a. Study design
The pilot trial described here was a randomized, triple-blind, trial conducted at the Central Institute of Mental Health (Mannheim, Germany). Details of the blinding are given below. In this study we compared two groups of patients with depressive episodes: an experimental group (tDCS) and a control group (sham stimulation). In addition, we measured salivary cortisol to measure HPAA activity in both groups of participants, evaluating diurnal cortisol pattern (DCP), cortisol awakening response (CAR), maximal cortisol decline (MCD), and variation of cortisol before and after stimulation. During the trial, participants continued to receive pharmacological and psychotherapeutic treatments. The trial schedule and measurements are presented in Table 1.
All participants were informed about the study and provided written informed consent before participation. This pilot trial was conducted following the Helsinki Declaration and approved by the Ethics Committee of the Medical Faculty Mannheim of the University of Heidelberg.
b. Randomization and blinding
Study participants were randomly assigned in approximately equal numbers (1:1) to an experimental (tDCS) and a control group. Each patient was assigned to one of the four stimulation devices (A, B, C and D), depending on the order of their inclusion in the study. Two of the four devices were tDCS (A and C) and the remaining two were sham stimulation devices (B and D). To ensure compliance with the quality criterion of reliability all stimulation treatments were carried out by the same practitioner (H.S.), with the same devices, in the same rooms, and always on weekdays (Monday to Friday) between 2:00 and 4:00 PM. The same physician always collected the assessment questionnaires. During the trial, neither the patients, nor the practitioner, nor the physician were informed about whether the treatment given was real or simulated. This blinding status was extended through the evaluation of the raw data, and disclosure only took place at the beginning of the final data analysis, reducing any possible experimental bias.
Initially, forty-seven patients (20 female and 27 male participants; mean age: 45.30 ± 14.20) with depressive disorders were recruited from the inpatient unit "affective disorders" (Central Institute of Mental Health, Mannheim) for this pilot trial between 15th August 2016 and 20th September 2017. The characteristics of the participants who completed the trial are shown in Table 2. To be included in the study participants had to meet the DSM-V criteria, evaluated through SCID-I interviews, for a depressive episode or major depression and sign the informed consent form. The interviews were carried out by Central Institute of Mental Health clinical experts, who were not involved in the study. Included patients had a Hamilton Depression Rating Scale (HDRS) of at least 18 points with baseline stability (i.e., less than 25% improvement one week before the beginning of stimulation). Before starting the study, we checked that the participants had an ECG sinus rhythm. Therapy-naïve status was not an inclusion criterion for this pilot trial so that the study participants could continue their prescribed treatments at the inpatient unit. During the trial medication was not changed but the doses were adjusted slightly if needed. We chose a three-week trial interval (21 days) to try to avoid the occurrence of essential changes in the pharmacological treatment. Finally, patients taking benzodiazepines as a pro re nata (PRN) treatment could participate with doses no greater than 1.5 mg/d of Lorazepam or equivalent, to keep additional pharmacological effects as low as possible. Participants taking sleep inducers such as Zopiclone and Zolpidem as a PRN treatment were not excluded from this pilot trial.
Patients with psychotic disorders, post-traumatic stress disorders, panic disorders, and borderline personality disorders were excluded from the study. We also excluded females with current pregnancies, participants with a conservatorship or with legal protectors, and patients who could not give their consent because of severe mental illness. Finally, we excluded patients complying with the following criteria: with metal implants or medical devices (e.g., pacemakers), illegal drug and alcohol dependency at the time of the study, acute suicide crisis, medical conditions that alter adrenal functions, use of glucocorticoids, intake of medication that changes the heart rate and its variability (e.g., beta-blockers), arrhythmias of any type, dementia syndrome (DSM-V/ICD-10 criteria), past medical history of severe cranioencephalic trauma, neurological diseases of any kind, severe decompensated medical conditions (e.g., therapy-resistant arterial hypertension, heart insufficiency, respiratory insufficiency), neoplastic diseases of any kind and infectious diseases of any kind.
Of the 47 patients included in the study, 40 completed the three-week treatment and seven participants left the trial (drop-out rate: 14.89%). We interrupted the trial for a 20-year-old participant on the third day of stimulation due to a newly identified skin rash all over the scalp. The trial had to be interrupted on the 7th day for a 78-year-old participant due to his discharge from the inpatient ward. For 3 participants (ages 48, 53 and 34) the trial was terminated before the first stimulation appointment due to new diagnostic assessments which triggered trial exclusion criteria. Similarly, one participant (age 38) left the study before the first stimulation appointment due to entering into a manic phase. Finally, a 27-year-old patient decided against further participation in the study because of side effects such as headaches which occurred during and after the first stimulation treatment.
