A simple version of resting‐state heart rate variability interpretation for patients with affective disorders: A four‐pattern perspective

Abstract

Heart rate variability (HRV) is often considered as a biomarker reflecting well‐being, but the clinical meaning of short‐term resting‐state HRV is not sufficiently defined. We assume that combining several common HRV indices as “HRV patterns” and using the patterns for screening purposes are meaningful approaches. Resting‐state 5‐min HRV data of 424 subjects were analyzed. Four of the most commonly used HRV indices were considered: standard deviation of normal–to–normal RR intervals, low‐frequency power, high‐frequency power and the ratio of low‐frequency to high‐frequency power. According to these indices, four HRV patterns were defined: normal pattern, low HRV pattern, relatively high sympathetic pattern, and relatively high vagal pattern. The associations between the demographics, lifestyles, personality traits, psychological states, and HRV patterns were explored: the low HRV pattern was positively associated with age, body mass index, and depression; the relatively high sympathetic pattern was positively associated with age and negatively associated with exercise habit; and the relatively high vagal pattern was negatively associated with having a steady job and novelty seeking. The pattern perspective may provide a convenient and evidence‐based way to interpret resting‐state HRV for patients with affective disorders.

INTRODUCTION

Heart rate variability (HRV), a noninvasive technique for measuring autonomic activity, is often considered as a biomarker reflecting well-being.¹ Both biological and psychological features were found to be associated with HRV and the clinical value of HRV in the psychiatric field has been discussed frequently.² There are several methods for measuring and calculating HRV. Two different time spans have been used for standardized HRV measurement: 5 min and 24 h.³ Separating the situations, resting-state HRV and HRV reactivity under specific physiological/psychological stimuli may have distinct meanings.^4–6 There are several approaches for analyzing HRV, including a method incorporating respiratory signals (respiratory sinus arrhythmia [RSA]) and methods using only heart rate signals. The latter include time domain, frequency domain, and nonlinear methods.¹ In general, HRV is considered to be robust for reflecting parasympathetic activity, with RSA, root mean square of successive differences (RMSSD), and high-frequency power (HF) usually viewed as parasympathetic-specific indices.^1,6,7 The indices representing overall variability, such as the standard deviation of normal–to–normal RR intervals (SDNN) and total power, also tend to reflect vagal activity.^5,7 On the other hand, low-frequency power (LF) are modulated by both sympathetic and parasympathetic systems; the ratio of low-frequency to high-frequency power (LF/HF) is suggested to represent the sympatho-vagal balance by some scholars, although this is a little controversial.^1,8 Based on previous studies, the clinical meaning of each HRV index is clear; however, whether incorporating the values of several indices is useful for clinical purposes still awaits clarification.

The clinical application of HRV in the literature includes aspects of evaluation and treatment.⁹ For treatment, because HRV is associated with an individual's tension/relaxation, the short-term measurement of HRV can be used for reflecting the real-time psychological status; this is the basis of HRV biofeedback therapy.⁹ Controversy exists with regard to evaluation because HRV is very sensitive and can be affected by many factors, including demographics (e.g., age, body mass index [BMI]), physical diseases (e.g., diabetes mellitus), and medications (e.g., tricyclic antidepressants).^4,7,10 Therefore, although HRV can be found associated with some features in group-based studies, it is unlikely to be used directly at an individual level. Moreover, even when interference of the above factors is excluded, low HRV is associated with various types of psychiatric disorders, including depression, anxiety, schizophrenia, alcohol use disorder, somatic symptom disorders and functional somatic syndromes, autism, and dementia.^{5–7,11–13} Because the resting-state HRV of various psychiatric disorders has similar features, it is suggested that resting-state HRV mainly reflects gross psychological discomfort or mental flexibility and therefore has limited meaning for differential diagnosis.^7,14 HRV reactivity designs related to specific psychopathologies of different psychiatric disorders may have more application for diagnostic purposes; however, further evidence is warranted.^5,15 In addition, the difference in resting-state HRV for psychiatric patients and healthy individuals often revealed a low to moderate effect size, which causes HRV to have unsatisfactory performance for diagnostic purposes.^6,7

