of individuals with CVDRF in aim 1. Qualitative examination of the psychosocial patterns within each profile revealed that CVDRF individuals possessed unique sets of psychosocial strengths and challenges. In other words, while each of these psychosocial factors were experienced by all individuals, the ways in which individuals with CVDRF experienced the same set of psychosocial factors were meaningfully different. Further, we also expanded the CVDRF literature in aim 2 by using those identified profiles to examine associations of the profiles with achievement of optimal CVDRF levels over fifteen years. We first present insight into key differences and complexities among patient psychosocial profiles.

The “Healthy” profile included the majority of participants in the study sample, and was marked by the best psychological health, personality traits, and social context factors. While this study is the first (to our knowledge) to examine psychosocial profiles among individuals with CVDRF, the pattern of a healthy majority group has been supported in other psychosocial latent profile studies. Specifically, a study by Chirnios and colleagues (2017) used LPA to derive depressive symptom profiles among participants in the MIDUS-II study. They used multiple psychological health measures, including the Mood and Anxiety Symptom Questionnaire, the CES-D scale, and the Pittsburgh Sleep Quality index, to identify profiles. The most common profile identified was the “No Symptoms” profile, which included over 60% of the sample (Chirinos et al., 2017).

Following the “Healthy” profile, the second largest profile was the “Distressed and Disadvantaged” profile group. This profile included about 20% of the participant sample. The emergence of this profile group supports constructs identified in the Clark and colleagues (2012) study. Clark and colleagues examined latent constructs of psychosocial factors among individuals in the Chicago Health and Aging project, and identified a “distressed group” which was marked by a cluster of poor psychological health and personality traits including elevated depressive symptoms, neuroticism, perceived stress, and low life satisfaction (Clark et al., 2012). Similarly, the “Distressed and Disadvantaged” profile in the current study was marked by the highest values of CES-D, and chronic burden stressors, lowest values of social support, in addition to the lowest values of positive psychological variables including perceived control and optimism. In addition, the “Distressed and Disadvantaged” profile further supports research showing that elevated depressive symptoms are often co-morbid with chronic conditions including hypertension (Cuevas et al., 2017; Hackett & Steptoe, 2016).

In addition, our “Discriminated Against” profile was primarily distinguished based the highest frequency of identity-based discrimination experiences, compared with the other two profiles. The identification of this profile supports the broader literature on psychosocial factors predicting the onset of CVDRF. All participants hold multiple identities which uniquely affect and inform experiences across the life course (Crenshaw, 1989). Individuals in the current study who embody marginalized identities, in particular the sample of black participants, face an increased risk of experiencing discrimination over time and across multiple sectors of society (Williams & Mohammed, 2009). Discrimination is a significant stressor, and evidence suggests experiences of discrimination are a potential risk factor for the onset of CVDRF (Dolezsar, McGrath, Herzig, & Miller, 2014; Lewis, Williams, & Clark, 2014). Our sample included exclusively individuals with CVDRF, and in support of this literature, we found that experiences of lifetime discrimination emerged as a defining set of psychosocial factors. However, it is critical to note that discrimination wasn’t the only stressor experienced in this profile. This profile was also marked by the highest level of chronic burden stressors and vigilance, while also experiencing poor social factors.

Therefore, the three profiles of individuals with CVDRF are psychosocially heterogeneous, and thus meaningfully different. In future CVD research, it may be necessary to account for this complexity in lived experiences (Meyer, Stanley, & Vandenberg, 2013; Murdoch & Detsky, 2013). Importantly, the identification of these profiles of individuals through this multivariate approach may promote emerging clinical and research initiatives, such as delivering personalized medicine, assisting the referral of patients to resources in the community (Allen et al., 2020; Andermann, 2018; Gajardo, Henriquez, & Llancaqueo, 2020), and improving the development and implementation tailored interventions among individuals with CVDRF. For instance, in our study individuals in the “Discriminated” profile and “Distressed and Disadvantaged” profile indicated similarly high levels of heightened chronic burden stressors. Through the traditional “single factor” approach, these individuals may be treated or conceptualized as the same. However, the presentation of the profiles as a whole underscores key differences that should inform the development of psychosocial interventions with a more effective reach (Flynn, Moran, Rash, & Campbell, 2019). A person in the “Distressed” profile may undergo cognitive behavioral therapy for reducing depressive symptoms and improving perceptions of control, stress management, or perhaps undergoing an intervention aimed at reducing social isolation. Interventions for individuals in the “Discriminated” profile should prioritize their high goal orientation, and could sponsor ties to community-level efforts, led by trusted community members who promote health and are knowledgeable of the needs of the community and challenges of persistent discrimination.

In our examination of these profiles with achievement of optimal CVDRF levels longitudinally, there were no significant differences in achievement of all 3 targets combined across the three profiles in the fully adjusted models. Investigating each target separately, we found individuals in the “Healthy” profile were more likely to meet the HbA1c target than those in the “Distressed and Disadvantaged” profile, after adjustment for socioeconomic and demographic factors. Findings were similar in magnitude in our sensitivity analyses using a target of HbA1c < 8.0, but findings were not statistically significant. This significant association in model 2 was attenuated in the fully adjusted model. Upon examination of each of the factors in the fully adjusted model, we found this was primarily driven by waist circumference. It is possible depressive symptoms and waist circumference may be part of a larger, more complex model predicting the achievement of optimal HbA1c. Research has demonstrated an indirect effect of depression on glycemic control via poor self-care behaviors, such as diet and exercise (Schmitt, Bendig, Baumeister, Hermanns, & Kulzer, 2021; Snoek, Bremmer, & Hermanns, 2015). These behaviors in turn may possibly inform waist circumference, and subsequently HbA1c. Future research should implement structural equation modeling to comprehensively examine these unique direct and indirect pathways to achieving optimal HbA1c.