Concerning medication, all 40 participants who completed the stimulation trial received the direct current stimulation as an add-on therapy to guideline-compliant pharmacotherapy for depression. Seven of these 40 participants received monotherapy with an antidepressant agent. Another 14 patients were treated with a combination of 2 antidepressants, and one patient was treated with a combination of 3 antidepressants. Six patients had been prescribed an antidepressant plus augmentation therapy (e.g., Aripiprazol), and five received a combination of 2 antidepressants and augmentation therapy. One patient received a quadruple combination of 3 antidepressants and quetiapine. Two received a triple combination of lithium, quetiapine and an antidepressant. Two patients received quetiapine monotherapy and two received dual therapy with lithium and quetiapine. There was no significant difference between the two groups concerning pharmacotherapy.
d. Transcranial direct current stimulation (tDCS)
Stimulation was performed using a CE-certified microprocessor-controlled device (Sooma Medical, Helsinki, Finland) which emitted a direct current with a maximum output of 40 mA. Direct current (DC) was applied via a pair of electrodes (anode and cathode) made of conductive rubber (dimensions: 5 x 7 cm, 35 cm²). Before placing the electrodes, these were wetted on sponges soaked in a 144 mol/L NaCl solution in order to lower the physiological skin resistance. Electrodes were placed based on the 10-20 international system, with the anode placement at EEG point F3 (left dorsolateral prefrontal cortex) and the cathode at EEG point Fp2 (right dorsolateral prefrontal cortex). A hood facilitated correct placement and this was adapted to the head of the participant.
The four stimulation devices (A, B, C and D) were previously programmed to perform the stimulation either in real (DC of 2.00 mA) or in sham modus (DC of 0.30 mA). In the sham stimulation modus, the devices produced an initial DC of 2.00 mA for a few seconds, ramping down to a DC of 0.30 mA and maintaining this current for a period of 30 minutes (treatment duration), giving sensations similar to real stimulation.
The stimulation was carried out according to the manufacturer's protocol, encompassing three weeks of treatment with daily 30-minute stimulation sessions. This protocol was used regardless of the device mode (real or sham stimulation). In our trial, participants received tDCS or sham stimulation until the end of week 3, from Monday to Friday, between 2:00 and 4:00 PM, with stimulation pauses at weekends.
e. Cortisol assessment
General procedure: To assess the activity of the HPAA we measured cortisol using saliva samples. The advantages of using salivary cortisol are mostly related to the practicability of the sample collection, with less invasiveness and less effort for the participant 9,25. Moreover, many studies have demonstrated a good correlation between plasma and salivary cortisol26,27, verifying the method's effectiveness for assessing plasma cortisol concentration. Salivary cortisol measurements were acquired as reliable indicators of total free plasma cortisol, with marginally lower concentration due to the presence of 11ß-HSD2 in saliva28. The advantages over blood cortisol measurements, such as laboratory independence and “stress-free” sampling, are more important for the repeated measurements used to determine the circadian rhythmicity of cortisol secretion.
Saliva collection occurred during the week when patients participated in daily ward routines. Participants received instructions on the saliva collection procedures. Immediately before the sample collection participants must not consume any nutrients or fluids, or smoke, or brush their teeth since cortisol levels in saliva may be increased by these actions. The collection was done using a kit based on cotton swabs and tubes (Salivette™, Sarstedt, Leicester). Participants were instructed to chew on the saliva collectors for 30 to 45 seconds immediately after awakening but when still in bed (A), 30 min. after awakening (B), 8 h. after awakening (C), and 14 h. (D) after awakening. The starting point for saliva sample collection was between 07:30 and 08:00 AM. After the saliva absorption, the cotton swab was removed from the oral cavity and placed into the corresponding tube. All samples were stored at -25˚C. After thawing, samples were centrifuged for 5 min at 3000 rpm, resulting in a clear low-viscosity supernatant. As published elsewhere, the samples of supernatant were analyzed using a time-resolved immunoassay with fluorescence detection29,30. The lower limit of detection was 0.43 nmol/L, with inter- and intraassay coefficients of variation of less than 10% across the expected range of cortisol levels. The concentrations were calculated in nmol/L29,30.
