Esketamine Versus Crisis Response Planning Versus Enhanced Treatment as Usual for Suicide Prevention: A Pragmatic Randomized Trial in a Brazilian Municipality (SAVE)

STUDY DESIGN AND OBJECTIVES: SAVE is a single-municipality, pragmatic, three-arm, parallel-group, superiority randomized controlled trial with 1:1:1 allocation and a Hybrid Type 1 effectiveness-implementation approach. It compares a single intravenous esketamine infusion plus enhanced treatment as usual (eTAU), a single Crisis Response Planning (CRP) session plus eTAU, and eTAU alone. The primary objective is to determine whether either active intervention reduces the hazard of a first suicide-related event during 12 months of follow-up compared with eTAU alone. The comparison between CRP and esketamine is exploratory. Secondary objectives concern suicidal ideation, associated symptoms, well-being, quality of life, service use, implementation, and costs. The pragmatic design integrates study procedures into municipal emergency care while retaining standardized eligibility, outcome assessment, and safety procedures. An independent Data and Safety Monitoring Board (DSMB) oversees safety. PILOT COHORT AND CONFIRMATORY COHORT: Recruitment began in June 2026. Following a temporary suspension for an internal data-quality and process audit, recruitment has resumed and the study is currently recruiting. The first 10 randomized participants constitute a pilot cohort enrolled before the current protocol procedures. The updated protocol has institutional ethics approval under CAAE 89878225.0.0000.0068 and specifies the revised substance-related eligibility criteria, visit windows, esketamine dose, EMA schedule, safety contact procedures, and external blinded endpoint adjudication. Pilot participants continue scheduled clinical and safety follow-up through 12 months under the applicable consent arrangements. Their characteristics, clinical observations, and operational findings are reported descriptively, with the procedures and data cutoffs identified. Pilot data are excluded from confirmatory efficacy analyses, and the pilot cohort will be identified separately in participant-flow reporting. The target confirmatory cohort comprises 468 participants randomized subsequently, with 156 per arm. Anticipated total enrollment is therefore 478. SETTING: The study is conducted in the public health network of Indaiatuba, São Paulo, Brazil. Participating services include urgent and emergency units, the reference hospital emergency department, the Medical Specialties Center (CEEM), Psychosocial Care Centers (CAPS), and primary care units (UBS), linked through the municipal Psychosocial Care Network (RAPS). Randomization and allocated interventions take place in emergency care settings, with continuing care through the municipal network. The single-municipality setting will be considered when interpreting generalizability. ELIGIBILITY AND STUDY FLOW: After clinical stabilization, attending clinicians refer potentially eligible patients to the research team. Eligible participants are aged 14 years or older and have either a suicide attempt within the previous 30 days, including an actual, interrupted, or aborted attempt, or current severe suicidal ideation with intent, corresponding to C-SSRS screening items 4 or 5. Non-suicidal self-injury alone does not satisfy the inclusion criteria. Participants must reside within the study catchment area, be able to maintain follow-up contact, and provide informed consent or, for participants aged 14-17 years, assent with guardian consent. A clinical decision that psychiatric inpatient admission is required after emergency evaluation precludes enrollment. Eligibility does not exclude participants on the basis of sex, gender identity, race/color, or socioeconomic status. Exclusions include contraindications to esketamine, pregnancy or breastfeeding, medical instability incompatible with study participation, a primary psychotic disorder, acute psychosis or mania precluding informed participation, inability to maintain follow-up contact, and the substance-related criteria described below. Contraindications to esketamine include relevant aneurysmal vascular disease, arteriovenous malformation, intracerebral hemorrhage, and hypersensitivity. Consent or assent and guardian consent are obtained before study-specific procedures. Screening includes sociodemographic, clinical, psychiatric, and substance-use assessments. Pregnancy testing is performed when applicable. Eligible participants complete baseline assessment on Day 0 before the allocated intervention and are randomized through REDCap. Interventions are delivered in the emergency department or urgent and emergency unit. Clinical treatment takes priority over research procedures. SUBSTANCE USE SCREENING: The Brazilian Portuguese ASSIST is administered during eligibility assessment. Substance-specific involvement scores follow WHO scoring rules: questions 2 through 7 contribute to each score, except that question 5 is omitted for tobacco. Questions 1 and 8 are excluded from these scores. Risk thresholds are 0-10, 11-26, and 27 or higher for alcohol, and 0-3, 4-26, and 27 or higher for other substance classes, corresponding to lower, moderate, and high risk. Substance-related exclusion applies when any of the following is present: 1. Acute intoxication sufficient to preclude valid consent, reliable clinical assessment, or safe delivery of the assigned intervention, as determined by the attending clinician. Re-screening is permitted after clinical recovery if all eligibility criteria remain satisfied. 