ADVERTISEMENT

Issue Navigator

Volume 15 No. 01
Earn CME
Accepted Papers





Scientific Investigations

Screening for Obstructive Sleep Apnea in Commercial Drivers Using EKG-Derived Respiratory Power Index

M. Melanie Lyons, PhD, ACNP1; Jan F. Kraemer, Dipl-Inform2; Radha Dhingra, MD3; Brendan T. Keenan, MS1; Niels Wessel, PhD2; Martin Glos, PhD4; Thomas Penzel, PhD4; Indira Gurubhagavatula, MD, MPH1,5,6
1Center for Sleep and Circadian Neurobiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; 2Department of Physics, Humboldt-Universitat zu Berlin, Berlin, Germany; 3Mahatma Gandhi Medical College and Hospital, Jaipur, India; 4The Centre of Sleep Medicine, Department of Cardiology, Charité Universitätsmedizin, Berlin, Berlin, Germany; 5Hospital of the University of Pennsylvania, Philadelphia, Pennsylvania; 6Sleep Disorders Clinic at the Philadelphia CMC VA Medical Center, Philadelphia, Pennsylvania

ABSTRACT

Study Objectives:

Obstructive sleep apnea (OSA) is common in commercial motor vehicle operators (CMVOs); however, polysomnography (PSG), the gold-standard diagnostic test, is expensive and inconvenient for screening. OSA is associated with changes in heart rate and voltage on electrocardiography (EKG). We evaluated the utility of EKG parameters in identifying CMVOs at greater risk for sleepiness-related crashes (apnea-hypopnea index [AHI] ≥ 30 events/h).

Methods:

In this prospective study of CMVOs, we performed EKGs with concurrent PSG, and calculated the respiratory power index (RPI) on EKG, a surrogate for AHI calculated from PSG. We evaluated the utility of two-stage predictive models using simple clinical measures (age, body mass index [BMI], neck circumference, Epworth Sleepiness Scale score, and the Multi-Variable Apnea Prediction [MVAP] score) in the first stage, followed by RPI in a subset as the second-stage. We assessed area under the receiver operating characteristic curve (AUC), sensitivity, and negative posttest probability (NPTP) for this two-stage approach and for RPI alone.

Results:

The best-performing model used the MVAP, which combines BMI, age, and sex with three OSA symptoms, in the first stage, followed by RPI in the second. The model yielded an estimated (95% confidence interval) AUC of 0.883 (0.767–0.924), sensitivity of 0.917 (0.706–0.962), and NPTP of 0.034 (0.015–0.133). Predictive characteristics were similar using a model with only BMI as the first-stage screen.

Conclusions:

A two-stage model that combines BMI or the MVAP score in the first stage, with EKG in the second, had robust discriminatory power to identify severe OSA in CMVOs.

Citation:

Lyons MM, Kraemer JF, Dhingra R, Keenan BT, Wessel N, Glos M, Penzel T, Gurubhagavatula I. Screening for obstructive sleep apnea in commercial drivers using EKG-derived respiratory power index. J Clin Sleep Med. 2019;15(1):23–32.


BRIEF SUMMARY

Current Knowledge/Study Rationale: In commercial motor vehicle operators (CMVOs), obstructive sleep apnea (OSA) is often undiagnosed because testing using the gold-standard measure, in-laboratory polysomnography, is expensive, not easily accessible, and time consuming for long-haul drivers who are regularly on the road. However, 28% to 60% of CMVOs have been found to have OSA and ∼30% of known CMVO accidents have been related to OSA.

Study Impact: We describe an approach that shows promise in identifying CMVOs who have severe OSA—who are likely to be at risk for a sleepiness-related crash—by using the spectral qualities of electrocardiography with portable diagnostic ability and without requiring overnight in-laboratory polysomnography. Future studies should include larger cohorts of patients.

INTRODUCTION

Characterized by repetitive breathing pauses during sleep, obstructive sleep apnea (OSA) occurs due to a collapse of the pharyngeal muscles, leading to cessation (apnea) or reduction (hypopnea) in airflow. These events are terminated by a sympathetic burst, which leads to a brief arousal from sleep. The resultant sleep fragmentation, intermittent hypoxia, and sympathetic activation have been linked with a number of downstream consequences. One symptom of OSA is daytime sleepiness, which is relevant for commercial motor vehicle operators (CMVOs) because of its association with sleepiness-related crashes.

Indeed, ∼30% of commercial vehicle accidents are due to sleepiness.13 OSA is common in CMVOs, with 28% to 60% having the disorder.4,5 This is because CMVOs tend to be obese, male, and middle-aged, the three most common risk factors for OSA.68 Despite this risk, in most CMVOs OSA is undiagnosed and, therefore, no treatment is implemented.5,810 The gold-standard test, polysomnography (PSG)11 is inappropriate for systematic screening because of its high expense, complexity, and relative inaccessibility, particularly among uninsured patients. Moreover, PSG requires the patient to sleep in a laboratory, which poses inconvenience to long-haul drivers. Therefore, simpler screening tools deserve investigation.

Symptom reporting has been shown to be unreliable in this group,9,10,12 and so may not be effective for screening. Objective tools include body mass index (BMI), age, and sex. These can be used alone, or in a two-stage fashion, with the second stage test being an ambulatory diagnostic recording that can be done in the CMVO's home or berth of the truck. These tools have included recordings of airflow, respiratory effort, and oxyhemoglobin saturation.

