Patients with major depression. This is a post hoc analysis of data from depressed patients treated with ECT in Rijnstate Hospital (Arnhem, The Netherlands) and who participated in the StudY of effect of Nimodipine and Acetaminophen on Postictal Symptoms after ECT (SYNAPSE; NCT04028596). SYNAPSE is a randomized controlled trial with a three-condition cross-over design. Patients received nimodipine, acetaminophen, or a placebo (water) in random and counterbalanced order at a maximum of 2 h before each ECT session. The order of the treatment conditions was randomized and differed across patients [16]. Repeated EEG measures before, during, and until one hour after the ECT sessions were collected. Inclusion criteria for patients were age ≥ 18 years and having a current clinical diagnosis of major depression (i.e., classified as unipolar, bipolar, schizoaffective disorder in DSM-5). The local medical ethical committee approved the study protocol (NCT04028596) and all included patients provided oral and written informed consent.

Healthy controls. To correct for test–retest variability of fE/I ratios, we included a public EEG dataset of healthy controls (HC) with repeated EEG recordings at two different time points [17]. None of these participants reported psychiatric or neurological symptoms. HC were aged ≥ 18 years and provided oral and written informed consent [18].

ECT was administered according to the standard treatment guidelines in The Netherlands [19], mostly twice weekly. The Thymatron System IV device (Somatics Incorporation Lake Bluff, Illinois, USA) was used to administer ECT stimuli with a constant-current (0.9 Ampère), bidirectional, square wave and brief pulse (1 ms). Electrodes were placed either unilateral (right [RUL]) or bi(fronto)temporal (BL). Dose titration or age-based dosage were used to choose the stimulus charge, which was adjusted during the course according to the psychiatrist’s discretion. In case patients did not improve after six sessions, unilateral placement was switched to BL during the ECT course. Anesthesia was mostly provided with etomidate (0.2–0.3 mg/kg body weight) for sedation and succinylcholine (0.5–1 mg/kg body weight) for muscle relaxation. Cessation of the ECT course was decided based on the treatment response, estimated by clinical judgement of the treating psychiatrist and clinically rated by using the Hamilton Depression Rating Scale (HDRS) [20].

Patients with major depression. Silver/silver chloride cup electrodes were placed on the scalp according to the International 10–20 system. To ensure enough space for placement of the ECT electrodes in patients receiving BL stimulation, EEG electrodes T3 and T4 were placed 10% behind and F7 and F8 above the pre-defined location. For patients receiving RUL stimulation, T4 and Cz were moved behind and F8 above. EEGs were recorded using a full-band DC amplifier (TMSi) and a NeuroCenter EEG recording system (Clinical Science Systems, Leiden). EEG recordings were sampled at 256 Hz. The impedance was kept below 5 kΩ. For the analyses, five-minute EEG recordings were used, measured with eyes closed prior to the first (or if missing, the second) ECT session (i.e., pre-ECT) and 2 weeks after the total ECT course (i.e., post-ECT). In case the 2-week post-ECT course measurement was missing (i.e., lost to follow-up), a resting-state EEG recording that was made before the last ECT session was used.

Healthy controls. Four-minute EEG recordings with eyes closed were used. EEG data were recorded using a 64-channel (silver/silver chloride) BioSemi ActiveTwo system (BioSemi B.V., Amsterdam) and were sampled at 1024 Hz [18], at two time points (i.e., T1 and T2). The available channels were reduced to the similar EEG electrodes of the patients. Here, EEG electrode T4 corresponded to T8, T3 to T7, T6 to P8, and T5 to P7.

Both EEG datasets were band-pass filtered (0.5–30 Hz; first-order Butterworth filter) and visually inspected for artifacts. Cz was used as reference electrode. EEG data of HC were resampled to 256 Hz. EEG electrodes containing (excessive) noise were rejected for analysis and segments containing artifacts were removed. This was considered justified, as removing segments from a signal and constructing the remaining together does not affect the scaling behavior of positively correlated signals [21]. All pre-processing steps and analyses were conducted with MATLAB R2022b (MathWorks, Natick, MA, USA).