2. High-risk hallucinogen involvement, defined as a substance-specific ASSIST score of 27 or higher for the hallucinogen category. 3. A documented history of ketamine use disorder identified during baseline psychiatric assessment. The ASSIST hallucinogen category does not identify ketamine exposure specifically. Relevant substance history is therefore also assessed clinically. The hallucinogen threshold is a conservative eligibility criterion and may exclude participants whose high-risk involvement concerns substances other than ketamine; exclusions and their reasons will be reported. Other ASSIST scores do not automatically exclude participation. High-risk involvement with alcohol or other non-tobacco substances is recorded for exploratory moderation analyses and to inform clinical care. RANDOMIZATION, ALLOCATION CONCEALMENT, AND BLINDING: A statistician not involved in outcome assessment generates a computerized sequence using variable permuted blocks and stratification by age group, recorded sex, and index presentation. Allocation is implemented through the REDCap randomization module with role-based permissions and an audit trail. The allocation table and block details are inaccessible to personnel enrolling participants, who cannot view future assignments. Participants and treating clinicians are unblinded. Outcome assessors, the external endpoint adjudicator, and confirmatory data analysts remain blinded to treatment allocation. Assessors do not ask about treatment received, and participants are asked not to disclose their allocation. Accidental disclosure is documented, and an alternative blinded assessor completes subsequent assessment when feasible. Confirmatory analysts receive coded treatment groups without treatment identifiers until the confirmatory analysis is finalized. All T1 assessments are conducted remotely by telephone or video using a centralized blinded assessor in every arm, including participants still in the clinical unit. Assessors do not access treating-team records during these assessments. This standardizes assessment modality and reduces contextual unblinding. Disclosure of allocation for urgent clinical management is limited to personnel who require that information. The reason, personnel involved, and timing are documented. Clinical safety procedures do not routinely disclose treatment allocation to blinded assessors or the adjudicator. INTERVENTIONS: Enhanced treatment as usual: eTAU comprises two study-specified components: early outpatient psychiatric consultation and lethal means safety counseling. Both are offered to participants in every randomized group alongside routine emergency care. The outpatient psychiatric consultation is arranged to take place within seven days of randomization at CAPS or CEEM. Lethal means safety counseling is delivered during the index emergency attendance by a trained research team member, who collaborates with the participant on a plan to reduce access to potentially lethal means. Treating clinicians retain responsibility for clinical decisions, and the municipal mental health network supports continuing care. Concomitant medication, psychotherapy, and service use are recorded. Suicide risk assessment and emergency procedures are applied consistently across groups. Esketamine plus eTAU: Participants allocated to this group receive one intravenous esketamine infusion at 0.375 mg/kg over 40 minutes. Administration takes place in a monitored clinical setting with pulse oximetry, non-invasive blood pressure measurement, and ECG monitoring. A physician is present, and an advanced cardiovascular life support-capable team is immediately available. Structured symptom and safety assessments occur during the infusion and for 240 minutes afterward. Clinical observation continues, with discharge after 24 hours if the participant is clinically stable. Protocol-defined safety thresholds trigger clinical assessment and intervention. These include sustained systolic blood pressure of 180 mmHg or higher or 90 mmHg or lower, diastolic blood pressure of 105 mmHg or higher or 50 mmHg or lower, a blood pressure change of 25 mmHg in either direction, or clinically significant dissociation or agitation. Treatment modification or discontinuation and the reason are documented. Crisis Response