Relationships between reduction in symptoms and restoration of function and wellbeing: Outcomes of the Oral Ketamine Trial on Suicidality (OKTOS)

Suicide is a major global cause of death and is frequently accompanied by major depressive disorder (MDD), which impairs mood, cognition and social/occupational functioning. Ketamine has demonstrated rapid reductions in suicidal ideation and depressive symptoms, but there is limited evidence on whether such symptomatic improvements extend to positively valenced outcomes such as functioning and subjective wellbeing. The authors situate this gap against patient-reported priorities: people with depression commonly value return to normal functioning and improvements in wellbeing as core indicators of recovery. Beaudequin and colleagues set out to examine how oral ketamine treatment for chronic suicidality affects both negatively valenced outcomes (suicidal ideation and depressive symptoms) and positively valenced outcomes (social/occupational functioning and wellbeing). Using data from the Oral Ketamine Trial on Suicidality (OKTOS), they applied Bayesian network (BN) modelling to quantify interactions among these four outcome domains and to explore which endpoint(s) best capture treatment effects in this sample.

This paper reports a secondary analysis of selected data from OKTOS, a single-arm, ten-week clinical trial conducted at the Thompson Institute (Queensland, Australia) between August 2018 and October 2019. The trial comprised six weeks of flexible-dose oral ketamine (one dose per week, six doses total) followed by a four-week no-treatment follow-up. Initial oral dosing began at 0.5 mg/kg and was titrated up or down at subsequent visits according to tolerance, with a maximum dose of 3.0 mg/kg at the sixth treatment. Safety and tolerability monitoring included a side-effect battery, urinalysis and vital signs; the extracted text reports no significant adverse events. Adult participants (18+) were referred for chronic suicidality and all met DSM criteria for MDD. Inclusion required suicidal ideation for more than six months and a Beck Scale for Suicidal Ideation (BSS) score ≥ 6 at screening; major exclusions included psychosis, bipolar mania/hypomania, acute suicidality requiring urgent intervention, uncontrolled cardiovascular disease, neurological pathology, abnormal liver function, pregnancy or breastfeeding. Participants continued their usual psychotropic medications during the trial. Four outcome measures were examined at baseline and at the four-week post-treatment follow-up: the Social and Occupational Functioning Assessment Scale (SOFAS) and World Health Organization Well-Being Index (WHO-5) as positively valenced outcomes, and the Beck Scale for Suicidal Ideation (BSS) and Montgomery–Åsberg Depression Rating Scale (MADRS) as negatively valenced outcomes. SOFAS was clinician-rated and categorised as: ≥ 80 healthy, 60–79 moderate difficulty, < 60 seriously impaired. WHO-5 was self-rated and converted to a 0–100% score with > 60% denoting good wellbeing. BSS > 6 indicated clinically significant suicidality; MADRS categories were used to classify depression severity. Analyses combined conventional non-parametric statistics and BN modelling. Data collection used tablet-based questionnaires (Qualtrics) with data prepared in SPSS v26 and Python SciPy. Due to non-normal distributions, Spearman's rho was used for correlations; categorical data were summarised as frequencies and percentages and two-tailed tests used a significance threshold of p < 0.05. Sample sizes for analysis were 32 at baseline and 30 at follow-up. For BN construction, the team discretised continuous measures according to predefined cutoffs, built simple four-node networks reflecting different directed-edge structures, and selected the model with highest overall accuracy and area under the ROC curve (AUC ROC) using GeNIe software. Scenario analyses were conducted by introducing new evidence to nodes and calculating percent change (Δ%) in node-state probabilities to quantify multivariate interactions.