Saliva samples for DCP, cortisol awakening response, maximal cortisol decline, and stimulation-related cortisol levels: Two sets of DCP samples were collected during the study. Participants were asked to collect saliva four times during the day before the first stimulation (baseline) and at the end of the third week31. Based on samples A and B (see above), the cortisol awakening response (CAR) was calculated from the difference between the two samples (i.e., CAR = B - A). Finally, the maximal cortisol decline (MCD) was calculated from the difference between the highest morning cortisol value (i.e., sample B) and the last value (i.e., sample D). Finally, salivary cortisol was collected "immediately before stimulation" (pre), "immediately after stimulation" (post), and "2 hours after stimulation" (2h-post) in W0 and W2 (Table 1). These samples were collected to record the short-term effects of tDCS on HPAA activity.
Implausible morning cortisol values (i.e., values of samples A or B ≤ 3 nmol/L) were replaced with half of the minimum morning cortisol value (in this case, 1.50 nmol/L), as suggested elsewhere32.
The primary outcome of this pilot study was any change in the DCP from the baseline to week 3 (W3) in the tDCS group in comparison with the control group. Additionally, with the data on the DCPs, the area under the curve with respect to the ground (AUCg) was calculated, using the formulae of Pruessner and colleagues33. Secondary outcomes included any change in the CAR from the baseline to W3, any change in the cortisol morning decrease from the baseline to W3, and any change in stimulation-related cortisol levels between W0 and W2.
Finally, we included in our analysis any variations of the MADRS scores through the trial period in both groups. In this case, as suggested elsewhere34, we defined clinical response based on MADRS scores if the MADRS total values showed a reduction to ≥ 50% of the initial value with a final score of ≤ 9 points. Finally, also as described elsewhere35, we defined partial response if the MADRS total values showed a reduction of ≥ 25% of the initial value.
g. Statistical analysis
Numeric variables that fitted a parametric distribution are specified in the text as mean (standard deviation). Numeric variables with a non-parametric distribution (i.e., median ≠ mean) are shown as median (interquartile range, 3rd quantile – 1st quantile). Category variables and count data are specified as numbers or fractions. Data with more than two decimals were rounded. Values smaller than 0.005 are presented as < 0.005 and values greater than 1 million are expressed in scientific notation4. Descriptive data in the text are presented in tables. To evaluate the significance of differences between groups, t-tests were used for continuous parametrically distributed data and the U-test for continuous non-parametrically distributed data. We used the χ2 or the exact Fisher test for the category and count data. For these differences, p values were calculated. This study defined statistical significance whenever the two-tail-p-value was less than or equal to 0.05.
Primary and secondary outcomes and MADRS scores were analyzed blindly by one of the authors who did not know the assignment of interventions to the trial groups (B.P.P). Intention-to-treat (ITT) and per-protocol analyses (PP) were performed for both primary and secondary outcomes using linear mixed models (LMM). We calculated the interaction time * daytime * group for the primary outcome. For the secondary outcome we calculated time * group (CAR and MCD) and time * daytime * group (stimulation-related cortisol levels). The interactions were corrected for gender, age, and weight confounding factors. We carried out multiple realizations with linear interactions and a maximum of 1000 iterations in case of missing values. For both trial groups, CAR, DCP, MCD, and stimulation-related cortisol levels were estimated in the multiple realizations using the variables gender and age of the participants. In the LMM, fixed effect omnibus tests were carried out to define the main effects of the variables in the model and to evaluate the variable in the model against the null model. The results of the ITT analysis were reported graphically, and the ITT and PP analyses’ results are described in the text using 95% confidence intervals (95CI). Cohen’s d was computed for ITT and PP analyses for the sample size effect. Post-hoc tests were carried out for the LMM when differences between the time * group (MCD, CAR) and the time * daytime * group (DCP and stimulation-related cortisol levels) were significant.
For the statistical analyses and the descriptive data, the SPSS software (International Business Machines Corporation, New York, United States of America), version 26.0, was used. For the LMM, we used the R-based software jamovi 2.0.036 and the toolbox GAMLj37. Finally, we generated the graphs using Prism 8 GraphPad (GraphPad Software Inc., California, United States of America).
h. Exploratory analysis
In addition to our principal objectives in this trial, we carried out an exploratory analysis with the baseline information of this trial to search for relationships between morning salivary cortisol, somatic symptoms, family history of depression and psychometric data (MADRS item 10 “suicide thoughts” and HRDS score). For that purpose, a generalized linear model was computed using the daytime B salivary cortisol concentrations (i.e., from samples collected between 8:00 and 8:30 AM). In the model, we included the dichotomic variables “family history of depression” and “somatic symptoms” as factors. Finally, we included “age”, “weight”, “BDI scores”, “HRDS scores” and “MADRS item 10” as covariates. For this analysis, we used the R-based software jamovi 2.0.0 36 and the toolbox GAMLj 37, and defined statistical significance whenever the two-tail-p-value was lesser than or equal to 0.05.