In our opinion, even though the short-term resting-state HRV reveals limited effect for psychiatric diagnosis, it still has clinical value. The advantage is its convenience, making it more helpful for screening than for diagnosing. Besides, in Taiwan, patients with affective disorders (such as somatic symptoms, anxiety, and depression) often consider themselves to have “autonomic dysregulation”; they show better adherence when autonomic measurements are applied and the results are adequately explained.¹⁶ However, when clinicians view the different HRV indices separately there are several common limitations. Although the biological meanings of distinct HRV indices are somewhat different, the indices reflecting mainly parasympathetic activity often show similar associations with psychiatric conditions, but the meanings to interpret them as a whole are lacking.^5,6 Also, if the levels of distinct HRV indices are different, it is difficult to explain them intuitively. A possible approach to overcoming the above problems is combining several common HRV indices as “patterns”; using this approach, the distinct meanings of these indices can be emphasized. The “pattern” approach can be used to classify all rational HRV data for clinical purposes; this is a common advantage of categorical data and has been used in other fields of psychiatry, such as personality.¹⁷ If these HRV patterns are associated with different psychological features, they can be viewed as probes for further detailed psychological evaluation. HRV has been found to be associated with both age and gender; these factors can be considered for interpreting HRV patterns rigorously. If we expect to provide a relatively simple version of HRV interpretation, then adopting the HRV data of general adults should be closer for our purpose; these types of data are available in the literature.^18,19

This study was designed according to the above background information. Based on the four most commonly used HRV indices (SDNN, LF, HF, and LF/HF) and the normative Asian HRV data in the literature, we defined four HRV patterns and explored their associations with psychological features.¹⁸ The psychological data in this analysis included recent psychological states, long-term personality traits, and lifestyle. According to the results of this study, we also proposed a simple protocol of HRV measurement and interpretation for clinical purposes.

METHODS Participants and procedure

The data of this analysis were gathered from the National Taiwan University Hospital (NTUH) and the NTUH Yunlin Branch between 2015 and 2020. The Institutional Review Board of the NTUH approved the design of this study (Study No. 201808047RINB), which is a secondary analysis of the data in a series of studies. The participants of this analysis are overlapped with those of several published HRV articles but the topics and analyzing approaches are totally different.^15,20 The inclusion criteria for entering this analysis were: patients with affective disorders (depression, anxiety, and somatic distress as their chief complaints); and healthy individuals without functional impairment. Exclusion criteria were: individuals without complete demographic, psychological, or HRV data; age younger than 20 years or older than 70 years; having psychotic symptoms; having overt cognitive impairment; having physical diseases that may interfere with HRV (e.g., diabetes mellitus, arrhythmia); taking medications that may affect HRV (e.g., tricyclic antidepressants, and antiarrhythmic agents); and persistently taking substances that may interfere with HRV (e.g., alcohol and tobacco).^4,7,20 The data of this series of studies were all gathered once, in the daytime. If a participant joined more than one study, only the data of the first participation were included in order to prevent repeated measurement. All participants gave their informed consent before entering the series of studies.

The data of 226 patients with affective disorders and 198 healthy individuals were compatible with the above criteria and entered this analysis; the origins of the data are described in Figure S1 and the demographic, psychological, and HRV data are shown in Table 1. The age and gender of the two populations were not significantly different. Patients with affective disorders had significantly higher levels of depression, anxiety, somatic distress and hypochondriacal ideation than the healthy individuals. Among the HRV indices, LF and HF were significantly lower in the patients with affective disorders than in the healthy individuals.

TABLE 1 Demographics, psychological features, and HRV of patients with affective disorders and healthy individuals

	Healthy individuals (n = 198)	Patients with affective disorders (n = 226)	Statistics
	Mean/n (SD/%)	Mean/n (SD/%)	t/χ2	p
Demographics
Gender (male)	59 (29.8%)	83 (36.7%)	2.274	0.132
Age (years)	43.04 (10.31)	44.29 (11.76)	−1.172	0.242
BMI (kg/m2)	23.86 (4.00)	22.87 (3.64)	2.642	0.009**
Occupational status (employed)	165 (83.3%)	130 (57.5%)	33.216	<0.001***
Exercise habit (hours/day)	0.28 (0.53)	0.21 (0.41)	1.482	0.139
Psychological states
BDI-II	5.78 (6.88)	20.92 (12.16)	−16.026	<0.001***
BAI	3.81 (5.03)	20.51 (11.85)	−19.295	<0.001***
PHQ-15	3.99 (3.67)	12.35 (5.79)	−17.975	<0.001***
HAQ	10.29 (7.53)	26.70 (13.62)	−15.575	<0.001***
Personality traits
Novelty seeking	14.67 (4.37)	13.99 (4.04)	1.653	0.099
Harm avoidance	13.96 (6.87)	22.66 (6.40)	−13.460	<0.001***
Reward dependence	13.18 (3.22)	12.73 (3.23)	1.410	0.159
Persistence	5.09 (1.63)	5.42 (1.80)	−2.005	0.046*
HRV
ln SDNN [ln(ms)]	3.59 (0.47)	3.49 (0.53)	1.937	0.053
ln LF [ln(ms2)]	5.82 (0.99)	5.59 (1.21)	2.124	0.034*
ln HF [ln(ms2)]	5.26 (1.15)	5.02 (1.35)	1.986	0.048*
ln LF/HF [ln(ratio)]	0.56 (0.84)	0.59 (0.96)	−0.391	0.696

