Blood loss is common during surgery; however, severe anemia is usually treated with transfusion of red blood cells. Whether transfusion is carried out, normovolemia is mandatory at all times. Electrocardiogram (ECG) monitoring is implemented as a fixed element in the perioperative period.
Electrocardiogram is easily accessible and noninvasive. Clinically, the ECG is used to monitor the heart rate (HR) and the occurrence of arrhythmias. Furthermore, ECG signals provide information regarding ischemic conditions of the heart. In contrast to myocardial ischemia and myocardial infarction, the depression or elevation of the ST segment as a marker pointing toward acute myocardial tissue hypoxia has been shown to be of little sensitivity and specificity during acute isovolemic anemia (1). However, within the last decades, clinical outcome studies revealed the prognostic value of other subtle changes in the ECG with respect to mortality (2-8). For example, prolongation of the QT interval and depression of the ST segment, as well as changes in HR variability and T-wave alternans, occur in chronic diseases with an increase in mortality (9-12).
It is unknown whether similar ECG changes also occur during acute normovolemic anemia and may indicate a critical restriction of oxygen transport, tissue oxygenation, and by that imminent tissue hypoxia. One might speculate that during progressing anemia, typical, subtle ECG changes appear before the individual critical hemoglobin (Hb) concentration has occurred.
We hypothesized that a low Hb concentration (Hbcrit) during normovolemia may cause alterations of the myocardial electrical current in a concentration-dependent fashion. Therefore, we investigated the impact of gradually increasing normovolemic anemia in an experimental setting on the descriptors of single-heartbeat traces of the ECG.
MATERIALS AND METHODS
After governmental approval, data were collected from eight healthy pigs (Deutsches Edelschwein) of either sex (female, n = 4; male, n = 4), weighing between 25 and 29 kg (mean body weight, 26.4 ± 1.2 kg). Animals were treated in accordance with the Principles of Laboratory Animal Care (National Institutes of Health publication 86-23, 1985).
Food was withheld for one night, but there was free access to water. Intramuscular premedication was performed using midazolam (2 mg·kg−1) and ketamine (10 mg·kg−1). Anesthesia was induced by intravenous injection of fentanyl (0.01 mg·kg−1), propofol (4 mg·kg−1), and vecuronium bromide (0.3 mg·kg−1) and maintained by continuous infusion of fentanyl (0.02 mg·kg−1· h−1) and propofol (20 mg·kg−1·h−1). Muscular relaxation was maintained by continuous infusion of vecuronium bromide (1 mg·kg−1·h−1) to minimize any potential disturbances of the ECG recordings.
Estimated insensible fluid losses other than those of blood withdrawal (e.g., perspiration, etc.) were compensated by intravenous infusion of Ringer's solution (3 mL·kg−1·h−1). A warming unit (Mallinckrodt WarmTouch 5200; Mallinckrodt Medical, Hazelwood, Mo) was used to maintain the body temperature constant at 36°C ± 0.5°C. Pigs were endotracheally intubated and mechanically ventilated at a rate of 12 min−1 and a positive end-expiratory pressure of 5 cm H2O (intermittent positive pressure ventilation). Tidal volume was adjusted to maintain normocapnia (36-40 mmHg), which was confirmed on the basis of multiple blood gas analyses.
Instrumentation and monitoring
All animals were placed in supine position. Several catheters were inserted using Seldinger technique to minimize blood loss. A 20-gauge catheter was placed into the left femoral artery for continuous arterial pressure recording and blood withdrawal during the hemodilution procedure. A Picco thermodilution catheter (PICCO Pulsiocath; PULSION Medical Systems AG, Munich, Germany) was placed into the right femoral artery for continuous measurement of cardiac output. A 16-gauge catheter was inserted through the left external jugular vein into the upper vena cava for infusion of hydroxyethyl starch (6% HES, 200,000/0.5; Braun, Melsungen, Germany) and for monitoring of central venous pressure (CVP). A pulmonary artery catheter (7.5F, Edwards Swan-Ganz; Baxter Healthcare, Irvine, Calif) was inserted for sampling of mixed venous blood and for monitoring of pulmonary artery pressure. Position of all catheters was verified by blood gas analysis and radioscopy.
