The dose-response effects of terbutaline on the variability, approximate entropy and fractal dimension of heart rate and blood pressure

Abstract

Aims

To study the dose-response effects of intravenous terbutaline on the cardiovascular and respiratory autonomic nervous regulation.

Methods

The study followed a randomized, placebo-controlled crossover design in six healthy adult volunteers. The terbutaline dose ranged from 10 to 30 μg min⁻¹. We continuously measured electrocardiogram, finger systolic arterial pressure (SAP) and flow-volume spirometry in supine and upright positions at baseline and during 3 h drug infusion. The periodic variability components of R-R intervals (time between successive heart beats) and SAP in relation to respiration were assessed using spectral analysis techniques. The regularity of the time series was assessed by approximate entropy (ApEn) and the convolutedness by fractal dimension (FD).

Results

Terbutaline dose-dependently decreased total variability of R-R intervals, low frequency (LF) variability of R-R intervals (10 s waves), high frequency (HF) variability of R-R intervals (respiratory variability), total variability of SAP, HF variability of SAP, baroreflex sensitivity, plasma potassium concentration, approximate entropy of R-R interval and of SAP as well as fractal dimension of R-R interval. Terbutaline dose-dependently increased heart rate, LF/HF ratios of R-R intervals and of SAP, LF variability of SAP, minute ventilation and plasma terbutaline concentration.

Conclusions

Terbutaline infusion decreases parasympathetic cardiovascular reactivity, baroreflex sensitivity, dimensionality of heart rate and plasma potassium concentration; it increases sympathetic dominance in cardiovascular autonomic balance, minute ventilation, and the regularity of heart rate and blood pressure time series.

Introduction

The adverse cardiovascular effects of β₂-adrenoceptor agonists have been discussed ever since their introduction. The heavy use of β₂-adrenoceptor agonists may cause circulatory disturbances by producing hypokalaemia, prolongation of the QTc interval and provoking sinus tachycardia, phenomena related to ventricular dysrhythmias [1]. An association has been found between the use of β₂-adrenoceptor agonists and the increased risk of fatal or near fatal asthma [2, 3]. Also, β₂-adrenoceptor agonists considerably interfere with cardiovascular and respiratory autonomic nervous regulation [4–6].

Quantitative data on various neurological, respiratory and cardiovascular diseases, as well as drug treatment which affects the sympathetic and parasympathetic control of cardiorespiratory organ functions can be obtained by spectral analysis of the beat-to-beat variability of R-R intervals and SAP as well as baroreflex sensitivity [7]. The spectrum displays relative densities of periodic components of signal variability on different frequency regions: HF variability at respiratory frequency is related to parasympathetic modulation and LF variability to both parasympathetic and sympathetic modulation.

The regularity of a signal, such as heart rate and SAP, can be assessed by ApEn and the convolutedness by FD [8–15]. The former measures the logarithmic likelyhood that a run of patterns that is close remains close on the next incremental comparison and the higher the entropy value, more the signal consists of random components of variability. FD measures the space-filling propensity of the signal. FD of a straight line is 1 and it approaches 2 for highly convoluted signals. An advantage of these methods is that the signals need not to be so stationary and sinusoidal as in the spectral analysis and these estimates may not be so strongly related to the mean and standard deviation of the signal [15].

The decreased heart rate variability and baroreflex sensitivity as well as increased regularity and the loss of dimensionality of heart rate have been associated with severely compromised physiological status and also mortality in humans [7, 10, 16–17]. However, to our knowledge there are no reports published on the dose-response effects of a β₂-adrenoceptor agonist medication on these variables. Therefore, we investigated the dose-response effects of intravenous terbutaline on the variability and complexity of R-R interval and arterial blood pressure signals in the cardiovascular system in healthy volunteers.

Methods

Test subjects

A total of six healthy male volunteers [mean age (s.d.) 24 (0) years, height 181 (6) cm and weight 76 (9) kg] participated in the study after informed written consent was obtained. The similar demography diminishes the physiological variation caused by age [18] or obesity [19]. Before commencing the study each subject underwent physical examination. No subjects with diabetes mellitus, cardiovascular, gastrointestinal, urinary tract, central or peripheral nervous system disease were included in the study. There was no medication for 2 weeks, no tea, coffee, cola-drink intake or heavy physical exercise for 24 h prior to the study or heavy meal within 2 h. The study protocol was approved by the ethics committee of the Turku University Hospital.

