A retrospective analysis of ketamine intravenous therapy for depression in real-world care settings

Ketamine intravenous therapy (KIT) is established as a rapid treatment for depressive symptoms, but most controlled evidence comes from single-infusion studies conducted at academic centres. Those trials typically show rapid symptomatic improvement within 24 hours and high short-term response rates after a single infusion, but relapse is common within two weeks. As community-based ketamine clinics have proliferated and commonly use repeated-infusion “induction” regimens (typically 4–8 infusions over 2–4 weeks) followed by variable maintenance dosing, there remains limited systematic time-course data using standardised clinical instruments from these real-world settings. Mcinnes and colleagues set out to quantify outcomes of KIT in community clinics across the United States by analysing a large, de-identified dataset collected via a measurement-based care software platform. The primary aim was to evaluate antidepressant response and remission after a KIT induction series using the Patient Health Questionnaire-9 (PHQ-9) and to estimate the short-term durability of response prior to any maintenance infusions. The analysis focuses on patients treated between 2016 and 2020 and emphasises treatment delivered in independent private practices rather than academic trials.

This was a retrospective analysis of de-identified patient-reported outcome data captured by a measurement-based care software platform across 178 independent community ketamine practices in 40 US states between 1 January 2016 and 30 December 2020. The full dataset comprised 9016 patients who received KIT; clinical records logged PHQ-9 responses, treatment dates, infusion labels (induction versus maintenance), doses and infusion durations, and free-text treatment notes. Demographic, diagnostic, medication, and detailed medical-history variables were not captured by the dataset used for this analysis. The study received approval from the Stanford University Institutional Review Board. The PHQ-9 (a nine-item self-report depression scale) was the primary outcome instrument; patients were prompted to complete the PHQ-9 electronically every 14 days via text-linked web portal. For this analysis, induction was operationally defined as 4–8 infusions administered over 7–28 days, with each infusion labelled as an induction infusion in the software. Baseline PHQ-9 was required within one month before induction and the post-induction PHQ-9 was required 14–31 days after the final induction infusion and before any maintenance infusion. When patients dropped out prior to completing induction, their last PHQ-9 was used if it fell at least two weeks after their final infusion. Key outcome definitions followed standard convention: response was a ≥50% reduction in total PHQ-9 score from baseline, remission was PHQ-9 < 5, and suicidal ideation (SI) was defined as a score ≥1 on item 9 of the PHQ-9. Baseline severity categories were 0–9 none/mild, 10–14 moderate, 15–19 moderately severe, and 20–27 severe. Statistical analyses included Chi-square tests for categorical relationships, Cohen’s d for effect size, point-biserial and Pearson/Spearman correlations for associations with baseline severity, and Kaplan–Meier survival analysis to estimate durability of response; significance was set at p < 0.05. Analyses were performed using standard scientific computing libraries and R.