Note: *p < 0.05; **p < 0.01; ***p < 0.001.

Abbreviations: BAI, beck anxiety inventory; BDI-II, beck depression inventory-II; BMI, body mass index; HAQ, health anxiety questionnaire; HF, high-frequency power; HRV, heart rate variability; LF, low-frequency power; LF/HF, ratio of low-frequency to high-frequency power; PHQ-15, patient health questionnaire-15; SDNN, standard deviation of normal–to–normal RR intervals.

Psychological measurements

The scores of five questionnaires were analyzed in this study. Regarding recent psychological state, the Beck Depression Inventory-II (BDI-II) was used for measuring depression, the Beck Anxiety Inventory (BAI) was used for measuring anxiety, the Patient Health Questionnaire-15 (PHQ-15) was used for measuring somatic distress, the Health Anxiety Questionnaire (HAQ) was used for measuring hypochondriasis. The Tridimensional Personality Questionnaire (TPQ) was adopted for personality traits.

Kroenke et al.²¹ developed the PHQ-15 to measure the severity of somatic distress. Three-point Likert scales (0, “not bothered at all”; 1, “bothered a little”; 2, “bothered a lot”) were used to indicate the severity of 15 types of somatic symptoms. Liao et al.²² examined the psychometric properties of the Chinese PHQ-15 and a three-factor structure was suggested, with internal consistency at a high level (Cronbach's alpha = 0.861). Lucock and Morley²³ designed the HAQ for estimating health anxiety or hypochondriacal ideation. The HAQ consists of 21 questions, for which four-point Likert scales were adopted: ranging from 0 (“not at all or rarely”) to 3 (“most of the time”). High internal consistency (Cronbach's alpha = 0.943) was found for the Chinese version of the HAQ,²⁴ in which a three-factor structure was suggested. The BDI-II and BAI were both developed by Beck et al.,^25,26 their purpose being to evaluate symptoms of depression and anxiety, respectively. Both the BDI-II and the BAI have 21 questions, scored using four-point Likert scales of 0–3; higher scores indicate higher severity of symptoms. The psychometric studies in Taiwan found the BDI-II and BAI to have satisfactory internal consistency (Cronbach's alpha = 0.94 for the BDI-II and 0.95 for the BAI).^27,28

The TPQ was developed by Cloninger²⁹ and measures long-term personality traits using 100 dichotomous questions. In its original version there are three major dimensions: novelty seeking (NS), which is behavioral activation; harm avoidance (HA), which is behavioral inhibition; and reward dependence (RD), which is the tendency to depend on social reward.²⁹ Originally, the three major dimensions each had four subdimensions but in the following studies “persistence,” a previous subdimension of RD, was considered independent. Therefore, the scores of persistence and RD are usually calculated separately.³⁰ The Chinese version of the TPQ in Taiwan was translated and examined by Chen et al.³⁰ and the mean values of NS/HA/RD were 13.2/13.8/13.5, respectively.

