Heart rate variability characteristics in a large group of active-duty Marines and relationship to posttraumatic stress

Abstract

Objective

Heart rate variability (HRV), thought to reflect autonomic nervous system function, is lowered in conditions such as posttraumatic stress disorder (PTSD). The potential confounding effects of traumatic brain injury (TBI) and depression in the relationship between HRV and PTSD have not been elucidated in a large cohort of military service members. Here we describe HRV associations with stress disorder symptoms in a large study of Marines, while accounting for well-known covariates of HRV and PTSD including TBI and depression.

Methods

Four battalions of male active-duty Marines (N=2430) were assessed 1-2 months prior to a combat deployment. HRV was measured during 5 minutes of rest. Depression and PTSD were assessed using the Beck Depression Inventory and Clinician Administered PTSD scale respectively.

Results

After accounting for covariates including TBI, a regression indicated that lower levels of high frequency (HF) HRV were associated with a diagnosis of PTSD (beta = -.20, p=.035). Depression and PTSD severity were correlated (r= .49, p <.001), however participants with PTSD but relatively low depression scores exhibited reduced HF compared to controls (p=.012). Marines with deployment experience (n=1254) had lower HRV than those with no experience (p = .033).

Conclusions

This cross-sectional analysis of a large cohort supports associations between PTSD and reduced HRV when accounting for TBI and depression symptoms. Future post-deployment assessments will be used to determine whether pre-deployment HRV can predict vulnerability and resilience to the serious psychological and physiological consequences of combat exposure.

Introduction

Heart rate variability (HRV), the quantitative assessment of variation in heartbeat intervals, can be used to detect alterations in autonomic nervous system (ANS) function (1). Heart rate is in part determined by influences on the sinoatrial node (SA) pacemaker, which is modulated by both the parasympathetic and sympathetic branches of the ANS (2). Spectral analysis of inter-beat intervals is used to derive the high-frequency (HF) peak of HRV, which is thought to reflect parasympathetic or vagal tone, though controversy exists about the sensitivity and specificity of widely-used HRV measures (3, 4).

HRV and regulation of the autonomic nervous system have been suggested to be useful in understanding cardiovascular and other health risks (4, 5). Decreased HRV has been associated with pathophysiology, psychopathology, and increased mortality (2, 6). Previous studies have reported lower HRV in psychiatric disorders such as schizophrenia, depression, bipolar disorder, panic disorder (7-10), and posttraumatic stress disorder (PTSD) (11-13). Accurate assessment of HRV can be done in a rapid (5-minute) period of time (14) using relatively non-obtrusive instrumentation; thus this physiological index has been used to study populations that may be at high risk for disrupted ANS functioning and the associated health complications.

One such high-risk group is represented by United States military service members who are deployed to combat situations such as Operation Enduring Freedom (Afghanistan post 2001), Operation Iraqi Freedom, or Operation New Dawn (OEF/OIF/OND). Since the onset of these conflicts, the prevalence of PTSD in returning veterans has been reported at 13-15% (15, 16). Prior studies have shown an association between PTSD and lower HRV (8, 11-13). PTSD is strongly associated with the occurrence of a traumatic brain injury (TBI) in OEF/OIF war veterans (17); furthermore TBI in and of itself has been related to lower HRV (18). Thus an unanswered question is whether HRV is associated with PTSD even when TBI is accounted for, in a population at elevated risk for both TBI and PTSD. Depression is also an important factor in understanding the relationship between trauma and HRV as illustrated by several studies on civilian trauma survivors (19, 20). Unknown, however, is the potential role of depressive symptoms in influencing the relationship between HRV and PTSD in military service members who are at high risk for both psychiatric conditions. Military personnel may constitute a unique population with respect to trauma and autonomic function due to a number of factors including prevalence of TBI, physical fitness, repeated exposure to severe combat-related trauma, and other features distinct to military service members.

The current study's objective was two-fold: 1) Assess the relationship between PTSD and HRV in a large group of active-duty Marines while accounting for important covariates of PTSD and HRV such as TBI and depression; and 2) Introduce the methodology used to assess HRV in this large participant group and assess the influence of potential covariates such as age (21, 22), ancestry (23), and use of nicotine (24) and caffeine (25, 26). This latter goal is in the service of studying HRV changes in the longitudinal portion of this study, i.e., after these service members return from deployment. The quantification of ancestry using genetic markers versus relying on participant self-report is a novel approach to replicating previous associations between HRV and ethnicity (23).

We hypothesized that lower HRV as quantified by the HF component would be associated with a PTSD diagnosis even when TBI was accounted for. We further hypothesized, based upon previous findings in civilian trauma survivors (20), that the co-occurrence of depressive symptoms and PTSD would result in lower HRV than PTSD alone.

Methods

Participants

Participants were active-duty Marines who were tested approximately one month prior to deployment to OEF, OIF, or OND as part of the Marine Resiliency Study (MRS), a prospective longitudinal study whose objective is to examine markers of risk and resilience to effects of combat stress in active-duty Marines. Four unique Infantry battalions (cohorts) of Marines were tested between July 2008 and August 2010 at one of two bases: Marine Corps Air Ground Combat Center (MCAGCC) in Twenty-nine Palms, CA (Cohorts 1, 2, and 3) and Camp Pendleton, CA (Cohort 4). See Table 1 for individual cohort sizes. This study was approved by the institutional review boards of the Veteran's Administration San Diego Health System IRB, the University of California San Diego, and the Naval Health Research Center.

Table 1.