Unlike HbA1c, we observed no significant effects for SBP or LDL as individual targets after partial or complete adjustment. It is possible that these psychosocial profiles may not predict optimal LDL target levels among younger and relatively healthy adults specifically. As for LDL, prior studies have found that hyperlipidemia at a younger age may be familial, and therefore we may not have been able to identify differences in achievement of optimal LDL levels among profiles (Chou et al., 2016). Further, these non-significant differences in meeting the optimal LDL target may also explain why there were no significant differences in achievement of optimal CVDRF levels of all 3 targets combined across the profiles. As for SBP, we observed that the majority of our study population had achieved the SBP target at each exam year. This may explain why we did not find significant differences in SBP achievement among the profile groups.

There are several other reasons why there were no significant differences in achievement of optimal CVDRF levels across profiles. First, it is possible these profiles may be more predictive of achieving optimal CVDRF levels only after an individual experiences a cardiovascular event, which may be a life-threatening and potentially traumatic experience (Cohen, Edmondson, & Kronish, 2015). Second, the profiles in the current study were developed using comprehensive measures at a single time point. The profiles from the current study do not account for the potential dynamic effects of psychosocial factors over time (Sutin et al., 2013). Evidence suggests that psychosocial factors may change over the life course, and remaining in the “Distressed and Disadvantaged” profile consistently over fifteen years may be more predictive of not achieving target levels as compared to being in that profile at just a single time point in life. It is possible there were unmeasured aspects of the psychosocial variables that were not fully captured as part of the profiles. For instance, the lifetime discrimination scale in the current study assessed discriminatory events at any point over an individual’s life. However, we did not have additional information regarding how the discriminatory events were perceived. There are critical interpersonal differences in how individuals appraise—or cognitively evaluate—discriminatory events as threatening, stable, and specific (Eccleston & Major, 2006). Some discriminatory events may be perceived as more damaging and negatively impactful than other events. Therefore, it is possible that the perceived severity of discriminatory experiences is more predictive of achievement, than the presence or absence of discriminatory events.

The current study adds to the body of research focused on psychosocial factors and achievement of optimal CVDRF levels. A major strength of the study includes the extensive range of psychosocial variables included simultaneously in the LPA. Most studies focus on a few variables, and we were able to identify profiles based on a more comprehensive group of variables. Our sample included the equal representation of non-Hispanic black and non-Hispanic white participants. Another strength included the fifteen-year longitudinal follow-up of younger adults. A notable implication of this study is the findings given the age of participants in our sample. Understanding associations with achievement of optimal CVDRF levels among this age group may be essential to preventing future cardiovascular events.

Our study also included limitations. The first was the participant sample size. Participants were included if they were knowledgeable of their diagnosis at baseline. We did not include individuals that solely met the clinical criterion of hypertension, hypercholesterolemia, and/or diabetes because individuals who are naïve to their condition may not actively be trying to reach target levels. Further, the inclusion criteria likely informed the percentage of participants with access to medical care. Being knowledgeable of a diagnosis, required access to a doctor or nurse.

The majority of the analytic sample (81.5%) indicated no perceived barriers to care. Individuals with barriers to care may not be as knowledgeable of a potential diagnosis, and these individuals are underrepresented in the analytic sample. In addition, achievement of optimal CVDRF levels was determined as met if participants met all three targets (i.e., HbA1c, SBP, LDL) even though most participants just indicated a history of one CVDRF (e.g., exclusively hypertension). This is because patients with CVDRF are generally encouraged by their doctor to achieve optimal levels for all three (Mayne et al., 2019). Likewise, while the rates of those who achieved optimal levels for HbA1c and SBP were high, these rates do not reflect the percentage of people who have that specific CVDRF (e.g., hypertension) and have met these levels. Due to small sample sizes, we were unable to focus exclusively on a sample of individuals with a single risk factor (e.g., hypertension) and examine associations between profile groups and a single target (e.g., SBP). It is possible that there may be important differences between the different CVDRF conditions, and it is necessary to examine the association between profile groups and specific target outcomes. Further, we were unable to include medication adherence in the model because this information was largely missing over the years. We suggest that future research incorporate this factor into statistical models. While the CARDIA study included black and white adults, future studies should include other racial and ethnic minorities who have a high burden of CVDRF.

Conclusions

Identifying distinct psychosocial profiles of patients provides an important lens by which CVDRF research can be empirically examined, and a way for healthcare providers to consider critical differences among patients. Response patterns of each profile indicate that the ways in which individuals with CVDRF experienced the same set of psychosocial factors were meaningfully different. The “Distressed and Disadvantaged” profile was less likely to meet the HbA1c target than those in the “Healthy” profile, after partial adjustment only. Further research should examine the association of profiles and achievement of optimal CVDRF levels in a population with higher rates of CVDRF and history of cardiovascular events.

Availability of Data and Material:

The data from the current study are from the CARDIA study.