Imbalance Model of Heart Rate Variability and Pulse Wave Velocity in Psychotic and Nonpsychotic Disorders

Abstract

Objectives

Patients with psychiatric disorders have an increased risk of cardiovascular pathologies. A bidirectional feedback model between the brain and heart exists widely in both psychotic and nonpsychotic disorders. The aim of this study was to compare heart rate variability (HRV) and pulse wave velocity (PWV) functions between patients with psychotic and nonpsychotic disorders and to investigate whether subgroups defined by HRV and PWV features improve the transdiagnostic psychopathology of psychiatric classification.

Methods

In total, 3448 consecutive patients who visited psychiatric or psychological health services with psychotic (N = 1839) and nonpsychotic disorders (N = 1609) and were drug-free for at least 2 weeks were selected. HRV and PWV indicators were measured via finger photoplethysmography during a 5-minute period of rest. Canonical variates were generated through HRV and PWV indicators by canonical correlation analysis (CCA).

Results

All HRV indicators but none of the PWV indicators were significantly reduced in the psychotic group relative to those in the nonpsychotic group. After adjusting for age, gender, and body mass index, many indices of HRV were significantly reduced in the psychotic group compared with those in the nonpsychotic group. CCA analysis revealed 2 subgroups defined by distinct and relatively homogeneous patterns along HRV and PWV dimensions and comprising 19.0% (subgroup 1, n = 655) and 80.9% (subgroup 2, n = 2781) of the sample, each with distinctive features of HRV and PWV functions.

Conclusions

HRV functions are significantly impaired among psychiatric patients, especially in those with psychosis. Our results highlight important subgroups of psychiatric patients that have distinct features of HRV and PWV which transcend current diagnostic boundaries.

Introduction

Cardiovascular morbidity in psychiatric disorders, such as schizophrenia and mood disorders, is common. This accounts for a shorter life expectancy in patients with severe mental disorders relative to the general population.^1,2 With the widespread use of psychiatric medications in the last 2 decades, risk factors for cardiovascular problems including diabetes, hypertension, dyslipidemia, metabolic syndrome, and obesity increased significantly.³ Both psychotic and nonpsychotic disorders are highly prevalent risk factors and are associated with morbidity and mortality of cardiovascular diseases.^4,5 Moreover, there is growing evidence that a significant rate of electrocardiogram (ECG) abnormalities among psychiatric patients could indicate an underlying cardiovascular pathology.^6,7

It is widely accepted that psychiatric disorders not only affect the brain but also involve the whole body, especially the heart.^8–10 The brain and heart are the 2 most vital organs of the body and interact very closely.¹¹ For example, when the mood (brain) is anxious or excited, the heart rate is instantly affected, and then the state of the brain will be re-affected by the heart through the activities of the sympathetic and parasympathetic nervous systems. During this mutual action and reaction interplay between the brain and heart, cardiac autonomic control plays a key role. Therefore, heart rate variability (HRV) has been widely used as a sensitive indicator of autonomic impairment in psychiatric disorders^12,13 as well as a prognostic risk factor for adverse cardiovascular events.^14–16

Another well-established cardiac risk factor is arterial wall stiffness; this can be determined by recording pulse wave velocity (PWV).¹⁷ With increased arterial wall stiffness mainly secondary to endothelial dysfunction and arteriosclerosis, the forward pulse wave travels faster and the reflected pulse wave returns earlier, influencing PWV. Therefore, PWV reflects properties of arterial stiffness and is associated with cardiovascular morbidity.^18,19 Along with increasing arterial stiffness, especially in the cerebrovascular microcirculation that can increase susceptibility to microvascular damage and remodeling in the brain,²⁰ it can result in structural brain damage and cognitive impairment.^21,22

Considering the growing body of evidence on the imbalance model of sympathetic and parasympathetic nervous systems and increased arterial wall stiffness in psychiatric patients, we speculated that cardiovascular differences in psychotic and nonpsychotic disorders might lead to the construction of subgroups of psychiatric patients. To the best of our knowledge, this study is the first to employ the canonical correlation analysis (CCA) method in defining subgroups by clustering psychiatric patients with different clinical diagnoses according to patterns of the relationship (canonical variates) between HRV and PWV. We further tested whether CCA-derived subgroups differed with regard to demographics, clinical classifications, and cardiovascular risks.

Aims of the Study

This study, therefore, used the indicators of HRV and PWV to test the following hypotheses: (1) There are significant cardiovascular differences among patients with psychotic and nonpsychotic disorders and controls without psychiatric disorder; (2) HRV and PWV functions are distinctly impaired between patients with psychotic and nonpsychotic disorders, and they may transcend current diagnostic boundaries.