Cohort sizes for heart rate variability (HRV) data collected in active duty Marines before deployment to OIF/OEF/OND.

	N eligible participants in cohort	N (%) participants with sufficient artifact-free HRV data
Cohort 1	315	298 (94.6%)
Cohort 2	721	699 (96.9%)
Cohort 3	670	603 (90.1%)
Cohort 4	886	830 (93.7%)
Total	2592	2430 (93.8%)

All active-duty Marines who were planning to deploy with their units were considered for inclusion into the study. Females were not included, since female Marines are not currently part of Infantry battalions. Since the US Marine Corps maintains specific guidelines that prohibit the severely mentally ill (i.e., Schizophrenia, Psychotic Disorder, Bipolar Disorder) from active-duty service, Marines with these pre-existing mental illness conditions were not included in the study.

The overall demographic composition of Marines in the MRS has been previously reported (27). Age, ancestry, and other covariate data for the participants are found in Table 2. 64.8% of participants reported graduating high school or receiving a GED, 32.1% reported some college or a college degree, and 0.4% reported a masters or doctoral-level degree. 61.8% of participants reported their marital status as never married, 36.1% reported being married, and 2.1% reported being separated or divorced. 68.3% had a junior enlisted rank (E1-E3), 28.9% were non-commissioned officers (E4-E9), and 2.6% were commissioned or warrant officers. Participants reported an average of 36.2 months (standard deviation=34.6) of military service. 48.2% reported at least one prior deployment.

Table 2.

Demographic and descriptive information for participants in each cohort.

	Cohort 1 (N=298)	Cohort 2 (N=699)	Cohort 3 (N=603)	Cohort 4 (N=830)	Total (N=2430)
Age (years)	21.9 (3.1)	22.4 (3.4)	23.2 (3.7)	23.1 (3.7)	22.8 (3.5)
Ancestry	62.8% European-American 3.7% African-American 19.1% Native-American/Mexican 14.4% Asian/Other	66.5% European-American 4.7% African-American 15.5% Native-American/Mexican 13.2% Asian/Other	60.7% European-American 6.5% African-American 17.2% Native-American/Mexican 15.6% Asian/Other	58.4% European-American 7.5% African-American 19.8% Native-American/Mexican 14.3% Asian/Other	61.9% European-American 6.0% African-American 17.9% Native-American/Mexican 14.3%Asian/Other
Body Mass Index (kg/m2)	26.8 (3.4)	27.3 (3.0)	28.1 (3.4)	27.8 (3.3)	27.6 (3.2)
Number (%) with a history of TBI	n=188 (63.1%)	n=422 (60.3%)	n=326 (54.1%)	n=413 (49.8%)	n=1349 (55.5%)
Hours since nicotine usea	n=133 4.0 (4.5)	n = 371 3.3 (4.2)	n = 284 2.7 (3.1)	n = 390 2.6 (3.3)	n = 1178 3.0 (3.7)
Hours since caffeine usea	n = 189 7.6 (6.7)	n = 460 6.6 (6.3)	n = 293 6.1 (2.7)	n = 419 5.0 (5.6)	n = 1361 6.1 (6.1)
AUDIT (alcohol use) total score	8.2 (6.3)	8.4 (7.4)	7.6 (6.7)	6.9 (6.3)	7.7 (6.8)
Number (%) using psychotropic medications	n=7 (2.3%)	n=12 (1.7%)	n=7 (1.2%)	n=4 (0.5%)	n= 30 (1.2%)
BDI (depression) total score	7.5 (8.3)	8.5 (8.8)	6.2 (7.3)	5.2 (6.5)	6.7 (7.8)
CAPS (PTSD) total score	16.9 (16.8)	16.8 (15.6)	14.6 (15.8)	13.2 (14.6)	15.0 (15.5)
Number (%) with PTSD	n=14 (4.7%)	n=41 (5.9%)	n=36 (6.0%)	n=29 (3.5%)	n=120 (5.0%)
Number (%) with a prior deployment	n=161 (54%)	n=386 (55.2%)	n=281 (46.5%)	n=426 (51.3%)	n=1254 (51.7%)

Note: History of TBI= self-reported history of a head injury accompanied by either loss of consciousness or altered mental status. AUDIT= Alcohol Use Disorders Identification Test. BDI= Beck Depression Inventory, CAPS=Clinician-Administered PTSD Scale. PTSD diagnosis was determined by meeting full DSM-IV criteria as assessed by the CAPS. Values in the table are means and (standard deviations) unless specified otherwise.

^acalculated only in subjects who self-reported use of this substance within 24 hours.

Procedure

All participating Marines provided voluntary written informed consent. Marines were informed about confidentiality relating to research data, and research records were rigorously protected. The entire MRS test battery was approximately four hours in duration and included a comprehensive evaluation of demographic information, history, and current symptoms with respect to military service, drug, alcohol and tobacco use, psychiatric conditions, head injuries, and psychological trauma (27). Height and weight were measured and blood samples were taken for genetic association studies.