The power spectral density (PSD) of each EEG electrode pair was estimated using Welch’s method in five second artifact-free segments with an overlap of 50%. To compare PSD values, these were averaged in frequency bins of 1 Hz in the frequency range 0.5–30 Hz.

A requirement for estimating fE/I ratios was the existence of a co-variation between the amplitude of the spectral power and the fluctuation function. To ensure this, only data consisting of LRTC were used to estimate fE/I values. Detrended fluctuation analysis (DFA) was done to check for the existence of LRTC with DFA exponent > 0.60 as threshold [22]. When performing DFA, the fluctuation function is computed, which is plotted on logarithmic axes. The fluctuation function is expressed as

$$\langle F(t)\rangle =\mathrm(\sigma \left(W\right))$$

(1)

with \(F(t)\) the fluctuation function and \(\sigma\) the standard deviation of the detrended signal of a set windows (\(W\)) of separate time series of length \(t\), with an overlap of 50%. The DFA exponent is the slope of the fluctuation function calculated using linear regression. After removing segments containing artifacts, the DFA exponent was fit between 2 and 25 s. DFA analysis was performed for each individual frequency band (i.e., bins of 1 Hz in the frequency range 0.5–30 Hz).

If the DFA exponent exceeded the threshold, a normalized fluctuation function nF(t) was computed by dividing each windowed signal profile by the mean amplitude of that window. Here, windows with a length of five seconds and an overlap of 80% were used. The fE/I ratio was defined as [15]

$$fE/I=1-_,W\mathrm(t)}$$

(2)

with \(_,W\mathrm(t)}\) being the Pearson correlation between the set of windowed amplitude values (\(_}\)) and the set of detrended amplitude-normalized signal profiles (\(_(t)}\)). Hence, fE/I ratios below 1 indicated inhibition dominated networks, above 1 indicated excitation dominated networks, while balanced, critical networks would have a value of 1. fE/I ratios were estimated in frequency bins of 1 Hz in the frequency range 0.5–30 Hz.

Participant, treatment, and EEG characteristics. After testing whether the data approximated a normal distribution (i.e., visualization of the histogram and Anderson–Darling test), we used mean ± standard deviation (SD) or median ± interquartile range (IQR) values to report participant, treatment, and EEG measures. In depressed patients, HDRS scores at baseline and after the ECT course were compared using the Wilcoxon signed-rank test. For spectral power analysis and estimation of fE/I ratios, we averaged all values over the available EEG electrodes (i.e., whole brain). Furthermore, to study regional effects on fE/I ratios three different regions were defined, i.e., the frontal region (including Fp1, Fp2, F3, F4, F7, F8, and Fz), centrotemporal region (including C3, C4, T3, and T4) and parieto-occipital region (including T5, T6, P3, P4, Pz, O1, and O2). For HC, we used the same descriptive statistics and EEG measures.

Comparisons at baseline. To test whether differences existed at baseline between fE/I ratios of patients (pre-ECT) and HC (T1), Wilcoxon rank-sum tests were performed.

Changes over time. Paired t-tests or Wilcoxon signed-rank tests were performed to test whether changes in spectral power and fE/I ratios occurred over the ECT course in patients (post-ECT–pre-ECT), as well as between the two time-point measurements in HC (T2–T1).

Comparisons of responders and non-responders. To study the association between changes in fE/I ratios and clinical outcome, patients were divided into the groups ‘responders’ (i.e., post-ECT HDRS score decrease ≥ 50% compared to baseline) and ‘non-responders’ after ECT. Next, these groups were analyzed regarding baseline EEG measures (pre-ECT) and changes in fE/I ratios over the ECT course (post-ECT–pre-ECT).

For all analyses, p-values < 0.05 were considered statistically significant. In case of multiple testing, also the false discovery rates (FDRs) were computed whereby pFDR < 0.05 values were considered statistically significant.