An alternative to these ambulatory tests that has not been evaluated before is heart rate variability (HRV) using electrocardiography (EKG), which can be summarized using time- or frequency-based metrics.13 Respiratory activity can affect EKG-derived variables. For example, respiratory movement can induce rotation in the electrical axis,14 as well as sinus arrhythmia.15 Additionally, OSA has been associated with cyclic heart rate variability,16 with patients experiencing a relative bradycardia during the apnea/hypopnea event, followed by a relative tachycardia at event termination, during the sympathetic burst. In addition, the effort to inhale against a closed glottis can lead to fluctuations in EKG voltage, as the chest wall moves away from the heart, followed by an increase in voltage when breathing resumes and the chest wall returns to neutral position. These data may be combined into a single metric that estimates the frequency of apneas and hypopneas per hour of sleep, or apneahypopnea index (AHI).

In the current study, we evaluated the utility of single-stage and two-stage risk prediction paradigms leveraging a three-channel EKG in identifying OSA in a cohort of CMVOs, using the AHI derived from PSG as the gold standard. Given that studies have associated severity of OSA with increased risk for motor vehicle crashes,17 we focused our observations on CMVOs with severe OSA (s-OSA; AHI of ≥ 30 events/h).

METHODS

The Institutional Review Board of the University of Pennsylvania approved the protocol. All participants provided signed informed consent.

Participant Selection

We recruited holders of active commercial driver's licenses through online advertisements from 2009–2011. Participants were required to reside within 40 miles from the Penn Sleep Center and be 18 to 65 years old. We excluded those using positive airway pressure or supplemental oxygen; those with nocturnal hypoxia because of any other illness; or medical or psychiatric conditions that prevented their ability to participate.

Confidentiality

Given that study data can affect employment, the National Institutes of Health granted a Certificate of Confidentiality,12 enabling us to resist court-ordered subpoenas to release personal health information. We shared this information with all participants.

Demographics, Symptoms, Examination, PSG and Case Definition

We collected demographic data,12 and self-reported data including medical history, apnea-related symptoms (snoring, choking/gasping, witnessed apneas), and daytime sleepiness (Epworth Sleepiness Scale [ESS]). Study personnel measured and recorded BMI, blood pressure, and neck circumference (NC). PSG tests were 14-channel studies conducted in the participant's home, as described previously,12 and included electroencephalography (C3, C4, Oz), three-channel EKG, eye, chin, and pretibial electromyography, chest and abdominal plethysmography, nasal and oral airflow by nasal cannula, and oral thermistor and finger oximetry. Scoring was performed by registered PSG technologists, using standards defined by the American Academy of Sleep Medicine (AASM). Per AASM scoring guidelines, hypopneas were defined as a ≥ 30% drop in signal amplitude for at least a 10-second duration and also associated with a ≥ 4% oxygen desaturation from baseline.18,19 Technologists set up and scored PSG tests while blinded to questionnaire data. We computed the AHI as the number of apneas plus hypopneas divided by hours of sleep time. s-OSA, our main case-definition of interest, was defined as an AHI of ≥ 30 events/h based on PSG.

Determining EKG-Derived Respiratory Power Index: A Surrogate for Apnea-Hypopnea Index

As briefly described previously,20 we integrated signals based on their estimated spectral power densities. We used the instantaneous respiratory rate to exclude noise in bands not relevant to respiration. We used the resulting integrated spectrum to estimate respiratory power, and from this, we calculated the respiratory power index (RPI), an estimate for the AHI.

To generate the initial estimations of the respiratory signal, we used prior algorithms, such as: amplitude of the R-Peak, P-Wave, and T-Wave, as well as the area of the QRS complex and the change in the heart rate, or respiratory sinus arrhythmia (RSA). We transformed each signal so that all signals show an identical average power. After averaging the calculated spectral power densities, the common respiratory component became dominant. Next, to remove the nonrespiratory parts of the spectrogram, we estimated the instantaneous respiratory frequency on a spectrum that is longitudinally rescaled on the frequency axis. This allowed us to filter out nonrespiratory components, and select the respiratory component. Summing the power density of this component over each time point provided an estimation of the respiratory power. We applied an adaptive20,21 filtering procedure to exclude artifacts and values related to ectopy.

For further details on this derivation, see Figure 1. The variables used in the EKG for calculating13 the HRV are summarized in Table 1.

Exemplary illustration of the methods behind the electrocardiographically derived respiratory power index (RPI).

The nighttime electrocardiograph recordings are preprocessed, including detection of fiduciary points and an estimation of signal-to-noise ratio (SNR), and are cut into segments of limited length (4–10 minutes) and high SNR (A). Segments of insufficient quality are discarded. Multiple embeddings of respiration into the ECG are used to derive electrocardiographically derived respiration (EDR) signals based on, e.g., the amplitudes of the peaks of the r-, p-, and t-wave (B), the respiratory sinus arrhythmia (RSA) (D). The spectrograms calculated from these signals (C,E) are normalized and averaged to amplify the common, ie, respiration based, component. In this case the RSA is not dominant as a source of modulation of the heart rate variability and shows mostly uncorrelated components that are reduced in the averaging process. The averaged spectrum is then analyzed to derive an estimate for the instantaneous respiratory frequency and masked to further reduce nonrespiration-related power in the spectrum (F). The power at each time-step is then calculated together with two levels used in the selection of events (G). The green line is the maximum level that determines the extent of an event if it lasts longer than a minimum amount of time and falls below the red level at least once. Time spans that indicate an event are underlaid blue on all time series. The amount of detected events, in relation to the analyzed time, is the RPI.

jcsm.15.1.23a.jpg

jcsm.15.1.23a.jpg
Figure 1

Exemplary illustration of the methods behind the electrocardiographically derived respiratory power index (RPI).

