Published on 21 September 2026

Analysis · Depression · Neuromodulation

After a successful course of ECT, should maintenance sessions continue alongside medication?

◆ Collection
Psychological Medicine · 2025; 55: e251 · Jelovac A et al.
DOI 10.1017/S0033291725101608
PMID 40874256
Scientific 74
Editorial 74

Editorial collections A Study Under the Microscope

The essentials

Four randomized trials, 254 patients randomized, 243 in the main analysis. In adults in remission or in response after acute electroconvulsive therapy, continuing properly dosed maintenance sessions in addition to medication reduces the risk of relapse at six months, with a relative risk of 0.57 (95% CI 0.37 to 0.88), a 43% relative reduction, and a number needed to treat of seven to prevent one relapse. No statistical heterogeneity is detected, and the result holds up across every sensitivity analysis, including the least favorable scenario for missing data. The limit lies in the evidence base itself: four trials, no blinding of participants, two of the four trials at high risk of bias and none at low risk, and a prediction interval that crosses the value of 1. The effect is consistent. It is not yet certain.

Context

The question arises at discharge, and it arises quickly. Electroconvulsive therapy that has worked leaves a patient improved but fragile: roughly half of patients maintained on continuation pharmacotherapy alone relapse within a year, and most of those relapses occur in the first six months. The handoff is usually left to medication alone.

Continuing spaced maintenance sessions is a known option whose use varies widely from one country to another. The authors report that fewer than 5% of acute courses were followed by maintenance in the United Kingdom and Ireland in 2022, 11% in Sweden in 2013, and 27% in a Finnish survey, yet 97% of centers surveyed in 2022, mostly North American, reported offering maintenance treatment, with a median of 50% of patients receiving it. Earlier syntheses, for their part, produced mixed and inconclusive results.

This is the point the meta-analysis reopens. It does not change the nature of the data available: it changes the criteria that decide which trials enter.

What changes relative to earlier syntheses

The benefit appears once the trials that reflect practice are retained

Authors’ choice	What this excludes or includes
Properly dosed maintenance sessions	Bitemporal or bifrontal placement at any dose relative to seizure threshold, right unilateral placement only at high dose, defined here as at least five times threshold
EffectOne trial is excluded for using low-dose right unilateral placement. In the two retained trials that used this electrode placement, the dose delivered was six times seizure threshold, with an ultrabrief pulse. The five-times-threshold floor is not measured data: it is an explicit assumption of the authors, carried over from the acute phase to the maintenance phase.
Maintenance added to medication	Both arms receive active continuation pharmacotherapy; maintenance treatment used alone is excluded
EffectThe question asked becomes the clinician’s own: does adding sessions change anything to what I am already doing. This is a more demanding contrast than a comparison between sessions alone and medication alone.
Primary outcome fixed at six months	Cumulative relapse at six months, the period of maximum vulnerability
EffectA common horizon across trials. Follow-up in the included trials ranges from six months to two years, but this synthesis analyzes nothing beyond six months. One-year durability is therefore not documented here, for lack of analysis rather than lack of data.

This point deserves stating plainly, because it is the heart of the article: what changed is not the quantity of data, it is the question being asked. The 2022 meta-analysis by the National Institute for Health and Care Excellence, which underlies the current UK recommendation, retained only two trials, one of which used low-dose right unilateral placement, and excluded two others that were available, one because more than 20% of participants had psychotic features, the other over the timing of randomization. A shifted inclusion criterion can make an effect appear or disappear without a single additional patient being enrolled. That is an argument for reading a meta-analysis’s inclusion criteria before reading its results, not the other way round.