For the assessment of HRV, participants were seated in quiet rooms. A finger photoplethysmograph (PPG, Pasco Scientific, Roseville, CA) was placed on the nail of the right fifth finger which rested on a desk. PPG is an optical technique used to detect beat-to-beat blood volume changes, for example as a result of pulse, in microvascular tissue. Fluctuations in the blood volume of the finger are directly related to the activity of the heart, thus the interval between peaks in the PPG signal, known as the PP interval, is considered a reasonably accurate reflection of the RR interval (28). PPG has been shown to be a sensitive and reliable peripheral instrument for the capture of cardiac activity (29), for example it is highly correlated with waveforms from simultaneous ECG recordings (30, 31). Frequency and time-domain measures of HRV derived from PPG were not significantly different from those derived by a simultaneous two-lead ECG recording (28). PPG was sampled at the rate of 1000 Hz. Using an oscilloscope display and amplification of the PPG signal (San Diego Instruments), the examiner ensured that the PPG was adequately capturing the heart beat waveforms without cutting off the peak of the R wave. The position of the PPG was adjusted until a visually clear heart-beat signal was obtained and each Marine was asked to keep his hand relatively motionless during the 5 min recording. Participants were asked to sit comfortably and direct their attention to a computer monitor where they were entertained with simple visual puzzles (e.g. locating hidden images in a photograph). The images were selected to be neutral and minimally affectively arousing or stress-inducing (e.g., dolphins, frogs). They were told that they did not need to memorize anything and that they would not be tested on the images afterward. The purpose of the hidden image task was to maximize the likelihood that participants remained stationary, awake, and alert for the duration of the recording. The images changed every 60 seconds (thus outside of the bandwidth for both LF and HF frequency ranges—see below) and the order of presentation of the images was the same for all participants. The 5-minute HRV recording session was simultaneous with the 5-minute acclimation period which is standardly used immediately prior to a session of eyeblink startle and prepulse inhibition measurement (32, 33). Thus, during the 5-minute PPG recording session participants wore headphones which delivered continuous broad-band noise at a decibel level of 70 dB, which is a standard level for an acclimation period prior to an eyeblink startle session. Participants were also prepared for electromyography (EMG) of the orbicularis oculi muscle recordings via the placement of two skin electrodes near the left eye for the purpose of subsequently assessing the eyeblink startle response and sensorimotor inhibition after the heart rate recordings presented here were completed (EMG data will be reported elsewhere). Only the constant 70 dB background noise was delivered during the 5-minute HRV recording.

Data Processing

Data files (one file per 5-minute session) from the PPG were extracted and processed through a multi-step procedure to generate HRV variables:

1. The systolic peaks of the PPG signal were identified and a tachogram representing intervals between heartbeats (the PP interval) was generated by measuring the time difference between successive peaks.

2. Tachograms were processed by the HRV analysis module of VivoSense 1.0 (Vivonoetics, 2011), which can process multiple files in an automated “batch process” format. See Supplemental Digital Content 1 for details of processing and derivation of HRV measures.

3. Following the batch process, trained scorers (VR, AM) visually inspected each file in accordance with the Task Force's recommendations that automated HRV analyses should be followed by visual inspection and manual correction. Additional corrections were required in approximately 8%-10% of files and each corrected file was re-processed to generate HRV variables.

4. Files for which the software determined that there was insufficient artifact-free data to accurately calculate frequency-domain variables were excluded from further HRV analysis. Typically this occurred when there was prominent motion artifact throughout the 5-minute session. Table 1 displays the percentages of sufficient artifact-free HRV data in each cohort.

HRV measures were generated per the recommendations of the Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology (14) and included [definitions and physiological correlates can be found in (3, 14) and specified citations]:

1. HR: Heart rate in beats/minute over the 5-minute session; a time-domain measure;

2. SDNN: standard deviation of the R-R intervals in ms; a time-domain measure influenced by both sympathetic and parasympathetic activity;

3. RMSSD: root mean square successive differences between R-R intervals; a time-domain measure primarily influenced by parasympathetic activity;

4. VLF: absolute power of the very low frequency (< 0.04 Hz) band in ms²; a frequency-domain measure whose physiological correlates are not well-understood;

5. LF: absolute power of the low frequency (0.04-0.15 Hz) band in ms²; a frequency-domain measure thought to reflect sympathetic activity and some parasympathetic activity (34, 35);

6. HF: absolute power of the high frequency (0.15-0.4 Hz) band in ms²; a frequency-domain measure thought to reflect primarily parasympathetic activity (34, 35);

7. LFnorm: LF power in normalized units calculated by LF/(total power-VLF); a frequency-domain measure. LFnorm reflects the percentage of total power that is accounted for by LF which reflects both sympathetic and parasympathetic activity;

8. HFnorm: HF power in normalized units calculated by HF/(total power-VLF); a frequency-domain measure. HFnorm is thought to reflect percentage of total power that is accounted for by HF which reflects primarily parasympathetic activity;

9. LF/HF ratio: ratio of LF over HF; a frequency-domain measure. Higher ratios have been proposed to reflect more sympathetic relative to parasympathetic activity (36). It is important to note, however, that use of the LF/HF ratio as a robust measure of sympathetic to parasympathetic balance has come under substantial scrutiny. Eckberg provides a review of the evidence that parasympathetic contributions to LF are significant and that HF may not reflect parasympathetic function when respiration is not controlled (4).

Of these nine variables, the LF/HF ratio and the transformed values for SDNN, RMSSD, LF, and HF were used in further analysis. HR was not used because, in relation to the other measures, is not a direct index of HRV. The LF/HF ratio is a widely used index that is the ratio of LFnorm and HFnorm, thus these latter variables were not further analyzed. Guidelines on the selection of appropriate epochs for assessing HRV suggest that a 5-minute recording window is sufficient to derive LF and HF as well as time-domain indices such as SDNN and RMSSD, but accurate assessment of VLF likely requires at least a 50-minute recording window (3, 14). Thus, VLF was not included in subsequent analyses.