(more ...)

Description of variables used in electrocardiography.

jcsm.15.1.23.t01.jpg

table icon
Table 1

Description of variables used in electrocardiography.

(more ...)

Respiratory Power Index

The method used to calculate the RPI has so far only been described preliminarily as a conference contribution; a detailed study will follow.20 Each of the metrics in Table 1 was calculated as a weighted average over all EKG segments, the weights being the length of the EKG segment. Multiple independent embeddings of the respiratory activity on the EKG were used to calculate the RPI, which has been used as a surrogate marker for AHI. These embeddings have the physical and rotational properties of the electrical axis, due to respiratory movement as well as neural sources in the case of RSA. As outlined earlier, the RSA depends on the breathing rate. Information from all embeddings was then integrated and controlled for noise, which is often a reason for infrequent sampling. The combination of spectral power densities for these signals along with instantaneous respiratory rate allowed for exclusion of noise not relevant to respiration. The integrated spectrum was used to estimate respiratory power from which RPI could be computed. Further information on extracting RPI and distinguishing RPI signals from artifact is presented in the supplemental material.

Validation of this technique was performed using an independent PSG data set.22 This information is summarized in the supplemental material. For determination of AHI ≥ 30 events/h, the coefficient of determination was calculated as, R2 = 0.90 with P < 10−10. Using a threshold of RPI > 12.23 for the determination of OSA (AHI ≥ 30 events/h) in this dataset, the RPI would show sensitivity of 1 and specificity of 0.75. Similarly, for determination of AHI > 15 events/h, the coefficient of determination was calculated as, R2 = 0.74. Using a threshold of 10 for the determination of OSA (AHI > 15 events/h) in this dataset, the RPI would show sensitivity of 0.93 and specificity of 0.75.

Computation of Risk

Single-Stage Strategies to Determine Risk

We first determined risk scores for single-stage models. These models included assessments of BMI, age, ESS, NC, and all 13 aforementioned EKG-derived metrics (Table 1). In addition, we assessed the performance of a previously derived Multi-Variable Apnea Prediction (MVAP) score23 which combines BMI, age, sex, and the average responses (ranging from 0–4) on three specific OSA-symptom frequency questions: (1) snorting or gasping; (2) loud snoring; and (3) breathing stops, choking, or struggling for breath. In addition to the overall MVAP score, we examined the predictive ability of the symptom score average (Index 1) alone.

Two-Stage Strategies to Determine Risk

We employed a two-stage risk prediction strategy, as described previously.8,24,25

We categorized participants into high, intermediate, or low probability groups for risk of s-OSA based on first-stage risk scores (BMI, age, NC, ESS, MVAP Index 1, and MVAP). Using a previously described, iterative, optimization method, we chose an upper-bound value of each risk score to separate participants who could be at high risk for s-OSA from those at intermediate risk, and a lower-bound value to separate intermediate risk from low-risk participants.26 We predicted that participants with a risk score that was greater than the upper-bound cutoff point would have s-OSA, whereas participants with a risk score of less than the lower-bound cutoff point would be predicted not to have s-OSA. In our model, CMVOs falling between these ranges would be expected to undergo the second-stage test (EKG), and have the RPI value compared against a threshold RPI. Those with RPI values at or above the RPI threshold would be deemed to be at high risk for s-OSA, and those whose values were below the threshold would be predicted not to have OSA.

Computation of Model Parameters

We computed predictive characteristics of our single- or two-stage prediction models based on derived optimal model parameters. Single-stage models classified s-OSA or no s-OSA based on a single parameter, which was the optimal cutoff point for the variable being examined. The two-stage strategies had three parameters: the upper and lower bound for the first-stage test, and the threshold value of RPI in the second stage. We defined the optimal parameter set for each model by using an iterative algorithm that minimized the sum of false positives and twice the false negatives (ie, missed cases). We weighted missed cases more heavily because of the potential risk of drowsiness-related crashes in this group. In the case that multiple cutoff points met our selection criteria, we chose as the optimal parameter set the one that which also maximized specificity. Our method has been detailed previously.2426

We determined area under the receiver operating characteristic curve (AUC) for single-stage models using continuous cutoff points for each single-stage predictor. For two-stage models, we calculated the AUC using logistic regression models with s-OSA as an outcome. We utilized the optimal parameter set for each model to compute sensitivity, specificity, positive likelihood ratio (+LR), negative likelihood ratio (−LR), positive posttest probability (PPTP), and negative posttest probability (NPTP). To estimate the expected accuracy of our estimates, we calculated nonparametric confidence intervals (CI) based on 1,000 bootstrapped samples of the dataset. That is, we generated 1,000 random samples (chosen with replacement) of equal size to the original population and obtained a distribution of 1,000 predictive characteristics; the nonpara-metric 95% CI for each measure were calculated as the 2.5th to 97.5th percentiles of these distributions.

