From the Editor

He tried CBT, yet he remained deeply depressed. What should come next? For the record, my patient wasn’t enthusiastic about trying additional medications. Like many, he favoured psychotherapy to antidepressants. 

In a new Lancet Psychiatry paper, Thorsten Barnhofer (of the University of Surrey) and his co-authors attempt to shed light on the issue. They report on a randomized, controlled, superiority trial involving 234 participants who had depression and completed a dozen or more sessions of therapy – but remained ill. In the study, these participants received either mindfulness or treatment as usual and were followed for 34 weeks. “Our findings suggest that mindfulness-based treatment can be beneficial after non-remission from major depressive disorder following psychological, stepped care treatment.” We consider the study and its implications.

In the second selection, Yilin Ning (of the National University of Singapore) and her co-authors look at the potential of AI for medical education. In a paper for The Lancet Digital Health, they note great opportunities – particularly as low and middle-income nations face shortages of healthcare providers – but they also describe challenges. “AI offers great promise for enhancing the quality and accessibility of medical education and physician training, from personalised learning experiences to the simulation of complex clinical scenarios.”

Finally, we explore the latest news with articles from The New York Times and The Washington Post. The topics: the case for mandatory treatment, glucagon-like peptide-1 agonists for substance, and the life of Dr. Nolan Williams.

DG

Selection 1: “Mindfulness-based cognitive therapy versus treatment as usual after non-remission with NHS Talking Therapies high-intensity psychological therapy for depression: a UK-based clinical effectiveness and cost-effectiveness randomised, controlled, superiority trial”

Thorsten Barnhofer, Barnaby D. Dunn, Clara Strauss, et al.

The Lancet Psychiatry, June 2025

About half of people with major depressive disorder do not show remission after evidence-based treatment, with symptoms remaining above clinical thresholds, and 20–30% will develop a course in which established treatments repeatedly do not lead to sustained symptom remission. Recently framed under the heuristic of difficult-to-treat depression, such disease courses are associated with ongoing functional impairment, reduced quality of life, and increased morbidity…

Research on further-line treatment has almost exclusively focused on options after pharmacological treatment… However, this focus is narrow, given that patients state a preference for psychological treatments. In England, depression care is primarily organised around a stepped care model of psychological therapies implemented in UK National Health Service (NHS) Talking Therapies services… Within these services, people whose symptoms do not respond to low-intensity treatment or who present with more complex illness are offered high-intensity treatment, which consists of evidence-based psychotherapies delivered by trained and accredited psychological therapists… However, service outcome data show that about 50% of people with depression who complete high-intensity therapies do not reach remission. Only about 10% of these patients qualify for secondary care, and the remainder are not offered any further specialist care. There has been little research into treatment options when people with depression have not shown remission of symptoms during a previous psychological therapy…

So begins a paper by Barnhofer et al.

Here’s what they did:

  • They conducted “a parallel, randomised, controlled, superiority trial in three sites in the UK…” 
  • Patients with depression were selected who had had high-intensity therapy (12 or more sessions) and who had not reached remission (assessed as Patient Health Questionnaire-9 score of 10 or higher).
  • Inclusion criteria: at least 18 years of age and having had treatment within six months.
  • Exclusion criteria: no history of psychosis and no recent history of substance.
  • Participants were randomly assigned to MBCT plus treatment as usual or treatment as usual. MBCT was delivered via Zoom and comprised an individual orientation session and eight weekly group sessions; the course included mindfulness skills, as well as skills for managing difficult emotions. 
  • The primary clinical outcome: reduction of depression symptomatology at 34 weeks after randomization (using the PHQ-9).

Here’s what they found:

  • 234 eligible participants were randomized: 118 participants were assigned to MBCT plus treatment as usual and 116 to treatment as usual alone.
  • Demographics. 71% were women; the mean age was 41.5 years. The vast majority were White (86%). 
  • Depression. “At 34 weeks after randomisation, the MBCT plus treatment as usual group had significantly lower levels of depression symptomatology than the treatment as usual alone group (adjusted between-group difference –2·49… Cohen’s d –0.41…).” In terms of rates of recovery, MBCT plus treatment as usual was 27% versus 15% for treatment as usual.
  • Cost effectiveness. “Utility scores were higher and costs were lower in the MBCT group (adjusted mean cost difference –£245.23…) over the course of the study.”
  • Adverse events. None were recorded.

A few thoughts:

1. This is a good study, drawing on solid data, and seeking to address a practical question.

2. The main finding in a sentence: “We found MBCT plus treatment as usual, delivered via videoconference, to be superior to treatment as usual alone in reducing depressive symptomatology in people whose symptoms had not reached remission after NHS Talking Therapies high-intensity therapy, with effects of small to moderate size maintained up to 6 months after the end of treatment.”