Methods

Sample and Setting

The study was approved by the Research Ethics Committee at the Shanghai Mental Health Center (SMHC). Participants gave written informed consent at the time of the recruitment stage. Participants younger than 18 years of age had their consent forms signed by their parents. The sample consisted of 3448 consecutive patients who visited or were hospitalized to the psychiatric or psychological health services at SMHC between 2017 and 2019.

In this study, 3448 patients with psychotic and nonpsychotic disorders were selected. Inclusion criteria were as follows: (1) drug-free for at least 2 weeks or 2 weeks after the end of the active period of the long-acting drug; (2) individuals or their parents with the capacity to provide informed consent; and (3) met the International Classification of Diseases 10th Revision (ICD-10) criteria of psychotic and nonpsychotic disorders (affective disorders, anxiety disorder, or obsessive-compulsive disorder). Exclusion criteria were as follows: (1) acute or chronic renal failure, liver cirrhosis, or active liver diseases; (2) severe or unstable physical diseases, including neurological disorders (delirium, dementia, stroke, epilepsy, etc.), congestive heart failure, angina pectoris, myocardial infarction, arrhythmia, hypertension (including untreated or uncontrolled hypertension), immunocompromised conditions, and blood glucose above 12 mmol/L; (3) drug (such as methamphetamine) abuse or dependence; (4) current pregnancy; (5) stroke within the last month; and (6) other situations judged by the investigators as not suitable for the current study. The grouping method for psychotic and nonpsychotic disorders is based not only on the diagnosis but also on the presence/absence of positive psychotic symptoms. Particularly, patients included in the psychotic disorder group were characterized by emerging or worsening psychotic symptoms, and patients included in the nonpsychotic disorder group had a current absence of positive psychotic symptoms.

In the same period of patient recruitment, a total of 57 participants without any mental disorder were selected as the control group, which followed the same inclusion and exclusion criteria, except for the diagnosis. They initially visited the mental health service and were interviewed and assessed by our clinicians. According to our diagnosis system, they were regarded to have “no mental disorder” (Chinese Classification of Mental Disorders, Third Edition, CCMD-3 Code. 92.1) in their medical records.

Measurements and Variables

The cardiac autonomic activity was measured in a period of 10–15 minutes in total. Participants were asked to remove cell phones and metallic objects and were given approximately 5–10 minutes to adapt to the examination environment. They were instructed to breathe evenly, avoid noticeable body movements, and lean back in a comfortable chair in a quiet and calming room. Talking was restricted. ECG was recorded by a 5-minute single channel (3-lead, red [R] on the right forearm and yellow [L] and black (N) on the left forearm) using a portable electronic box and computerized analysis system (QHRV Pro+ HW6, Medeia Co., Ltd.). A computer that enabled the assessment of HRV and PWV variables was used. The 5-minute ECG signal was transformed to R-R intervals with an AC converter online (QRS detector and timer; resolution time: 2224 samples/s), which ensure the same procedure for all participants. A standard R-R interval-cleaning algorithm (exclude R-R intervals < 0.25 s and >2.5 s and those that differ by >15% from the previous one) was employed in the analysis system to detect gross artifacts or noise, and HRV and PWV analyses were performed on the cleaned data. As for the frequency-domain analysis, spectral powers of R-R intervals in respective frequency bands were calculated using Fast Fourier Transform. These were assessed using the software QHRV health manager (Medeia Co.), and assessments were based on the widely used time-domain analysis, frequency-domain analysis,^23,24 and pulse wave properties analysis.¹⁷

HRV Variables

HRV assesses the function of the autonomic nervous system. It relies largely on the balance between activities of both the sympathetic and parasympathetic nervous systems.

• 1) Time domain: Time-domain parameters were recorded and calculated to include long- and short-term assessments. For long-term assessment, the SD of all normal R-R intervals (SDNN, ms)²⁵ was used as a measure of variability in general autonomic nervous system function. For short-term assessment, the percentage of pairs of consecutive R-R intervals that differed by more than 50 milliseconds (pNN50, %)²⁶ was applied to reflect variability in cardiac parasympathetic tone.