A blood gas analyzer (Instrumentation Laboratory, Lexington, Ky) was used to determine arterial, mixed venous, and central venous PO2, PCO2, and pH. Hemoglobin, arterial oxygen saturation, and mixed venous oxygen saturation were measured by spectrophotometry adjusted to pig Hb (682 CO-Oximeter; Instrumentation Laboratory).
After completion of instrumentation, a 60-min stabilization period was allowed to achieve baseline conditions. Baseline data were recorded, and subsequently, acute normovolemic anemia was induced by simultaneous replacement of blood with HES (6% HES, 200,000/0.5) at a rate of 1 mL·kg−1·min−1 until the individual critical Hbcrit was achieved.
At Hbcrit, the compensatory mechanisms of the organism (increase in cardiac output and peripheral oxygen extraction) become exhausted, and the organism is endangered by general tissue hypoxia. It has been demonstrated that at this time point oxygen delivery is insufficient to fulfill oxygen demands of the body. As a consequence, less oxygen than actually needed can be consumed. Thus, Hbcrit can be detected as a decline in previously stable V˙O2 values measured by indirect calorimetry or by means of a pulmonary artery catheter.
Detection of Hbcrit was automated with computer software particularly designed for this purpose in animal experiments (DeltaCrit System, programmed by J.M.) (13). Oxygen consumption values (V˙O2) collected by indirect calorimetry during a stable 60-min observation period were included in an online regression analysis. Every V˙O2 value measured during the subsequent hemodilution process was compared with the mean value predicted by the DeltaCrit system: if the actual value was outside a predefined range (3 * SD of regression line), a significant decrease in V˙O2 was assumed, and the computer alerted visually and acoustically. The hemodilution procedure was interrupted for collection of a complete data set after exchange of 10% of blood volume, respectively.
Six-channel ECG was recorded with a sampling rate of 500 Hz via adhesive skin electrodes (Kendall Arbo; Tyco Healthcare, Neustadt, Germany) at electrode positions I, II, III, aVL, aVF, and V5 (Fig. 1A) and stored directly on a PC notebook with a commercial hardware interface (Cardiax v.3.41.1; Mesa Medizintechnik GmbH, Benediktbeuern, Germany).
Electrocardiographic data of five episodes were used for calculations. These episodes were defined post hoc when 0% (baseline), 30%, 50%, 80%, and 100% exchangeable blood volume (i.e., blood volume exchanged until Hbcrit is reached) had been exchanged (Fig. 1B). The duration of these episodes were 5 min. These dilutional steps corresponded to the following mean Hbcrit values (95% confidence intervals): baseline, 9.5 (8.2/10.7) g·dL−1; 30%, 8.0 (5.7/9.2) g·dL−1; 50%, 5.5 (3.8/6.2) g·dL−1; 80%, 3.8 (2.9/4.8); and 100%, 3.3 (1.9/3.7) g·dL−1 and are depicted in figures and tables as Hb 9.5 g·dL−1 (baseline), 8.0 g·dL−1, 5.5 g·dL−1, 3.8 g·dL−1, and 3.3 g·dL−1 (Hbcrit) for clarity.
We used a wavelet bandpass-filtering method (Daubechies 4 wavelet, 0.03-50 Hz, 8 dB) for all channels at each episode (Fig. 1C). S points of each QRS complex were automatically classified (Nevrokard, Nevrokard Kiauta, Slovenia), and data were cut such that for each heartbeat data pieces started at 300 ms before the S point and lasted until 500 ms after the S point (Fig. 1D).