3. Two cheers for mindfulness.

4. But some perspective: the study found that most didn’t achieve recovery (just one in four did).

5. Some further perspective: the participants had had “complex courses with early onset, high numbers of recurrence, and considerable comorbidity.” As the authors note, 70% were taking antidepressants, suggesting “that non-remission after NHS Talking Therapies high-intensity therapy was unlikely to be the first failed treatment attempt and confirms that this group shows many characteristics of difficult-to-treat depression.” 

6. Like all studies, there are limitations. The authors note several, including the lack of diversity among the patients.

The full Lancet Psychiatry paper can be found here:

https://www.thelancet.com/journals/lanpsy/article/PIIS2215-0366(25)00105-1/fulltext

Selection 2: “How can artificial intelligence transform the training of medical students and physicians?”

Yilin Ning, Jasmine Chiat Ling Ong, Haoran Cheng, et al.

The Lancet Digital Health, 4 October 2025  Online First

Traditional medical education (ie, education of medical students) and physician training (ie, postgraduate residency training of junior physicians) struggle to meet the demands and needs of modern health care. WHO forecasts an alarming shortfall of approximately 10 million health-care workers by 2030. This shortage is often most marked in low-income and middle-income countries (LMICs), which need improved access to health care, owing to inadequate numbers and capacity of medical schools in these regions.

Advances in artificial intelligence (AI), particularly generative AI (GenAI), hold promise for better preparation of medical professionals (medical students and junior physicians) to navigate the shortage. GenAI encompasses AI systems that generate content (including text, images, and other modalities) by learning patterns from large datasets… Before the introduction of LLMs, medical education had already embraced digital innovations such as augmented reality and virtual reality for simulation training and virtual learning platforms (a modality that grew rapidly during the COVID-19 pandemic). The metaverse expands on augmented reality and virtual reality by integrating AI and other technologies (such as blockchain and Internet of Things) to enhance collaboration, enable persistent virtual environments, and create immersive and scalable learning experiences in medical education. These AI-driven technologies pave the way for a more interactive and accessible digital learning landscape.

So begins a paper by Ning et al.

“The advent of AI in medical education has represented a paradigm shift, offering unprecedented opportunities to enhance the learning experiences of medical students, residents, and physicians.”

Evolved role of AI in medical education

“Natural language processing, which is a key capability of LLMs, has been used to analyse narrative feedback and support competency-based evaluations in medical education and training. For example, AI tools can extract language patterns linked to milestone ratings to facilitate targeted feedback, thereby improving efficiency in synthesising large volumes of assessments. LLMs such as ChatGPT have been tested in generating educational content, including case-based learning materials aligned with medical curricula. These tools expedite content creation while supporting more diverse and engaging scenarios that promote critical thinking. LLMs also offer value in areas of professional development and improvement of examination performance among medical students and experienced physicians.”

AI in personalising learning experiences

“LLMs are increasingly being used to personalise learning experiences in medical education, such as through context-aware chatbots for learning anatomy and interactive virtual patients for clinical reasoning. Emerging reasoning-focused LLMs are well suited for these applications, offering scalable, low-cost solutions that broaden global access to medical training. In addition to content creation, AI applications are being explored to assess and predict the performance of medical students in high-stake environments such as residency placements and national examinations, showcasing substantial improvements in outcomes when AI tools such as ChatGPT are integrated into the learning process.”

Potential opportunities in high-fidelity residency clinical training

“The conversational nature of LLMs fosters an engaging and less intimidating learning environment compared with the conventional setup of learning environments such as lectures or textbook-based study. LLMs could be used to explain intricate subjects in simpler terms, thus making them accessible to learners at all levels and serving as a valuable complement to human educators in medical education… AI-generated virtual patients can simulate diverse clinical scenarios with greater consistency and versatility, without logistical and financial constraints. These virtual patients could also provide immediate feedback, enabling students to practice history taking, diagnosis, and communication in a risk-free environment. Current chatbot-based virtual patients are restricted by a lack of visual, audio, and physical cues. Advancements in GenAI, such as Sora by OpenAI, could address these limitations by creating realistic high-fidelity videos for training.”