Heart rate variability

We used the standard procedure for 5-min HRV measurements.³ The participants were asked to sit and to breathe freely, not falling asleep during the measurements. The measurement was in the daytime; the subjects were asked to rest for several minutes before the HRV recording. An HRV analyzer (SS1C, Enjoy Research Inc., Taiwan) was adopted for signal acquisition, storage and processing.³¹ An 8-bit analog-to-digital converter with a sampling rate of 512 Hz was used for recording lead I electrocardiographic signals. The digitized electrocardiographic data were analyzed online and simultaneously stored on a hard disk for offline verification. Each QRS complex was examined by computer algorithm according to the likelihood of a standard QRS template; any noise, artifact, or ventricular premature complex was rejected in this step. To ensure temporal continuity, normal and stationary R-R interval values were re-sampled and interpolated at a rate of 7.11 Hz. A total of 2048 data points in 288 s were generated via the interpolation. The time domain index SDNN, representing the overall variability, was calculated first¹ and then we performed fast Fourier transformation to produce frequency domain data. After deleting the baseline shift, a Hamming window was used for attenuating the leakage effect. An algorithm was then adopted to estimate the power spectrum density. We captured the signals in different frequency bands. HF corresponded to the 0.15–0.4 Hz frequency band and is considered to be a specific index for parasympathetic activity.^4,19 The 0.04–0.15 Hz frequency band belonged to LF, which is modulated by both sympathetic and parasympathetic systems.^4,19 The LF/HF ratio, an index for sympatho-vagal balance (with a little controversy), was then calculated.^1,8 The normality of the HRV data Indices was tested using the Shapiro–Wilk test; because SDNN, LF, HF, and LF/HF were not distributed normally, they were logarithmically transformed to correct for skewness. RMSSD was not included in this analysis because its biological meaning is similar to that of HF.

Definition of HRV patterns

We chose four HRV indices to construct the patterns: SDNN, LF, HF, and LF/HF. The main reason is that the meanings of the four indices are overlapped but not totally equal, so their similarities and differences may provide additional information. Furthermore, these four HRV indices are the most common in the psychiatric field and interpreting them may provide direct instructions to clinicians. SDNN, LF, and HF all have parasympathetic components and their general distribution may be used to judge the vagal activity.¹ HF is vagal-specific; SDNN tends to reflect vagal activity; and LF represents both vagal and sympathetic activities but is still vagal predominant in some conditions (such as slow breathing; in this condition, LF is even more suitable for reflecting vagal activity than HF).¹ Although HF is parasympathetic-specific, the major part of vagal activity is not necessarily located in the HF band; therefore, considering LF and SDNN concurrently is sometimes meaningful.^1,11 For example, if the levels of LF and HF are different, the value of SDNN may be additionally adopted for adjudicating the parasympathetic activity. The meaning of SDNN is equal to which of total power. If SDNN, LF, and HF are generally located in the moderate to high range, LF/HF may be helpful for judging the relative strength of sympathetic and vagal systems.¹

According to the above concepts, we defined the four patterns as: normal pattern (two or three of SDNN, LF, and HF at a moderate or high level, with a moderate level of LF/HF); low HRV pattern (none or one of SDNN, LF, and HF at moderate or high level, with any level of LF/HF; this pattern represented low parasympathetic activity); relatively high sympathetic pattern (two or three of SDNN, LF, and HF at moderate or high level, with a high level of LF/HF); and relatively high vagal pattern (two or three of SDNN, LF, and HF at moderate or high level, with a low level of LF/HF). The most important meaning of considering HF, LF, and SDNN together is to detect a “low HRV pattern”; if not belonging to a low HRV pattern, use of LF/HF to judge the sympathetic/vagal predominance may have some meaning. The descriptions “high sympathetic” and “high vagal” were relative and focused on the sympatho-vagal balance; they did not reflect absolute neural activities.

Regarding the cutoff values for judging the HRV levels, among the normative HRV values in the literature the data described by Kim and Woo¹⁸ were comprehensive; they had a relatively high sample size (n = 3408) and the features of the participants (East Asian, age 18–65 years) were similar to those in our study. We therefore viewed the first and third quartiles (25% and 75% percentiles) of SDNN, LF, HF, and LF/HF described in that article as the lower/upper limits of the moderate level. The distributions of our data are very similar to those of the above-mentioned article and a detailed description is presented in Table 2.

TABLE 2 Definition of the four HRV patterns and the adopted HRV levels

HRV patterns	Definition
Normal pattern	Among SDNN, LF, and HF, at least two indices are in moderate to high level
	Moderate level of LF/HF
Low HRV pattern	Among SDNN, LF, and HF, none or only one index is in moderate to high level
	Any level of LF/HF
Relatively high sympathetic pattern	Among SDNN, LF, and HF, at least two indices are in moderate to high level
	High level of LF/HF
Relatively high vagal pattern	Among SDNN, LF, and HF, at least two indices are in moderate to high level
	Low level of LF/HF
HRV levels	Range of the moderate levela
SDNN (ln SDNN)	28.2–46.3 ms (3.3–3.8 ln[ms])
LF (ln LF)	132.6–451.4 ms2 (4.9–6.1 ln[ms2])
HF (ln HF)	75.1–287.9 ms2 (4.3–5.7 ln[ms2])
LF/HF (ln LF/HF)	0.9–3.1 ratio (−0.1–1.1 ln[ratio])

Note: In our sample, the 25th–75th quantiles of ln SDNN, ln LF, ln HF, and ln LF/HF were 3.2–3.8, 5.0–6.5, 4.3–5.9, and 0.0–1.2, respectively, which were quite similar to the above ranges.