Study design

The study followed a randomized, placebo-controlled crossover design. The subjects visited the laboratory twice. The wash-out period was 2 weeks. The studies were started 16.00 h to avoid circadian variation [20].

Study drug

The infusion concentration was 5.0 mg terbutaline in 50 ml physiological saline. The control treatment was physiological saline (50 ml). The drug infusion was divided in three 60 min phases. The infusion rates were 6 ml h⁻¹ (= 10 μg terbutaline/min), 12 ml h⁻¹ (= 20 μg terbutaline min⁻¹) and 18 ml h⁻¹ (= 30 μg terbutaline min⁻¹) in each phase. The heart rate levels that were not exceeded in each phase were baseline+20 beats min⁻¹, +40 beats min⁻¹ and +60 beats min⁻¹. The protocol could be performed at designed rates in almost all subjects. However, because volunteers had different sensitivity to the heart rate increase, the infusion rate was reduced in a few phases for safety reasons. The measurement in the supine position was started 15 min after the beginning of each infusion phase followed by the measurement in the upright position and the stabilization of haemodynamics.

Data acquisition

First the subject rested supine for 15 min to stabilize haemodynamics before the cardiopulmonary signal recordings at baseline and in each infusion phase in the supine and standing position for 15 min. During the measurements the subject had a controlled breathing rate (0.25 Hz) paced using a sound signal given by a microcomputer sound generator. At the end of the baseline measurements and each infusion phase a venous blood sample was taken for plasma potassium (4 ml ammonium heparin tube) and terbutaline [10 ml EDTA (K3) tube] concentration measurements. The tubes were immediately placed on ice and centrifuged in a refrigerated centrifuge within 15 min of collection. Ammonium heparin tubes were stored at 4° C in refrigerator and analysed within 12 h and EDTA (K3) tubes were stored at −70° C prior to analysis. Potassium concentrations were calculated from all six test subjects and terbutaline levels from four test subjects.

A continuous precordial three-channel ECG was obtained via a cardioscope (Olli Monitor 432, Kone Instrument Division, Finland). A noninvasive arterial blood pressure signal was simultaneously recorded from the left middle finger kept at the heart level using an Ohmeda 2300 BP Monitor (Ohmeda, Inc., U.S.A.). A flow-volume spirometer signal (M901, Medikro Oy, Finland) was recorded to compute the respective pulmonary air flow and volume changes. The analogue outputs of the monitors were connected to an analogue interface (M9401 Transducer Interface, Medikro Oy, Finland) of an IBM PC/AT compatible microcomputer equipped with an extended cardiovascular autonomic function test software package (CAFTS, Medikro Oy, Finland) developed for computing heart rate, blood pressure, air flow and air volume variabilities. The signals were analogue-to-digital converted with a temporal resolution of 200 Hz. The programme utilizes a modified algorithm for QRS detection [21] followed by interpolation of R peaks by second order polynomial fitting. The R-R intervals were detected with a temporal resolution of 2 ms and then systolic, mean, diastolic arterial and pulse pressure were estimated for each R-R interval. Continuous signals of ECG, blood pressure and respiration were synchronized for time domain and spectral analysis. Blood pressure at the level of brachial artery was also measured before and after each cardiovascular measurement (Sphygmomanometer/printer BP-203 M II, Instrumentarium, Finland).

Analysis

The variabilities of R-R intervals and SAP as well as tidal volume and minute ventilation were computed for user-defined stationary data sets of 400 R-R intervals using the fast Fourier transformation method for spectral analysis (Figure 1). The total variability spectra (from 0.004 Hz to 0.40 Hz) for the R-R interval and SAP time series were computed after linear detrending of the signals. These variability spectra were integrated over frequency in two frequency bands (LF band from 0.04 Hz to 0.15 Hz and HF band from 0.15 Hz to 0.40 Hz). Sympathovagal balance was estimated by the computation of the LF/HF ratio for the R-R interval and SAP spectra. Baroreflex sensitivity was assessed utilising fast Fourier transformation-based cross-spectral analysis [22]. The parameter was defined as the square root of the ratio the variability of R-R intervals/the variability of SAP. The frequency bands chosen were those in which SAP changes occur prior to the respective R-R interval changes with a coherence coefficient >0.5 in the frequency band 0.04–0.20 Hz.