The study at a glance

Population
Adults aged 18 and older with a major depressive episode, unipolar or bipolar, in remission in three trials or in response in one trial, following acute electroconvulsive therapy. Three of the four trials included only unipolar forms. The authors describe the population as prototypical of the indication: mean age around 70 in three of the four trials, substantial proportions with psychotic features, and a female predominance throughout. Per-trial samples ranged from 33 to 128 randomized participants. Total of 254 randomized, 127 per arm.
Intervention
Properly dosed continuation electroconvulsive therapy, combined with continuation pharmacotherapy. Two trials used bitemporal brief-pulse placement, the other two a right unilateral ultrabrief-pulse placement delivered at six times seizure threshold, one of them allowing a change of placement during treatment for a minority of patients.
Comparator
Continuation pharmacotherapy alone: individualized regimens in two trials, nortriptyline alone or combined with risperidone in one trial, and a combination of venlafaxine and lithium in the fourth.
Outcomes
Primary: cumulative relapse at six months, defined by each trial’s own criteria, all relying on a validated depression scale. Secondary: acceptability, measured as all-cause discontinuation. Cognitive tolerability was not examined in this synthesis.
Design
Systematic review and meta-analysis of four randomized controlled trials, three European and one North American. Search conducted March 11, 2025 across PubMed, Embase, Web of Science, and CENTRAL, with no date or language restriction: 770 records identified, 423 after deduplication, eight full texts assessed, four trials retained. PROSPERO registration CRD420251000113, reporting compliant with PRISMA 2020, risk of bias assessed with RoB 2. Random-effects model with restricted maximum likelihood estimation, prediction intervals, worst-case missing-data imputation. The Hartung, Knapp, Sidik and Jonkman adjustment was used only in sensitivity analysis: the authors did not retain it for the primary inference, because it paradoxically narrows intervals when heterogeneity is very low.

Quality control

Point checked	Verdict
Protocol registered before analysis	Yes
FindingPROSPERO registration CRD420251000113 and reporting compliant with PRISMA 2020. Selection and extraction were done in duplicate, by two independent reviewers, with trial investigators contacted for missing data.
Risk-of-bias assessment	RoB 2
FindingAn appropriate tool for randomized trials, applied trial by trial. Two trials are rated high risk for deviations from intended interventions, missing data, or outcome measurement. The other two raise concerns for lack of information on allocation concealment. No trial is rated low risk of bias.
Blinding of participants	Absent
FindingNo trial maintained blinding of participants, which would have required repeated sham sessions under general anesthesia. Three of the four trials did, however, use blinded raters. Greater medical attention toward the group continuing sessions cannot be ruled out: the authors acknowledge this explicitly.
Heterogeneity between trials	Not detected
FindingI² at 0% in every analysis. The four trials point in the same direction with comparable magnitudes, despite very different stimulation parameters, session schedules, and medication regimens.
Sensitivity analyses	Consistent
FindingThe result is significant in the intention-to-treat analysis (254 participants) as in the modified intention-to-treat analysis (243 participants), with last observation carried forward as with worst-case imputation, with and without the Hartung and Knapp adjustment. The numerical detail of these analyses appears in a supplementary table that we were unable to consult.
Subgroup analyses	Not conducted
FindingNeither subgroup analyses nor meta-regression, the number of trials being too small. Nothing is therefore documented on the influence of stimulation modality, dosing strategy, or patient characteristics. The authors judge that these analyses would have added little since heterogeneity is null; the argument is weak, since I² has very low power with only four trials.
Publication bias	Not assessable
FindingFour trials allow no formal test of asymmetry. This is not an absence of publication bias, but an inability to look for it.
Funding of the review	None
FindingThe authors declare having received no specific funding, public, commercial, or charitable, for this synthesis.
Authors’ conflicts of interest	Two of four authors
FindingOne author declares honoraria from UpToDate, honoraria from Northwell Health, and copyright royalties from Cambridge University Press. The last author, also the corresponding author, declares speaker honoraria from MECTA, a manufacturer of electroconvulsive therapy devices, as well as from Otsuka and Janssen, and a Janssen honorarium for an advisory board on esketamine. The other two authors declare no ties. These ties concern the intervention under evaluation.

Results

43%
Relative risk reduction of relapse at six months when properly dosed maintenance sessions are added to pharmacotherapy, compared with pharmacotherapy alone.