Data pieces were visually controlled for ectopic beats or artifacts. We used data only when at least 150 consecutive heartbeats (representing a period of 1-1.5 min, depending on the actual HR of the pig) were undistorted by visually identifiable artifacts. For each of these single-heartbeat traces, we determined the latencies of the time points P, Q, S, and the end of the T wave as well as the maximum amplitude of the T wave with a software routine programmed by the authors in MATLAB (MATLAB; The MathWorks, Natick, Mass). To determine the end of the T wave, we fitted a tangent in the maximal downslope of the T wave using a polynomial second order. The crossing of the fitted tangent with the baseline, determined as the mean from 20 data points before the beginning of the P wave, was defined as the end of the T wave (14).
A summarizing statistical analysis including all channels and all animals was performed via a time-frequency analysis on each of these single heartbeats. A fast Fourier analysis based on a window of 20 ms at shifting bandpass-filtered signals (1-Hz bandwidth) was used to estimate the absolute amplitude (square root of the power) and the instantaneous phase of each time-frequency bin. For each episode, the mean of phase and absolute amplitude was calculated across heartbeats at each time-frequency bin. To answer the question whether there is a significant difference between the episodes, we hypothesized that the difference between baseline and episodes is based on statistical variance, which means that all ECG heartbeats are drawn from the same distribution independent of the episode.
Hemodynamic data (cardiac index [CI], CVP, mean arterial pressure [MAP], etc.) were analyzed with the Friedman test (Scheffé corrected for multiple comparison, test level P = 0.01). To quantify the mean effect across all animals, we performed a robust regression analysis. The slope of the regression line served as an estimate for quantifying mean changes. For the results of the time-frequency analysis, we implemented a permutation test (for details, see Appendix).
Hemodynamic data are presented in Table 1 for the five episodes where Hbcrit values were 9.5, 8.0, 5.5, 3.8, and 3.3 g·dL−1. Heart rate, CVP, and PaO2 did not change throughout the experiments (P > 0.01). Compared with the baseline measurements, MAP dropped significantly at time points Hb 3.8 g·dL−1 and Hb 3.3 g·dL−1. Also, compared with the baseline measurements, CI steadily increased, whereas Hbcrit values significantly decreased for each episode. Lactate concentrations were within reference range at all time points.
During hemodilution, none of the eight animals had macroscopic ST-segment changes. None of the animals presented oxygen supply dependency of V˙O2 before time point Hbcrit had been reached.
We show in Figure 2 representative results of one channel in one animal across the Hbcrit 9.5, 8.0, 5.5, and 3.8 g·dL−1 and at Hbcrit (3.3 g·dL−1). On the left-hand side, 150 single heartbeats and the determined time points for the beginning of the P wave, Q, R, S, and R′ as well as the end of the T wave are illustrated for each dilution level. On the right-hand side, the statistics of the QT, QTc, and ST intervals and the amplitude of the T wave are shown as box plots, indicating the median (green line), upper and lower quartile range (indicated by the box size), and 1.5 times the interquartile range. Green crosses represent outliers. It is clearly visible that the intervals of QT, QTc, and ST increase with progressing normovolemic anemia, whereas the amplitude of the T wave decreases significantly.
The population statistics for all animals and all channels is described in Table 2. The PQ interval prolonged with the progression of normovolemic anemia. However, changes are small and within the temporal resolution defined by the sampling rate used. The length of the QRS complex was unchanged across the different dilution levels. However, the amplitude of S decreased steadily with increasing hemodilution. The QT interval and its frequency-corrected derivative (Bazett) and the ST interval show a significant prolongation with progressing anemia starting at Hb 8.0 g·dL−1 to Hb 3.3 g·dL−1 for the whole population. The amplitude of the T wave also declines already for the dilution step from baseline to Hb 8.0 g·dL−1 for the whole population and in single channels with the dilution step from Hb 8.0 g·dL−1 to Hb 5.5 g·dL−1. All these changes are significant according to the regression analysis.