Figure 1.

R-R interval (RRI) time series and variability spectra during terbutaline and placebo infusion in supine position (30 μg terbutaline min⁻¹, Phase 3). Note especially the dramatically decreased low frequency (LF) and high frequency (HF) variability, approximate entropy (ApEn) and fractal dimension (FD) as well as the increased LF/HF ratio. PSD = power spectral density

The calculation of ApEn for the time-series of the R-R interval and SAP makes use of two input parameters, the dimension of the pseudo phase-space vectors formed from the original signal, and the filter parameter, i.e. the maximal distance of the vectors embedded in the multidimensional space [8–11]. The dimension parameter was fixed at 2 which appears to be the best choice in order to have maximal sensitivity to changes in the complexity of the signal. The filter parameter was chosen as 20% of the standard deviation of the signal. This ensured that the approximate entropy did not correlate strongly with the total variability of the source signals. Approximate entropy is a biased statistic: the estimate of this quantity increases asymptotically with the number (N) of the data points. Therefore N must be the same for each data set to ensure appropriate comparisons. We found that 800 sample points gave statistically meaningful results. These points were not pre-selected but simply the first 800 points of the time-series were used. The direct current component of the signal was removed before computing approximate entropy.

The FD of the R-R interval and SAP was calculated by the formula [12, 13]: FD = ln(N)/[ln(N)+ln(D/L)]. N is the number of the data points, L is the total length of the curve and D is the planar extent of the curve. If the lengths and distances of the R-R interval curve are measured in seconds both on the cumulative time axis and the interval axis, the fractal dimension will be very close to 1, especially with long data sets. Therefore we scaled the R-R intervals so that the basic measurement unit was 0.005 s. In the case of the SAP signal we used the pressure values in mmHg as a unit. These (arbitrary) scales naturally affected the absolute values of FD but we were only interested in the relative changes in FD. Fractal dimensions were computed using all the data points available (800–1600 points) since this variable is not dependent on the number of data points. The linear trend of the signal set was removed before the computations of ApEn and FD.

The plasma potassium levels were analysed by flame photometry in the Central Laboratory of Turku University Hospital (Turku, Finland) and plasma terbutaline levels by solid phase extraction (BondElut® CBA) with liquid chromatography mass spectrometry system in AstraDraco AB (Lund, Sweden).

The normal distribution of the variables was verified using the Shapiro-Wilks’ W test and homogenity of variances using the Levene's test. The differences between placebo and terbutaline treatments, carry-over effects, and period effects were analysed by t-test for independent samples [23] (Statistica for Windows, release 4.5, StatSoft Inc., USA). The variabilities were transformed to ten-base logarithmic values to normalise data distribution. Statistical significance was established at the level of P < 0.05.

Results

Heart rate and heart rate variability

The terbutaline infusion increased heart rate dose-dependently from 57 beats min⁻¹ to 109 beats min⁻¹ [terbutaline vs placebo, P < 0.00001, 95% confidence interval (95% CI) for mean treatment differences, 48 to 57] in the supine position and from 76 beats min⁻¹ to 126 beats min⁻¹ (P < 0.01, 31 to 70) in the standing position when compared with placebo treatment (Table 1 and Figure 2).

Table 1.

The effect of intravenous terbutaline on heart rate, R-R interval variability, baroreflex sensitivity as well as fractal dimension of R-R interval and systolic arterial pressure

Figure 2.

Heart rate (a), total variability (TV, b), low frequency variability (LFV, c), high frequency variability (HFV, d) and LF/HF ratio of R-R intervals (e) as well as baroreflex sensitivity (f) at baseline and in each terbutaline (▴) and placebo (▪) infusion phase in supine position. Data are expressed as mean (s.e. mean). Variabilities are expressed as ten-base logarithm values. The level of P was calculated for mean treatment difference (n = 6): *P < 0.05, **P < 0.01, ***P < 0.001 and *****P < 0.00001

The total variability of R-R intervals decreased dose-dependently from 3.96 ms² to 2.17 ms² (P < 0.001, −2.42 to −1.13), the LF variability of R-R intervals from 3.17 ms² to 1.73 ms² (P < 0.01, −2.08 to −0.80) and the HF variability of R-R intervals from 3.76 ms² to 1.68 ms² (P < 0.001, −2.48 to −1.68) after terbutaline infusion in the supine position (Figure 2). Terbutaline increased the LF/HF ratio of R-R intervals from 0.84 to 1.04 in the supine position (0.06 to 0.34, Figure 2). In the upright position the total variability of R-R intervals decreased from 3.45 ms² to 1.97 ms² (P < 0.01, −1.99 to −0.97), the LF variability of R-R intervals from 3.12 ms² to 1.66 ms² (P < 0.01, −2.12 to −0.80) and the HF variability of R-R intervals from 2.39 ms² to 1.13 ms² (P < 0.05, −2.14 to −0.38, Table 1).