Participant flow and baseline characteristics: Of 64 people screened, 40 met inclusion criteria for OKTOS and eight of those withdrew prior to completing the protocol, leaving 32 participants analysed at baseline and 30 at follow-up. Mean age at baseline was 45.7 ± 14.2 years and 50% were female. Nine participants (28%) were employed, 41% were partnered and 59% had tertiary education. All baseline participants met criteria for chronic suicidal ideation (BSS) and the majority (28/32, 87%) had severe depression on the MADRS; all had moderately or seriously impaired functioning on SOFAS prior to treatment. Change in outcomes: Pre–post trajectories showed improvements across domains but effect sizes differed. At baseline no participant met the SOFAS healthy-function threshold (minimum SOFAS 40, maximum 70, median 60). At four-week follow-up 43% (13/30) had SOFAS ≥ 80 (minimum 30, maximum 90, median 70), though seven participants had lower SOFAS scores at follow-up than at baseline. For wellbeing, 6% (2/32) met the WHO-5 > 60% threshold at baseline (minimum 0, maximum 80, median 16); at follow-up 27% (8/30) met the threshold (minimum 0, maximum 88, median 20), and seven participants recorded lower WHO-5 scores at follow-up than at baseline. The authors note that effect sizes for functioning and wellbeing were smaller than for suicidality and depression. Associations and correlations: At baseline there were no significant associations among the four outcomes except a moderate positive correlation between SOFAS and WHO-5 (Spearman's rho = 0.4, p = 0.027). By follow-up, stronger relationships emerged: wellbeing correlated negatively with depression (rho = -0.8) and suicidality (rho = -0.6); functioning correlated negatively with depression (rho = -0.6) and suicidality (rho = -0.5). Bayesian network findings and scenario modelling: The team evaluated simple four-node BNs with each outcome treated as a potential target node; model accuracies were 70% (functioning as target), 77% (depression), 90% (suicidality) and 97% (wellbeing). The BN with wellbeing as the outcome node achieved the highest accuracy (97%) and an AUC ROC of 98% for both 'good' and 'poor' wellbeing states, leading the authors to identify wellbeing as the data-derived treatment endpoint in this sample. Scenario analyses quantified probabilistic effects: for example, no depression after treatment increased the chance of good wellbeing by 139% and healthy functioning by 134%; mild depression reduced the chance of good wellbeing by 50% and healthy functioning by 39%; absence of clinically significant suicidality increased the chance of good wellbeing by 100% and healthy functioning by 24%. Other multivariate scenarios showed complex dependencies, such as residual moderate functional difficulty despite no depression being associated with a 63% higher chance of no clinically significant suicidality but a 72% reduced chance of good wellbeing, while persistent suicidality together with moderate functional impairment produced 100% reductions in chances of good wellbeing and no depression.

Beaudequin and colleagues interpret their findings as evidence that oral ketamine treatment for chronic suicidality can improve both negatively valenced outcomes (depression and suicidal ideation) and positively valenced outcomes (functioning and wellbeing), but that the magnitude and timing of these changes differ. They highlight a notable lag and smaller effect sizes for functional recovery and wellbeing compared with symptomatic remission, consistent with prior literature indicating that symptom improvement does not automatically translate into full restoration of function or quality of life. The Bayesian network analyses are presented as illuminating the complex interplay among domains and identifying wellbeing as the most informative treatment endpoint in this dataset. The authors suggest potential mechanisms for dissociation between symptom change and functional/wellbeing recovery, including ketamine's procognitive effects and modulation of reward-related neural circuits (orbitofrontal cortex, ventral striatum, ventral tegmental area), and propose that cognitive impairment may mediate functional recovery. They further note that residual symptoms, duration and quality of remission, treatment maintenance and patient adherence may influence longer-term functional outcomes. The investigators acknowledge several limitations that constrain interpretation: the small sample size (intentional for this feasibility trial), absence of a control arm which limits causal inference, potential insensitivity of SOFAS as a measure of domain-specific functioning, short follow-up that precludes assessment of durability, and loss of information inherent in discretising continuous variables for BN modelling. Given these caveats, the authors caution that conclusions about ketamine's role in improving positive outcomes should be tentative. They argue, however, that inclusion of functional measures and subjective wellbeing in antidepressant trials and in shared decision-making is important, and they propose BN modelling as a useful tool for identifying data-driven endpoints and for visualising multivariate interactions that may support personalised clinical care.