General Statistical Methods

Unless otherwise noted, we summarized continuous variables using mean and standard deviation (SD) and categorical variables using frequency and percent. We used a natural log or square-root transformation, when appropriate and if needed for continuous variables to meet parametric modeling assumptions, and compared between s-OSA and non-s-OSA groups using t tests. Categorical variables were compared between groups using chi-square or Fisher exact tests.

RESULTS

Sample Characteristics

Of the 95 participants who met our inclusion criteria and had EKG data available, 89 (93.7%) were male. We present demographic data in Table 2. The mean (SD) BMI, age, NC, and systolic and diastolic blood pressure were: 34.2 (8.2) kg/m2, 43 (8.5) years, 43 (3.9) cm, 133.2 (11.5) mmHg and 78.8 (14.6) mmHg, respectively. Most participants (70.5%) had obesity or were considered to have morbid obesity. These data are consistent with studies in prior truck driver cohorts5,8,27 and with established OSA risk factors.2830

Demographic characteristics of the sample, overall and stratified by AHI.

jcsm.15.1.23.t02.jpg

table icon
Table 2

Demographic characteristics of the sample, overall and stratified by AHI.

(more ...)

Compared to individuals with AHI < 30 events/h, participants with s-OSA were older, more likely to have obesity, had larger NC, higher diastolic blood pressure, and worse sleepiness (see Table 2). In this group, mean (SD) AHI was 53.8 (19.2) events/h, compared to 14.5 (8.2) events/h in the nons-OSA group. Participants with s-OSA had a higher RPI than those without s-OSA (24.3 [13.1] versus 11.5 [4.9] events/h; P < .0001), supporting the strong association between the RPI and AHI from PSG.

Electrocardiogram Derived RPI Metrics Are Associated With Severe OSA

We observed statistically significant or suggestive associations between select EKG-derived metrics and severe OSA (see Table 3). Specifically, CMVOs with s-OSA had significantly lower high frequency band to total power ratio (P = .0304) and average beat-to-beat interval (P = .017) compared to individuals with AHI < 30 events/h. In addition, s-OSA individuals had suggestively higher percentages of NN (normal to normal) beat-to-beat interval differences < 10 ms (P = .063) and occurrence of NN sequences < 10 ms (P = .070). As shown in Table 3, RPI was the most significantly associated metric with severe OSA; thus, this was chosen as our primary EKG-derived measure in our modeling strategies to identify s-OSA.

Summary statistics for electrocardiogram-derived respiration and heart rate variability characteristics.

jcsm.15.1.23.t03.jpg

table icon
Table 3

Summary statistics for electrocardiogram-derived respiration and heart rate variability characteristics.

(more ...)

Single-Stage Strategy: RPI Is Associated With Severe OSA

The discriminatory power of all single-stage models is shown in Table 4 and Table S1 in the supplemental material, along with optimal cutoff points, AUC, sensitivity, specificity, +LR, −LR, PPTP, and NPTP. When examining the AUC statistic, the RPI was the best predictor of s-OSA, followed by the MVAP score. Specifically, RPI had an overall AUC of 0.872 (95% CI: 0.791–0.942). The optimal cutoff point of RPI ≥ 12.23 yielded a high sensitivity of 88.5% (75.0% to 100.0%) and specificity of 63.8% (52.4% to 75.0%). The MVAP score was the second best performing model in terms of AUC (0.813 [0.702–0.910]) and showed similar sensitivity (87.5% [74.6% to 100%]) and specificity (63.6% [51.7% to 75.0%]) as the RPI, based on an optimal cutoff point of 0.616. The improved performance of both the RPI and MVAP when compared to other single-stage models (eg, BMI, age, ESS, NC, or MVAP Index 1) appears to be driven by higher specificity, as several of these single-stage models had higher sensitivity estimates, but lower AUC. When examining posttest probabilities, the RPI model showed a NPTP of 0.064 (0.00–0.146), suggesting only a 6.4% probability of s-OSA if the screening test predicts an absence of OSA. However, the PPTP was 0.479 (0.333, 0.627), suggesting that slightly < 50% of individuals screening positive will have s-OSA. This difference reflects the overweighting of missed cases when we chose our optimal cutoff point. Similarly, the MVAP showed an NPTP of 6.7% and a PPTP of 46.7% in our sample at the optimal cutoff point.

Predictive characteristics of single variable models with bootstrap confidence intervals.

jcsm.15.1.23.t04.jpg

table icon
Table 4

Predictive characteristics of single variable models with bootstrap confidence intervals.

(more ...)

EKG-derived metrics other than RPI had relatively low AUCs (ranging from 0.511 to 0.668), indicating that they are less useful for predicting s-OSA (see Table S1).

Two-Stage Strategy: RPI Increases Discriminatory Power for Predicting Severe OSA When Used In Tandem With Established Risk Factors

Table 5 shows the discriminatory power of RPI when used in tandem with the six variables known for their established association with OSA risk: BMI, age, NC, ESS, MVAP Index 1, and MVAP. Specifically, RPI was used as a second-stage screen among individuals predicted to have intermediate risk using the clinical measures as a first-stage screen.