Blood pressure and blood pressure variability

Blood pressure at the level of brachial artery did not change significantly during the terbutaline infusion (Table 2). However, the variability indices of SAP changed significantly in the supine position. The LF variability of SAP increased from 2.06 mmHg² to 2.49 mmHg² (P < 0.05, 0.06 to 0.80). The HF variability of SAP first increased from 1.85 mmHg² to 1.99 mmHg² (P < 0.01, 0.07 to 0.21) then decreased from 1.84 mmHg² to 1.43 mmHg² (P < 0.01, −0.67 to −0.15). Respectively the LF/HF ratio of SAP increased from 1.10 mmHg² to 1.83 mmHg² (P < 0.001, 0.55 to 0.91) after the terbutaline infusion. No significant changes were found in the absolute level of blood pressure or in the variability indices of SAP in the upright position (data not shown).

Table 2.

The effect of intravenous terbutaline on the blood pressure at the level of brachial artery and on the finger systolic arterial pressure variability in supine position

Baroreflex sensitivity

The baroreflex sensitivity decreased dose-dependently from 25.0 ms/mmHg to 2.4 ms/mmHg (P < 0.001, −29.7 to −15.5) in the supine position (Fig, 2) and from 7.8 ms/mmHg to 1.2 ms/mmHg (P < 0.01, −9.3 to −3.9) in the upright position (Table 1).

Approximate entropy and fractal dimension

ApEn of R-R intervals decreased dose-dependently from 1.41 to 1.09 (−0.50 to −0.14, P < 0.01) in the supine position and from 1.21 to 0.73 (−0.54 to −0.04, P < 0.05) in the upright position (Figure 3). Respectively, ApEn of SAP decreased dose-dependently from 1.24 to 0.76 (−0.71 to −0.25, P < 0.01) in the supine position and from 1.20 to 0.96 (−0.38 to −0.10, P < 0.05) in the upright position.

Figure 3.

Approximate entropy of R-R interval and of SAP in supine (a and b) and upright position (c and d) at baseline and in each terbutaline (▴) and placebo (▪) infusion phase. Data are expressed as mean (s.e. mean). The level of P was calculated for mean treatment difference terbutaline vs placebo (n = 6, except phase 3 in upright position n = 5): *P < 0.05 and **P < 0.01

FD of R-R interval signal also decreased from 1.73 to 1.30 (−0.55 to −0.31, P < 0.001) in the supine position and from 1.34 to 1.17 (−0.29 to −0.05, P < 0.05) in the upright position (Table 1). The fractal dimension of SAP did not change significantly.

Ventilation

The minute ventilation increased in the supine position (terbutaline vs placebo, mean±s.e. mean (l min⁻¹): baseline, 11.1±1.6 vs 10.2±0.9; Phase 1, 11.3±1.5 vs 10.1±1.5; Phase 2, 12.3±1.1 vs 11.0±2.1 (all insignificant); Phase 3, 13.1±1.2 vs 10.4 1.3 (P < 0.01, 1.9 to 3.7). The increase was not significant in the upright position (respectively, 10.6±0.9 vs 12.3±1.2; 12.2±1.1 vs 12.2±1.4; 12.5±1.0 vs 11.5±0.9; 13.2±1.4 vs 11.5±0.7).

Plasma potassium and terbutaline concentrations

The plasma terbutaline concentration increased dose-dependently (respectively, (nmol l⁻¹): <4.0 vs <4.0; 52.8±1.1 vs <4.0 (P < 0.001); 118.3±5.8 vs <4.0 (P < 0.001); 155.3±7.1 vs <4.0 (P < 0.001, 132.5 to 170.1). Concentrations <4.0 nmol l⁻¹ could not be detected. The plasma potassium concentration decreased (respectively, (mmol l⁻¹): 4.0±0.1 vs 4.2±0.1 (non significant); 3.1±0.1 vs 4.1±0.1 (P < 0.001); 2.6±0.1 vs 3.9±0.1 (P < 0.0001); 2.5±0.1 vs 4.0±0.1 (P < 0.001, −1.9 to −1.0).