Time-frequency representations of the single-beat segments show significant changes between episodes compared with baseline (0% replaced blood volume). Figure 3A shows the differences in the time-frequency distributions of single-heartbeat ECG segments between baseline and Hb 8.0, 5.5, and 3.8 g·dL−1 and at Hbcrit (3.3 g·dL−1), respectively, as lambda maps. The number of experiments that express the same trend in the appropriate time-frequency region is coded in color. In addition, all the subplots of Figure 3A-C include the mean across six channels. It is clearly visible that, with increasing normovolemic anemia, an area of significant differences (P < 0.01) for stable episodes compared with baseline can be identified at a time range between approximately 220 and 280 ms after the S wave of the QRS complex and a frequency range of 1 to 12 Hz in up to five experiments. This area expands to higher frequencies with more experiments (up to seven of eight) showing the same significant trend with increasing normovolemic anemia. In addition, an area in a time range between 90 and 150 ms after R of QRS complex shows a significant decrease in absolute amplitude for the episodes of normovolemic hemodilution (beginning at Hb 5.5 g·dL−1) compared with baseline. This trend is seen in up to seven of eight pigs. For a more intuitive understanding, we have included two examples of the mean of single ECG beats at the appropriate episode (pig 5, channel 7; and pig 7, channel 5) as subplots in Figure 3, row D. Furthermore, we have reprojected the significant regions of the time-frequency maps on to the averaged single-beat ECG segments of the according pig, episode, and channel; these regions are marked in red in the averaged ECG beats. The visualization depicts that the significant changes in the time-frequency maps can be located in the phase of early and late repolarization.
Row B of Figure 3 shows the results of the calculations of chance modulation. The single-beat ECGs of each episode were permutated with the single-beat ECGs at baseline. Thus, the structure introduced in the data by the process of normovolemic anemia is destroyed. In fact, all the subplots in row B of Figure 3 show that no pig exhibits a pattern in the time-frequency plots as soon as the data are permutated between episodes of normovolemic anemia and baseline. This is the proof that the results visualized in row A of Figure 3 are not based on a statistical variation but are caused by normovolemic anemia.
Finally, row C of Figure 3 visualizes how strong the effect seen in the time-frequency representation is compared with a modulation expected just by chance. For the regions described in the subplots in row A, the effect is up to 80 times larger than an effect that is expected by chance modulation. The lambda maps of the z score imply that the effect is strongest in the period of early and late repolarization at the critical hematocrit. Furthermore, the area for which more experiments show the same trend is more extended at critical hematocrit (100% exchanged blood volume [EBV]). To visualize where in the single-heartbeat ECGs the significant regions between baseline episode and episode Ei are localized, we reprojected the significant regions of each time-frequency difference map onto the single-heartbeat ECGs for each channel and each experiment. In Figure 3D, we show as an example the significant regions for channel 5 of pig 7 and for channel 7 of pig 5. Please note that because of different HRs and individual anatomies the T wave might occur at different latencies; nevertheless, the reprojection shows that for all pigs the regions of significant differences are localized in the phase of early and late repolarization.
We show that from moderate to extreme acute anemia with concomitant normovolemia, ECG changes appear in single heartbeats in a pig model. These alterations are attributed to the phase of early and late repolarization.
Significant changes in the QT and ST intervals and in T-wave amplitude occur at a mean Hbcrit of 8 g·dL−1. The overall statistics reveals a significant reduction in the absolute amplitude in four of eight pigs for the period of early repolarization and for five of eight pigs for the period of late repolarization in a frequency range of 2 to 10 Hz. With increasing anemia, seven of eight pigs show the effect of reduction in absolute amplitude compared with baseline, with the frequency spectrum involved extending across 30 Hz, reaching 45 Hz at critical hematocrit. The effect size normalized on a random effect shows the highest effect for 80% and 100% of the EBV, corresponding to a mean Hbcrit of 3.8 and 3.3 g·dL−1, respectively. At this point, we emphasize that the significant changes evaluated for the PQ interval need to be interpreted with care. The sampling rate of 500 Hz resolves 2 ms for each data point; this limits the interpretation of statistical significance. For further studies, sampling rates greater than 2 kHz seem to be advisable to resolve possible changes in the 1-ms range.
