Vasculopathy related to manic/hypomanic symptom burden and first generation antipsychotics in a sub-sample from the Collaborative Depression Study (CDS)

Abstract

Background

Mood disorders substantially increase risk of cardiovascular disease, though the mechanisms are unclear. We assessed for a dose-dependent relationship between course of illness or treatment with vasculopathy in a well-characterized cohort.

Methods

Participants with mood disorders were recruited for the National Institute of Mental Health Collaborative Depression Study (CDS) and followed prospectively. A cross-sectional metabolic and vascular function evaluation was performed on a sub-sample near completion after a mean follow-up of 27 years.

Results

A total of 35 participants from the University of Iowa (33) and Washington University (2) sites of the CDS consented to a metabolic and vascular function assessment at the Iowa site. In multivariate linear regression, controlling for age, gender, and smoking, manic/hypomanic, but not depressive, symptom burden was associated with lower flow-mediated dilation (FMD). Cumulative exposure to antipsychotics and mood stabilizers was associated with elevated augmentation pressure and mean aortic systolic blood pressure. This appeared specifically related to first generation antipsychotic exposure and mediated by increases in brachial systolic pressure. Although second generation antipsychotics were associated with dyslipidemia and insulin resistance, they were not associated with vasculopathy.

Conclusions

These results provide evidence that chronicity of mood symptoms contribute to vasculopathy in a dose-dependent fashion. Patients with more manic/hypomanic symptoms had poorer endothelial function. First generation antipsychotic exposure was associated with arterial stiffness, evidenced by higher augmentation pressure, perhaps secondary to elevated blood pressure. Vascular phenotyping methods may provide a promising means of elucidating the mechanisms linking mood disorders to vascular disease.

INTRODUCTION

The National Institute of Mental Health Collaborative Depression Study (CDS) investigated the nature, course, and etiology of mood disorders [1, 2]. The five participating centers of the clinical studies component (Massachusetts General Hospital and Harvard University in Boston, Rush Presbyterian – St. Luke’s Medical Center in Chicago, the University of Iowa in Iowa City, New York State Psychiatric Institute and Columbia University in New York, and Washington University in St. Louis) recruited 955 individuals with mood disorders from 1978–1981 and followed them for up to 31 years. Publications from the CDS have contributed greatly to understanding the nosology and prognosis of mood disorders. Recent work has quantified the impact of mood disorders on disability [3, 4], suicide [5], and cardiovascular mortality [6].

The relationship between mood disorders and vascular disease has been seen across specific mood diagnoses in representative samples [7–9] and with vascular outcomes in those with coronary artery disease [10, 11]. Potential mechanisms have been divided into direct pathways involving physiological changes secondary to mood, indirect pathways involving traditional risk factors for vascular disease, or through background factors that contribute to developing both mood disorders and vascular disease [12]. Vascular phenotyping may be a useful tool to identify relevant mechanisms although there has been limited study with available research demonstrating significant impairments in vascular function or stiffness with depression and/or bipolar disorder in most [13–17] but not all studies [18]. These cross-sectional studies have focused on the presence or absence of illness without dosing mood [15–17, 19], apart from assessing duration of illness [13].

The identification of a dose-response supports a causal relationship between mood and vascular outcomes. Lesperance et al. found severity of depressive symptoms at baseline and 1 year following myocardial infarction to be associated with mortality in a dose-dependent fashion [20]. With an episodic course common to mood disorders, reliable methods of dosing mood quantitatively using longitudinal data may better classify symptom burden as an exposure, though the requisite data are seldom available. Prospective studies of symptom burden have found manic [6], but not depressive symptoms [6, 21], associated with subsequent cardiovascular mortality. Treatments for mood disorders may also adversely impact vascular risk factors and thus events [22, 23]. Pharmacoepidemiological studies have associated antipsychotics with sudden cardiac death [24], but otherwise there are limited data directly linking treatments with adverse vascular outcomes apart from cerebrovascular events in dementia [25].