• 2) Frequency Domain: Frequency-domain parameters were quantified frequency components of HRV that included 3 frequency regions (or bands)^27,28: very-low-frequency (VLF) power (0.003–0.04Hz), low-frequency (LF) power (0.04–0.15 Hz), and high-frequency (HF) power (0.15–0.40 Hz). The HF is an index of parasympathetic system functioning; conversely, the LF primarily reflects the functioning of the sympathetic nervous system. Moreover, a previous study suggested that the LF is associated with the modulation of cardiac autonomic outflows through baroreflexes.²⁹ Therefore, the LF/HF ratio³⁰ is assumed to be a balance index of the sympathetic/parasympathetic nervous system. The physiologic meaning of VLF is disputed but is possibly related to the parasympathetic system.³¹

PWV Variables

PWV is used to determine arterial stiffness; it is related to age, blood pressure, and arteriosclerosis and reflects the general condition of cardiovascular health.^17,19

• 1) Cardiac ejection elasticity index (EEI): EEI represents the left ventricle ejection capacity and compliance/elasticity of large arteries. EEI < 0.3 indicates vasoconstriction, arteriosclerosis, or left ventricle ejection insufficiency; EEI > 0.7 indicates vasodilation of large arteries, anemia, increased ejection power, or hyperthyroidism.

• 2) Dicrotic dilatation index (DDI): DDI represents elasticity in small arteries. DDI < 0.3 indicates vasoconstriction or arteriosclerosis (arterial stiffness); DDI > 0.7 indicates vasodilation.

• 3) Dicrotic elasticity index (DEI): DEI represents compliance/elasticity of small arteries/arteriole or venous blood flow. DEI < 0.3 indicates arteriolar constriction, arteriosclerosis; DEI > 0.7 indicates arteriodilation.

Data Analysis

Comparison

Patients were first divided into 2 groups: the psychotic disorder (diagnosis with psychosis) and nonpsychotic disorder (diagnosis with affective disorders, anxiety disorder, or obsessive-compulsive disorder) groups. Demographic features are presented with quantitative variables expressed as means (SD), and qualitative variables expressed as frequencies (%). The 2 groups were compared using χ² tests for comparison of categorical variables and independent t-tests for comparison of continuous variables. Effect sizes were calculated using Cohen’s d for mean comparisons among psychotic, nonpsychotic, and control groups (table 1). Effect sizes were classified as small (d = 0.2), medium (d =  0.5), and large (d ≥ 0.8).³² For further comparison of HRV and PWV variables between the 2 groups, age, gender, and body mass index (BMI) were controlled using multivariate analysis of variance (MANOVA) because of potential confounding factors (figure 1A). A histogram was created for further graphical display of the frequency in the LF/HF among the psychotic, nonpsychotic, and control groups (figure 1B). The 1-way ANOVA test was performed for comparisons of LF/HF among 3 groups.

Table 1.

Comparison of Demographic, Clinical, HRV, and PWV Characteristics Between Patients With Psychotic and Nonpsychotic Disorders