Predictive characteristics of two-stage models with bootstrap confidence intervals.

jcsm.15.1.23.t05.jpg

table icon
Table 5

Predictive characteristics of two-stage models with bootstrap confidence intervals.

(more ...)

We observed that using the two-stage strategy improved the discriminatory power of each clinical single-stage model in predicting s-OSA. The best model used MVAP in the first stage, followed by RPI (cutoff point of 16 events/h) for the group with intermediate values of MVAP (scores between 0.22 and 0.87). The model yielded an AUC of 0.883 (95% CI: 0.767– 0.924), as well as high sensitivity (0.917 [0.706–0.962]) and specificity (0.848 [0.767–0.936]) in our sample. Importantly, the two-stage model involving the MVAP and RPI had a low NPTP, suggesting only 3.4% (1.5% to 13.3%) of individuals screening negative have s-OSA; PPTP was moderately high at 68.8% (51.9% to 84.2%). Although the combination of MVAP and RPI showed the highest AUC, this model was followed closely by the two-stage model using only BMI as a first-stage screen. Specifically, lower and upper bounds for BMI of 25 kg/m2 and 39 kg/m2, respectively, and an RPI of 16 events/h, yielded an AUC of 0.860 (0.787–0.924), 92.3% sensitivity, and 79.7% specificity, as well as a similar NPTP (3.5% [0.00% to 9.0%]) and slightly lower PPTP (63.2% [47.6% to 79.0%]) when compared to the MVAP two-stage model. Given that BMI is included as part of the MVAP, these results suggest that BMI may be a primary driver for the strong associations in the two-stage MVAP model.

DISCUSSION

This study had two important novel findings. First, a metric derived from spectral analysis of the EKG, which we called the RPI, had strong ability to discriminate s-OSA in a sample of CMVOs. This metric can be obtained easily using ambulatory EKG recordings to determine s-OSA risk. Second, although the RPI provided excellent discriminatory power to screen for s-OSA, this metric, when used in tandem with first-stage clinical measures of BMI alone or the MVAP score, also worked effectively.

Previous studies from our group12,27,31 and others5,32 have observed that despite a high prevalence of s-OSA in CMVOs and its association with sleepiness-related vehicular crashes,17,32 numerous challenges persist in screening and identifying this at-risk population. The use of EKG-derived signals, in particular, HRV, to classify sleep apnea has vastly improved over the past four decades. In particular, recent advances in the utility of cardiorespiratory coupled nonlinear signals in addition to clinical parameters have been shown to determine both sleep stages and sleep disorders.3338

Ours is the first study to investigate screening for s-OSA based on estimations of an EKG-based RPI in CMVOs. Further, to strengthen our screening tool we analyzed multiple metrics in stages to determine the best measure for development of a more robust model for screening for s-OSA in this high-risk population. Across our analyses, the optimal single-stage model was based on RPI alone, while the best performing two-stage models utilized BMI or MVAP as a first-stage screen, followed by RPI among those that had intermediate risk after the initial screen. We note that AUC values were similar across all three models (ranging from 0.860 to 0.883), as were the model sensitivities (ranging from 88.5% to 92.3%). Despite these similarities, importantly, the two-stage model has the added benefit of limiting the number of individuals who require more intensive EKG by utilizing more easily obtainable clinical measures as an initial screen; this would be expected to improve efficiency in both diagnosis and treatment, as well as reduce the costs associated with diagnosing OSA in CMVOs. In addition, we observed a 15% to 20% improvement in specificity for our two-stage models when compared to the single-stage RPI model, suggesting that the two-stage did a better job in correctly identifying CMVOs who did not have OSA.

Previous studies have examined time-dependent and spectral analysis of HRV in OSA and found significant associations with severity of OSA.35,3840 In a study of males with mild and severe OSA, after controlling for age and BMI, Aydin and colleagues showed that metrics described in Table 1 such as standard deviation of all NN intervals (SDNN) and high frequency (HF) were attenuated, whereas the ultra-low frequency (ULF), very low frequency (VLF), low frequency (LF), and the ratio of low-frequency and high-frequency powers (LF/HF) were increased during a 24-hour timeframe compared to patients without OSA.35 These findings were present in the presence of common comorbidities such as hypertension, which is prevalent in CMVOs.35 Roche and colleagues analyzed seven HRV metrics, and found that observed differences between daytime and nighttime values of SDNN index and the square root of the mean of the sum of the squares of differences between adjacent NN intervals were significant, independent predictors for OSA.38

Busek et al.39 observed that select HRV metrics varied by sleep stage. Total spectrum power, VLF, LF/HF ratio, and the LF spectral component were higher during rapid eye movement (REM) compared to stage N2 and stage N4 of nonrapid eye movement (NREM) sleep, whereas the HF spectral band was increased in stage N2 and N4 of NREM sleep. Whether HRV metrics, including RPI and severity of s-OSA, are influenced by the disrupted nature of sleep architecture and circadian dysregulation that may be prominent in evening/night shift work for CMVOs remain unknown. Another method, which combines oximetry with HRV indices,41 may be a potential screening tool as well, given that oximetry has also been shown to be an effective second-stage screening tool in CMVOs.8