Planning plus eTAU: Participants allocated to this group receive one 20-to-45-minute CRP session with a trained clinician. The plan is developed collaboratively and includes personal warning signs, internal coping strategies, reasons for living, social support contacts, and professional and emergency contacts. Participants receive a printed plan and a digital photograph. Clinicians complete a standardized three-hour training program with didactic content, role-play, certification, and continuing supervision. Fidelity checklists are used, and a consented subset of sessions, approximately 10-20%, is audio-recorded for adherence and competence assessment. Fidelity audits target inter-rater agreement of kappa 0.80 or higher. RESCUE TREATMENT AND CONTINUING FOLLOW-UP: After a participant experiences an event meeting the primary endpoint, open-label rescue combining esketamine and CRP may be offered according to clinical eligibility and safety assessment. Participants remain in scheduled outcome and safety follow-up. Rescue does not remove a participant from the confirmatory intention-to-treat population or remove the first qualifying event from the primary analysis. The first qualifying event is analyzed under the original randomized allocation. Subsequent events, rescue exposure, and reasons for incomplete or modified rescue delivery are recorded. Recurrent-event analyses describe burden under the assigned care strategy, including access to rescue, and do not isolate the effect of rescue treatment. Clinical management does not wait for endpoint adjudication. PRIMARY ENDPOINT AND EXTERNAL BLINDED ADJUDICATION: The primary endpoint is time from randomization to the first qualifying suicide-related event during 12 months of follow-up. Qualifying events are an actual, interrupted, or aborted suicide attempt; a suicide-preventive psychiatric admission; suicide death; or self-injury requiring emergency department care. Events are identified through structured C-SSRS interviews, routine clinical surveillance, participant contacts, and medical record linkage. Suicidal intent and individual event components are recorded separately so that non-suicidal self-injury is distinguished from a suicide attempt. Multiple reports describing the same episode are linked rather than counted as independent events. Individual components are also analyzed separately. An external adjudicator blinded to randomized allocation reviews suspected endpoint events using the study definitions and source documentation from interviews, clinical records, and event reports. A study-team member prepares dossiers with treatment-identifying information removed. The adjudicator determines whether each event meets the endpoint definition, its component and occurrence date, and whether multiple reports describe the same episode. Event occurrence, study-team awareness, and adjudication dates are recorded separately. Requests for clarification, inconclusive decisions, and accidental unblinding are documented. Adjudication is separate from symptom assessment, clinical management, and adverse-event reporting. Safety reporting and necessary care proceed without awaiting an adjudication decision. ASSESSMENT SCHEDULE AND VISIT WINDOWS: Assessments occur at baseline (T0), 24 hours (T1), seven days (T2), and weeks 2, 4, 8, 16, 24, 32, 40, and 52 (T3-T10). Monthly event surveillance continues between scheduled visits. Day 0 is the date of randomization and the time origin for the primary endpoint and nominal follow-up schedule. Baseline assessment, also called T0 or V0, is completed on Day 0 before the allocated intervention. T0/V0 has no additional day-level window. T1 is the protocol-defined 24-hour contact after delivery of the assigned intervention. No additional day-level window or late replacement is permitted. An uncompleted T1 assessment is recorded as missing and is not administered retrospectively. The remaining target days and allowed windows are anchored to randomization: T2: Day 7, plus or minus 2 days. T3: Day 14, plus or minus 3 days. T4: Day 28, plus or minus 5 days. T5: Day 56, plus or minus 7 days. T6: Day 112, plus or minus 10 days. T7: Day 168, plus or minus 14 days. T8: Day 224, plus or minus 14 days. T9: Day 280, plus or minus 14 days. T10: Day 364, plus or minus 21 days. Nominal visit labels, intervention timestamps, and actual assessment dates and times are retained. Later visit windows do not shift after a delayed assessment. Assessments outside their permitted windows are retained and identified for sensitivity analyses; uncompleted assessments are recorded as missing. Calendar-month EMA periods are scheduled separately from week-based visits. CLINICAL AND PATIENT-REPORTED MEASURES: Measures include the C-SSRS, Beck Scale for Suicide Ideation (BSI), Patient Health Questionnaire-9 (PHQ-9), Montgomery-Asberg Depression Rating Scale (MADRS), Generalized Anxiety Disorder-7 (GAD-7), Snaith-Hamilton Pleasure Scale