From the Editor

He had several antidepressant trials. rTMS was helpful but the improvement faded quickly. Should he try ketamine? My patient had read good things and asked.

In a new paper for JAMA Psychiatry, Ana Jelovac (of Trinity College Dublin) and her co-authors attempt to answer that question. 62 hospitalized patients with depression were randomly assigned to receive either repeated ketamine or midazolam treatment and were followed for 24 weeks afterwards. “Serial adjunctive ketamine infusions were not more effective than serial midazolam infusions in reducing depressive symptoms in inpatients receiving usual psychiatric care.” We consider the paper and its implications.

How has the supply of US psychiatric beds changed with time? In the second selection, from JAMA Psychiatry, Karen Shen (of Johns Hopkins University) and her co-authors drew on US databases, finding a slight reduction in overall beds but perhaps an increase in acute care supply, albeit with an increase in beds from large for-profit hospital chains. “Given reports of safety concerns at large for-profit chains, our findings also underscore the need for research on the effects of growing corporatization of inpatient mental health care on patient outcomes.”

And in the third selection, published on their website, ChatGPT staff write about recent controversies involving those with mental health problems, suggesting that the organization has been moved to action. The essay describes their efforts to make advice safer and more appropriate for users who are psychotic, suicidal, or becoming emotionally reliant on AI. “We worked with more than 170 mental health experts to help ChatGPT more reliably recognize signs of distress, respond with care, and guide people toward real-world support – reducing responses that fall short of our desired behavior by 65-80%.”

DG

Selection 1: “Serial Ketamine Infusions as Adjunctive Therapy to Inpatient Care for Depression: The KARMA-Dep 2 Randomized Clinical Trial”

Ana Jelovac, Cathal McCaffrey, Masashi Terao, et al.

JAMA Psychiatry, 22 October 2025 Online First

As many as one-third of patients with major depression do not achieve remission with oral antidepressants. Although single ketamine infusions produce rapid antidepressant effects, these benefits typically dissipate within days. Esketamine nasal spray has received regulatory approval in multiple jurisdictions, yet off-label intravenous racemic ketamine use is widespread and increasing, with some evidence suggesting that it may be more effective than esketamine… 

A key challenge in evaluating ketamine’s efficacy is the difficulty in maintaining blinding due to its distinctive dissociative effects. Meta-analytic evidence shows substantially larger effect sizes in saline-controlled trials (Cohen d, 1.8) compared to those using midazolam as a psychoactive placebo (Cohen d, 0.7). While a number of saline-controlled or single-infusion ketamine vs midazolam comparisons are available, the evidence from midazolam-controlled trials of serial adjunctive ketamine infusions, reflecting contemporary real-world clinical practice, remains sparse. In the largest such trial to date, Shiroma et al randomized 54 outpatients with depression to receive 6 thrice-weekly infusions of ketamine or midazolam (with midazolam participants crossing over to ketamine after 5 infusions) and found no significant difference in Montgomery-Åsberg Depression Rating Scale (MADRS) scores 24 hours after the final infusion. However, there was a significant advantage for ketamine after 5 infusions, before the midazolam arm crossover. In a pilot trial, Gallagher et al found no difference in 24-item Hamilton Depression Rating Scale outcomes between 4 once-weekly ketamine or midazolam infusions in 25 inpatients at end-of-treatment assessment.

So begins a paper by Jelovac et al.

Here’s what they did:

  • They conducted a “double-blind, randomized, midazolam-controlled, pragmatic trial conducted at an academic center in Ireland between September 2021 and August 2024.” 
  • Participants were adults experiencing a major depressive episode with a baseline Montgomery-Åsberg Depression Rating Scale (MADRS) score of 20 or higher. 
  • “Participants were randomized 1:1 to receive up to 8 twice-weekly intravenous infusions of either ketamine (0.5 mg/kg) or midazolam (0.045 mg/kg) as an adjunct to usual-care pharmacotherapy and other aspects of routine inpatient psychiatric care.” 
  • Participants were followed up for six months.
  • Primary outcome: change in depression severity measured by the MADRS score from baseline to end of treatment.

Here’s what they found:

  • Of 371 eligible patients invited to participate, 65 were randomized and 62 were included in the analysis.
  • Demographics & illness experience. The mean age was 53.5 years; the majority were male (59.7%) and all were White. The mean MADRS at baseline was 28.8 and the vast majority had MDD (90.3%).
  • MADRS scores. “Adjusting for baseline score, there was no statistically significant difference in the primary outcome, end-of-treatment MADRS scores, favoring ketamine (n = 32) over midazolam (n = 30) (adjusted mean difference, −3.16…).” (!!)
  • Secondary outcomes. The analysis showed no significant advantage of ketamine over midazolam. 
  • Adverse events. They occurred in 10% or more of patients including fatigue, sleep disturbance, and abnormal liver function test results.