Studies have shown that the most noticeable differences in HRV in patients with OSA are observed in stage N3/N4 sleep and REM sleep stages in human3,39,42 and animal models.43 Given the decrease in metabolic activity, stages N3 and N4 sleep are associated with a substantial decline in the VLF component of HRV.39,43 In participants with OSA, VLF values during wake did not differ from those during light sleep. This equalization of values (known as “non-dipping”44 during sleep) may be a consequence of OSA. After 3 months of continuous positive airway pressure (CPAP) therapy, the difference in VLF values between wake and light sleep was restored, resembling values seen in control participants without OSA.42 This supports the view that long-term CPAP treatment could be detected by EKG markers of parasympathetic activity.45

Our study had several strengths, including its prospective design and confirmation of all OSA cases with full PSG. The tool we propose, in contrast to PSG, is more amenable to mass screening because it is inexpensive and allows self-application. Moreover, EKG may be more acceptable to long-haul drivers, as it can be applied in the berth or at home, and can be dispensed and retrieved by mail, thus avoiding reliance on overnight sleep in a laboratory. Our study limitations include small sample size, which limits the ability to control for potential confounding variables, such as the use of beta-adrenergic blockers, which can restrict HRV. The RPI may not be suitable for use in such groups, nor in groups with significant comorbidities such as sick sinus syndrome, myocardial infarction, diabetic neuropathy, and cardiac failure. These conditions, however, may not pertain to many employed CMVOs, who must meet fitness-for-duty requirements mandated by the Federal Motor Carrier Safety Administration during a medical evaluation that must occur every 2 years.31,45,46 Thus, CMVOs with predetermined diseases such as uncontrolled diabetes, hypertension or heart disease are disqualified from obtaining a license.31,45 Nevertheless, any disease condition that may parlay a high sympathetic tone could change HRV and respiration. However, the RPI method utilized in this study includes only noise-free episodes, and a decreased RSA does not alter the estimated EKG-derived respiration (Figure 1). Another issue that could arise with our model is that of chain of custody, that is, ensuring that the intended participant is the one who actually wore the device. Solutions to chain-of-custody issues have already been introduced and their utility analyzed with respect to portable home sleep apnea testing devices; these solutions may also be applicable to EKG.11,47 Further, independent studies should evaluate the predictive characteristics and performance of optimal cutoff points described here within independent populations.

Future studies should also explore the use of HRV/RPI as a screening tool using in a prospective, randomized design and evaluate its cost-effectiveness in the diagnosis of s-OSA. Such an evaluation should account for not only the costs of screening, but also the cost savings through prevented crashes, and the costs related to crashes that may result from missed cases. Other at-risk groups should also be explored, such as CMVOs who engage in night-shift driving and may be at even higher risk for crashes.48,49

In conclusion, the model presented here provides an efficient and accurate paradigm for detecting severe OSA within a commercial driver population. This can improve screening and diagnosis, facilitate timely treatment, prevent negative health consequences to individual drivers and reduce the risk of sleepiness-related crashes. Additional prospective studies to assess this approach in the commercial driver and other populations are warranted.

DISCLOSURE STATEMENT

All authors have seen and approved the manuscript. MML is funded by NIH grant T32HL07713 and IG is funded by grant NIOSH/CDC R01-OH009149. JFK reports consulting for BIOTRONIK Vertriebs GmbH & Co. KG; BTK for Biomedical Statistical Consulting; NW for BIOTRONIK Vertriebs GmbH & Co. KG; IG for Embla Inc. TP reports grant/research support from Cidelec, France, Itamar, Israel, and Löwenstein Medical Technology; Speakers' Bureau for UCB Pharma. The other authors report no conflicts of interest.

ABBREVIATIONS

−LR

negative likelihood ratio

+LR

positive likelihood ratio

AASM

American Academy of Sleep Medicine

AHI

apnea-hypopnea index

AUC

area under the curve

BMI

body mass index

CI

confidence intervals

CMVO

commercial motor vehicle operator

CPAP

continuous positive airway pressure

EKG

electrocardiography

ESS

Epworth Sleepiness Scale

HF

high frequenc

HRV

heart rate variability

LF

low frequency

LF/HF

ratio of low-frequency and high-frequency powers

MVAP

Multi-Variable Apnea Prediction

NC

neck circumference

NPTP

negative posttest probability

NREM

non-rapid eye movement

OSA

obstructive sleep apnea

PPTP

positive posttest probability

PSG

polysomnography

REM

rapid eye movement

RPI

respiratory power index

RSA

respiratory sinus arrhythmia

s-OSA

severe OSA

SD

standard deviation

SDNN

standard deviation of all NN intervals

ULF

ultra-low frequency

VLF

very low frequency

REFERENCES

1 

Philip P. Sleepiness of occupational drivers. Ind Health. 2005;43(1):30–33. [PubMed]

2 

Radun I, Summala H. Sleep-related fatal vehicle accidents: characteristics of decisions made by multidisciplinary investigation teams. Sleep. 2004;27(2):224–227. [PubMed]

3 

Akerstedt T. Consensus statement: fatigue and accidents in transport operations. J Sleep Res. 2000;9(4):395[PubMed]

4 

Kales SN, Straubel MG. Obstructive sleep apnea in North American commercial drivers. Ind Health. 2014;52(1):13–24. [PubMed]

5 

Howard ME, Desai AV, Grunstein RR, et al. Sleepiness, sleep-disordered breathing, and accident risk factors in commercial vehicle drivers. Am J Respir Crit Care Med. 2004;170(9):1014–1021. [PubMed]