Adverse effects

The symptomatic adverse effects reported during the terbutaline infusion were tremor and restlessness (severe), palpitation and weakness (moderate) as well as peripheral vasodilatation, headache, lack of concentration in paced breathing, heavyness in calves and nose bleeding (mild). Two subjects had cardiac arrhythmias: one had two ventricular premature beats in the upright position in the Phase 3, the other had increasing numbers of both atrial and ventricular premature beats (Phase 2 upright position, 35 atrial and 2 ventricular; Phase 3 supine position, 31 ventricular; Phase 3 upright position, 75 atrial and 17 ventricular during the 15 min measurement). Two subjects also had vasovagal reactions (i.e. hypotension, dizziness and cold sweat) in upright position. One experienced this during the first baseline measurement and the other during the final measurement with terbutaline infusion. The latter subject did not complete the measurement.

Discussion

Our study demonstrated that intravenous terbutaline dose-dependently increased heart rate, minute ventilation, LF variability of SAP, LF/HF ratios of R-R interval and of SAP as well as plasma terbutaline concentration. The drug decreased the total, LF and HF variability of R-R intervals, baroreflex sensitivity, ApEn, FD and serum potassium level.

The terbutaline infusion markedly decreased the variability indices of R-R intervals which are related to vagal tone. The decrease in the total and especially in the HF (respiratory) variability reflects the attenuation of respiratory sinus arrhythmia and evidence of a vagal withdrawal [24–26]. The periodic changes in the vagal activity have been suggested to be generated by several ractors: baroreflex, Bainbridge reflex, lung stretch reflexes and central cardiorespiratory coupling [27]. We considered the drug effect marked, because there was such a strong increase in heart rate, and because it was also evident in the upright position, which is physiologically dominated by increased sympathetic activity. The reduction in the HF variability of R-R intervals was likely to cause the simultaneous diminution in the HF variability of SAP [24]. The decrease in the HF variability of SAP in our study cannot be produced by a decrease in ventilatory activity [28], since both tidal volume and minute ventilation increased during the terbutaline infusion.

The LF variability of R-R intervals and SAP (10 s waves) may be a consequence of the baroreflex mediated control of blood pressure and reflect peripheral vasomotor tone [24–26, 29]. These are considered to be indices of cardiovascular sympathetic modulation although they are also affected by cardiovagal modulation. The increased LF variability of SAP is evidence for more clearly adrenergic modulation than the respective variability of R-R intervals because of the profound adrenergic control of arterioles [30]. The decrease in the LF variability of R-R intervals is most likely due to vagal withdrawal. The LF/HF ratio of R-R intervals has been proposed as an index of sympathovagal balance e.g. it has been shown to increase after passive orthostatic tilt and the increase appears to be reduced after β-receptor blockade in humans [31, 32]. Both LF/HF ratios for R-R intervals and SAP increased during the terbutaline infusion suggesting increased sympethetic dominance in the cardiovascular autonomic balance. The separation of sympathetic and parasympathetic influences in the spectrum can be explained by the the response time of the parasympathetic nervous system (100–300 ms) which is much shorter than that of the sympathetic nervous system (2.5–7.5 s for β- and 5–15 s for α-adrenergic mechanisms) [29].

The arterial baroreflex is an important negative feed-back system in readjusting acute deviations in blood pressure to its physiological set point. Increased blood pressure activates vagal and inhibits adrenergic mechanisms which decrease heart rate and total peripheral vascular resistance [29]. When blood pressure is below the target level baroreflex stimulation is attenuated preventing furher decrease. The baroreflex sensitivity estimated by the cross-spectral analysis decreased dose-dependently during the terbutaline infusion. The attenuated baroreflex sensitivity was not caused by decreased blood pressure although it may have prevented the decrease in blood pressure. The significant differences in the variability indices of R-R interval and SAP as well as in baroreflex sensitivity started to appear in the second Phase of the infusion where heart rate was 35 beats min⁻¹ above baseline and plasma terbutaline concentration 118 nmol l⁻¹. The findings described above are supported by our previous studies on the effects of inhaled salbutamol in asthmatic children [5].