Variables studied in relationship to HRV were as follows: age in years, ancestry, hours since last nicotine use (for participants who reported using nicotine within 24 hours), hours since last caffeine use (for participants who reported using caffeine within 24 hours), body mass index (BMI) in kg/m², history of traumatic brain injury (TBI) as defined by a self-report of head injury that was accompanied by either a loss of consciousness or altered mental status, and current use of psychotropic medications (Table 2). This latter category was defined broadly as the current use of one or more of the following classes of medications: antidepressants, benzodiazepines, sleep aids, mood stabilizers, prescription stimulants, or antipsychotic medications. Beck Depression Inventory-II (BDI) (37) scores and the Clinician-Administered PTSD Scale (CAPS) (38) scores were also assessed in relation to HRV. The CAPS is a clinician-administered structured interview and is considered the “gold standard” for assessment of PTSD symptoms and ascertainment of a PTSD diagnosis using DSM-IV criteria (see below).

Ancestry was determined using genetic information as described in (39). In brief, genotypes of 1,783 ancestry-informative markers (AIMs) were used to determine a participant's ancestry at the continental level for the 7 geographic regions Africa, Middle East, Europe, Central/South Asia, East Asia, Americas, and Oceania. Ancestry estimates were determined using STRUCTURE v2.3.2.1. (40) at K=7, including prior population information of the HGDP reference set (41). Based on these ancestry estimates, MRS participants with HRV data were placed into 4 groups: participants with >95% European ancestry were grouped with European-Americans (N=1503); participants with >5% African ancestry and <5% each Native American, Central Asian, East Asian, and Oceanic ancestry as African-American (N=145), participants with >5% Native American and <10% African, and <5% each Central Asian, East Asian and Oceanic ancestry as Hispanic and Native Americans (N=434), and all others, including 43 East Asians with >95% East Asian ancestry as Others (N=348).

Statistical Analysis

Frequency and descriptive statistics were generated for each HRV measure (Table 3). In cases where a variable showed significant skew, transformations were applied per the recommendations of Tabachnick and Fidel (42). Natural log transformations of frequency-domain measures such as LF, HF, and the LF/HF ratio are widely used (11, 13, 21, 43-45). Following transformation of the appropriate variables, outliers defined as greater than 3 standard deviations from the mean were excluded from subsequent data analyses (Table 3). SDNN and RMSSD showed a moderate positive skew and square root transformations of these variables were generated (42). LF, HF, and the LF/HF ratio showed a substantial positive skew and natural log transformations were generated. After the log transformations, distributions were inspected and skew was assessed. The log-transformed HRV values were normally distributed. Distributions of HRV were relatively similar across the four cohorts (though differences in mean HRV measures were observed between cohorts; see Table 4 and below).

Table 3.

Descriptive information for heart rate variability variables in the entire sample (N=2430).

	Heart Rate	SDNN	RMSSD	VLF	LF	HF	LFnorm	HFnorm	LF/HF Ratio
Mean (SD)	67.1 (10.2)	64.1 (26.9)	58.9 (34.5)	2380.2 (2470.9)	5144.1 (5467.4)	4153.8 (5074.4)	.56 (.17)	.41 (.16)	1.9 (1.7)

Note: Heart rate is in beats/min. SDNN =standard deviation of the R-R intervals in ms. RMSSD =root mean square successive differences between R-R intervals. VLF= absolute power of the very low frequency (< 0.04 Hz) band in ms². LF = absolute power of the low frequency (0.04-0.15 Hz) band in ms². HF = absolute power of the high frequency (0.15-0.4 Hz) band in ms². LFnorm= LF power in normalized units calculated by LF/(total power-VLF). HFnorm= HF power in normalized units calculated by HF/(total power-VLF). LF/HF ratio= ratio of LF over HF. For a given variable in a given cohort, a range of 0-9 subjects were moved due to outlier scores, for an average 0.34% removed due to outlier scores.

Table 4.

Relationship of heart rate variability variables to demographic and other factors.

	SDNN	RMSSD	LF	HF	LF/HF ratio
Cohort	F(3,2419)=45.68***, ηp2 = .054 Cohort 1>Cohort 3,4 > Cohort 2	F(3,2411)=34.65***, ηp2 = .041, Cohort 1>Cohort 3,4 > Cohort 2	F(3,2408)=25.59***, ηp2 = .031 Cohort 1>Cohort 3,4 > Cohort 2	F(3,2414)=15.56***, ηp2 = .019 Cohort 1>Cohort 3,4 > Cohort 2	F(3,2419)=2.06, ηp2 = .003 Cohort 1>Cohort 3b,
Age (years)	Pearson r = −.11***	Pearson r = -.12***	Pearson r = −.11***	Pearson r = −.16***	Pearson r = .09***
Ancestry	F(3,2419)=2.54*, ηp2 = .003 Caucasian> African-American, Asian/Other	F(3,2411)=.37, ηp2 =.000	F(3,2408)=8.48***, ηp2 = .010 Caucasian>African-American, Native-American/Mexican, Asian/Other	F(3,2414)=1.74, ηp2 =.002 Caucasian>Asian/Otherb,	F(3,2419)=7.10***, ηp2 =.009 Caucasian>African-American, Native-American/Mexican, Asian/Other
Body Mass Index (kg/m2)	Pearson r = −.04	Pearson r = .01	Pearson r = −.07**	Pearson r = −.01	Pearson r = −.08***
History of TBI	t(2413)=.03, d = .001	t(2405)=.44, d =.02	t(2402)=.17, d=.01	t(2408)=1.91, d = .08	t(2413)=2.44*, d = .10 TBI history> no TBI
Hours since nicotine usea	Pearson r = .10***	Pearson r = .09**	Pearson r = .09**	Pearson r = .07*	Pearson r = −.01
Hours since caffeine usea	Pearson r = .05	Pearson r = .03	Pearson r = .02	Pearson r = .02	Pearson r = .01
AUDIT total score	Pearson r = .01	Pearson r = .02	Pearson r = −.01	Pearson r = .01	Pearson r = −.01
Use of psychotropic medications	t(2422)=1.50, d=.28 med users> non-users	t(2412)=1.28, d=.24 med users>non-users	t(2409)=2.03*, d=.38 med users> non-users	t(2415)=1.47, d=.27 med users> non-users	t(2420)=.17, d=.03
History of a prior deploymentc	F(1,2407)=2.52, ηp2 = .001	F(1,2399)=2.95, ηp2 = .001	F(1,2396)=2.02, ηp2 = .001	F(1,2402)=5.13*, ηp2 =.002, prior < no prior	F(1,2407)=0.45, ηp2 <.001