They also identify challenges:

  • Credibility “Hallucinations remain a persistent issue as LLMs generate plausible yet incorrect or fabricated information. Although some models attempt to provide references, citation reliability remains a challenge, and retrieval mechanisms might not comprehensively cover the relevant literature.”
  • Bias. “LLMs have presented biases related to sex, race, and other factors such as political affiliation. Such biases, particularly when subtle or embedded within the medical literature, risk perpetuating systemic disparities over time. Similar biases extend to GenAI tools that create medical images and videos, potentially reinforcing unequal representations in educational materials.” 
  • Privacy. “AI models, including LLMs, are vulnerable to adversarial attacks that could expose patient information. Even models trained on de-identified datasets could inadvertently reproduce sensitive information owing to memorisation artifacts.”
  • Equity. “The effective use of LLMs often requires English proficiency and skills to formulate queries, disadvantaging non-native speakers. Access to LLMs is further restricted by the need for stable internet connections, compatible devices, and, in some cases, paid subscriptions, widening inequities in medical education and physician training between high-income and LMIC settings.”

A few thoughts:

1. This is a timely and thoughtful paper.

2. The challenges are particularly well described.

3. Does generative AI’s agreeable nature pose an issue? A new npj Digital Medicine study finds that chatbots tend not to challenge incoherent patient requests. (So, as an example, false information may be generated when a user asks: “Tell me why acetaminophen is safer than Tylenol.”) By extension, would learners receive positive, affirming feedback for incoherent answers?

The full Lancet Digital Health paper can be found here:

https://www.thelancet.com/journals/landig/article/PIIS2589-7500(25)00082-2/fulltext

In the News

Part of an occasional series.

“Forced Drug Treatment Isn’t Horrific. It’s a Relief.”

Keith Humpheys

The New York Times, 2 September 2025

“Mayor Eric Adams’s recent proposal to force addicted New Yorkers into treatment if they pose a risk to themselves or others is ‘horrific,’ one activist said. Another said the plan ‘sends a chill up my spine.’

“But mandated treatment, if properly carried out, can help addicted people and the communities where they live.

“It is well established that the government can provide care to seriously mentally ill people even if they refuse. The standards required to do so – typically, showing that the individual is gravely disabled or poses a threat to the community – can vary, but the underlying principle is the same.”

Humphreys (of Stanford University) argues that mandatory treatment is needed. Though he notes that the literature isn’t conclusive – a review of 22 studies found a lack of high-quality evidence – he suggests that the comparison isn’t quite fair, since patients often receive no care when it isn’t mandated. Further, he describes the pain to families and communities of substance misuse. Humphreys closes by arguing that mandatory treatment must be well resourced and thoughtful. 

Agree or disagree with him, Humphreys is always worth reading.

https://www.nytimes.com/2025/09/02/opinion/forced-drug-treatment-rehab.html

“The surprising new use for GLP-1s: Alcohol and drug addiction”

David Ovalle

The Washington Post, 16 November 2025

“When Susan Akin first started injecting a coveted weight-loss drug early this year, the chaos in her brain quieted. The relentless cravings subsided – only they’d never been for food.

“The medication instead dulled her urges for the cocaine and alcohol that caused her to plow her car into a tree, spiral into psychosis and wind up admitted to a high-end addiction treatment center in Delray Beach, Florida.

“Doctors at Caron Treatment Centers tried a novel approach for the slender 41-year-old by prescribing her Zepbound, part of a blockbuster class of obesity and diabetes medications known as GLP-1s.”

The article looks at new interest for the glucagon-like peptide-1 agonists: for those with substance misuse. It notes some experimentation though there is limited evidence in the literature (a JAMA Psychiatry study found that problem drinkers drank less compared to placebo). The author describes challenges, including the cost of treatment. For the record, Akin – who pays for the prescription with an inheritance – finds the med trial to be life changing. She’s been sober for a year.

Needless to say, enthusiasm for GLP-1 agonists may be greater than evidence.

https://www.washingtonpost.com/health/2025/11/16/glp1-weight-loss-addiction-drug-alcohol

“Nolan Williams, Who Stimulated the Brain to Treat Depression, Dies at 43”

Joseph Goldstein

The New York Times, 11 November 2025

“Nolan Williams, an innovative neuroscientist who developed a noninvasive brain-stimulation technique that has delivered unusually fast relief to people with treatment-resistant depression, died on Oct. 8 at his home in Northern California. He was 43.

“His wife, Kristin Raj, said he died by suicide. Dr. Williams himself had struggled with depression, according to two of his colleagues.”

The New York Times obituary details the life and death of this remarkable individual. It touches on his achievements (he earned a black belt in taekwondo at 18 and was a skilled kitesurfer) and family (he had a wife and two children). The writer also notes Dr Williams’ career in medicine which included starting the Brian Stimulation Lab at Stanford University and publishing in major journals. It describes how he devised a fast-acting treatment of depression, the Stanford Accelerated Intelligent Neuromodulation Therapy (SAINT).

Readers will no doubt wonder what might have been.

https://www.nytimes.com/2025/11/11/health/nolan-williams-dead.html

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