				Comparisons
Variables	Overall Sample	Psychotic Disorder	Nonpsychotic Disorder	t/χ 2 a	P	Effect Size Cohen’s d
Cases (n)	3448	1839	1609	—	—	—
Age (years) [Mean (SD)]	41.3 (17.9)	41.8 (15.8)	40.8 (19.9)	t = 1.7	.089	0.056
Age range (years)	7–93	10–86	7–93	—	—	—
Female [n (%)]	1746 (50.6)	845 (45.9)	901 (56.0)	χ 2 = 34.7	<.001	—
Weight (kg) [Mean (SD)]	65.6 (13.9)	67.1 (13.9)	63.9 (13.6)	t = 6.8	<.001	0.233
Height (cm) [Mean (SD)]	166.9 (8.2)	167.6 (8.0)	166.0 (8.4)	t = 5.6	<.001	0.195
BMI [Mean (SD)]	23.5 (4.0)	23.8 (4.1)	23.1 (3.9)	t = 5.3	<.001	0.175
Number of smokers [n (%)]	1102 (32.0)	688 (37.4)	414 (25.7)	χ 2 = 53.9	<.001	—
Number of drug-naïve [n (%)]	409 (11.8)	193 (10.5)	216 (13.4)	χ 2 = 7.0	.008	—
Number of inpatient [n (%)]	1883 (54.6)	1263 (68.7)	620 (38.5)	χ 2 = 284.0	<.001	—
Diagnostic classification [n (%)]
Schizophrenia	1771 (51.4)	1771 (96.3)c	0	—	—	—
Depression	345 (10.1)	0	345 (21.4)	—	—	—
Bipolar disorder (nonpsychotic)	704 (20.4)	0	704 (43.8)	—	—	—
Anxiety disorder	475 (13.8)	0	475 (29.5)	—	—	—
Obsessive-Compulsive disorder	85 (2.3)	0	85 (5.3)	—	—	—
HRV and PWV profiles
Total cardiac pulsation [Mean (SD)]	265.8 (45.0)	268.6 (45.7)	262.6 (44.1)	t = 4.0	<.001	0.134
Number of pseudo-morph [Mean (SD)]	1.7 (3.6)	1.7 (3.7)	1.7 (3.5)	t = 0.2	.864	0
Heart rate [Mean (SD)]	88.8 (14.2)	89.8 (14.5)	87.7 (13.8)	t = 4.2	<.001	0.148
Time domain
SDNN [Mean (SD)] (ms)	28.3 (17.2)	27.3 (17.1)	29.4 (17.2)	t = 3.6	<.001	0.122
pNN50 [Mean (SD)] (%)	6.0 (12.0)	5.5 (11.4)	6.5 (12.6)	t = 2.4	.016	0.083
Frequency domain
VLF [mean (SD)]	94.7 (77.7)	89.8 (74.0)	100.2 (81.3)	t = 4.0	<.001	0.134
LF [mean (SD)]	109.8 (105.9)	104.7 (106.2)	115.6 (105.3)	t = 3.0	.002	0.103
HF [mean (SD)]	119.8 (109.1)	114.8 (105.4)	125.4 (113.0)	t = 2.9	.004	0.097
LF/HF [mean (SD)]	1.004 (0.506)	0.980 (0.517)	1.031 (0.491)	t = 3.0	.003	0.101
Total power	923.4 (750.1)	886.6 (737.7)	965.5 (762.1)	t = 3.1	.002	0.105
PWVb
EEI	0.584 (0.160)	0.582 (0.154)	0.586 (0.167)	t = 0.7	.457	0.025
DDI	0.612 (0.154)	0.616 (0.153)	0.607 (0.153)	t = 1.8	.078	0.059
DEI	0.618 (0.291)	0.613 (0.284)	0.623 (0.300)	t = 1.1	.282	0.034

Note: BMI, body mass index; HRV, heart rate variability; PWV, pulse wave velocity; SDNN, SD of all normal R-R interval; pNN50, percentage of normal consecutive interbeat intervals differing by more than 7 ms; VLF, very low frequency; LF, low frequency; HF, high frequency; EEI, ejection elasticity index; DDI, dicrotic dilatation index; DEI, dicrotic elasticity index.

^a t/χ ² : t for the independent t-test; χ² for kappa test. Significant P values are bolded.

^bThree cases were lost in the psychotic disorder group (n = 1836), whereas 9 were missed in the nonpsychotic disorder group (n = 1600).

^cThe remaining 68 patients with psychotic disorder included 53 with schizoaffective psychosis, 8 with delusional disorder, 5 with acute and transient psychotic disorder, and 2 with paranoid psychosis.

Fig. 1.

HRV and PWV profiles of psychotic and nonpsychotic groups adjusted for age, gender, and BMI. (A) Marginal means from multivariate analysis of variance were standardized with means (SDs) from nonpsychotic group to convert to z scores. (B) Distribution of LF/HF in psychotic, nonpsychotic, and control groups. The mean LF/HF ratio in the control group is 1.5. Note: BMI, body mass index; HRV, heart rate variability; PWV, pulse wave velocity; SDNN, SD of all normal R-R interval; pNN50, percentage of normal consecutive interbeat intervals differing by more than 7 ms; VLF, very low frequency; LF, low frequency; HF, high frequency; EEI, ejection elasticity index; DDI, dicrotic dilatation index; DEI, dicrotic elasticity index.

Factor Analysis

Exploratory factor analysis was performed using principal components analysis and varimax rotation with Kaiser normalization. The number of factors retained in the analysis was based on factors that accounted for greater than 10% of the common variance as well as interpretability (table 2).

Table 2.

Standardized Factor Loadings Obtained From Exploratory Factor Analysis, Using Varimax Rotation of 8 Variables

Variables	HRV (Factor 1)	PWV (Factor 2)
SDNN	0.942	0.037
pNN50	0.786	0.009
VLF	0.890	−0.124
LF	0.937	−0.038
HF	0.922	−0.112
EEI	−0.037	0.924
DDI	−0.181	0.910
DEI	0.063	0.775

Note: HRV, heart rate variability; PWV, pulse wave velocity; SDNN, SD of all normal R-R interval; pNN50, percentage of normal consecutive interbeat intervals differing by more than 7 ms; VLF, very low frequency; LF, low frequency; HF, high frequency; EEI: ejection elasticity index; DDI, dicrotic dilatation index; DEI, dicrotic elasticity index. Significant factor loadings are bolded.