Vasculopathy, defined as impaired function or structure of blood vessels, can be identified prior to adverse cardiovascular events using validated techniques. Assessment of endothelial function, in particular, provides a useful intermediate phenotype for the study of atherosclerosis and may serve as a clinically relevant, quantitative and sensitive surrogate outcome [26, 27]. Endothelial function is impaired by every known risk factor for atherosclerosis, precedes the development of atherosclerotic lesions, and predicts progression and complications of atherosclerosis [28], even independent of conventional risk factors [29]. Flow-mediated dilation (FMD), a measure of endothelial function, prospectively predicts vascular events in a manner that allows the clinical relevance of findings to be translated. In population-based samples of older adults, an effect size of 2/3 standard deviations (SD) in FMD has been associated with a 25–30% increased risk of cardiovascular events [30, 31].

To ascertain whether lifetime symptom burden or medication exposure are predictive of vascular dysfunction, we assessed vascular function cross-sectionally in a sample of individuals with affective disorders whose phenomenological and treatment histories have been rigorously assessed over roughly one quarter century through participation in the CDS. Based on our prior findings with cardiovascular mortality in this sample [6], we hypothesized that manic, but not depressive, symptom burden would be associated with poorer vascular function. Based on metabolic profile, we further hypothesized that treatment with second generation antipsychotics and valproic acid derivatives would be associated with poorer vascular function.

METHODS AND MATERIALS

Sample

The CDS involved a large, prospective cohort of individuals followed for up to 31 years to test hypotheses related to the nosology, phenomenology, and etiology of mood disorders [1]. The sample included Caucasian (genetic hypotheses were proposed), English-speaking individuals who had knowledge of their biological parents and provided written informed consent. We recruited participants from those who completed the CDS study at two of five CDS sites and resided near Iowa City. Participants with a history of diabetes, hyperlipidemia, hypertension, or heart disease at intake into the CDS were excluded. We additionally excluded any participants over 80 years of age. A total of 59 former participants from The University of Iowa in Iowa City, IA and 7 former participants from Washington University in St. Louis, MO, representing all of those meeting inclusion criteria were contacted, of whom 44 (75%) were female. Two of the male former participants were deceased. Procedures were approved by the respective Institutional Review Boards.

CDS Assessments

On intake into the CDS, participants underwent interviews for the Schedule for Affective Disorders and Schizophrenia [32] and the Personal History of Depressive Disorders (available upon request). Follow-up assessments categorized severity of affective psychopathology from Psychiatric Status Ratings (PSRs) using the Longitudinal Interval Follow-up Evaluation, administered every six months in the first five years and annually thereafter [33–35]. The PSRs provided weekly ratings of symptom levels for each Research Diagnostic Criteria syndrome [36], as previously described [37]. The PSRs have demonstrated intraclass correlation coefficients of 0.9 [38]. Intake and follow-up ratings facilitated rigorous ascertainment of mood disorder diagnosis using previously outlined methods [6, 37, 39].

PSRs were used to quantify longitudinal affective morbidity. Using previously published methods [6], a week of clinically significant affective symptoms was defined as a PSR cutoff score of > 2/6 (required at least obvious evidence of the disorder) on the major depression, schizoaffective depression scale, mania, or schizoaffective mania scales or a score of 3/3 (definite criteria) for minor depression, intermittent depression, or hypomania. Burden of clinically significant depressive or manic/hypomanic morbidity was expressed as the proportion of weeks during follow-up with depressive or manic/hypomanic symptoms, respectively, exceeding these thresholds. These measures have been previously shown to have stability over long-term follow-up for both major depression and bipolar disorder [40, 41]. Treatment exposure was recorded during follow-up and included the following classes of medications: first-generation antipsychotics, second-generation antipsychotics, lithium, valproic acid derivatives, carbamazepine, lamotrigine, tricyclic antidepressants, selective serotonin reuptake inhibitors, monoamine oxidase inhibitors, other antidepressants, and benzodiazepines [6]. These classes were pooled into two broad medication classes: mood stabilizers/antipsychotics and antidepressants. In this observational CDS study, study investigators observed but did not direct treatment.

Chart Review

Weight or other metabolic assessments were not obtained at baseline through the CDS. We systematically reviewed the medical records within two years of CDS intake of all participants to abstract height, weight, total cholesterol, blood pressure, and heart rate.

Metabolic and Vascular Assessments

Participants were instructed to fast for 12 hours and not to smoke 2 hours prior to assessment and this was verified by direct questioning. Upon arrival, a research nurse measured vital signs, height, weight, and waist circumference. A blood sample was drawn. Participants then provided a health history and underwent physical examination.