Cohort derivation and sample characteristics: From 9016 patients in the measurement-based care platform, 3518 had completed a 4–8 infusion induction within 7–28 days; of these, 537 patients met the additional requirement of having both a baseline PHQ-9 within one month pre-induction and a post-induction PHQ-9 14–31 days after induction, and therefore comprised the analytic induction cohort. The investigators note substantial missing outcome data across the broader sample and present analyses comparing baseline PHQ-9 distributions for excluded versus included patients to assess selection bias. Primary depression outcomes after induction: In the n = 537 cohort the mean baseline PHQ-9 was 18.1 (SD = 5.3, 95% CI = 17.6–18.5) and the mean post-induction PHQ-9 at 14–31 days was 9.4 (SD = 6.5, 95% CI = 8.9–10.0). The mean raw reduction in PHQ-9 score was 8.7 points (SD = 6.6, median = 8.5), corresponding to a mean percent reduction of 47.1% (median 52.0%). The response rate (≥50% reduction) was 53.6% and the remission rate (PHQ-9 < 5) was 28.9% (n = 155/537). The Cohen’s d effect size for induction was reported as 1.5. A minority of patients worsened after induction: 8.4% (45/537) had an increased PHQ-9 score, including nine patients with an increase ≥5 points. Relationships with baseline severity: The magnitude of PHQ-9 score reduction correlated with baseline severity (Pearson r = 0.44, p < 10^-25; Spearman r = 0.42), indicating larger absolute drops for patients with higher baseline scores. However, response rates were similar across baseline severity categories (none/mild 56.1%, moderate 54.1%, moderately severe 52.2%, severe 53.9%; χ2 (8) = 0.24, p > 0.999). By contrast, remission likelihood declined with greater baseline severity (58.5%, 43.9%, 28.7%, 17.8% across increasing severity categories; χ2 (8) = 42.6, p < 10^-5), and remission status showed a negative point-biserial correlation with baseline severity (r = -0.27, p < 10^-9). The tendency to worsen did not meaningfully correlate with baseline severity. Suicidal ideation outcomes: Two-thirds of the cohort (66.3%, n = 356) endorsed some SI at baseline (PHQ-9 item 9 > 0). Of these, 73.0% (260/356) reported a reduction in SI after induction and 42.7% (152/356) reported no SI at the end of induction. Improvement rates varied by baseline SI intensity (for example, 79.6% improvement among those averaging item-9 = 3). Overall, 6.0% of the full induction cohort reported an increase in SI after induction; 16 individuals experienced increases in both overall PHQ-9 and SI, and another 16 had an increase in SI without an increase in total PHQ-9. Durability of response and maintenance patterns: A Kaplan–Meier analysis among responders estimated the probability of retaining responder status (loss of response defined as return below the ≥50% reduction threshold). Probability of retaining response was 78.3% at 28 days after induction (95% CI = 73.3–83.7%) and 59.9% at 56 days (95% CI = 51.7–69.4%); patients were censored when they began maintenance or stopped reporting PHQ-9s at least every two weeks. Regarding maintenance practice patterns, 52.5% of patients who completed induction entered maintenance care. Responders who entered maintenance received on average 3.6 maintenance infusions (SD = 4.9), remitters 3.7 (SD = 5.0), and the larger group of 3518 patients who completed induction averaged 2.6 maintenance infusions (SD = 5.0). Median time from final induction infusion to first maintenance infusion was 48 days for remitters and 43 days for responders; across all patients entering maintenance the median interval was 28 days, with considerable variation. The investigators highlight uncertainty about whether treatment discontinuation reflects clinical improvement or financial/access barriers.

Mcinnes and colleagues interpret these findings as evidence that a standard multi-infusion KIT induction delivered in community clinics can produce a robust antidepressant response that is reasonably durable over several weeks without immediate maintenance. They emphasise that the observed response rate of 53.6% and remission rate of 28.9% at 14–31 days post-induction compare favourably with some prior induction studies conducted in academic settings and exceed a previously published small community outpatient report; differences between studies may relate to timing of outcome measurement, patient populations, circadian factors related to infusion timing, or other practice differences. The investigators acknowledge real-world sources of bias that could inflate apparent effectiveness, including lack of blinding and patients paying out-of-pocket who may have strong expectancy effects. They also note that the induction response measured 14–31 days after the final infusion may underestimate peak early effects documented immediately post-infusion in some trials. A key safety and clinical concern raised is that a small subset of patients worsened after induction (8.4%) and 6.0% experienced increased suicidal ideation; potential contributors include dysphoric/anxious ketamine responses, comorbidities, adverse effects leading to intolerance, or psychological reactions when a last-resort treatment does not meet expectations. Several important limitations are highlighted by the authors: only 15.3% of patients who completed induction in the platform had complete pre- and post-induction PHQ-9 data and were therefore included in the primary analysis; the measurement platform did not capture demographics, medical or psychiatric history, concomitant medications, or psychotherapy engagement; dosage data were incomplete; and there is clear economic and access selection bias in privately funded KIT. The authors suggest that richer electronic health record datasets that include demographic and clinical history (for example Osmind’s EHR) would permit more informative analyses, including predictors of response, the impact of treatment refractoriness, medication interactions, and integration with psychotherapeutic approaches such as ketamine-assisted psychotherapy. Finally, they call for further long-term outcome data and replication studies to address the limitations inherent to retrospective real-world analyses.

The investigators conclude that KIT induction in community clinics was associated with a substantial antidepressant effect at 14–31 days post-induction (response 53.6%, effect size d = 1.5; remission 28.9%) and that responders had an approximately 80% probability of sustaining response at 4 weeks and approximately 60% at 8 weeks without maintenance infusions. They also note that 42.7% of patients with baseline suicidal ideation no longer endorsed SI after induction, that roughly half of completers elected to continue into maintenance treatment (receiving on average 2–3 maintenance infusions), and that a small but clinically important minority worsened after induction. The authors state that worsening and increased suicidal ideation warrant further study and that better, more comprehensive point-of-care data collection is needed to characterise long-term outcomes and predictors of benefit and harm in real-world KIT practice.