This Week in Psychiatry — Aug 17, 2026

Welcome to This Week in Psychiatry. This week we're covering 10 notable papers spanning precision treatment in mood disorders, the risks and realities of antidepressant use and discontinuation, the physical health gap in severe mental illness, and new models for scaling and targeting care. Let's dive in.

We start with two very different attempts to make mood disorder treatment more precise. In the BMJ, De Prisco and colleagues published a living umbrella review that is essentially a map of the entire bipolar disorder evidence base [1]. They pulled together 77 systematic reviews with network or pairwise meta-analyses, covering 116 distinct interventions, from drugs to nutraceuticals to brain stimulation to circadian-based therapies, across 133 outcomes, and ended up grading roughly 2,500 meta-analyses. The sobering headline is the certainty of that evidence: only 236 estimates rated high certainty under GRADE, against nearly 1,450 rated low or very low. Within that, the picture is still clinically usable. For adult bipolar depression, interventions effective across outcomes included cariprazine, divalproex, fluoxetine, lamotrigine, lumateperone, lurasidone, olanzapine alone or with fluoxetine, quetiapine, and ketamine as augmentation, but in children and adolescents only lurasidone and olanzapine with fluoxetine survived. For mania the adult list was long, including lithium, valproate, carbamazepine, haloperidol and most second-generation antipsychotics, while the paediatric list narrowed to aripiprazole, asenapine, olanzapine, quetiapine and risperidone. For maintenance, lithium, valproate, quetiapine, aripiprazole including the long-acting injectable, asenapine, risperidone long-acting injectable, and group psychoeducation as augmentation all held up. The authors also built a free open-access platform with a preference-based tool covering 12 interventions and 17 safety outcomes, which is designed to be used in the room with a patient who cares more about weight gain than about sedation, or vice versa.

Alongside that, the American Journal of Psychiatry published a single-centre, randomised, double-blind, three-arm trial from Li and colleagues asking whether we should stop siting transcranial magnetic stimulation with a tape measure [2]. A total of 123 adults with major depressive disorder or bipolar II depression and at least one antidepressant failure received 20 sessions of intermittent theta-burst stimulation over two weeks, delivered by robotic neuronavigation to either the conventional 5-centimetre site, a functional-connectivity-guided site showing negative resting-state connectivity with the subgenual anterior cingulate, or a structural-connectivity site defined by tractography to the same region. Structural-connectivity targeting beat the 5-centimetre rule on the primary outcome at two weeks, with about a nine-percentage-point greater reduction in Hamilton depression scores and an effect size around 0.7. At six weeks both connectivity-guided arms were ahead of conventional targeting, with the structural arm showing an effect size above 1. Importantly, by week 12 the groups had converged and the differences were no longer statistically significant, and adverse events were comparable with no seizures or mania. So this is promising preliminary evidence for faster and deeper early response with individualised targeting, not yet proof of a durable advantage.

Our second theme is antidepressants — when to keep them and how to come off them. JAMA Psychiatry published a review from Wisner and colleagues that reframes the perinatal SSRI conversation around maternal health [3]. They note that in the United States around five to six percent of pregnant patients take an SSRI, while major depressive disorder affects roughly one in eight, and psychiatric illness is a leading cause of maternal morbidity and mortality. The core methodological argument is confounding by indication: once you disentangle drug effects from the effects of the depression itself and its sequelae, the accumulated evidence suggests SSRIs carry little or no risk of serious adverse outcomes, while several studies document harm from discontinuation in women with moderate to severe illness. Their prescription for practice is a structured decision-making process that presents the risks of untreated or undertreated illness alongside the risks of exposure, and they frame access in maternity and psychiatric care deserts as a public health imperative.

That sits interestingly next to a small but conceptually pointed study in Psychotherapy and Psychosomatics from Hengartner and colleagues on what is actually happening when patients feel worse during tapering [4]. They followed 32 Swiss primary care patients over 26 weeks of antidepressant tapering, generating 187 repeated measurements per variable. Within individuals, changes in depression and anxiety scores tracked closely with changes in the affective withdrawal subscale, correlations around 0.6, and more modestly with discriminatory physical withdrawal symptoms. Crucially, affective, depressive and anxiety scores rose after dose reductions and fell again during intervals when the dose was held steady, and adjusting for physical withdrawal symptoms substantially attenuated those rises. The authors are careful that causality cannot be inferred from 32 patients, but the dose-linked, oscillating pattern accompanied by physical withdrawal symptoms looks much more like withdrawal than like relapse. Practically, if affective symptoms spike within days of a dose reduction and settle when you hold the dose, consider slowing the taper rather than reinstating a relapse diagnosis.