Regression

To evaluate the discrimination value of HRV and PWV variables for psychotic and nonpsychotic disorders, we performed a binary logistic regression analysis to explore which factor or variable best discriminated between the 2 groups. The regression models were analyzed separately for factor and individual variable levels (table 3).

Table 3.

Logistic Regression for Discrimination of Psychotic and Nonpsychotic Disorders

Discriminator Variables	Beta	SE	Odds Ratio	95% CI	Wald Statistic	P-Value
Factor level
Sex	−0.401	0.072	0.670	0.581–0.772	30.707	<.001
BMI	−0.037	0.009	0.964	0.947–0.981	17.380	<.001
HRV	0.106	0.035	1.111	1.038–1.190	9.125	.003
PWV	0.923	0.036	1.072	0.999–1.150	3.754	.053
Individual level
Sex	−0.440	0.074	0.644	0.557–0.745	35.376	<.001
BMI	−0.038	0.009	0.963	0.946–0.980	17.425	<.001
VLF	0.002	0.001	1.002	1.000–1.003	4.928	.026
LF	−0.002	0.001	0.998	0.996–1.000	4.451	.035
HF	0.002	0.001	1.002	1.000–1.003	3.657	.056
LF/HF	0.336	0.096	1.400	1.161–1.688	12.375	<.001
EEI	2.061	0.479	7.851	3.070–20.074	18.507	<.001
DDI	−1.413	0.458	0.243	0.099–0.597	9.531	.002

Note: CI, confidence interval; HRV, heart rate variability; PWV, pulse wave velocity; SDNN, SD of all normal R-R interval; pNN50, percentage of normal consecutive interbeat intervals differing by more than 7 ms; VLF, very low frequency; LF, low frequency; HF, high frequency; EEI, ejection elasticity index; DDI, dicrotic dilatation index; DEI, dicrotic elasticity index.

Correlation

Correlation between HRV and PWV characteristics was investigated using parametric and nonparametric correlations as well as the CCA. Traditional correlations were used to test for associations between HRV and PWV (supplementary material 3). Since the relationship between HRV and PWV is extremely close, they were affected by psychotic or nonpsychotic disorders. The present study applied, for the first time, a CCA method to define subgroups according to the patterns of the relationship (canonical variates) between HRV and PWV. CCA was used to identify the linear combinations of these 2 sets of variables to be maximally correlated with each other. CCA determines pairs of linear combinations, termed canonical variables, from 2 sets of variables (HRV and PWV; figure 2A), such that the correlation between canonical variables is maximized (figure 2B).

Fig. 2.

(A) Hierarchical cluster analysis of 2 canonical variates and scatterplot for the 2 subgroups and (B) a canonical variate pair. Support vector machine model for 2-subgroup solution, (C) train model (n = 2836) and (D) validation model (n = 600). Note: HRV, heart rate variability; PWV, pulse wave velocity.

Classification

To find subgroups in this 2-dimensional space of data points (each point represents an individual case), hierarchical cluster analysis was applied using MATLAB’s pdist, linkage, cluster, and cluster data functions. In the present study, an average-linkage algorithm was applied to group our samples. Euclidean distance was applied as a metric to evaluate sample similarity. This method was applied to identify subgroups among data points according to inter-point and inter-cluster distances. Euclidean distance between every pair of subjects in this 2-dimensional feature space was calculated, and Ward’s minimum variance method was applied to select a specific clustering from the dendrogram (figure 2A), iteratively linking pairs of subjects in closest proximity. Hierarchical clustering analysis was used to delineate clusters of subjects in a 2-dimensional space defined by these 2 canonical variates. This 2-cluster solution was optimal for summarizing relatively homogeneous subgroups that were maximally dissimilar from each other. We further assessed the utility of the extracted subgroups of samples. To evaluate whether subgroups were identical, the support vector machine (SVM) model was trained and validated using a linear kernel (figures 2C and 2D). Finally, the distribution of subgroups across psychotic and nonpsychotic groups was illustrated in figure 3. Supplementary material 4 (s-table 5) depicts the demographic, clinical Axis-I diagnosis, HRV, and PWV profiles of the 2 subgroups defined by the CCA.

Fig. 3.

Distribution of HRV and PWV features by subgroup and clinical diagnosis. Note: HRV, heart rate variability; PWV, pulse wave velocity.