Brazilian version (SHAPS-BR), 13-item short Pittsburgh Sleep Quality Index (shortPSQI; five components, global score 0-15), 12-Item Short Form Health Survey (SF-12), 7-item Beck Hopelessness Scale (BHS-7), Short Warwick-Edinburgh Mental Well-Being Scale (SWEMWBS), and Clinical Global Impression scales (CGI-S and CGI-I). Instrument-specific administration follows the study assessment schedule; not every instrument is administered at every visit. Baseline covariates and exploratory moderators include impulsivity (BIS-15), substance involvement (ASSIST), borderline personality features (MSI-BPD), religiosity and spirituality (DUREL), ADHD symptoms (ASRS-S), and manic symptoms (ASRM). ASRM is also repeated at T2 for symptom monitoring. Brazilian Portuguese versions are used, with adapted or abbreviated forms identified in the assessment materials. Sex and gender identity are recorded separately; self-reported race/color is not treated as genetic ancestry. VERIFICATION OF eTAU DELIVERY: Delivery of early outpatient psychiatric consultation and lethal means safety counseling is documented separately in every arm, including dates, completion status, and reasons for non-completion. At T2, structured questions record whether the consultation occurred, its date, and any reason for non-attendance. Municipal scheduling and attendance records from CAPS, CEEM, and UBS are reviewed monthly to verify service contacts. Appointment booking alone is not counted as consultation delivery. Time from randomization to the completed consultation is reported by randomized group as an implementation measure. ECOLOGICAL MOMENTARY ASSESSMENT: The MindLogger smartphone application delivers three prompts per day during each of three periods of 30 consecutive days beginning at baseline and at calendar months 4 and 8 of follow-up. Two prompts occur at fixed times, 09:00 and 21:00. The third occurs at a randomly selected time between 10:00 and 20:00. All times refer to local time in Indaiatuba. Each period contains 90 scheduled prompts, giving 90 assessment days and 270 scheduled prompts per participant across the study. Each prompt has a 60-minute response window and one reminder. The month 4 and month 8 periods are not treated as exact equivalents of the week 16 and week 32 visits. Brief surveys assess momentary affect, self-evaluative states, cognitive and emotional regulation, interpersonal context, recent substance use, wish to live or die, suicidal ideation intensity and controllability, perceived ability to resist acting on suicidal thoughts, and recent self-harm. Item-specific ordinal and categorical response formats are used. Scheduled and actual prompt and response times are recorded. Completion is calculated as answered prompts divided by delivered prompts, with delivered prompts also reported against the planned schedule. Completion is described separately by sampling period and prompt type. Days outside the three sampling periods are planned non-observation days and are not counted as missed prompts. EMA analyses are exploratory and account for repeated observations within participants, sampling period, prompt type, and actual response timing. They examine day-level trajectories and prospective associations with subsequent outcomes. A completion threshold of 50% of delivered prompts defines the main analyzable EMA sample, with completion and sensitivity analyses reported by period. EMA analyses are not included in confirmatory family-wise error control. Pilot EMA data collected under earlier schedules are described separately. EMA SAFETY CONTACT AND ESCALATION: Responses indicating the highest level of current suicidal ideation, self-harm since the previous prompt, or inability to resist acting on suicidal thoughts trigger a safety alert. The study team manages alerts through a dedicated mobile phone with WhatsApp and contacts the participant within 24 hours of the alert. The 24-hour interval refers to elapsed time, including weekends and holidays. Contact includes clinical risk assessment and referral to emergency services when indicated. The alert, contact attempts, assessment, and resulting action are documented in REDCap. The application displays crisis contact information. Participants are informed that the application and study WhatsApp contact do not provide continuous emergency monitoring and that urgent care should be sought immediately when needed rather than awaiting the study response. BIOMARKERS AND LABORATORY PROCEDURES: Venous blood is collected at baseline for exploratory biomarker analyses. Collection tubes, sample handling, and processing follow assay-specific laboratory requirements. Samples are analyzed through the hospital laboratory and are not retained in a research biobank; residual samples are disposed of after reporting in accordance with the approved procedures. The planned panel includes high-sensitivity C-reactive protein, homocysteine, vitamins B1, B2, B6, B12 and