Introduction

Psychophysiological states are dynamic processes that vary to individual adaptation to the demands from the internal and external environments. These processes can be reflected in different physiological variables, such as cardiac activity. The length of the heart inter-beat intervals (IBI) are not constant, as it changes according to the behavioral and physiological conditions of individuals. The study of the variations of these intervals is performed by means of the heart rate variability (HRV) analysis, which can provide information about psychophysiological states present in individuals under different circumstances (^{Billman, 2011}). The HRV analysis can be addressed through different methods that are classified into time, frequency, and nonlinear domains (^{Task Force, 1996}). It has been established that some methods provide information on the activity of the autonomic nervous system (ANS) under different physiological or behavioral conditions (^{Allen, Chambers, & Towers, 2007}). In order to perform these analyses, it is required to obtain a measure of the IBI; by means of different methods using per example the electrocardiogram (EKG), whose intervals between R waves (RRI) are calculated (Figure 1A). Usually, RRIs are processed in two different ways; the first one consists in analyzing them directly, by means of their averages or standard deviations. The second way is to consider the RRIs as time series, each RRI being a value in the data sequence (Figure 1B). To represent them as time series these data are re-sampled at a fixed rate, usually using the interpolation method (^{Task Force, 1996}). However, there are methods that process the data without re-sampling (^{Laguna, Moody, & Mark, 1998}).

In the field of psychophysiology, HRV methods that provide measures of the level of activation of the ANS are widely used (^{Bernston et al., 1997}; ^{Laborde, Mosley, & Thayer, 2017}). These measures have revealed differences between the Non-REM (NREM) and REM sleep (^{Lanfranchi, Pépin, & Somers, 2017}; ^{Mikoteit, De Witte, Holsboer-Trachsler, Hatzinger, Beck, & Pawlowski, 2019}; ^{Stein & Pu, 2012}; ^{Tobaldini, Nobili, Strada, Casali, Braghiroli, & Montano, 2013}). Likewise, it has been also described the presence of HRV autonomic differences between light and deep NREM sleep (^{Ako, Kawara, & Uchida, 2003}; ^{Toscani, Gangemi, & Silipo, 1996}; ^{Trinder et al., 2001}). However, NREM sleep division into two sleep kinds is not a standard procedure in the assessment of HRV.

There are other HRV measures that are also related to other phenomena, such as physiological variability, which has been proposed as a health indicator (^{Goldberger, 1997}). Since NREM sleep is a heterogeneous phenomenon consisting of different phases, it is important to know which HRV measures also manifest this heterogeneity. Therefore, the objective of this study was to analyze and to compare the HRV during light and deep NREM sleep of healthy individuals.

Figure 1: A) It represents the extraction of the intervals between each R wave (RRI) of the EKG. B) Representation as time series of the RRI obtained from the EKG, in the abscissa axis each RRI is placed and in the ordinate axis the time in ms between each R wave.  

Method

Results

A total of 435 blocks of 5 minutes of cardiac activity were collected, 162 corresponded to light sleep and 273 to deep sleep. Total sleep time observed during the 8 hr recording was 457.5 minutes (SD 37.7). The percentages of each sleep phase were for N1 of 6.9 (SD 2.7), N2 of 52.8 (SD 7), N3 of 19.7 (SD 5.8) and REM sleep of 20.6 (SD 3.9).

In relation to cardiac activity, when comparing the medians of each measure (Table 1), it was found that in the measures belonging to the non-linear domain higher values predominated during deep sleep (with the exception of α1 and SD2). Comparing the medians of each measure (Table 1), it was found that in the measures belonging to the non-linear domain higher values predominated during deep sleep (with the exception of α1 and SD2). On the other hand, three of the five time domain measures (HR, SDNN and logHRV) had higher values in light sleep. After the Holm-Bonferroni correction at the significance level, significant differences were found in 4 of the measures (Table 2), where median values were higher in light sleep.

Table 1: Median HRV measurements during light and deep sleep 

Table 2: Significant differences between light and deep sleep 

Discussion

The measures in which significant differences were found correspond to indicators of general variability of cardiac functioning as described by other authors (^{Allen, Chambers, & Towers, 2007}). In the case of the SDNN and logHRV measures, significant differences were observed between both types of sleep, with greater variability evident in light sleep. The above suggests that there are functional processes during light sleep reflexed as cardiac activity less stable, for example it has been reported that the presence of K complexes modifies blood pressure and cardiac activity (^{Silvani, 2008}). However, the RMSSD measure, which also evaluates variability, showed higher values during deep sleep without reaching significant levels.

The procedure followed for the calculation of the RMSSD is different from that of SDNN and logHRV (^{Shaffer & Ginsberg, 2017}), while the latter only differ in the way they represent the IBI values. This indicates that, of the three measures used within the time domain, there are at least two different ways to characterize the variability associated with cardiac functioning.

In the case of the non-linear domain measures related to variability, we found that SD2, from Poincaré's graph, showed significant differences between both sleep stages. This measure is also considered to reflect total variability and has been related to SDNN (^{Brennan, Palaniswami, & Kamen 2001}). According to ^{Goldberger (1997}) another way to characterize cardiac variability is to evaluate its fractal characteristics such as self-similarity (when a part of the object is similar to the whole object). In this case, the DFA analysis quantifies fractal correlation properties of IBI series. In this study we found that the measure α1, which represents short-term fluctuations, was greater in light sleep. P^{enzel, Kantelhardt, Becker, Peter and Bunde (2003}) using discriminant analysis described that the measures α1 and α2 were effective in differentiating sleep stages. These authors, in their analysis, included stage N1 within light sleep, in addition to adding REM sleep and wakefulness. The same authors observed that by including SDNN in the discriminant analysis, phase separation was more effective. In our study, we did not include phase N1 because few participants presented it for more than 5 minutes, which is common in this stage of NREM sleep (^{Carskadon & Dement, 2017}). In addition, some authors propose that N1 is a transitional stage between wakefulness and sleep (^{Ogilvie, 2001}).

Because the regulation of the cardiovascular system involves different mechanisms that interact with each other in a non-linear way (^{Ernst, 2017}), the use of non-linear methods allow us to know characteristics of HRV that are not obtained with linear methods (^{Young & Benton, 2015}). Entropy measures quantify the regularity and predictability of a system (^{Richman & Moorman, 2000}), in this case no differences were found between both types of sleep with the two entropy measures analyzed (ApEn, SamEn). No significant differences were found in measure D2 either, this measure provides information on the minimum number of variables required to model the underlying system (^{Acharya, Joseph, Kannathal, Lim, & Suri, 2006}).

With the measures used in this work it can be concluded that the differentiation between light and deep sleep depends on the type of analysis, based on the measures associated with cardiac variability, the division of NREM sleep could be justified in terms of cardiovascular activity. One factor that could contribute to this difference was the presence of compensatory rebound from deep sleep, although the sleep percentages of the second PSG were within normal parameters (^{Carskadon & Dement, 2017}), in future studies it would be advisable to perform more than two PSG to compare HRV during light and deep sleep. Additionally, studies with larger samples are required for better characterization of light and deep sleep, as well as the inclusion of HRV measures that allow inference of ANS activity.