Results

Sample Characteristics

Characteristics of 3448 patients are presented in table 1, including demographics, clinical diagnosis, HRV, and PWV. In the psychotic group, patients had a lower proportion of females, weighed more, had a higher BMI, and a greater proportion of smokers and inpatients than the nonpsychotic group. HRV variables were significantly different between the 2 groups with a small effect size. Patients were further stratified by the severity level of psychotic and nonpsychotic disorders, which assumed that inpatients suffered more severely than outpatients. HRV variables were significantly reduced in inpatient groups compared with outpatient groups in both psychotic and nonpsychotic disorders (supplementary material 1).

HRV and PWV Profile Comparison

Adjusted for age, sex, and BMI in the MANOVA, differences between psychotic and nonpsychotic groups were significant for SDNN, VLF, LF, and EEI. The distribution of LF/HF was highly identical between the psychotic and nonpsychotic groups but significantly differed for the control group. The control group had a significantly higher LF/HF score compared with the psychotic and nonpsychotic groups (1-way ANOVA test for 3 groups: F = 35.2, df = 2, P < .001; post hoc Bonferroni test control vs psychotic/nonpsychotic group: P < .001). Detailed comparisons of HRV and PWV characteristics between controls and patients with psychotic/nonpsychotic disorders can be found in supplementary material 2 (s-table 3).

Exploratory Factor Analysis

Exploratory factor analysis of the 8 selected variables resulted in 2 factors (table 2). The first factor, with an eigenvalue of 4.12, and high loading coefficients (>0.7) for SDNN, pNN50, VLF, LF, and HF, was labeled “HRV.” The second factor, with an eigenvalue of 2.2 had high loading coefficients for EEI, DDI, and DEI, was labeled “PWV.”

Logistic Regression Model

Exploratory binary logistic regression for patients with psychotic and nonpsychotic disorders used HRV and PWV factor scores or individual variables as discriminators. For the factor level, the overall model achieved a low classification accuracy rate of 56.5%. For the individual level, the accuracy rate only reached 53.4%. However, table 3 presents that most of these factors and variables were found to significantly discriminate between psychotic and nonpsychotic disorders in this model.

CCA Classification

Since HRV and PWV features are 2 sets of co-related variables (s-table 4), CCA-based subgroup clustering has a significant advantage over logistic regression modeling. CCA identified 2-dimensional representations of HRV and PWV features. To explore clusters in our clinical sample, hierarchical cluster analysis was applied to assign outpatients to nested subgroups with similar patterns of relationship between HRV and PWV features. Our analysis revealed 2 subgroups defined by distinct and relatively homogeneous patterns along 2 dimensions (figures 2A and 2B), comprising 19.0% (subgroup 1, n = 655) and 80.9% (subgroup 2, n = 2781) of the 3436 outpatients, respectively. The 2-subgroup SVM model was trained by 2836 cases and validated by 600 cases using linear kernel. Results of the full analysis (confusion matrix, accuracy, sensitivity, and specificity) are presented in figures 2C and 2D. Figures 2C and 2D also show the confusion matrix for the 2 subgroups, achieving an overall accuracy of 98.3% and 96.3%.

Distribution of Subgroups Between Psychotic and Nonpsychotic Disorders

Supplementary material 4 (s-table 5) depicts demographic, clinical, HRV, and PWV profiles of 2 subgroups defined by the CCA. Features of the 2 subgroups are summarized as follows: all HRV variables, especially in frequency domains, were significantly reduced with a large effect size in patients in subgroup 2 compared with those in subgroup 1. As illustrated in figure 3, there was considerable mixing across subgroups between psychotic and nonpsychotic disorders. Two subgroups suggest that more distinct HRV features correlate with PWV manifestations than were captured by clinical phenomenological diagnostic definitions.

Discussion

To our knowledge, this is the first large-scale study to investigate factorial levels of cardiovascular risk and CCA-driven subgroups of patients by applying both HRV and PWV indexes. Although previous studies have repeatedly shown that lower HRV, thought to reflect impaired autonomic nervous system function, is associated with psychotic and nonpsychotic disorders, it remains largely unknown whether this condition is universal or specific to a particular psychiatric disorder. Our results support that the imbalance model of sympathetic and parasympathetic nervous systems is prevalent in both psychotic and nonpsychotic disorders but is slightly pronounced in psychotic disorders (the effect size was small). In addition, the current findings provide the first evidence of transdiagnostic subgroups through the correlation between HRV and PWV at the level of the clinical population. We found such subgroups independent of specific clinical diagnostic classifications. Each subgroup included both psychotic and nonpsychotic disorders. Subgroup 2, which included the majority of the sample, had significantly decreased HRV function compared with that of subgroup 1.