Note: SDNN and RMSSD were square root-transformed. VLF, LF, HF, and the LF/HF ratio were natural log-transformed. Direction of findings is reported if statistical significance is reached in either the overall test or planned comparisons, or at least a small effect size is achieved. History of TBI= self-reported history of a head injury accompanied by either loss of consciousness or altered mental status. AUDIT= Alcohol Use Disorders Identification Test. SDNN =standard deviation of the R-R intervals in ms. RMSSD =root mean square successive differences between R-R intervals. LF = absolute power of the low frequency (0.04-0.15 Hz) band in ms². HF = absolute power of the high frequency (0.15-0.4 Hz) band in ms². LF/HF ratio= ratio of LF over HF.

η_p² = partial Eta squared. d= Cohen's d effect size.

^acalculated only in subjects who self-reported use of this substance within 24 hours.

^boverall ANOVA did not achieve statistical significance but this planned comparison was significant at p <. 05.

^cANCOVAs for history of prior deployment were conducted with age as a covariate and PTSD group (no PTSD vs. partial PTSD vs. full PTSD) as a factor.

^***p < .001

^**p < .010

^*p < .050

A high proportion of Marines had BDI and CAPS scores of zero (22.3% for BDI, 15.7% for CAPS); hence categorical variables were created for BDI and CAPS scores. For the BDI, three categories were generated: BDI scores of 0, BDI scores between 1 and 19 suggesting minimal/mild depression, and BDI scores of 20 or greater suggesting moderate/severe depression. For the CAPS, Marines were categorized as having no symptoms of PTSD, meeting partial criteria for PTSD, or meeting full criteria for PTSD. Full criteria are derived from the DSM-IV diagnosis for PTSD and require the following: at least 1 “B” symptom (the traumatic event is persistently re-experienced), 3 “C” symptoms (persistent avoidance of stimuli associated with the trauma and numbing of general responsiveness), and 2 “D” symptoms (persistent symptoms of increased arousal). Participants were deemed to have met partial criteria for PTSD if they did not meet full criteria but did endorse: 1) at least 1 “B” symptom, 2 “C” symptoms, and 2 “D” symptoms; or endorsed 2) at least 1 “B” symptom, 3 “C” symptoms, or 2 “D” symptoms.

In order to assess the relationship between HRV and PTSD after accounting for important covariates, a single multinomial logistic regression was conducted with PTSD category (comparing no PTSD to partial PTSD and full PTSD) as the outcome variable and HRV and TBI history as predictors as well as age, cohort, and ancestry, because these variables were shown to have significant and consistent associations with HRV (see Table 4). A regression approach was chosen in order to account for multiple covariates while preserving power, and PTSD category was chosen as the outcome with the hope that this approach could most sensitively detect potential HRV changes associated with different severity levels of PTSD. Cases with no PTSD were used as the reference category. As above, a large proportion of zero CAPS scores precluded the use of CAPS scores as a continuous measure of PTSD severity. In order to minimize multicolinearity among the highly intercorrelated HRV variables, one HRV index, log-transformed HF, was entered as a predictor. The assumptions of the model were tested with goodness-of-fit Pearson and Deviance Chi-Squares. Significance values were both p > .100, indicating that the data were consistent with model assumptions.

To assess whether there was a PTSD-by-depression interaction on HRV, a two-way analysis of variance (ANOVA) was conducted with PTSD category (no PTSD versus full or partial PTSD) and BDI category (no or minimal/mild depression vs. moderate/severe depression) as the independent variables and log-transformed HF as the dependent variable.

To provide simple descriptive statistics across diagnostic groups, HRV differences with respect to presence or absence of a DSM-IV diagnosis of PTSD as defined by the CAPS were assessed using an independent samples t-test, and HRV differences between the three BDI categories (no depression, minimal/mild, and moderate/severe) were assessed using a Univariate ANOVA.

Relationships between HRV and symptoms scores as well as the continuous variables of age, BMI, hours since nicotine use, hours since caffeine use, and scores on the Alcohol Use Disorders Identification Test (AUDIT) (46) were assessed with Pearson R correlation coefficients. HRV differences with respect to presence or absence of a TBI, history and presence or absence of psychotropic medication usage were analyzed using Independent sample t-tests. HRV differences between cohorts and ancestry categories were analyzed using Analyses of Variance (ANOVA). HRV differences between those with a history of a prior deployment and those without a deployment history were analyzed using Analysis of Covariance (ANCOVA) with age as a covariate and PTSD group as a factor. It should be noted that cohort (1-4) had significant associations with both HRV variables and outcome measures of interest (BDI, CAPS); hence our descriptive statistics are presented by cohort. The four cohorts were tested during four different periods and uncontrolled factors such as season, differences in training, and differences in specific deployment destination may have occurred. To account for this potential variance, cohort was consistently kept as a factor in our statistical models.