D, folate, magnesium, zinc, selenium, thyroid-stimulating hormone, free T3, free T4, testosterone, DHEA-S, prolactin, and sex hormone-binding globulin. Exploratory analyses examine associations between baseline biomarkers and subsequent outcomes, including potential moderation of treatment effects. These analyses are not powered to establish definitive biomarker-defined subgroup effects. ENROLLMENT TARGET AND STATISTICAL PLANNING: The target confirmatory sample is 468 participants, with 156 per group, in addition to the 10 pilot participants. For the primary time-to-first-event endpoint, statistical information depends on the number of participants experiencing a first qualifying event. Recurrent episodes in the same participant do not count as additional first events for sample-size planning. Planning for the two active-versus-eTAU comparisons uses a conservative two-sided alpha of 0.025 per comparison, a nominal power target of 90%, and equal allocation within each pairwise comparison. Using Schoenfeld's method, target hazard ratios of 0.50 for esketamine and 0.39 for CRP correspond to approximately 103 and 56 required first events, respectively. These are assumed planning effects, not estimates from the pilot. Initial planning assumptions include a 30% 12-month event risk in eTAU, 24 months of accrual, 12 months of individual follow-up, and allowance for 20% attrition. The enrollment target alone does not establish that the required event totals or nominal power will be achieved. The statistical planning documentation will specify the translation from required events to enrollment and examine sensitivity to control risk, accrual, follow-up, and loss to follow-up. Any change to the enrollment target will require a documented revised calculation and corresponding protocol and registry updates. Pilot observations may inform recruitment, follow-up, and event-ascertainment assumptions. A pooled short-term event proportion from 10 participants is not substituted for the 12-month control-group risk and is not used to estimate the target treatment effect. STATISTICAL ANALYSIS: Confirmatory analyses include all participants randomized in the post-pilot cohort and follow the intention-to-treat principle, retaining participants in their assigned groups regardless of adherence, treatment discontinuation, or subsequent rescue. Pilot participants are reported descriptively and excluded from confirmatory efficacy analyses. The two confirmatory comparisons are esketamine plus eTAU versus eTAU and CRP plus eTAU versus eTAU. Holm's sequential procedure controls the two-sided family-wise type I error at 0.05: the smaller P value is compared with 0.025 and, if that hypothesis is rejected, the remaining P value is compared with 0.05. Effect estimates are reported with 95% confidence intervals, with nominal intervals distinguished from multiplicity-adjusted inference. CRP versus esketamine is exploratory. Primary analysis: Cox proportional-hazards models estimate hazard ratios, adjusting for age group, recorded sex, index presentation, baseline suicidal ideation, and baseline PHQ-9 score. Kaplan-Meier curves describe event-free follow-up. Schoenfeld residuals are used to assess proportional hazards; time-varying effects or restricted mean survival time estimates over the specified follow-up horizon are considered when appropriate. No municipality-level covariate is included because the trial is conducted in one municipality. Death from a cause other than suicide is addressed as a competing event in sensitivity analyses. Recurrent-event and component analyses: An exploratory negative-binomial model with a person-time offset estimates incidence rate ratios for distinct suicide-related episodes, including recurrences, using the same covariate set. Episode definitions and adjudication rules are consistent with the primary analysis. Further exploratory analyses include recurrent-event models, mean cumulative event functions, and analyses of individual endpoint components. Competing-risk analyses explicitly distinguish the event of interest from competing events. These analyses are outside confirmatory family-wise error control. Secondary outcomes: Continuous symptom outcomes, including suicidal ideation, depression, anxiety, anhedonia, sleep, hopelessness, well-being, and quality of life, are analyzed using mixed-effects models with participant random intercepts and fixed effects for randomized group, time, and group-by-time interaction. Actual assessment timing is retained, and sensitivity analyses are restricted to assessments within permitted windows. Repeated binary outcomes are analyzed using generalized estimating equations with a logit link. Economic evaluation: Incremental costs and quality-adjusted life years (QALYs) will be compared between each active intervention and eTAU from a societal perspective. Costs include direct medical expenditure and productivity losses. QALYs will be estimated using SF-6D