Our findings are consistent with evidence that indicators of HRV are not associated with alterations in specific psychiatric disorders but rather with broad abnormalities in both psychotic and nonpsychotic disorders. In addition, the discrepancies observed were not limited to specific indicators of HRV but were found in almost all of the time and frequency domains. Two meta-analysis reports^13,33 in patients with depression clearly demonstrated that reductions in HRV are prevalent in depressed patients, an effect not fully attributable to antidepressant medication use. Studies and meta-analysis³⁴ on HRV in psychotic patients are also consistent with the current findings. All of this evidence supports the idea that the imbalance of sympathetic and parasympathetic nervous systems in psychiatric patients is robust. Altered or dysregulated parasympathetic/sympathetic modulation is a posited mechanism underlying the disruption of homeostasis in several major body systems, which is also associated with a risk for a wide range of diseases (such as cardiovascular disease, hypertension, and diabetes mellitus) and all-cause mortality.³⁵

The question remains if whether such impairments in HRV differ across psychiatric categories. We found a higher level of reduction in HRV for patients with psychotic disorders than for patients with nonpsychotic disorders. This may indicate a difference in cerebrovascular risk factors between the 2 types of categories. Patients with psychotic disorders have more severe symptoms and functional declines, experience more stressful events, and face more environmental challenges that might lead to impaired cardiac vagal tone compared with those with nonpsychotic disorders. Consistent with the neurovisceral integration theory,³⁶ low HRV is an index of impaired central-peripheral neural feedback mechanisms that leads to a lack of psychophysiological resources when an individual is overloading with stressful experiences and environmental challenges. In addition, it had been reported that obesity can reduce HRV,³⁷ and BMI index was much higher in patients with psychotic disorders than those with nonpsychotic disorders in the current sample, which might cause lower HRV in psychotic patients. Compared with controls, our results revealed that the LF/HF ratio was a significant parameter for discriminating patients with psychotic/nonpsychotic disorders. Particularly, the LF/HF ratio was much lower in patients than in controls. When comparing the LF and HF band between psychotic and nonpsychotic patients (table 1), the LF band was found to be significantly lower in patients with psychotic disorder, but the HF band was higher in patients with nonpsychotic disorder. Therefore, we assumed that the lower LF/HF ratio reported in the current study may be mainly due to a lower LF band for psychotic disorder while due to a higher HF band for nonpsychotic disorder. We found a reduced VLF in patients with psychotic disorders. The VLF component reflects possibly efferent sympathetic activity³⁸ and is modulated by the parasympathetic system,³⁹ although the physiologic mechanism underlying the VLF is disputed. Decreased VLF is associated with increased inflammatory parameters,⁴⁰ and increased inflammation is associated with the severity of psychosis.⁴¹

However, psychiatric disorders are considerably heterogeneous, and the more important question is whether there is a specific model or subgroup featured by cerebrovascular risk indicators independent of psychiatric classification. This result is linked to the specific units of an Research Domain Criteria (RDoC) domain of “arousal and regulatory systems: circadian rhythms” ⁴² and stands in contrast to other RDoC studies that have identified subgroups via biological domains.^43–45Subgroup 1 was characterized by less severe and lower impairments in HRV. In contrast, patients in subgroup 2 were characterized by severe and higher impairments in HRV, which potentially implies a higher risk of cardiovascular problems. These differential patterns of HRV and PWV across subgroups suggest an explanation for the marked diagnostic disagreement⁴⁶ in psychiatric disorders that is routinely observed across clinicians.⁴⁷ According to Cohen’s classification of effect size,³² the significant differences of HRV variables between psychotic and nonpsychotic groups had a small effect size (table 1), while the significant differences between subgroups 1 and 2 had a large effect size (s-table 5). Thus, our findings provide evidence of a subgroup of psychiatric patients (transdiagnostic rather than diagnostic specific) reflecting severely impaired parasympathetic/sympathetic activities and lend support for efforts to develop parasympathetic/sympathetic modulation-targeted treatment that extends to clinical diagnosis-based interventions.