Vascular function was assessed using pulse wave analysis, pulse wave velocity, and conduit vessel function by individuals blinded to diagnosis, current medications, and any CDS data. Our primary outcomes of interest were FMD (endothelium-dependent) and nitroglycerin (NTG; endothelium-independent)-mediated vasodilation of the brachial artery, assessed non-invasively via ultrasound measurement of brachial artery diameter using previously described methods [42] with a 10 MHz linear array transducer (Biosound Esaote AU5) by a trained sonographer. These measures were selected as the primary outcome because they represent the most extensively studied and validated non-invasive measures of endothelial function though are seldom used clinically because of the cost and need for highly trained operators [43]. A longitudinal section of the brachial artery was imaged above the antecubital fossa where baseline images of diameter and Doppler velocities were recorded. Flow-mediated dilation was assessed one minute after release of a distal occluding forearm cuff that had been inflated to 50 mg Hg above systolic pressure for 5 minutes. After 10 minutes of rest and restoration of baseline flow and diameter, nitroglycerin 400 mcg was administered by sublingual spray. NTG-mediated vasodilation was assessed four minutes later.

Vascular damage (such as endothelial dysfunction) leads to increased vascular stiffness, and this translates into accelerated return of peripheral pulse waves to the aortic, leading to increases in augmentation pressure, augmentation index, and overall aortic systolic pressure. Secondary outcomes included pulse wave velocity, aortic augmentation pressure, augmentation index (adjusted for heart rate of 75), and systolic aortic pressure. These were assessed using Sphygmocor arterial tonometry system [44], which measures peripheral arterial pulse wave timing and characteristics. For the measurement of pulse wave velocity, the pressure pulse waveform was recorded simultaneously with an electrocardiogram signal, which provides an R-wave timing reference. Recordings were performed consecutively at the carotid and femoral artery sites to measure the pulse wave velocity in a section of artery that includes the aorta. Information from peripheral pressure pulse waveform in addition to systolic and diastolic pressures were used to derive central aortic pressure waveform and a range of central indices of ventricular-vascular interaction.

Exploratory outcomes included risk factors for vascular disease: body mass index (BMI), waist circumference, fasting cholesterol (high density lipoprotein (HDL) cholesterol, low density lipoprotein (LDL) cholesterol, Lp(a)), fasting triglycerides (total, LDL, very low density lipoprotein (VLDL)), apolipoprotein B, brachial (clinic) arterial pressure, insulin resistance (calculated from fasting insulin and glucose by homeostatic model assessment for insulin resistance (HOMA-IR)[45]), highly sensitive C-reactive protein, and interleukin-6.

Statistical Analyses

All analyses were conducted using SAS 9.2. Descriptive statistics were compiled on a variety of sociodemographic and clinical variables, which were compared to that of the full CDS sample of those who completed at least 24 years of follow-up (N=362) using t-tests and chi-square tests for continuous and categorical variables respectively.

The proportion of weeks with clinically significant manic, hypomanic, minor depressive, or major depressive symptoms during follow-up quantified symptom burden. To minimize risk of Type I error on multiple comparisons and given overlapping metabolic adverse effects, treatment exposures were lumped into two broad classes: mood stabilizers/antipsychotics and antidepressants. Analyses of more refined medication classes followed any significant findings. Given the infrequent use of these agents, carbamazepine and lamotrigine were not included in these follow-up analyses for mood stabilizer / antipsychotic exposure though exposure to these agents was included in the broad classification. Individuals on combinations of medications were considered exposed if any one of those medications belonged to an exposure group of interest. To minimize the impact of outliers and to maximize the likelihood that residuals were normally distributed in this small sample, exposures of interest were modeled as the logit transformation of the proportion of time exposed. Linear regression models determined the effect of symptom burden and treatment exposure on vascular function, controlling for age, gender, and tobacco exposure in pack*years. Because symptom burden and respective treatments were highly correlated in this small sample, symptom burden and treatment exposure were modeled separately to mitigate multicollinearity and overfitting. Linear regression models followed the following format:

• Y = β₀ + β₁X₁ + β₂X₂ + β₃X₃ + β₄X₄ + β₅X₅ + ε

where Y represents our dependent outcome variable, X₁ is age (linear effect), X₂ is gender, and X₃ is tobacco exposure in pack years. In symptom burden models, X₄ is proportion of follow-up with clinically significant depressive symptoms and X₅ is the proportion of follow-up with clinically significant manic or hypomanic symptoms. In the treatment models, X₄ is proportion of follow-up treated with antidepressants and X₅ is the proportion of follow-up treated with antipsychotics or mood stabilizers. To explore for any potential overfitting, sensitivity analyses using reduced models, including only age with symptom burden or treatment exposure variables, were created for any models producing significant results.