A few thoughts:

1. This is a good study published in a major journal. There is much to like here, including the comparison with midazolam (an attempt to maintain blinding – a challenge given ketamine’s dissociative effects, meaning that past work may have been tainted by expectancy bias).

2. The key finding in a sentence: “In this randomized clinical trial, serial intravenous ketamine was not significantly more effective than serial midazolam in reducing depressive symptoms, and no significant benefit was observed on any secondary efficacy, cognitive, cost-effectiveness, or quality-of-life outcome.”

3. More: “Despite widespread clinical enthusiasm based on early reports of large rapid antidepressant effects of single ketamine infusions, our finding of a small between-group difference falls below both the effect sizes reported in prior open-label or saline-controlled studies, and the minimal clinically important difference for the MADRS.”

4. Interesting.

5. How to explain the improvement seen with midazolam? Placebo? The general benefits of an inpatient setting?

6. There are clear limitations with this study, including the small sample size. The authors note the challenges of recruitment during the pandemic. As well, blinding was problematic (most participants receiving ketamine speculated that).

7. How should we interpret the conflicting results across different studies? Dr. Richard Braithwaite (of the Sussex Partnership NHS Foundation Trust) and his co-authors reviewed recent research comparing ketamine and ECT. “Close scrutiny shows that this literature includes methodologically flawed trials and meta-analyses that threaten the integrity of evidence-based medicine and risk steering patients and clinicians towards what might be a less effective treatment.” That Comment can be found here:

https://www.thelancet.com/journals/lanpsy/article/PIIS2215-0366(25)00300-1/abstract

8. Is part of the discrepancy in results owing to the biases of the study authors, a case of ECT versus ketamine camps? (For the record, Dr. Braithwaite is the lead psychiatrist for ECT at his hospital.)

9. A quick word of thanks to Drs. Tyler Kaster and Daniel Blumberger (both of the University of Toronto) in helping me better understand this paper. The podcast interview of Dr. John Torous (of Harvard University) with the corresponding author is particularly lucid and clear. It can be found here:

https://edhub.ama-assn.org/jn-learning/audio-player/19011950. 

The full JAMA Psychiatry paper can be found here:

https://jamanetwork.com/journals/jamapsychiatry/fullarticle/2840552



Selection 2: “National Trends in Inpatient Psychiatric Bed Supply, 2011 to 2023”

Karen Shen, Mark Olfson, Adam Sacarny

JAMA Psychiatry, 10 September 2025  Online First

While most mental health care is delivered in outpatient settings, patients in crisis may require inpatient psychiatric care. In the US, patients receive this care in freestanding psychiatric hospitals and psychiatric units of general hospitals. However, there are ongoing concerns that demand for inpatient beds has outpaced supply. Psychiatric patients are disproportionately boarded in emergency departments, and many general hospitals have closed psychiatric units. Yet, there are signs of concurrently increasing for-profit activity, particularly expansions of large for-profit hospital chains. While for-profit entry may alleviate access problems, it may also exacerbate concerns about safety and quality of care. However, there is limited research on recent trends in psychiatric bed supply, including trends in large for-profit chains.

So begins a research letter by Shen et al.

Here’s what they did:

  • They combined 2011 to 2023 data from databases (the Healthcare Cost Report Information System and the American Hospital Association Annual Survey). 
  • They also used these databases to identify freestanding psychiatric hospitals and psychiatric beds within general hospitals.
  • They assessed total beds, then looked at ownership.

Here’s what they found:

  • Total psychiatric bed capacity fell slightly from 115 350 beds to 112 986 beds. 
  • Composition. The composition of beds shifted, with “shrinking capacity at general hospitals and growth of the freestanding hospital sector.” The share of beds at freestanding facilities rose from 65.3% to 74.4%.
  • Large for-profit chains. Bed supply in this group rose by 14 474 beds, from 11% of all freestanding beds in 2011 to 27% in 2023. 

A few thoughts:

1. This brings good data to the discussion.

2. How to understand the overall changes? “The decline in total bed capacity was driven by general hospitals and government-owned freestanding psychiatric facilities. Concurrent expansion of bed supply at freestanding facilities owned by large for-profit chains nearly offset that decline. Because government-owned facilities largely serve long-term patients, these results may imply an expansion of overall capacity for patients needing acute inpatient psychiatric care.”

3. That said, they offer a cautionary note: “The supply of psychiatric beds is increasingly concentrated in large chains that have received media attention related to quality of care.”