6 

Xie W, Chakrabarty S, Levine R, Johnson R, Talmage JB. Factors associated with obstructive sleep apnea among commercial motor vehicle drivers. J Occup Environ Med. 2011;53(2):169–173. [PubMed]

7 

Smith B, Phillips BA. Truckers drive their own assessment for obstructive sleep apnea: a collaborative approach to online self-assessment for obstructive sleep apnea. J Clin Sleep Med. 2011;7(3):241–245. [PubMed Central][PubMed]

8 

Gurubhagavatula I, Maislin G, Nkwuo JE, Pack AI. Occupational screening for obstructive sleep apnea in commercial drivers. Am J Respir Crit Care Med. 2004;170(4):371–376. [PubMed]

9 

Parks P, Durand G, Tsismenakis AJ, Vela-Bueno A, Kales S. Screening for obstructive sleep apnea during commercial driver medical examinations. J Occup Environ Med. 2009;51(3):275–282. [PubMed]

10 

Talmage JB, Hudson TB, Hegmann KT, Thiese MS. Consensus criteria for screening commercial drivers for obstructive sleep apnea: evidence of efficacy. J Occup Environ Med. 2008;50(3):324–329. [PubMed]

11 

Collop NA, Tracy SL, Kapur V, et al. Obstructive sleep apnea devices for out-of-center (OOC) testing: technology evaluation. J Clin Sleep Med. 2011;7(5):531–548. [PubMed Central][PubMed]

12 

Platt AB, Wick LC, Hurley S, et al. Hits and misses: screening commercial drivers for obstructive sleep apnea using guidelines recommended by a joint task force. J Occup Environ Med. 2013;55(9):1035–1040. [PubMed]

13 

Heart rate variability. Standards of measurement, physiological interpretation, and clinical use. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Eur Heart J. 1996;17(3):354–381. [PubMed]

14 

Bacharova L, Triantafyllou E, Vazaios C, Tomeckova I, Paranicova I, Tkacova R. The effect of obstructive sleep apnea on QRS complex morphology. J Electrocardiol. 2015;48(2):164–170. [PubMed]

15 

Bonsignore MR, Romano S, Marrone O, Insalaco G. Respiratory sinus arrhythmia during obstructive sleep apnoeas in humans. J Sleep Res. 1995;4 S1:68–70. [PubMed]

16 

Guilleminault C, Connolly S, Winkle R, Melvin K, Tilkian A. Cyclical variation of the heart rate in sleep apnoea syndrome. Mechanisms, and usefulness of 24 h electrocardiography as a screening technique. Lancet. 1984;1(8369):126–131. [PubMed]

17 

Tregear S, Reston J, Schoelles K, Phillips B. Obstructive sleep apnea and risk of motor vehicle crash: systematic review and meta-analysis. J Clin Sleep Med. 2009;5(6):573–581. [PubMed Central][PubMed]

18 

Rechtschaffen A, Kales A. A Manual of Standardized Terminology, Techniques and Scoring System For Sleep Stages of Human Subjects. Bethesda, MD: US Department of Health, Education, and Welfare, National Institutes of Health; 1968.

19 

Berry RB, Budhiraja R, Gottlieb DJ, et al. Rules for scoring respiratory events in sleep: update of the 2007 AASM Manual for the Scoring of Sleep and Associated Events. Deliberations of the Sleep Apnea Definitions Task Force of the American Academy of Sleep Medicine. J Clin Sleep Med. 2012;8(5):597–619. [PubMed Central][PubMed]

20 

Kraemer JF, Gurubhagavatula I, Penzel T, Wessel N. Estimation of electrocardiogram-derived respiratory power for apnea detection and continuous monitoring. J Crit Care. 2013;28(6):e33.

21 

Wessel N, Malberg H, Bauernschmitt R, Kurths J. Nonlinear methods of cardiovascular physics and their clinical applicability. Int J Bif Chaos. 2007;17:3325–3371.

22 

Bartels W, Buck D, Glos M, Fietze I, Penzel T. Definition and importance of autonomic arousal in patients with sleep disordered breathing. Sleep Med Clin. 2016;11(4):435–444. [PubMed]

23 

Gurubhagavatula I, Nkwuo JE, Maislin G, Pack AI. Estimated cost of crashes in commercial drivers supports screening and treatment of obstructive sleep apnea. Accid Anal Prev. 2008;40(1):104–115. [PubMed]

24 

Gurubhagavatula I, Fields BG, Morales CR, et al. Screening for severe obstructive sleep apnea syndrome in hypertensive outpatients. J Clin Hypertens (Greenwich). 2013;15(4):279–288.