Outcome	Result
Cumulative relapse at six months	RR 0.57 (95% CI 0.37 to 0.88)
ReadingPrimary outcome, modified intention-to-treat analysis on 243 participants. The confidence interval excludes the value of 1, so the result is statistically significant at the conventional threshold.
Number needed to treat	7 (95% CI 5 to 24)
ReadingSeven patients receiving maintenance treatment to prevent one additional relapse. The upper bound of 24 is a reminder that precision remains modest.
Heterogeneity	I² = 0%
ReadingNo variability detected between trials beyond chance. With four trials, this value is consistent but not very informative: it does not prove homogeneity, it fails to demonstrate heterogeneity.
Prediction interval	0.28 to 1.16
ReadingThe most important figure in the article after the first one. It estimates the effect expected in a future trial, and it crosses the value of 1: a subsequent trial might not reach significance.
Acceptability, all-cause discontinuation	RR 1.12 (95% CI 0.48 to 2.62), prediction interval 0.15 to 8.38
ReadingNo significant difference detected. The confidence interval is wide and the prediction interval considerably more so: this analysis does not demonstrate the absence of an added burden of discontinuation, it simply fails to detect one.

Critical appraisal

Domain	Judgment
Conduct of the review	Solid
FindingProspective registration, dual selection and dual extraction, RoB 2, prediction intervals, a worst-case scenario tested, unpublished data obtained from investigators. The synthesis work follows the rules, and it exposes its own weaknesses rather than concealing them.
Question asked	Relevant
FindingAdequate dose and use in combination, that is, the real clinical situation. This refocusing is also what makes comparison with earlier syntheses difficult: they were not answering the same question. The dose floor retained nonetheless remains an assumption of the authors, not trial data.
Size and quality of the evidence base	Fragile
FindingFour trials, 254 randomized, two trials at high risk of bias and none at low risk, no blinding of participants, a prediction interval crossing 1. A fifth, moderately sized trial could shift the estimate substantially.
External validity	Narrow
FindingOlder patients, often with psychotic features, very predominantly unipolar. Yet older age and psychotic features are associated with better response and lower relapse risk after electroconvulsive therapy. The authors themselves write that transposing these findings to younger, chronic, treatment-resistant patients or those with complex presentations may be limited, and that nothing transposes beyond unipolar and bipolar depression.
Conflicts of interest	Worth flagging
FindingThe declared ties concern directly the technique under evaluation, including a device manufacturer. They are disclosed, and no funding was received for this review, which is the minimum expected, but they justify awaiting replication by a team with no ties to the equipment.

Level of evidence

Scientific74

Editorial74

Confidence is high regarding the conduct of the review and the internal consistency of the result: the direction of the effect is the same across all four trials, and it survives deliberately pessimistic assumptions about missing data. It is low regarding the generalizability of the figure itself. Four trials without blinding of participants, two of them at high risk of bias and none at low risk, do not fix a value: they indicate a direction. The prediction interval says so plainly, and the authors do not try to work around it.

What is demonstrated: in these four trials, adding properly dosed maintenance sessions is accompanied by fewer relapses at six months. What is suggested: that this benefit is found in routine practice, among comparable patients. What amounts to expert opinion: the dose floor retained as adequate, five times seizure threshold for right unilateral placement, which the authors explicitly derive by analogy with the acute phase and which structures the entire selection of trials.

The colleague test

What an experienced colleague might say about this study in two minutes, between two consultations.

“A careful synthesis on a thin evidence base. A 43% relative reduction in relapse risk at six months, seven patients to treat, no heterogeneity detected, that holds together. But four trials, nobody blinded, no trial at low risk of bias, and a prediction interval that climbs back above 1. I’ll offer it to the patient, I won’t promise it.”

Translated for practice: continuation electroconvulsive therapy again becomes an option to discuss explicitly at discharge, not a modality reserved for repeated treatment failures. The discussion belongs with the patient and the team that delivered the acute course, and it belongs at the moment of discharge, not three months later.

What you can do with this

Raise the question of the handoff as soon as the acute course ends, before discharge, rather than at the first sign of returning symptoms.
Keep the order of magnitude rather than the exact value: roughly one patient in seven spared a relapse at six months, with real uncertainty around that figure.
Confirm with the team that delivered the acute course that the maintenance parameters under consideration match those evaluated here, particularly for right unilateral placement, retained only at high dose. An underdosed maintenance session has not been shown to be effective.
Do not present the lack of difference in discontinuation as evidence of equivalent tolerability: the analysis is too imprecise for that, and cognitive tolerability was not examined in this synthesis.
Keep in mind that the population studied is older, often psychotic, and almost exclusively unipolar: the figure does not transpose mechanically to a young, chronic patient or one with multiple comorbidities.
Know how to answer a patient who asks why continuation is being proposed when they are already feeling better: it is precisely because they are feeling better, and the period that follows is the period of risk.