Significance levels were set at p < 0.050 and Cohen's d and partial Eta squared (η_p²) effect sizes were calculated when relevant. Statistical analyses were conducted with PASW/SPSS 18.

Results

Description of HRV

The range of the HRV measures (Table 3) were generally consistent with HRV values reported in a recent review of 44 short-recording HRV studies (47), with the exception of values of VLF, LF and HF, which are substantially higher in the current population than what has been previously reported.

Relationship of HRV to depression and PTSD

Across all four cohorts, 1678 (69%) participants were categorized as having minimal to mild depression while 189 (7.8%) participants were categorized as having moderate to severe depression. The ANOVA yielded no significant differences or trends among the three BDI categories on any of the HRV measures.

Across all four cohorts, 120 Marines (4.9%) were categorized as meeting the full criteria for PTSD while 195 Marines (8.0%) were categorized as meeting partial PTSD criteria. The multinomial logistic regression with PTSD category as the outcome and HRV, TBI group and other covariates as predictors was statistically significant (Chi-Square = 77.7, p < .001, pseudo-R²=.03). HRV reached statistical significance in its association with partial PTSD and full PTSD (Table 5) such that log-transformed HF values were lower in individuals with partial and full PTSD compared to individuals without PTSD. As a confirmatory post-hoc analysis, an ordinal regression was conducted with PTSD category (no PTSD, partial PTSD, full PTSD) as the ordinal outcome. The proportional odds assumption was met (Test of Parallel lines Chi-Square = 11.8, p < .100). The model was statistically significant (Chi-Square = 65.5, p < .001), and again HF was statistically significant in its association with PTSD (ordered log-odds estimate = -.18, p = .004) when other covariates were also in the model.

Table 5.

Parameter estimates for the high frequency index (HF) in the multinomial logistic regression predicting no PTSD diagnosis (n=2115) versus partial PTSD diagnosis (n= 195) versus full PTSD diagnosis (n=120).

PTSD group	B	Std. Error	Wald χ2	df	Significance	Exp (B) Odds Ratio	95% CI lower for Exp (B)	95% CI upper for Exp (B)
Partial PTSD	−.16	.08	4.44	1	0.035	.85	.94	1.02
Full PTSD	−.20	.09	4.44	1	0.035	.82	.67	.99

Note: The following covariates were first entered into the model: age, ancestry, cohort, and history of TBI. HF values were log-transformed. PTSD groups were determined by the CAPS. Partial PTSD= did not meet full criteria and had one of two CAPS profiles: 1) at least 1 “B” symptom, 2 “C” symptoms, and 2 “D” symptoms; or 2) at least 1 “B” symptom, 3 “C” symptoms, or 2 “D” symptoms. Full PTSD = at least 1 “B” symptom, 3 “C” symptoms, and 2 “D” symptoms. χ² = Wald Chi-Square

The two-way ANOVA with PTSD group and depression group as independent variables and log-transformed HF as the dependent variable indicated no significant PTSD-by-depression interaction [F(1,2387)=.18, ns, η_p²<.001)] nor a main effect of depression group [F(1,2397)=.05, ns, η_p² < .001). Figure 1A displays back-transformed HF means comparing subjects with no PTSD and no depression, subjects with depression but no PTSD, subjects with full or partial PTSD and depression, and subjects with full or partial PTSD but no depression. The main effect of PTSD group did not reach statistical significance [F(1,2397)=2.55, p=.111, η_p² = .001), but post-hoc comparisons of the four groups indicated that the PTSD-only group had significantly lower HF than the group without PTSD and depression (p = .012).

Figure 1.

Mean high frequency index (HF) in Marines with and without PTSD and depression or TBI, all cohorts combined.

Note: Error bars are standard errors of the mean. Values in figure are back-transformed absolute HF means. Left Panel (A): * p < .050, PTSD only versus no PTSD/no depression in post-hoc comparison with transformed HF values as the dependent variable. Sample sizes are as follows: no PTSD and no depression n = 1965, PTSD only n = 248, depression only n = 124, PTSD and depression n = 64. “PTSD” refers to partial or full PTSD cases as determined by the CAPS. “Depression” refers to moderate/severe depression as determined by the BDI. Right Panel (B): * p < .050, main effect of PTSD versus no PTSD in ANOVA with transformed HF values as the dependent variable. Sample sizes are as follows: no PTSD and no TBI history n = 981, PTSD and no TBI history = n = 87, no PTSD and TBI history, n=1113, PTSD and TBI history n= 228. “PTSD” refers to partial or full PTSD cases as determined by the CAPS. History of TBI= self-reported history of a head injury accompanied by either loss of consciousness or altered mental status.

Follow-up Pearson correlations within full and partial PTSD cases indicated that HF was not significantly associated with any specific symptom CAPS symptom domains (re-experiencing r = -.02, ns, avoidance r = -.001, ns, or arousal r= .02, ns). A Pearson correlation also indicated that total CAPS scores were significantly positively correlated with BDI scores (r= .54, p < .001), even when subjects with BDI and CAPS scores of zero were excluded from the correlation (n= 1655, r = .49, p < .001).