Exploratory analyses investigated the relationship between symptom burden and treatment exposure and potential physiological mediators: body fat percentage, body mass index, waist circumference, brachial systolic blood pressure, laboratory measures, and HOMA-IR. Any potential mediators found to have a significant association with exposures of interest (symptom burden and treatment exposure) were added to regression models in which these exposures were associated with primary or secondary outcomes. The reduction in R² for the exposure of interest after inclusion of the potential physiological mediator was used to estimate possible impact. Mediation was further explored in a sub-sample of participants with available data from chart abstraction for changes in relevant mediators over the course of CDS follow-up.

RESULTS

Baseline Demographic and Clinical Characteristics

35 participants from the University of Iowa (33) and Washington University (2) sites of the CDS consented for cross-sectional vascular and metabolic assessment at the University of Iowa. The enrolled participants had been followed for a mean (SD) of 26.8 (1.2), median of 27, at least 24 and up to 30 years. Sociodemographic characteristics of the sample are outlined in Table 1. The mean (SD) age of participants was 61(8) (range 50 to 76), 83% were female, and 57% had a diagnosis of bipolar disorder. The mean (SD) body mass index was 30 (6) and 63% smoked tobacco. Participants significantly differed from others in the full CDS sample that completed 24 years of follow-up only regarding gender (χ²=5.04, df=1, p=0.03) though did not significantly differ from those recruited based on our inclusion criteria.

Table 1.

Sociodemographic and clinical characteristics of sample as available from CDS (N=35).

	Mean (SD)
Age	61 (8)
Age of onset for mood disorder	25 (7)
Percent of time with clinically significant symptoms
Depressive	30.0 (26.6)%
Manic/hypomanic (if bipolar)	7.1 (19.9)%
	# (%)
Female Sex†	29 (83%)
Unipolar Major Depression	15 (43%)
Bipolar disorder	20 (57%)
Bipolar I	14 (40%)
Bipolar II	6 (17%)
History of Tobacco Use	22 (63%)
Medication Treatment (any history of)
First Generation Antipsychotics	25 (71%)
Second Generation Antipsychotics	11 (31%)
Valproic acid derivatives	13 (37%)
Lithium	22 (63%)
Carbamazepine	7 (20%)
Lamotrigine	3 (9%)
Antidepressants	31 (89%)

^†Significantly higher than expected from the Collaborative Depression Study participants with at least 24 years of follow-up (χ²=5.04, df=1, p=0.025).

Clinical characteristics of the sample as assessed are detailed in Table 2. FMD data for one participant could not be interpreted due to motion. Pulse wave velocity could not be reliably obtained on four participants. Augmentation index could not be calculated to a heart rate of 75 for two participants due to unusually low resting heart rates. The total number of participants included in the analyses varies accordingly for models on these outcomes described below.

Table 2.

Clinical characteristics of sample from cross-sectional metabolic and vascular assessment.

	Mean (SD)
Body Mass Index (kg/m2)	29.8 (6.2)
Waist Circumference (cm)	100.2 (15.4)
FMD Diameter % Change	4.6 (4.3)
NTG-mediated Diameter % Change	11.0 (5.4)
Augmentation Pressure (mmHg)	14.9 (8.0)
Augmentation index adjusted for HR=75	25.6 (9.2)
Systolic aortic pressure (mmHg)	122 (17)
Pulse wave velocity (m/s)	10.8 (3.8)
Triglycerides (mg/dL)
Total	122.2 (91.6)
LDL	41.8 (12.9)
VLDL	65.5 (54.9)
Cholesterol (mg/dL)
Total	204.0 (31.9)
LDL	121.6 (27.9)
HDL	55.1 (12.5)
VLDL	21.6 (12.0)
Apolipoprotein B (mg/dL)	92.3 (20.8)
Glucose (mg/dL)	100.5 (16.6)
Insulin, total (uU/mL)	10.9 (9.0)
Leptin (ng/mL)	22.3 (20.1)
Highly sensitive C-reactive protein (mg/L)	4.0 (4.8)
Interleukin 6 (pg/mL)	2.4 (1.8)
Cumulative tobacco exposure (pack*years)	13.0 (20.0)