Frequently asked questions

How long should sessions continue, and at what pace?

This synthesis does not answer that question. Its primary outcome is fixed at six months, and it does not compare different durations or schedules against one another. The authors state explicitly that the optimal frequency schedule is not established: the included trials used very different protocols, ranging from progressive spacing to a monthly schedule, at a pace never less frequent than every two weeks. The decision remains individual, discussed with the team following the patient.

Is the benefit the same in patients with bipolar disorder?

This synthesis cannot say. Three of the four trials included only unipolar forms, and no analysis by polarity was conducted, the number of trials being too small to allow subgroups. The absence of an analysis is not the same as the absence of a difference.

Should medication treatment be maintained in parallel?

Yes, that is the very condition of inclusion. Maintenance sessions were evaluated in combination, never alone, and both arms received active continuation pharmacotherapy. What this synthesis compares is medication plus sessions against medication alone, not sessions against medication.

Why are existing guidelines more cautious?

Because they rely on a set of trials selected under different criteria. The 2022 UK meta-analysis retained only two trials, one of them using low-dose right unilateral placement, and excluded two others for reasons related to the proportion of psychotic features and the timing of randomization. The disagreement concerns which data were selected, not how they were analyzed.

What should be taken from the prediction interval?

That one should be cautious about quoting a figure to the patient. The prediction interval estimates what a subsequent trial would show, and it includes the possibility that no significant benefit would be found. The direction of the effect is more solid than its size.

Annotated bibliography

Source study. Jelovac A, Braithwaite R, Kellner CH, McLoughlin DM. Continuation electroconvulsive therapy combined with pharmacotherapy for depression relapse prevention: A systematic review and meta-analysis. Psychological Medicine, 2025, volume 55, article e251. DOI 10.1017/S0033291725101608. PMID 40874256. No specific funding. Conflicts of interest declared by two of the four authors, bearing on the technique under evaluation. No supplementary material accompanied the document reviewed: the supplementary figures and tables of this publication were not consulted, and no data appearing only there is used here.

Comparison cohort, cited by the authors. Jørgensen A, Gronemann FH, Rozing MP, Jørgensen MB, Osler M. Clinical outcomes of continuation and maintenance electroconvulsive therapy. JAMA Psychiatry, 2024, volume 81, issue 12, pages 1207 to 1214. DOI 10.1001/jamapsychiatry.2024.2360. A Danish national cohort of 19,944 patients who received a course of electroconvulsive therapy over twenty years, across all psychiatric indications. In propensity-score matched analyses, maintenance treatment was associated with fewer psychiatric rehospitalizations (hazard ratio 0.69, 95% CI 0.56 to 0.84) and fewer suicidal behaviors (0.47, 95% CI 0.27 to 0.82) within six months. Its observational design precludes drawing a causal relationship from it: it makes the hypothesis more plausible, it does not demonstrate it. The figure specific to unipolar depression cited by the authors comes from a personal communication rather than from the publication.

Comparison guideline, cited by the authors. National Institute for Health and Care Excellence. Depression in adults: treatment and management. NICE Guideline NG222, June 29, 2022. This is the two-trial synthesis that underlies the current UK position on maintenance treatment, and the main point of disagreement with the present meta-analysis.

Regulatory and organizational context, to be established locally. The organizational arrangements for outpatient continuation electroconvulsive therapy, and the actual availability of the medications used as continuation pharmacotherapy in the included trials, namely nortriptyline alone or combined with risperidone, the combination of venlafaxine and lithium, and individualized regimens, vary from one country to another. This point deserves verification before any transposition: the documents consulted here do not address the availability of these medications in any given jurisdiction.

Editorial collections

A Study Under the Microscope