The 120 participants across all four cohorts who met full diagnostic criteria for PTSD were compared to participants without full or partial PTSD (n=2115) on the selected HRV measures using independent samples t-tests. Levene's tests for Equality of Variances were non-significant for all five HRV measures. Marines with PTSD had significantly lower RMSSD [t(2210)=2.2, p = .027, d = .21], lower LF [t(2207)=2.6, p = .010, d = .24], and lower HF [t(2213)=2.5, p = .013, d = .23] than study participants without full or partial PTSD. The LF/HF ratio did not significantly differ among the two groups [t(2217)=.82, ns, d = .08], nor did SDNN [t(2218)=1.7, p = .087, d = .16).

Relationship of HRV variables to covariates

The relationship between at least one of the selected HRV measures and the following variables reached statistical significance: cohort, age, ancestry, hours since nicotine use, BMI, history of TBI, psychotropic medication use, and history of prior deployments (Table 4). Because cohort had significant associations with many HRV variables, we consistently kept it in as a factor in our subsequent regression models. Variables associated with lower HRV included older age, non-Caucasian ancestry, BMI, recency of nicotine use, and TBI history. Figure 1B displays back-transformed HF means of no PTSD versus partial or full PTSD subjects with and without a TBI history. There were small effect sizes for HRV differences between participants who were taking psychotropic medications and those who were not such that psychotropic medication users had, on average, higher HRV. Finally, when age was used as a covariate, Marines with a history of prior deployment had significantly lower HF than those without a deployment history. Although the deployment history-by-PTSD-group interaction was not statistically significant [F(2,2402=.72, ns], planned comparisons indicated that HF was significantly lower in previously deployed participants without PTSD versus participants without PTSD who did not have a previous deployment (p= .033). This difference did not reach statistical significance in those with partial PTSD or full PTSD (Figure 2). It is detailed in Figure 2 that the sample sizes for the comparisons in the partial and full PTSD groups were substantially smaller than those for the no PTSD comparisons. Figure 2 displays back-transformed HF means for Marines with and without a deployment history delineated by PTSD groups and adjusted for age.

Figure 2.

Mean high frequency index (HF) for no history of prior deployments versus prior deployment history in Marines with no PTSD, partial PTSD, and full PTSD, all cohorts combined.

Note: Errors bars are standard errors of the mean. Values in figure are back-transformed absolute HF means. *p < .050 vs. No PTSD/No Prior Deployment group using ANCOVA with transformed HF values as the dependent variable and age as a covariate. Sample sizes are as follows: no PTSD and no prior deployment n = 1031, no PTSD and prior deployment n = 1063, partial PTSD and no prior deployment n = 80, partial PTSD and prior deployment n = 116, full PTSD and no prior deployment n = 51, full PTSD and prior deployment n = 68. See results for details.

Discussion

We have described a relatively non-obtrusive and rapid methodology to assess, process, and analyze HRV in a large population of military service members who, because of their eventual deployment to combat zones, are at risk of developing stress-related conditions. Furthermore, we have tested the relationship between HRV and PTSD while accounting for a history of traumatic brain injury and depression symptoms. TBI is highly prevalent in military service members of the recent conflicts in the Middle East, and the presence of TBI and PTSD together have been identified as the “signature injury” of these wars (48). The overlap between TBI and PTSD in military service members is further exemplified by the recent observation that deployment-related TBI is a strong predictor of deployment-related PTSD (49). Further, there are known effects of head injuries on HRV (albeit not strongly observed in our study, see below). Despite this knowledge, no study on PTSD and HRV in a military population has controlled for TBI history. The current results suggest that, even when a TBI history is accounted for, lower HRV is significantly associated with PTSD.

Depressive symptoms, however, were not related to HRV in this sample, and contrary to our hypothesis, the co-occurrence of PTSD and depressive symptoms was not associated with lower HRV than either condition alone even though PTSD symptom severity and depressive symptom severity were highly related. Rather, the group with PTSD in the absence of moderate or severe depressive symptoms had the lowest HRV. This finding is in contrast to what has been previously observed. Studies in non-military trauma survivors suggest that a trauma history and depression interact in their influence on autonomic arousal; for example the presence of depression with a trauma history was associated with lower respiratory sinus arrhythmia than either condition alone in females with a history of trauma related to crime, natural disaster, or assault (19). In a study of survivors of Hurricane Katrina, depression was more strongly associated with lower HRV than was PTSD (20), a finding in notable contrast to the current results. While depression was not related to lower HRV in the current study, a prior history of deployment was, even in Marines without a PTSD diagnosis, reminiscent of a recent report of an association between combat exposure and decreased HRV (13). Thus, the relationships among depressive symptoms, trauma symptoms, and autonomic nervous system function may vary depending on the population of trauma survivors and is highly dependent on the nature and context of the traumatic event.

Several factors could explain the absence of a relationship between depressive symptoms and HRV, one of which is the inherent limitation of a self-report instrument. Furthermore, our classification of BDI scores into just three categories, while necessary to achieve adequate sizes of groups, may have been too coarse to detect more subtle HRV differences. Lastly but importantly, the BDI is sensitive in capturing the severity of acute depressive symptoms but is less informative about the chronicity of depression. Most findings relating depression to HRV have been conducted on individuals with a chronic depression condition (7, 19). Nevertheless, the current findings suggest that there remains a high co-occurrence of symptoms of depression with post-traumatic symptoms, which is an immediate public health concern in active-duty military personnel. The causal directions of these relationships have not been well-elucidated.