Abbreviations: HF, high-frequency power; HRV, heart rate variability; LF, low-frequency power; LF/HF, ratio of low-frequency to high-frequency power; SDNN, standard deviation of normal–to–normal RR intervals.

^aAccording to Kim and Woo,¹⁸ these are HRV values of general adults in an Asian sample, not separated according to different age/gender. Values falling outside of these ranges were coded as “high level” or “low level.”

Statistical analysis

After generating the four HRV patterns, we made an intergroup comparison of the demographics, recent psychological states, personality traits, and HRV. Analysis of variance (ANOVA) and the Scheffe method (for posthoc analysis) were adopted for continuous variables and the χ² test for categorical variables. We used multinomial logistic regression analysis to clarify the associations between the HRV patterns and the demographics and psychological features. The normal pattern was set as the reference. Two regression models were considered. In the first model, we examined the main effects of demographics, recent psychological states, and personality traits on HRV patterns using the “enter” method. In the second model, the possible interaction between age/gender and psychological features was explored using the “backward” method; the interaction items with significant effects were entered into this model. Possible multicollinearity was examined during the analysis. The above analyses were all two-sided and the alpha value was set as 0.05. These analyses were completed using SPSS 25 (IBM, USA).

RESULTS

According to the definitions, the numbers of participants belonging to the normal pattern/low HRV pattern/relatively high sympathetic pattern/relatively high vagal patterns were 176/115/69/64, respectively. The low HRV pattern had the highest age and the normal pattern had the lowest age among the four patterns; the intergroup difference was significant. There was also a significant intergroup difference for BMI and occupational status: BMI was significantly higher in the low HRV pattern than in the normal pattern, but the trend of occupational status was the reverse of this. Exercise habit had a significant intergroup difference in ANOVA but the post-hoc analysis did not reveal a significant difference between any two patterns.

Regarding psychological features, the scores of the BDI-II, BAI, PHQ-15, and HAQ for the four groups revealed no significant intergroup difference. The distribution of patients with affective disorders and healthy individuals in the four patterns also was not statistically different. Regarding personality traits, NS in the relatively high sympathetic pattern and normal pattern was significantly higher than in the low HRV pattern and relatively high vagal pattern; HA, RD, and persistence did not show any significant intergroup difference. HRV was the foundation for separating the four patterns and SDNN, LF, HF, and LF/HF all revealed a significant intergroup difference. Detailed information can be found in Table S1.

Table 3 gives the results of multinomial logistic regression regarding the main effects of the independent variables. The low HRV pattern was significantly associated with age, BMI and depression (positive associations). The relatively high sympathetic pattern had a significant association with age (positive) and exercise habit (negative). The relatively high vagal pattern revealed significant negative associations with occupational status and novelty seeking. In Table 3, two interaction items were entered into the model: age × RD and age × persistence. These two items were only significantly associated with the relatively high vagal pattern: the former was positive and the latter was negative. But in the model of Table 3, the main effects of age did not reach significant levels. Considering that age was positively associated with the relatively high vagal pattern in Table 3, the two interaction items can be understood as follows: during the aging process, items with low persistence or high RD were more likely to reveal the relatively high vagal pattern. The factors significantly associated with the HRV patterns are also illustrated in Figure 1.