FMD = flow-mediated dilation, HDL = high density lipoprotein, HR = heart rate, LDL = low density lipoprotein, NTG = nitroglycerin, VLDL = very low density lipoprotein

Depressive symptom burden was highly correlated with cumulative exposure to antidepressants (r=0.86, p<0.0001) as was manic symptom burden with cumulative exposure to mood stabilizers and antipsychotics (r=0.72, p<0.0001). For this reason, symptom burden and treatment variables were modeled separately. There was a negative correlation between manic and depressive symptom burden (r=−0.47, p<0.004).

Symptom Burden and Vascular Function

Table 3 presents results of symptom burden and treatment exposure models on primary outcome measures. In multivariable linear regression, controlling for age, gender, and tobacco exposure, manic symptom burden was associated with lower flow-mediated dilation (t=−2.21, partial R²=0.15, p=0.035). This would represent an absolute reduction of 2.2% (0.52 SD) in FMD (baseline=4.6%) across the interquartile range of those with bipolar disorder (0.6 to 5.6% manic/hypomanic symptom burden). Using data from the entire CDS sample, this would represent an absolute difference of 2.3% dilation (0.54 SD) on FMD (or about a 50% relative change in FMD) between those with 0.4% and 4% manic/hypomanic symptom burden, the previously reported median for bipolar II and bipolar I respectively [6]. Manic symptom burden was not associated with NTG-mediated vasodilation. Depressive symptom burden was not associated with any of the tested measures as reported. Symptom burden measures were not associated with any of the secondary outcomes. Exploratory analyses associated manic symptom burden with hypertriglyceridemia (t=2.12, partial R²=0.13, p=0.043), but not specifically with LDL-triglycerides or VLDL-triglycerides. Manic symptom burden was also associated with higher BMI (t=2.55, partial R²=0.18, p=0.016), waist circumference (t=2.79, partial R²=0.21, p=0.009), and HOMA-IR (t=2.94, partial R²=0.23, p=0.006). Depressive symptom burden was associated with a higher heart rate (t=2.11, partial R²=0.13, p=0.044). However, in the 28 participants with available data, manic symptom burden was not correlated with change in BMI nor was depressive symptom burden with change in heart rate.

Table 3.

Results of Multivariate Linear Regression Models for Exposures of Interest on Primary Outcomes.

	β	SE	t	p
Dependent Variable = FMD (N=34)
Symptom Burden Model
Depressive symptom burden	−0.576	0.385	−1.50	0.15
Manic/hypomanic symptom burden	−0.993	0.449	−2.21	0.035*
Treatment Model
Antidepressant exposure	−0.221	0.276	−0.80	0.43
Mood stabilizer/antipsychotic exposure	−0.225	0.173	−1.31	0.20
Dependent Variable = NTG-mediated dilation (N=35)
Symptom Burden Model
Depressive symptom burden	0.808	0.465	1.74	0.09
Manic/hypomanic symptom burden	0.308	0.536	0.57	0.57
Treatment Model
Antidepressant exposure	0.411	0.320	1.29	0.21
Mood stabilizer/antipsychotic exposure	−0.111	0.204	−0.54	0.59

^*p < 0.05. Coded such that greater manic/hypomanic symptoms were associated with poorer endothelial function.

All models controlled for age, gender, and tobacco exposure (pack*years)