Third theme, and this is where the largest datasets are: the physical health and mortality gap in severe mental illness. The Lancet Psychiatry published a statewide Queensland hospital cohort from Halstead and colleagues linking more than 178,000 people over 24 years, comparing about 30,000 people with schizophrenia-spectrum or bipolar disorder against age- and sex-matched hospital comparators [6]. The gradient is the story. The relative incidence of physical disease rose steadily with the number of organ systems involved, from roughly 1.9 times for at least one affected system up to about 3.2 times for at least five. In absolute terms at 20 years, there were around 18 additional cases per 100 persons with at least two, and still nearly 12 additional cases per 100 with five or more systems affected. The disparity was largest in the youngest strata — in females under 25 with at least five affected systems the hazard ratio was above nine. Multimorbidity here is not a late complication of middle age; it accumulates early.

And the pandemic data in the same journal show that this gap is not closing. Chilman and colleagues analysed linked whole-country records for England, covering over 13.4 million people with COVID-19, of whom about 160,000 had severe mental illness [7]. All-cause mortality after infection was roughly 56 percent higher in people with severe mental illness overall — and rather than shrinking, the excess was smaller in the first pandemic stage, at about 27 percent, and larger during and after vaccine roll-out, at around 55 percent. Vaccination uptake began ahead of the general population, at 73 percent versus 67 percent during roll-out, but by the end of the study fell behind, 79 percent versus 87 percent fully vaccinated. Vaccination only partially attenuated the excess COVID-19 mortality. The message for practice is unglamorous and concrete: proactive metabolic and cardiovascular screening in young patients, and owning vaccination status as part of psychiatric follow-up.

Our final theme is how we target and scale scarce care. In JAMA Psychiatry, Edwards and colleagues built machine-learning models on nearly two million periodic health assessments from more than 668,000 regular United States Army soldiers to identify who should be offered a remote dialectical behaviour therapy skills intervention [5]. Discrimination was moderate for suicide, with an area under the curve of 0.72, and better for suicide attempt at 0.81. But the concentration of risk was limited: suicide risk was meaningfully elevated only in the top five percent of assessments, capturing under a fifth of deaths, and attempt risk only in the top ten percent, capturing about 47 percent of attempts. Suicide and attempt had distinct predictors and only modestly correlated predictions, so this is a resource-allocation tool requiring explicit cost-benefit judgement, not a clinical risk verdict.

Two trials tackle scale from the delivery side. In Psychotherapy and Psychosomatics, Chen and colleagues randomised 1,281 Chinese adults with significant distress on the Kessler-10 to an eight-week hybrid digital mindfulness programme delivered by trainee instructors under a train-the-trainer model, plus usual services, or usual services alone [9]. Distress fell significantly more in the intervention arm, with a moderate effect size around 0.5 at week eight, adherence averaged six of eight sessions, and from a societal perspective the intervention cost less than usual care, with a high probability of cost-effectiveness. No severe adverse events occurred and meditation-related adverse experiences declined over time. Follow-up beyond eight weeks remains unknown. In the Journal of Child Psychology and Psychiatry, Cook and colleagues ran a large-scale feasibility trial of the SPECIFiC seven-session online psychoeducation course for 119 parents and carers of children with fetal alcohol spectrum disorder [10]. Recruitment and retention met thresholds and satisfaction was high, but the candidate primary outcome, parenting stress, did not differ significantly between groups, and there were no clear differences in child behaviour, parental mental health, or quality of life. The one clear signal was a large improvement in parenting self-efficacy, at a cost of a few hundred pounds per participant, so a definitive trial is justified but efficacy is unproven. Finally, a Personal View in The Lancet Psychiatry from Roth and Ayers argues that ambient artificial intelligence scribes bundle three functionally distinct operations — generating narrative history, generating mental state observations that were never spoken aloud, and generating diagnostic formulation — each with different failure modes, liability exposure and consent implications [8]. Their point is that the mental state exam in particular has no verbal source text to summarise, so an AI-generated version is an inference dressed as an observation.

If you only have time for one paper this week, make it the BMJ living umbrella review of bipolar disorder treatments [1]. It is the single most directly usable output of the week — a graded, phase-specific and age-specific map of what works, plus a free preference-based tool you can open during a consultation.

Here are the key takeaways from this week in Psychiatry. First, for bipolar disorder, effective options differ sharply by phase and by age, and the paediatric evidence base is far narrower than the adult one, so extrapolating from adults is not safe. Second, connectivity-guided theta-burst targeting produced faster and larger early antidepressant improvement than the 5-centimetre rule, but the advantage was gone by three months. Third, in perinatal depression the risk of untreated illness now deserves as much weight in the discussion as the risk of SSRI exposure. Fourth, affective symptoms that rise after a dose reduction and settle when the dose is held should raise suspicion of withdrawal, not automatic relapse. Fifth, physical multimorbidity in schizophrenia-spectrum and bipolar disorder accumulates early and across multiple organ systems, and excess mortality after COVID-19 infection persisted through and after vaccine roll-out, alongside falling vaccination uptake. And sixth, suicide-risk prediction models can help allocate preventive programmes, but their sensitivity remains limited and they cannot substitute for clinical assessment.

That's your roundup for This Week in Psychiatry. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. See you next week.

If this weekly briefing is useful, follow the show in your podcast app so new episodes arrive automatically. And a quick rating on Apple Podcasts or Spotify helps other physicians discover the show.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.