TABLE 3 Multinomial logistic regression models of the HRV patterns^a

	Low HRV pattern		Relatively high sympathetic pattern		Relatively high vagal pattern
(A) The model considering only the main effect
Cox and Snell R2	0.280
	Odds ratio (95% CI)	p	Odds ratio (95% CI)	p	Odds ratio (95% CI)	p
Genderb	1.304 (0.700–2.429)	0.403	1.690 (0.881–3.241)	0.115	0.866 (0.434–1.724)	0.681
Age	1.112 (1.076–1.149)	<0.001***	1.040 (1.008–1.073)	0.014*	1.030 (0.998–1.064)	0.069
BMI	1.145 (1.062–1.234)	<0.001***	1.040 (0.960–1.127)	0.334	1.048 (0.962–1.141)	0.281
Occupational statusb	0.671 (0.340–1.322)	0.249	0.646 (0.321–1.299)	0.220	0.494 (0.244–0.998)	0.049*
Exercise habit	0.943 (0.499–1.780)	0.856	0.397 (0.162–0.969)	0.043*	1.388 (0.727–2.650)	0.321
BDI-II	1.058 (1.019–1.099)	0.004**	0.995 (0.956–1.035)	0.796	1.012 (0.970–1.056)	0.577
BAI	1.004 (0.957–1.054)	0.861	1.002 (0.953–1.053)	0.952	1.017 (0.965–1.071)	0.538
PHQ-15	0.964 (0.890–1.044)	0.371	1.027 (0.945–1.115)	0.532	0.961 (0.877–1.052)	0.387
HAQ	0.984 (0.953–1.016)	0.326	0.984 (0.952–1.016)	0.315	0.987 (0.953–1.022)	0.449
Novelty seeking	0.951 (0.884–1.024)	0.183	1.046 (0.975–1.121)	0.208	0.917 (0.846–0.994)	0.035*
Harm avoidance	0.979 (0.932–1.029)	0.403	1.011 (0.961–1.064)	0.668	1.005 (0.953–1.060)	0.859
Reward dependence	0.945 (0.864–1.033)	0.215	0.987 (0.898–1.084)	0.783	0.974 (0.884–1.073)	0.591
Persistence	1.076 (0.906–1.277)	0.403	1.007 (0.848–1.194)	0.940	0.971 (0.809–1.166)	0.754
(B) The model with interaction items having significant effects
Cox and Snell R2	0.313
	Odds ratio (95% CI)		p	Odds ratio (95% CI)	p	Odds ratio (95% CI)	p
Genderb	1.186 (0.629–2.234)		0.598	1.707 (0.886–3.287)	0.110	0.742 (0.367–1.502)	0.407
Age	1.132 (0.969–1.322)		0.118	1.040 (0.889–1.217)	0.623	0.981 (0.842–1.142)	0.801
BMI	1.157 (1.072–1.249)		<0.001***	1.042 (0.960–1.130)	0.323	1.066 (0.978–1.161)	0.145
Occupational statusb	0.680 (0.343–1.351)		0.271	0.638 (0.317–1.283)	0.207	0.541 (0.261–1.121)	0.098
Exercise habit	0.978 (0.516–1.853)		0.946	0.401 (0.165–0.977)	0.044*	1.506 (0.778–2.914)	0.225
BDI-II	1.053 (1.013–1.093)		0.008**	0.995 (0.956–1.035)	0.805	1.003 (0.959–1.048)	0.898
BAI	1.007 (0.959–1.057)		0.784	1.001 (0.952–1.052)	0.973	1.025 (0.972–1.082)	0.356
PHQ-15	0.965 (0.890–1.046)		0.384	1.028 (0.947–1.117)	0.507	0.952 (0.868–1.044)	0.296
HAQ	0.984 (0.953–1.017)		0.336	0.983 (0.952–1.016)	0.313	0.990 (0.955–1.026)	0.565
Novelty seeking	0.938 (0.870–1.010)		0.090	1.043 (0.973–1.119)	0.232	0.894 (0.822–0.973)	0.009**
Harm avoidance	0.978 (0.930–1.028)		0.390	1.010 (0.961–1.062)	0.686	1.005 (0.951–1.063)	0.850
Reward dependence	0.744 (0.467–1.185)		0.213	0.967 (0.645–1.449)	0.869	0.513 (0.325–0.809)	0.004**
Persistence	2.184 (0.928–5.413)		0.074	1.041 (0.503–2.151)	0.914	3.014 (1.367–6.643)	0.006**
Age × reward dependence	1.006 (0.996–1.016)		0.265	1.001 (0.991–1.010)	0.873	1.015 (1.005–1.026)	0.005**
Age × persistence	0.984 (0.966–1.002)		0.078	0.999 (0.981–1.016)	0.872	0.973 (0.956–0.991)	0.004**

Note: *p < 0.05; **p < 0.01; ***p < 0.001.

Abbreviations: BAI, beck anxiety inventory; BDI-II, beck depression inventory-II; BMI, body mass index; HAQ, health anxiety questionnaire; HRV, heart rate variability; PHQ-15, patient health questionnaire-15.

^aNormal pattern was set as reference.

^bUsing dummy variables. Gender: women = 0, men = 1; occupational status: unemployed = 0, employed = 1.