25 

Morales CR, Hurley S, Wick LC, et al. In-home, self-assembled sleep studies are useful in diagnosing sleep apnea in the elderly. Sleep. 2012;35(11):1491–1501. [PubMed Central][PubMed]

26 

Gurubhagavatula I, Maislin G, Pack AI. An algorithm to stratify sleep apnea risk in a sleep disorders clinic population. Am J Respir Crit Care Med. 2001;164(10 Pt 1):1904–1909. [PubMed]

27 

Berger M, Varvarigou V, Rielly A, Czeisler CA, Malhotra A, Kales SN. Employer-mandated sleep apnea screening and diagnosis in commercial drivers. J Occup Environ Med. 2012;54(8):1017–1025. [PubMed Central][PubMed]

28 

Epstein LJ, Kristo D, Strollo PJ Jr, et al. Clinical guideline for the evaluation, management and long-term care of obstructive sleep apnea in adults. J Clin Sleep Med. 2009;5(3):263–276. [PubMed Central][PubMed]

29 

Hla KM, Young TB, Bidwell T, Palta M, Skatrud JB, Dempsey J. Sleep apnea and hypertension. A population-based study. Ann Intern Med. 1994;120(5):382–388. [PubMed]

30 

Young T, Peppard P, Palta M, et al. Population-based study of sleep-disordered breathing as a risk factor for hypertension. Arch Intern Med. 1997;157(15):1746–1752. [PubMed]

31 

Pack A, Dinges D, Maislin G. A Study of Prevalence of Sleep Apnea Among Commercial Truck Drivers. Publication no DOT-RT-02–030. Washington, DC: Federal Motor Carrier Safety Administration; 2002.

32 

Burks SV, Anderson JE, Bombyk M, et al. Nonadherence with employer-mandated sleep apnea treatment and increased risk of serious truck crashes. Sleep. 2016;39(5):967–975. [PubMed Central][PubMed]

33 

Penzel T, Glos M, Schobel C, Lal S, Fietze I. Estimating sleep disordered breathing based on heart rate analysis. Conf Proc IEEE Eng Med Biol Soc. 2013;2013:6571–6574. [PubMed]

34 

Stein PK, Pu Y. Heart rate variability, sleep and sleep disorders. Sleep Med Rev. 2012;16(1):47–66. [PubMed]

35 

Aydin M, Altin R, Ozeren A, Kart L, Bilge M, Unalacak M. Cardiac autonomic activity in obstructive sleep apnea: time-dependent and spectral analysis of heart rate variability using 24-hour Holter electrocardiograms. Tex Heart Inst J. 2004;31(2):132–136. [PubMed Central][PubMed]

36 

Hayano J, Watanabe E, Saito Y, et al. Screening for obstructive sleep apnea by cyclic variation of heart rate. Circ Arrhythm Electrophysiol. 2011;4(1):64–72. [PubMed]

37 

Ravelo-Garcia AG, Saavedra-Santana P, Julia-Serda G, et al. Symbolic dynamics marker of heart rate variability combined with clinical variables enhance obstructive sleep apnea screening. Chaos. 2014;24(2):024404.

38 

Roche F, Gaspoz JM, Court-Fortune I, et al. Screening of obstructive sleep apnea syndrome by heart rate variability analysis. Circulation. 1999;100(13):1411–1415. [PubMed]

39 

Busek P, Vanková J, Opavský J, Salinger J, Nevsímalová S. Spectral analysis of the heart rate variability in sleep. Physiol Res. 2005;54(4):369–376. [PubMed]

40 

Penzel T, Kantelhardt JW, Grote L, Peter JH, Bunde A. Comparison of detrended fluctuation analysis and spectral analysis for heart rate variability in sleep and sleep apnea. IEEE Trans Biomed Eng. 2003;50(10):1143–1151. [PubMed]

41 

Raymond B, Cayton RM, Chappell MJ. Combined index of heart rate variability and oximetry in screening for the sleep apnoea/hypopnoea syndrome. J Sleep Res. 2003;12(1):53–61. [PubMed]

42 

Gapelyuk A, Riedl M, Suhrbier A, et al. Cardiovascular regulation in different sleep stages in the obstructive sleep apnea syndrome. Biomed Tech (Berl). 2011;56(4):207–213

43 

Akselrod S, Gordon D, Madwed JB, Snidman NC, Shannon DC, Cohen RJ. Hemodynamic regulation: investigation by spectral analysis. Am J Physiol. 1985;249(4 Pt 2):H867–H875. [PubMed]

44 

Suzuki M, Guilleminault C, Otsuka K, Shiomi T. Blood pressure “dipping” and “non-dipping” in obstructive sleep apnea syndrome patients. Sleep. 1996;19(5):382–387. [PubMed]

45 

Hedner J, Darpo B, Ejnell H, Carlson J, Caidahl K. Reduction in sympathetic activity after long-term CPAP treatment in sleep apnoea: cardiovascular implications. Eur Respir J. 1995;8(2):222–229. [PubMed]

46 

Thiese MS, Moffitt G, Hanowski RJ, Kales SN, Porter RJ, Hegmann KT. Commercial driver medical examinations: prevalence of obesity, comorbidities, and certification outcomes. J Occup Environ Med. 2015;57(6):659–665. [PubMed Central][PubMed]

47 

Zhang C, Berger M, Malhotra A, Kales SN. Portable diagnostic devices for identifying obstructive sleep apnea among commercial motor vehicle drivers: considerations and unanswered questions. Sleep. 2012;35(11):1481–1489. [PubMed Central][PubMed]

48 

Lee ML, Howard ME, Horrey WJ, et al. High risk of near-crash driving events following night-shift work. Proc Natl Acad Sci U S A. 2016;113(1):176–181. [PubMed]

49 

Haraldsson PO, Akerstedt T. [Drowsiness--greater traffic hazard than alcohol. Causes, risks and treatment]. Lakartidningen. 2001;98(25):3018–3023. [PubMed]


Supplemental Material

Supplemental Material

(.pdf | 280 KB)