Marines with a history of TBI with associated altered mental status or loss of consciousness demonstrated lower HRV, but the effect size for this finding was very small. The proportion of Marines reporting a previous TBI was relatively high (55.5%) but was consistent with the published characteristics of this cohort of Marines (27). Many of these cases are likely “mild” head injuries given that either altered mental status or loss of consciousness was sufficient to identify a TBI. It is also important to note that participants were asked about a history of any head injury, not solely those related to prior military deployments, which likely accounted for the high percentage of TBIs in this sample as compared to other studies (17). Head injury has been associated with alterations in the regulation of the cardiac system and lower HRV (18, 50, 51). Interestingly and potentially relevant to the current findings, HRV abnormalities in athletes with a recent concussion were only observed during an exercise session, but not at rest (52). The severity of a TBI as well as time elapsed since the injury are also likely to be important factors in the normalization of HRV (51).

Other factors that were associated with HRV and are relevant to consider in future analyses with these cohorts included age, ancestry, body mass index, and nicotine use. The relationship between age and HRV has been widely observed. The genetic determination of ancestry underscores previous observations that genetic factors are thought to contribute to a substantial proportion of the variance (13-23%) in HRV, albeit less so than the combined influence of non-genetic variables such as age, sex, and environmental factors (53). BMI was only weakly related to one HRV measure, the high-frequency index. BMI is known to be a relatively less reliable index of fat accumulation in athletes with high muscle mass (54). Active-duty Marines readying for deployment to a combat zone fall into this category. The high fitness levels and relative youth of these participants may also account for the higher values of HRV observed in this study compared to other published reports. Recent use of nicotine was related to lower HRV, supporting previous work that nicotine use may alter autonomic functioning (24).

Mechanisms underlying lower HRV in PTSD has been postulated to reflect reduced vagal or parasympathetic tone (14). Diminished parasympathetic tone may accompany changes in amygdala and medial prefrontal cortex activation, brain regions that have been implicated in PTSD and are thought to underlie fear and threat responses (55, 56). Changes in LF and HF have been associated with altered connectivity between the aforementioned brain regions and structures such as the anterior cingulate and insula that are implicated in orienting attention and vigilance (56, 57). In any interpretation of findings related to LF, HF, and the LF/HF ratio, however, it is important to consider the following caveat. There has traditionally been an overreliance on these indices as being direct reflections of sympathetic/vagal balance. As Eckberg (4) reviews, parasympathetic contributions to LF are significant, changes in HF may not always be explained by changes in parasympathetic activity, and under certain conditions sympathetic and parasympathetic changes occur in parallel to one another and not reciprocally. Thus caution should be used in making firm conclusions about the physiological underpinnings of sympathetic versus parasympathetic functioning in the absence of rigorous experimental controls.

Other limitations of this study include a restricted age range and a lack of female participants, limiting the generalizability of our findings to a relatively young, athletic group of males. This sample is, we would argue, highly representative of US service members currently at greatest risk for combat-related PTSD. Another limitation is that there were small but significant differences (η_p² = .003-.054) between cohorts in HRV and other demographic variables that may have been due to chance or may have been attributable to a number of random factors such as differences in season of testing, individual battalion demographics, and physical training courses leading up to or prior to the data collection. The relationship between HRV measures and PTSD caseness, however, was robust enough to be significant even when controlling for cohort. Several potential experimental factors which can affect HRV are relevant to mention. First, participants were asked to attend to video images of hidden pictures. Previous studies have reported cardiac deceleration during reaction time and response inhibition tests [see (58) for a summary]. Whereas participants in this study were told that they would not be required to respond to the visual images, the paradigm could arguably represent a cognitive challenge that affected HRV. We cannot entirely rule out that cognitive activity may have affected HRV, which would be a consistent phenomenon across all participants in the study. Second, no measure of respiration was obtained during HRV assessment, which is a notable limitation since changes in breathing rates are directly related to respiratory sinus arrhythmia (3) and can also be associated with different mood states, e.g., higher respiratory rates in anxious patients (59). The collection of large sample sizes of Marines in short time frames rendered PPG a practical rapid method as opposed to use of ECG Holter monitors plus respiratory band application and recordings. Thus we were not able to assess to what extent breathing rates in participants with PTSD may have moderated HRV. A final limitation is that this study is cross-sectional, however prospective analysis of these cohorts upon return from deployment and after onset of PTSD symptoms is ongoing.

In conclusion, we have described our methodology for the collection and analysis of short-term HRV in a large population of Marines readying for deployment. Lower HRV was observed in participants with a full or partial PTSD diagnoses at the pre-deployment MRS time frame even when TBI history was accounted for. Previous deployments were associated with lower HRV while depression was not, but depression was more strongly related to PTSD than it was to HRV. Future longitudinal analyses of these military service members will include the consideration of the HRV factors and covariates elucidated here. The ultimate aim of this research is to uncover whether ANS functions can predict who is vulnerable and who is resilient, or whether ANS functions emerge in tandem with mental health effects of combat exposure.

Abbreviations

AIMS

ancestry-informative markers

ANS

autonomic nervous system

AUDIT

Alcohol Use Disorders Identification Test

BDI

Beck Depression Inventory

BMI

body mass index

CAPS

Clinician-Administered PTSD scale

EMG

electromyography

HF

high frequency index

HR

heart rate

HRV

heart rate variability

LF

low frequency index

MRS

Marine Resiliency Study

OEF

Operation Enduring Freedom

OIF

Operation Iraqi Freedom

OND

Operation New Dawn

PPG

photoplethysmograph

PTSD

posttraumatic stress disorder

SA

sinoatrial node

SSRI

selective serotonin reuptake inhibitor

TBI

traumatic brain injury

VLF

very low frequency index