FIGURE 1. Illustration of the four HRV patterns and their associated factors. HF, high-frequency power; HRV, heart rate variability; LF, low-frequency power

DISCUSSION

The features associated with the HRV patterns were revealed in this analysis. The low HRV pattern was significantly associated with aging, increasing BMI, and depression; the findings indicated that physiological and psychiatric conditions were related to this pattern. The relatively high sympathetic pattern is also associated with aging; the other feature of this pattern was lack of exercise. The relatively high vagal pattern was related to being jobless and to several personality features, including low novelty seeking, high RD with aging, and low persistence with aging. How to understand these results and apply them for clinical purposes is worthy of discussion.

The effect of aging is a reduction of SDNN, LF, and HF but it is also associated with polarization of LF/HF (mainly elevation). A decrease of SDNN, LF, and HF with aging has been reported in many studies; in the psychiatric field, the aging effect has an even higher association with HRV than psychiatric phenomena.³² On the other hand, the association between LF/HF and aging is not simply negative. Kuo et al.'s¹⁹ study revealed that in the age interval of 40–50 years LF/HF was positively associated with aging; however, the association changed to negative at above 50 years of age. This phenomenon is clearer in men than in women. The sample in our study included a population aged 40–50 years, which may explain the positive association between aging and the relatively high sympathetic pattern. On the other hand, the negative association between BMI and HRV has been widely reported; BMI was hence considered to be an important potential confounder in HRV studies.³³ In summary, the influence of aging and BMI on HRV patterns is likely to be via biological mechanisms.

Among the several recent psychological features, depression had the highest association with the low HRV pattern. This finding is similar to the result of our previous study (the sample is not overlapped with the present one): in that study, relative to the BAI, PHQ-15, and HAQ scores, the BDI-II scores revealed a higher negative correlation with HF.³² Low SDNN, LF, and HF levels in patients with depression have been found in several meta-analyses.^7,12 An issue worthy of discussion is that anxiety and somatic symptoms were found to be associated with low vagal activity in many studies but only the effect of depression reached a significant level in our analysis.⁶ One explanation is that the depression/anxiety/somatic symptoms were highly comorbid in our sample (Table S2). In this condition, if depression had the highest effect on lowering HRV, it may hide the effects of the other two symptoms. If depression and age were removed from the independent variables, anxiety would show a significant effect on the HRV pattern; its effect was similar to that of depression in the original model (Table S3). We hence supposed that depression, anxiety, and somatic symptoms had similar effects on lowering HRV in our sample; depression had the largest influence among them. The same reason can also be used to explain the nonsignificant association between the low HRV pattern and HA, which was reported to be negatively associated with HRV in several studies.³⁴ Among the dimensions of the TPQ, HA had the highest correlation with depression/anxiety/somatic symptoms, therefore its effect was hidden.

Relatively high vagal activity is often understood to be related to resting; this was supported in our analysis.⁴ The relatively high vagal pattern is negatively associated with novelty seeking, which is a personality dimension about behavioral activation. Because personality is a time-steady feature, the association is likely to have a long-term tendency. Joblessness can also be directly connected with resting. As with the effect of age × persistence, it can be understood from the behavioral aspect. Persistence is a personality dimension about personality maintenance. High persistence often presents as devotion to responsibility or tasks, whereas high novelty seeking is related to pleasure seeking.³⁵ This finding can be interpreted as follows: an individual with a low tendency toward behavioral maintenance shows relatively high vagal activity during the aging process and is more likely to be jobless. The result on RD can be explained using polyvagal theory.³⁶ This theory indicates that there are two vagal systems: the dorsal motor nucleus is associated with vagal shutdown or “freeze” and the nucleus ambiguus is related to social interaction.³⁶ High activity of the nucleus ambiguus may present with high RD.³⁷ Our result further implies that the association between RD and relatively high vagal activity may increase during aging.

Relatively high sympathetic tone is sometimes understood as high stress or anxiety, but this was not supported by our results.³⁷ In our analysis, anxiety is associated with the low HRV pattern, but the effect was lower than with depression. In other studies, anxiety/stress also had a higher association with vagal indices than with LF/HF.³⁸ An explanation is that the sympathetic component in HRV is not clear; skin conductance level, a sympathetic-specific index, has been found to be positively associated with anxiety.³⁹ However, even adopting the viewpoint that “LF/HF is not suitable to be viewed as a marker of sympatho-vagal balance,” our result still indicated that using this index in the psychiatric field is valuable. In our sample, the feature of the relatively high sympathetic pattern is a low exercise level. A possible interpretation is that regular exercise elevates vagal activity. Previous studies have disclosed that exercise can increase the values of vagal-based HRV indices such as HF and LF; LF/HF was reduced by exercise,⁴⁰ which is identical to our finding.