Furthermore, in the current study, we found no significant difference in PWV functions between patients with psychotic and nonpsychotic disorders. This negative result is in agreement with studies showing that PWV indicators were not sensitive to discrimination in psychiatric disorders, such as schizophrenia⁴⁸ or depression,⁴⁹ but contradict other studies reporting positive associations between arterial stiffness and depression.^50,51Although nonsignificant, this finding provides some rationales for future examination of arterial stiffness in the psychiatric population. First, it could be assumed that unlike HRV indicators related to the severity of mental disorders, PWV indicators may be related to cardiovascular morbidity, which may be considered a more advanced form of vascular damage.⁵² Second, our sample was free of medicine, but previous studies showed that the use of antipsychotics was associated with increased arterial stiffness.⁵³ Third, PWV indicators may be more sensitive to psychiatric symptoms in the elderly patients than in adult or adolescent patients, because of the ceiling effect. In other words, the younger population had a greater uniformity of arterial stiffness and a lower risk for cardiovascular diseases.⁴⁹

Our findings have several important clinical implications. First, cardiovascular risk indicators were increased widely and comprehensively in all psychiatric patients, regardless of the presence of psychosis. In contrast to the high prevalence of cardiovascular risk factors reported in previous studies, lower levels of monitoring and less prevention paradigms targeting cardiovascular problems have been implemented in psychiatric practice. In addition, psychiatric patients, especially those with severe psychotic disorders, are significantly less likely than the general population to receive preventive therapies of proven benefit, including aspirin and β-blocker therapy for cardiovascular diseases.⁵⁴ Second, the prevalence of metabolic risk factors for cardiovascular diseases, such as diabetes, obesity, and metabolic syndromes, in patients with severe mental disorders is approximately 1.5 to 2 times higher than in the general population.^55,56 All of these unfavorable effects on various metabolic risk factors were partly side effects of medications used for treatment. However, psychiatric diagnoses are treated as a single label for heterogeneous cardiovascular risks, which is problematic as it is clear that treatment should vary by subgroup. We determined different subgroups of patients associated with different patterns of cardiovascular risks, which have also been broadly distributed across psychiatric diagnoses. However, we are still in the initial stages of this approach; therefore, it would be premature to suggest definitive claims until more preventive strategies for cardiovascular problems targeting subgroups of patients have been conducted.

The strengths of this study include its large sample size and its novelty in developing CCA-driven subgroups for a psychiatric population and using canonical variates as indicators. However, given the cross-sectional design and magnitude of heterogeneity across the clinical population, our results should be interpreted with caution. For example, although the current sample was drug-free when recruited, our study results might be confounded by different prior usages of psychiatric medications in the 2 groups. Patients in the psychotic group were more likely to be treated with antipsychotic drugs that could adversely affect adiposity and glucose and lipid metabolism.³ Although the numbers in the 2 different groups were large and sufficient for statistical analysis, the higher proportion of males, smokers, and inpatients and the lower proportion of drug-naïve patients in the psychotic group are potential confounders, which need to be taken into account. Although it is reasonable to speculate that inpatients are more serious compared with outpatients, this inference is not supported by clinical data. Given this study’s cross-sectional design, it remains unknown whether the 2 subgroups of patients could be used to predict the onset of and mortality from cardiovascular disease. Considering the age influences on HRV parameters, the age difference between the 2 subgroups could be responsible for the reported effects concerning differences in autonomic and vascular parameters. Future longitudinal studies on the preventive manipulation of new approaches and models will be important to clarify the precise role of these 2 subgroups. Finally, although the purpose of the current study was not to compare the cardiovascular risks between patients and controls, the sample size of the control group was significantly smaller compared with that of the patient groups; therefore, one should be cautious when interpreting the results of the comparison between cases and controls.

Conclusions

Our results suggest that HRV function is reduced among psychiatric patients. Compared with nonpsychotic patients, psychotic patients had a higher extent of HRV impairment. HRV is an important physiological marker in psychiatric disorders and may provide valuable clinical information about specific phenotypes as well as cast light on increased cardiovascular risks. These avenues of research could help inform diagnosis and treatment protocols in psychiatric practice and could aid in the promotion of cardiovascular health by facilitating collaborative efforts of cardiologists with psychiatrists.

Funding

This study was supported by the Ministry of Science and Technology of China, National Key R&D Program of China (2016YFC1306800), National Natural Science Foundation of China (81671329 and 81671332), Science and Technology Commission of Shanghai Municipality (19441907800, 17411953100, 19ZR1445200, and 19411950800), Construction, Shanghai 3-Year Public Health Action Plan (GWV-10.1-XK18), Shanghai Clinical Research Center for Mental Health (19MC1911100), and the Clinical Research Center at Shanghai Mental Health Center (CRC2018ZD01, CRC2018ZD04, CRC2018YB01, and CRC2019ZD02).