The full JAMA Psychiatry research letter can be found here:

https://jamanetwork.com/journals/jamapsychiatry/article-abstract/2838698

Selection 3: “Strengthening ChatGPT’s responses in sensitive conversations”

Open AI

OpenAI.com, 27 October 2025

We recently updated ChatGPT’s default model to better recognize and support people in moments of distress. Today we’re sharing how we made those improvements and how they are performing. Working with mental health experts who have real-world clinical experience, we’ve taught the model to better recognize distress, de-escalate conversations, and guide people toward professional care when appropriate. We’ve also expanded access to crisis hotlines, rerouted  sensitive conversations originating from other models to safer models, and added gentle reminders to take breaks during long sessions.

We believe ChatGPT can provide a supportive space for people to process what they’re feeling, and guide them to reach out to friends, family, or a mental health professional when appropriate. Our safety improvements in the recent model update focus on the following areas: 1) mental health concerns such as psychosis or mania; 2) self-harm and suicide; and 3) emotional reliance on AI. Going forward, in addition to our longstanding baseline safety metrics for suicide and self-harm, we are adding emotional reliance and non-suicidal mental health emergencies to our standard set of baseline safety testing for future model releases.

So begins an unsigned essay. 

They note their problem-solving approach, beginning with an attempt define and measure the problem.

“Mental health symptoms and emotional distress are universally present in human societies, and an increasing user base means that some portion of ChatGPT conversations include these situations. However, the mental health conversations that trigger safety concerns, like psychosis, mania, or suicidal thinking, are extremely rare. Because they are so uncommon, even small differences in how we measure them can have a significant impact on the numbers we report.”

They explain their success: “We estimate that the model now returns responses that do not fully comply with desired behavior under our taxonomies 65% to 80% less often across a range of mental health-related domains.”

They review the three problems that they have focused on, commenting that few users have conversations: 0.07% of users in any given week for psychosis, mania, and severe health symptoms; 0.15% for self harm and suicide; and 0.15% for emotional reliance on AI. They then note their improvements and training.

Psychosis, mania, and other severe mental health symptoms 

“On challenging mental health conversations, experts found that the new GPT‑5 model, ChatGPT’s default model, reduced undesired responses by 39% compared to GPT‑4o (n=677)… On a model evaluation consisting of more than 1,000 challenging mental health-related conversations, our new automated evaluations score the new GPT‑5 model at 92% compliant with our desired behaviors under our taxonomies, compared to 27% for the previous GPT‑5 model.”

Self harm and suicide 

“On challenging self harm and suicide conversations, experts found that the new GPT‑5 model reduced undesired answers by 52% compared to GPT‑4o (n=630)… On a model evaluation consisting of more than 1,000 challenging self harm and suicide conversations, our new automated evaluations score the new GPT‑5 model at 91% compliant with our desired behaviors, compared to 77% for the previous GPT‑5 model.”

Emotional reliance on AI 

“On challenging conversations that indicate emotional reliance, experts found that the new GPT‑5 model reduced undesired answers by 42% compared to 4o (n=507)… On a model evaluation consisting of more than 1,000 challenging conversations that indicate emotional reliance, our automated evaluations score the new GPT‑5 model at 97% compliant with our desired behavior, compared to 50% for the previous GPT‑5 model.”

They write about the help of experts. “More than 170 of these clinicians (specifically psychiatrists, psychologists, and primary care practitioners) supported our research over the last few months…” They include some sample dialogues; people expressing suicidal thoughts are told about helplines.

“We’ve made meaningful progress but there’s more to do.”

A few thoughts:

1. This is fascinating – a look at the attempts to make an AI chatbot safer. 

2. The (brief) description of the evaluation model is particularly interesting.

3. They have made progress.

4. But is it enough progress? 91% complaince with desired behaviours is a step forward – but still leaves many unfortunate answers, especially worrisome give the incredible volume of ChatGPT conversations. As Dr. Niall Boyce (of Wellcome) observes in his Substack: “even small percentages add up to large absolute numbers with such a widely-used tool.”

5. Past Readings have looked at the problems of AI-provided therapy. For instance, Dr. Andrew Clark (of Boston University) analyzed chatbot responses to clinical situations in a JMIR Mental Health paper, posing as an adolescent and forwarding three detailed, fictional vignettes. “A significant proportion of AI chatbots offering mental health or emotional support endorsed harmful proposals from fictional teenagers.” That Reading can be found here:

https://davidgratzer.com/reading-of-the-week/reading-of-the-week-ai-therapy/

The full OpenAI.com essay can be found here:

https://openai.com/index/strengthening-chatgpt-responses-in-sensitive-conversations

Reading of the Week. Every week I pick articles and papers from the world of Psychiatry.