utility values derived from SF-12. Incremental cost-effectiveness ratios and uncertainty analyses will be reported, with interpretation across a range of willingness-to-pay thresholds specified in the economic analysis plan. Sensitivity, moderation, and missing-data analyses: Per-protocol analyses are supplementary to intention-to-treat analyses. Exploratory moderation analyses examine treatment interactions with prior attempts, baseline severity, ASSIST high-risk status, and biomarkers, with effect estimates and confidence intervals interpreted cautiously. These analyses do not establish definitive subgroup effects. Participants without an observed endpoint event are censored at the last date on which endpoint status can be established, subject to the defined follow-up horizon. Inverse probability weighting assesses sensitivity to dropout, and multiple imputation addresses missing baseline covariates when appropriate. Longitudinal analyses examine missing-at-random assumptions and use sensitivity analyses, including pattern-mixture approaches, to assess departures from those assumptions. Missed assessments, discontinuation of an intervention, loss to follow-up, and withdrawal of consent are distinguished. Analyses will use R and Stata. The detailed statistical analysis plan will be finalized before treatment identities are revealed to confirmatory analysts. IMPLEMENTATION EVALUATION: The implementation evaluation uses ImpRes-BR to organize planning and evaluation, CFIR 2.0 to assess contextual determinants, ERIC to specify implementation strategies, and a Power/Interest matrix for stakeholder mapping. The Acceptability of Intervention Measure, Intervention Appropriateness Measure, and Feasibility of Intervention Measure (AIM/IAM/FIM) are administered to participating professionals at weeks 4, 24, and 52 of service implementation. Formative assessments immediately after training are distinguished from these scheduled outcomes. Strategies include clinician training and capacity building; stakeholder engagement with municipal managers, emergency services, and CAPS; audit and feedback using quarterly service reports; contextual adaptation through interviews and focus groups; and integration of study procedures into routine clinical workflows through safety checklists, documentation tools, and defined responsibilities. The evaluation combines professional surveys, fidelity audits, service-use measures, and semi-structured interviews. Qualitative sampling is purposive, with an anticipated approximately 20 professionals per participating unit, adjusted according to thematic saturation. Qualitative analysis uses a documented coding framework and procedures for resolving coding disagreements. DATA MANAGEMENT AND CONFIDENTIALITY: Study data are captured in REDCap with role-based access, range checks, and audit trails. Direct identifiers are stored separately from analytic datasets. Treatment-identifying information is restricted according to study roles, and adjudication dossiers are prepared to preserve allocation blinding. Data handling follows applicable Brazilian data protection requirements, including the Lei Geral de Proteção de Dados. Participant reports and publications use aggregate or appropriately de-identified information. SAFETY MONITORING AND OVERSIGHT: Suicide risk assessment is incorporated into clinical contacts across all groups. Adverse events and serious adverse events, including suicide attempts, hospitalizations, and deaths, are documented and reported to the responsible ethics committee and other authorities as required. Clinical management, safety reporting, and endpoint adjudication serve distinct purposes; none of the required clinical responses is delayed pending research classification. The independent DSMB reviews cumulative safety data quarterly and when additional review is needed. It may recommend continuation, modification, or suspension. Unblinded safety information is restricted to personnel with an oversight need and is not routinely shared with blinded assessors, the adjudicator, or confirmatory analysts. ETHICS AND REGULATORY COMPLIANCE: The study received approval from the Ethics Committee of the Hospital das Clínicas, Faculdade de Medicina, Universidade de São Paulo, under CAAE 89878225.0.0000.0068, approval number 7.723.561. The updated protocol has ethics approval under the same CAAE. Municipal authorizations were obtained for conduct within the Indaiatuba public health network. The study follows the approved consent procedures, applicable Brazilian requirements for research involving human participants, and the Declaration of Helsinki. Participants receive updated information or renewed consent when required by approved procedures. Withdrawal from research does not restrict access to usual clinical care. Substantive protocol changes are communicated to the ethics committee, registry, and participating teams in accordance with applicable requirements.