FMD = flow-mediated dilation, NTG = nitroglycerin, SE = standard error

Medication Exposure and Vascular Function

Antipsychotic / mood stabilizer and antidepressant exposure were not associated with FMD or NTG-mediated vasodilation. Cumulative exposure to antipsychotics and mood stabilizers was associated with vascular stiffness as evidenced by elevated aortic systolic augmentation pressure (t=2.66, partial R²=0.20, p=0.013) and total aortic systolic blood pressure (t=2.37, partial R²=0.16, p=0.025). Treatment exposure models for these measures are reported in Table 4. As evidenced in the table, first generation antipsychotics were the only medication class significantly associated with these findings. An increase of 5.8 mmHg (0.73 SD) in augmentation pressure and 16.8 mmHg (1.00 SD) in aortic systolic blood pressure was observed across the interquartile range of those exposed to first generation antipsychotics (3 to 44% exposure). In exploratory analyses, first generation antipsychotics were associated with elevated peripheral brachial systolic arterial pressure (t=3.41, partial R²=0.28, p=0.002). Second generation antipsychotics were associated with hypertriglyceridemia (t=3.11, partial R²=0.24, p=0.004) and insulin resistance (t=2.16, partial R²=0.13, p=0.039). Valproic acid derivative exposure was associated with hypertriglyceridemia (t=2.96, partial R²=0.23, p=0.006), insulin resistance (t=2.28, partial R²=0.15, p=0.030), and low HDL-cholesterol (t=−4.33, partial R²=0.38, p=0.0002). For second generation antipsychotics and valproic acid derivatives, similar results were observed for VLDL-triglycerides as total triglycerides, but LDL-triglycerides were not related to exposure. In the subset of 28 individuals with available data, exposure to first generation antipsychotics was significantly associated with change in systolic blood pressure, even after controlling for age, gender, and tobacco exposure.

Table 4.

Results of Multivariate Linear Regression Models for Treatment Exposure on Augmentation Pressure and Systolic Aortic Pressure (N=35).

	β	SE	t	p
Dependent Variable = Augmentation Pressure
Treatment Model
Antidepressant exposure	−0.364	0.452	−0.80	0.43
Mood stabilizer/antipsychotic exposure	0.769	0.290	2.66	0.013*
Individual mood stabilizer/antipsychotic classes (modeled separately)
First generation antipsychotics	1.137	0.457	2.49	0.019*
Second generation antipsychotics	0.522	0.684	0.76	0.45
Lithium	0.613	0.353	1.74	0.09
Valproic acid derivatives	0.267	0.597	0.45	0.66
Dependent Variable = Systolic Aortic Pressure
Treatment Model
Antidepressant exposure	−0.846	1.016	−0.83	0.41
Mood stabilizer/antipsychotic exposure	1.539	0.650	2.37	0.025*
Individual mood stabilizer/antipsychotic classes (modeled separately)
First generation antipsychotics	3.275	0.930	3.52	0.001*
Second generation antipsychotics	1.776	1.486	1.20	0.24
Lithium	0.745	0.803	0.93	0.36
Valproic acid derivatives	0.331	1.319	0.25	0.80

^*p < 0.05. Coded such that greater manic/hypomanic symptoms were associated with poorer endothelial function.

All models included age, gender, and tobacco exposure (pack*years)

SE = standard error

The 25 individuals who received first generation antipsychotics were taking them for 23% of the observed follow-up period and were more likely to have bipolar I disorder (χ²=5.25, df=1, p=0.02) though otherwise did not differ from the remainder of the sample. Controlling for diagnosis did not substantially alter the aforementioned findings. In an exploratory follow-up analysis, there was no relationship between lithium exposure and our primary and secondary vascular function outcomes.

Reduced Models

Very similar results were obtained for significant findings in the primary and secondary analyses above in the sensitivity analysis using reduced models that included only age as a covariate. In these models, manic symptom burden no longer crossed the threshold for statistical significance on FMD (p=0.06).

Exploratory Mediation Analyses

When manic symptom burden was modeled individually with its physiological correlates, BMI produced the greatest reduction in the partial R² for manic symptom burden on FMD (52%). Mediation of manic symptoms on FMD by BMI, however, was not supported by the lack of an association between manic symptom burden and change in BMI in the subsample with available data with baseline (N=28). Further, those with a higher manic symptom burden tended to have a higher BMI at baseline though this correlation was not significant. Adding peripheral brachial systolic pressure dramatically reduced the partial R² between first generation antipsychotic exposure and central aortic augmentation pressure (by 82%). In the sub-sample with available baseline data (N=28), change in systolic blood pressure reduced the partial R² between first generation antipsychotic exposure and central aortic augmentation pressure as well (by 33%), rendering the initial association no longer significant. Models with peripheral brachial systolic blood pressure on aortic systolic blood pressure were not produced as the latter is derived from the former. In the sub-sample, inclusion of change in systolic blood pressure reduced the partial R² for first generation antipsychotic exposure on aortic systolic blood pressure by 42% while first generation antipsychotic exposure remained significantly associated.

