Reporting of harms in clinical trials of esketamine in depression: a systematic review

Esketamine is an intranasal N-methyl-D-aspartate (NMDA) receptor antagonist authorised in 2019 for adults with treatment‑resistant major depressive disorder. Regulatory approvals were based largely on a small set of phase III and maintenance trials, but only one pivotal trial reached statistical significance and longer‑term efficacy and safety remain debated. Previous systematic reviews of conventional antidepressants have documented repeated inadequacies and inconsistencies in how adverse events (AEs) are reported in trial publications, raising concerns about the reliability of harms information available to clinicians and regulators. Taillefer De Laportalière and colleagues set out to evaluate the quality and completeness of harms reporting in all published clinical trials of intranasal esketamine for depressive disorders. The review had two linked objectives: to score published trial reports against the CONSORT Extension for Harms (using an expanded 21‑item checklist) and to compare AE data presented in journal articles with the results posted on ClinicalTrials.gov, where reporting of trial results is mandated in many jurisdictions. The authors frame this work as necessary to judge whether published trial reports provide accurate information for benefit–risk assessment of esketamine.

This is a systematic review registered on PROSPERO and reported according to PRISMA. The authors searched Medline (PubMed) and ClinicalTrials.gov for clinical trials of intranasal esketamine in people with depressive disorders, using the search terms esketamine AND (depression OR major depressive disorder OR depressive disorder) for PubMed and condition = depression, other term = esketamine, status and study results = with results for ClinicalTrials.gov. Only published trials were included; trials had to involve human participants with depressive disorders treated by intranasal esketamine for more than seven days. Presence of a comparator was not required. Screening proceeded by title, abstract and then full text review against these eligibility criteria. Data extraction was manual from ClinicalTrials.gov entries and full‑text published articles. Extracted items included trial characteristics (publication date, trial name, phase, centres, arms, blinding, randomized sample size, primary outcome, inclusion/exclusion criteria, funding, first author affiliation) and detailed AE information per arm (AE type classified by MedDRA, number of patients with at least one AE, discontinuations due to AEs, and reporting thresholds). Risk of bias was assessed with the Cochrane RoB‑2 tool for randomized trials and ROBINS‑I for non‑randomized studies. Harms reporting quality was scored independently by two authors using a 21‑item expansion of the CONSORT Extension for Harms (each item scored 1 if adequately reported, 0 if not), producing a Total Harm Reporting Score (THRS) graded as high (17–21), moderate (12–16), low (7–11) or very low (0–6). For AE comparisons between publications and ClinicalTrials.gov, serious and non‑serious AEs were analysed separately and considered by trial phase (induction weeks 1–4, maintenance from week 5, and discontinuation/follow‑up). Descriptive statistics reported proportions and counts; the percentage of missing AEs was calculated trial by trial by comparing totals across the two sources. The authors state there were no deviations from the registered protocol.

The search identified 436 references, of which 10 trials met inclusion criteria and together enrolled 2,597 participants. Nine studies were double‑blind, parallel‑group, placebo‑controlled randomized controlled trials (RCTs); one (ID5, NCT02497287) was an open‑label, non‑randomized trial designed specifically for longer‑term safety assessment. Seven of the ten trials were Phase III. Nine trials enrolled adults aged 18–64; one included older adults (>64). All ten trials reported lead authors affiliated with the pharmaceutical company marketing esketamine. Risk‑of‑bias assessment using RoB‑2 rated the nine randomized trials as having ‘‘some concerns’’. These concerns principally arose in domain 2 (deviations from intended interventions), because dissociative and other AEs could have allowed participants or investigators to infer treatment allocation and thereby potentially undermine blinding. For several trials (ID6–ID9), domain 4 (measurement of the outcome) also raised some concerns because outcome raters might have been unblinded by AEs; other trials used blinded, independent raters via telephone. Using the 21‑item CONSORT for Harms checklist, the median THRS across trials was 10 (range 9–14). By the prespecified grading, nine trials were classified as 'low quality' for harms reporting and the open‑label ID5 trial as 'moderate quality'. Several checklist items were commonly missing across publications: 3a (clear definition of AEs), 4a (mode of AE data collection), 4c (methods for attributing AEs to the trial drug), 4d (plans for monitoring harms and stopping rules), 5b (handling of recurrent AEs), and 10a (balanced discussion placing benefits and harms in perspective). Item 8b (severity grading of AEs) was met in only one trial (ID10), with most trials providing only vague statements such as 'most adverse events were mild or moderate'. Item 8c (reporting both number of AEs and number of patients with AEs) showed inconsistent compliance. Comparing registries and publications, ClinicalTrials.gov entries listed 9,464 AEs in total versus 5,859 reported in the corresponding journal articles. For serious AEs, ClinicalTrials.gov recorded 179 events compared with 130 in published articles. The authors report that 41.5% of serious AEs and 39% of non‑serious AEs recorded on ClinicalTrials.gov were not reported in journal publications. The registry generally used a 5% frequency threshold for reporting AEs, whereas three published reports (ID5, ID8, ID9) used a 10% threshold, which the authors note could affect which events appear in articles. The extracted text indicates that the majority of serious AEs that were not mentioned in published articles affected patients in the esketamine arms; one passage gives a count of 88/94 but this figure is not fully consistent with the totals reported elsewhere in the extraction. Among the unreported serious events attributed to esketamine in the registry, the authors highlight psychiatric events (including two suicide attempts and one completed suicide), cardiovascular events (for example cerebral haemorrhage, hypertensive crisis), and renal/urinary events (nephrolithiasis, tubulointerstitial nephritis). The authors also note that AEs occurring outside the induction phase were often omitted from publications, with seven of ten trials lacking follow‑up AE reporting in the published articles despite reporting withdrawal syndromes.

Taillefer De Laportalière and colleagues interpret their findings as evidence that harms reporting in published clinical trials of intranasal esketamine for depression is generally poor and incomplete. The authors emphasise that a substantial fraction of AEs recorded on ClinicalTrials.gov were absent from the corresponding journal articles: about two‑fifths of serious AEs and roughly 39% of non‑serious AEs according to the extracted data. They argue that missing information disproportionately concerned events in esketamine treatment groups and included psychiatric, cardiovascular and renal/urinary harms; the presence of suicide attempts and a completed suicide among registry‑reported events is singled out as particularly concerning given interest in rapid‑acting NMDA antagonists for suicidality. The discussion situates these results alongside earlier work showing suboptimal harms reporting in antidepressant trials and other therapeutic areas, and notes that the esketamine literature appears comparable to prior findings. The authors highlight recurrent methodological shortcomings identified by the CONSORT harms checklist: lack of clear AE definitions, absence of description of data collection and attribution methods, inconsistent severity grading, variable frequency thresholds for reporting events, and almost universal failure to describe methods for analysing AEs. They caution that using arbitrary frequency thresholds (commonly 5% in registries but sometimes 10% in publications) may obscure less frequent but clinically important harms and that only one trial described methods for presenting or analysing AEs, limiting the ability to detect rare events. Limitations acknowledged by the authors include restricting searches to Medline and ClinicalTrials.gov and excluding unpublished data, and the fact that all included trials were conducted by the same pharmaceutical sponsor which may have homogenised reporting practices. Based on their findings, the authors recommend more consistent use of the CONSORT Extension for Harms by investigators and journal editors and call for post‑marketing, real‑world studies to better characterise the safety profile of esketamine. They also note that incomplete reporting in publications complicates benefit–risk assessment and raises questions about the transparency of the evidence base underpinning regulatory approval.