Our pig model is a well-established model to study normovolemic anemia-induced tissue hypoxia (15). In this study, the experiment was terminated around 30 min after having achieved the critical hematocrit. However, it is known from previous studies that all animals die within about 3 h due to asystole, if no further intervention takes place at Hbcrit (15).
To our knowledge, time-frequency analysis on a series of single-heartbeat ECG signals as carried out here has not yet been published. The method, nevertheless, is well established in the field of neuroscience for the analysis of evoked and spontaneous oscillations in neuroelectric brain signals [interested readers, see Le Van Quyen et al. (16)]. Here, time-frequency analysis allows the detection of otherwise invisible fluctuations in the electric field expressing the electrical activity of neuronal populations. The permutation test has proven to be effective in determining significant differences on an empirical distribution without the need of further assumptions on the structure of the data (i.e., normally distributed) (17).
A mechanistic explanation for the changes observed in the current study is not readily at hand. Normovolemia was achieved by the isovolemic replacement of an iso-oncotic colloidal solution for whole blood. Previous studies with identical hemodilution protocol have shown that normovolemia is preserved exchanging whole blood by HES (18).
During normovolemic hemodilution to Hb 3.8 g·dL−1, none of the hemodynamic parameters indicated hypovolemia or a critical restriction of myocardial oxygen supply in any of the eight animals. Neither blood pressure nor HR changed significantly, and CVP remained stable during the hemodilution procedure. Weiskopf et al. (19) reported a linear increase in HR in response to acute normovolemic anemia. However, these data were collected in conscious humans, for anesthetized humans, and animals; other data suggest no increase in HR due to normovolemic anemia (20). Our data support the hypothesis that general anesthesia adds to HR rigidity in the setting of acute normovolemic anemia. The anemia-typical increase in CI is caused by a reduction in blood viscosity (21).
The criterion standard for the detection of the critical Hbcrit is the determination of the onset of O2 supply dependency of V˙O2, reflected as the sudden decline in V˙O2 (22). All animals in this experiment were hemodiluted until the decline in V˙O2 occurred. Such our approach reflects supply dependency of the whole organism, because indirect calorimetry determines a decline in V˙O2 of the whole body. However, novel data suppose that anemia tolerance is different for several organs. Generally, it has been supposed that the myocardium is the organ with the lowest tolerance to acute anemia because an increase in cardiac output with a concomitant increase in myocardial oxygen consumption is the major compensatory mechanism of acute anemia. However, there is emerging evidence that other organs reach their outer limit of anemia tolerance at an earlier time point (23). The experimental setting as described here does not allow resolving tissue-specific supply dependency of oxygen consumption in the time course of acute normovolemic hemodilution. As a consequence, hypoxia of these organs might be missed by the approach presented. In the following, we draw a hypothetical line between the prolongation of the QT interval as observed here and clinical knowledge and outcome about QT prolongation.
QT prolongation as observed in the context of hemodilution is a well-known phenomenon associated with increased risk of potentially lethal ventricular arrhythmias (24). In fact, the human long-QT syndrome is a paradigmatic arrhythmogenic disease and has been described as congenital or acquired condition (25). In a speculative analogy, the coincidence of flattening of the T wave and QT prolongation as observed here reconciles in part the congenital human long-QT syndrome type 2 (26).
Hypoxia typically leads to intracellular ATP depletion, which in turn activates IKATP channels that abbreviate repolarization indirectly leading to less intracellular accumulation of calcium (27). However, such activation would more readily be correlated to shortening of the QT interval.
Acute normovolemic anemia results in a depression of the T-wave amplitude and an elongation of the QT and ST intervals. No physiological transfusion triggers (lactate concentration, ST-segment elevation, arrhythmias, etc.) were present at the time when the ECG changes described by our model occurred for the first time. The results are highly consistent across animals and channels. Thus, we propose that further studies are needed to clarify the mechanisms that lead to the observed changes in the ECG descriptors.