DISCUSSION

Measures of mania chronicity were inversely proportional to endothelial function (FMD). With data linking an effect size of 0.66 SD in FMD to a 25–30% increased risk of cardiovascular events [30, 31], our finding of 0.54 SD FMD from the manic symptom burden of 0.4% for a typical bipolar II to the manic symptom burden of 4% for a typical bipolar I [6] would be expected to translate to roughly a 20–25% increased risk of such events. First generation antipsychotic exposure was associated with greater arterial stiffness, evidenced by higher augmentation pressure and aortic systolic pressure. In exploratory mediation models, this appears to be mainly due to elevation of peripheral systolic arterial pressure with first generation antipsychotics. Long-term exposure to second generation antipsychotics and valproic acid derivatives were associated with several vascular risk factors of the metabolic syndrome.

The demonstration of an association between first generation antipsychotics and arterial stiffness is of particular interest and was not anticipated by our hypotheses or previously studied. The alpha-1 adrenergic antagonism of many first generation antipsychotics would be expected, if anything, to reduce total peripheral resistance and subsequently blood pressure. Through this mechanism, antipsychotics can induce orthostatic hypotension [46]. Unlike most clinical trials, our analysis assesses medication exposure over several decades. Chronic administration of haloperidol induces vasoconstriction in animal models [47]. D₂ receptors are involved in the inhibition of sympathetic nerve activity [48] and D₂ antagonism, which is characteristic of first generation antipsychotics, could perhaps increase sympathetic tone and subsequently blood pressure. Hypertension is arguably the most robust cause of increased vascular stiffness [49]. The potential mediation of the induction of arterial stiffness with antipsychotics by systolic blood pressure appears biologically plausible and amenable to confirmatory study.

Methodological strengths of our approach include application of the concept of symptom burden, which overcomes mere evaluation of the presence/absence of a disorder and the inherent selection bias therein. Symptom burden also overcomes the limitations of dosing severity at a single time point, which fails to capture the burden of affective symptomatology over the long-term course of an often episodic illness that fluctuates in its severity. Interview-based ratings of symptoms were obtained, which may be more sensitive in detecting cognitive symptoms of depression [50]. We also utilized vascular function as a primary outcome, facilitating mechanistic investigation. There are, nonetheless, several limitations to our study. Participants were not randomized to treatment. Our relatively high acuity clinical sample may have underrepresented those with milder forms of depressive illness, restricting our range of exposure and subsequently decreasing our ability to detect an effect of depressive symptom burden. The relatively small sample size similarly increases risk of Type II error and limits generalizability. Risk of type I error was mitigated by defining primary outcomes a priori and grouping exposures classes. Our primary finding is further consistent with our previous report of a link between manic/hypomanic symptom burden and cardiovascular mortality in the CDS [6]. Exploration of mediation was limited by sample size and design. Temporal relationships between the potential mediators and vascular outcomes cannot be established with a single cross-sectional metabolic and vascular assessment so we acquired baseline data in a sub-sample from chart review. The design focused on potential physiological mediators relevant to vasculopathy, assuming that behavioral and psychological factors mediate their effects through physiological measures. Subsequent study of potential behavioral and psychological mediators would add value. Several of these factors could be related to symptom burden and have been linked with vascular outcomes, including but not limited to hostility [51], anxiety [52, 53], and social isolation [54].

The use of existing data from the CDS allowed us to rigorously classify symptom burden and treatment exposures to assess their impact on vascular function, adding considerably to the developing literature on vasculopathy in mood disorders. Our cross-sectional vascular phenotyping methods were similarly rigorous and allowed us to identify associations between manic symptom burden and endothelial dysfunction and first generation antipsychotics with arterial stiffness. In exploratory analyses, we were able to delve into potential mechanisms, such as elevated blood pressure on the relationship between first generation antipsychotics and arterial stiffness. Future studies may target other mediators within a causal pathway, such as behavioral or psychological variables that may link manic symptom burden with flow-mediated dilation and first generation antipsychotics with autonomic dysregulation. Due to their quantitative nature and sensitivity to vascular risk factors, vascular phenotyping methods provide a promising means of elucidating the mechanisms linking mood disorders to vascular disease.