The cardiovascular effects were expected because terbutaline is an effective and selective β₂-adrenoceptor agonist [33]. The effects are produced directly by the binding of terbutaline on β₂-adrenoceptors on and remote from adrenergic nerve terminals but also indirectly by enhancing norepinephrine release after activation of presynaptic β₂-receptors [34, 35]. In addition, norepinephrine has been suggested to inhibit acetylcholine release from postganglionic cholinergic nerve terminals in canine tracheal smooth muscle [36]. This is probably mediated by presynaptic β-receptors and may in part explain the decreased cardiovagal effects. Very small amounts of terbutaline have also been found to penetrate the blood–brain barrier in rabbits suggesting that the stimulation of central β-adrenoceptors may account for some of the drug effects [37]. The tachycardia after the terbutaline infusion has been suggested to be caused by the direct effects on heart rather than systemic effects e.g. decreased blood pressure [38]. Increased heart rate, whether induced by β₂-adrenergic stimulation or atropine, has associated with decreased LF and HF variabilities of R-R intervals and arterial blood pressure [5, 24]. Therefore, our findings may not be specific for β₂-adrenergic stimulation. However, no direct comparison between the effects of β₂-adrenergic stimulation and cholinergic blockade on the cardiovascular variability indices has yet been reported to our knowledge. The marked hypokalaemia induced by terbutaline most likely contributes to the tachycardia and decreased variability cardiovascular indices by increasing the slope and amplitude of action potential in sinoatrial and atrioventricular nodes.

The variability of heart rate and blood pressure is produced by complex interactions of multiple physiological control mechanisms. It has been hypothesized on the basis of clinical observations that the loss of complexity in healthy organ system may lead to a seriously impaired ability to adapt to physiologic stress [10, 14]. The decreased ApEn reflects dramatically increased regularity in the cardiovascular signals in our test subjects [8–11]. ApEn has been shown to be closely related to the HF variability of R-R intervals and therefore is also considered to reflect vagal modulation heart rate [15]. This may suggest that the regularity in the cardiovascular signals increases as the fast and possibly irregular vagal modulation decreases and the more sinusoidal adrenergic modulation increases. The approximate entropy of R-R interval and of SAP clearly dose-dependently decreased except in the last phase in the supine position. We speculate that the overall blood pressure control system may have reached the most regular state where almost all irregular feedback loops were heavily damped.

Very similar kind of behaviour could be found in FD of R-R interval. Decreased FD reflects decreased dimensionality and convolutedness of the heart rate time series suggesting a smaller number of independent dynamical variables affecting heart rate [12, 13]. However, FD of SAP did not depend on the terbutaline dose. In addition, the baseline of FD of SAP was on a higher level in the upright than in supine position. This is a totally reverse effect when compared with the respective value for R-R interval. Previously it has been assumed that ApEn and FD of R-R interval can be used interchangeably, because of their close correlation with each other [15]. Our results indicate that this is not appropriate in the case of SAP. The unchanged FD of SAP suggests more stable regulatory processes in blood pressure which are not so easily damped by terbutaline when compared with the effects on the heart rate.

In the spectral analyses with fast Fourier transformation a periodic continuous function of time is represented as a summation of a series of sine and cosine waves [39]. However, heart rate and SAP signals may not necessarily display such regular periodicies, especially under external stress. Therefore ApEn and FD, which do not depend strongly on stationarity, sinusoidal periodicity, mean and standard deviation of the signals, may provide important additional information of the cardiovascular signal regulation [15].

In conclusion, an intravenous terbutaline infusion decreases markedly and dose-dependently cardiovagal nervous reactivity. It also increases sympathetic dominance in the cardiovascular autonomic balance and decreases baroreflex sensitivity. The decreased blood plasma potassium level as well as the changes in the variability indices and baroreflex sensitivity demonstrated in this study are similar to the risk factors known to predipose to malignant ventricular tachyarrhythmias and to increase susceptibility to sudden death in adults with heart disease [7, 16, 17]. In addition to spectral estimates of periodic variability, approximate entropy and fractal dimension may provide useful complexity estimates in predicting adverse drug effects, although further trials are needed to understand better their application to clinical medicine.