1. Leung JM, Weiskopf RB, Feiner J, Hopf HW, Kelley S, Viele M, Lieberman J, Watson J, Noorani M, Pastor D, et al: Electrocardiographic ST-segment changes during acute, severe isovolemic hemodilution in humans. Anesthesiology
2. Atiga WL, Calkins H, Lawrence JH, Tomaselli GF, Smith JM, Berger RD: Beat-to-beat repolarization lability identifies patients at risk for sudden cardiac death. J Cardiovasc Electrophysiol
3. Christensen PK, Gall MA, Major-Pedersen A, Sato A, Rossing P, Breum L, Pietersen A, Kastrup J, Parving HH: QTc interval length and QT dispersion as predictors of mortality in patients with non-insulin-dependent diabetes. Scand J Clin Lab Invest
4. Huikuri HV, Castellanos A, Myerburg RJ: Sudden death due to cardiac arrhythmias. N Engl J Med
5. Malik M, Batchvarov VN: Measurement, interpretation and clinical potential of QT dispersion. J Am Coll Cardiol
6. Oikarinen L, Nieminen MS, Viitasalo M, Toivonen L, Jern S, Dahlöf B, Devereux RB, Okin PM: QRS duration and QT interval predict mortality in hypertensive patients with left ventricular hypertrophy: the Losartan Intervention for Endpoint Reduction in Hypertension Study. Hypertension
7. Pathak A, Curnier D, Fourcade J, Roncalli J, Stein PK, Hermant P, Bousquet M, Massabuau P, Sénard J, Montastruc J, et al: QT dynamicity: a prognostic factor for sudden cardiac death in chronic heart failure. Eur J Heart Fail
8. Rossing P, Breum L, Major-Pedersen A, Sato A, Winding H, Pietersen A, Kastrup J, Parving HH: Prolonged QTc interval predicts mortality in patients with type 1 diabetes mellitus. Diabet Med
9. Algra A, Tijssen JG, Roelandt JR, Pool J, Lubsen J: QTc prolongation measured by standard 12-lead electrocardiography is an independent risk factor for sudden death due to cardiac arrest. Circulation
10. Goldberg RJ, Bengtson J, Chen ZY, Anderson KM, Locati E, Levy D: Duration of the QT interval and total and cardiovascular mortality in healthy persons (The Framingham Heart Study experience). Am J Cardiol
11. Schouten EG, Dekker JM, Meppelink P, Kok FJ, Vandenbroucke JP, Pool J: QT interval prolongation predicts cardiovascular mortality in an apparently healthy population. Circulation
12. Zabel M, Klingenheben T, Franz MR, Hohnloser SH: Assessment of QT dispersion for prediction of mortality or arrhythmic events after myocardial infarction: results of a prospective, long-term follow-up study. Circulation
13. Meier J, Wolkhammer S, Habler O: The DeltaCrit System (DCS): a computer program for standardized bedside detection of critical oxygen delivery using the Deltatrac II TM metabolic monitor. Comput Biol Med
14. Shimizu M, Ino H, Okeie K, Yamaguchi M, Nagata M, Hayashi K, Itoh H, Iwaki T, Oe K, Konno T, et al: T-peak to T-end interval may be a better predictor of high-risk patients with hypertrophic cardiomyopathy associated with a cardiac troponin I mutation than QT dispersion. Clin Cardiol
15. Meier J, Kemming GI, Kisch-Wedel H, Wölkhammer S, Habler OP: Hyperoxic ventilation reduces 6-hour mortality at the critical hemoglobin
16. Le Van Quyen M, Foucher J, Lachaux J, Rodriguez E, Lutz A, Martinerie J, Varela FJ: Comparison of Hilbert transform and wavelet methods for the analysis of neuronal synchrony. J Neurosci Methods
17. Wu W, Wheeler DW, Staedtler ES, Munk MHJ, Pipa G: Behavioral performance modulates spike field coherence in monkey prefrontal cortex. Neuroreport
18. Pape A, Meier J, Kertscho H, Steche M, Laout M, Schwerdel F, Wedel M, Zwissler B, Habler O: Hyperoxic ventilation increases the tolerance of acute normovolemic anemia
in anesthetized pigs. Crit Care Med
19. Weiskopf RB, Feiner J, Hopf H, Viele MK, Watson JJ, Lieberman J, Kelley S, Toy P: Heart rate increases linearly in response to acute isovolemic anemia