When the HRV data of the patients with affective disorders and healthy individuals were compared, LF and HF were significantly lower in the former, but the effect size was small. Therefore, from the scope of diagnosing depression/anxiety/somatic symptom disorders, the value of HRV seems much lower than that of the questionnaires (Figure S2). Furthermore, the distribution of individuals with affective disorders was not significantly different in the four HRV patterns. This indicates that (1) HRV cannot provide accurate estimation to the status of affective disorders; (2) if the clinicians want to use HRV for interpretation, the proposed four patterns can be considered; but even in this condition, they should be used for the screening purpose. A suitable application of HRV patterns is viewing them as a probe for psychological phenomena. For example, if a low HRV pattern is found, depression and anxiety should be clarified; personality features and levels of activity should be noticed when an individual reveals a relatively high vagal pattern. A prerequisite of this approach is that the biological factors that may affect HRV have been excluded.

The ANOVA results did not reveal that the psychopathologies (PHQ-15, HAQ, BDI-II, and BAI scores) were significantly different among the four HRV patterns, but some significant associations between the patterns and psychopathologies were found in multiple regression models. An explanation is that HRV is very sensitive; therefore, if potential confounders are not considered, the analysis is unlikely to reveal significant associations between HRV and psychopathologies.⁴ This may explain the discrepancy.

Our results showing that HRV is associated with various kinds of physiological and psychological conditions are compatible with those in the literature.⁴ Although the association between HRV and a specific construct (such as depression) can be measured in well-designed studies, it is unlikely to directly connect psychological features and HRV in clinical practice without considering common confounders. The pattern approach is an attempt to remind clinicians that HRV can be affected by many factors; the factors with significance in this study may be the most common ones and should be considered in practice. In other words, the goal of this study is to provide useful and evidence-based information for clinicians, rather than to present brand new findings.

Several limitations of this study should be noticed. First, the data we adopted for defining normal HRV values were the first and third quartiles of healthy Korean adults.¹⁸ This is based on the possible skewness of the absolute values of SDNN, LF, HF, and LF/HF. If other principles were adopted for defining the normal ranges, the results may be different. Furthermore, the normative HRV data in the surveys of different countries showed discrepancies.^8,18 If normative HRV data were available in Taiwan they would be more suitable to adopt, but to our knowledge such data are still lacking, hence we used the data of East Asian people. Second, we did not stratify our data according to age and gender. HRV was often considered to be affected by age and gender, therefore the “normal range” of HRV in people with different age/gender may be distinct.¹⁹ However, we expected to build a simple and clinically oriented model and a detailed stratification may reduce the convenience for clinical purposes. Furthermore, the sample size of each group after stratification will decrease and this may influence the analytical quality. Third, our psychiatric sample was based on the population with affective disorders (depression, anxiety, and somatic symptoms). However, the results cannot be extended to populations with other psychiatric disorders, such as substance use disorders and psychotic disorders; specific HRV features have been found in these psychiatric conditions.^7,11

CONCLUSION

In summary, our results indicate that when the resting-state HRV reveals a specific pattern it is possible to associate this with one's personality, recent psychological state and demographic or lifestyle features. The suggested protocol of resting-state HRV measurement and interpretation is illustrated in Figure 2. We believe that this approach could generate more reliable HRV data and provide more evidence-based interpretation for the multiple HRV indices. Our results may also remind clinicians of two issues: common confounders (nonpsychological factors) should be noted when interpreting HRV findings; and the HRV patterns should be viewed as useful for screening purposes rather than for making accurate psychiatric diagnoses. Whether this method can help clinicians find psychological problems efficiently and whether suggestions based on the HRV patterns can increase the level of health await further clarification.

FIGURE 2. (A) Proposed flowchart of 5-min HRV measurement for psychiatric purposes. (B) Proposed algorithm of 5-min HRV measurement for clinical purposes, focusing on interpreting the results of evaluation. BMI, body mass index; DM, diabetes mellitus; HF, high-frequency power; HRV, heart rate variability; LF, low-frequency power; LF/HF, ratio of low-frequency to high-frequency power; SDNN, standard deviation of normal–to–normal RR intervals; TCA, tricyclic antidepressant

ACKNOWLEDGMENTS

The authors thank Huei-Mei Ma and Yi-Ling Lin for their administrative work during manuscript preparation.

CONFLICT OF INTEREST

All authors declare no conflict of interest.