20. Van der Linden P, Wathieu M, Gilbart E, Engelman E, Wautrecht JC, Lenaers A, Vincent JL: Cardiovascular effects of moderate normovolaemic haemodilution during enflurane-nitrous oxide anaesthesia in man. Acta Anaesthesiol Scand
21. Kreimeier U, Messmer K: Hemodilution in clinical surgery: state of the art 1996. World J Surg
22. Cain SM: Oxygen delivery and uptake in dogs during anemic and hypoxic hypoxia. J Appl Physiol
23. Nanduri J, Wang N, Bergson P, Yuan G, Ficker E, Prabhakar NR: Mitochondrial reactive oxygen species mediate hypoxic down-regulation of hERG channel protein. Biochem Biophys Res Commun
24. van Bommel J, Siegemund M, Henny CP, Ince C: Heart, kidney, and intestine have different tolerances for anemia
. Transl Res
25. Roden DM: Drug-induced prolongation of the qt interval. N Engl J Med
26. Gorgels AP, Al Fadley F, Zaman L, Kantoch MJ, Al Halees Z: The long QT syndrome with impaired atrioventricular conduction: a malignant variant in infants. J Cardiovasc Electrophysiol
27. Gögelein H, Hartung J, Englert HC: Molecular basis, pharmacology and physiological role of cardiac K(ATP) channels. Cell Physiol Biochem
Taken two episodes i and j, the relationship of the ECG beats to the episode is mutually destroyed, i.e., ECG-heartbeats are randomly drawn from the distributions of episode i and j. For each permutation, we again computed the time-frequency distributions as described above for the two episodes and recomputed the difference between these two mixed episodes. This permutation was performed 100 times. Based on these 100-difference maps, we calculated an empirical distribution of the differences at each time-frequency bin. This empirical distribution allows to determine whether the difference values between the true episodes 1 and 2 are significantly larger or smaller than the random difference with the values 0 or 1. To show identical trends in experiments, we calculated the sum of each time-frequency bin across all experiments and channels. To allow for variation of exact time and frequency localization, we used an additional Gaussian smoother of 20 data points. The result of this procedure is the so-called lambda map, visualizing in how many experiments the effect is observed for a certain time-frequency interval.
To that point, we used many statistical tests for each time-frequency point. To prevent an increase in likelihood of a false rejection of H0, we performed an additional step that compares the amount of modulation in the lambda map of the data with an amount that is expected just by chance. To this end, we used the same permutation test to derive the distribution of chance differences in the difference matrix between two conditions. To estimate the modulation in lambda expected by chance, we calculated the difference between a time-frequency distribution based on the permutation of sweeps between the two respective episodes and the distribution of time-frequency maps based on 100 permutated difference vectors. Hence, we compare a random difference with the distribution of random differences. The result is a lambda map for each time-frequency bin and across all sessions and pigs that shows the level of chance modulations in case there are no differences between conditions. We use this modulation to express the modulation of the real lambda map by computing a z score. The z score expresses how much stronger the real modulations are in comparison with a modulation expected just by chance. Please note that we corrected for multiple comparison by using a z score for each time-frequency bin. To also correct for multiple comparison for different frequencies, we used a Bonferroni correction, leading to a test level corresponding to a critical z score of 4.7.