BackgroundThyroid function and glycolipid metabolic alterations are often associated with major depressive disorder (MDD), yet their roles in suicide risk among adolescents with MDD remain unclear. This study aimed to investigate thyroid-stimulating hormone (TSH) levels, glycolipid metabolism parameters, and their associations with suicidal ideation (SI) in adolescents with MDD.MethodsThis cross-sectional study was conducted at one general hospital and one psychiatric hospital in Anhui Province, China. Socio-demographic data and laboratory parameters were collected from participants, and the patients’ depressive symptoms and SI severity were assessed using the 24-item Hamilton Depression Rating Scale (HAMD-24) and the Positive and Negative Suicide Ideation Inventory (PANSI), respectively. TSH levels and glycolipid metabolism parameters were also measured.ResultsA total of 146 adolescents with MDD and 70 healthy controls (HCs) were enrolled in this study. Compared with HCs, patients had lower fasting blood glucose (FBG) levels (P < 0.001). Logistic regression analyses showed that a worse relationship with family, a higher HAMD-24 total score, and higher TSH and FBG levels were independently associated with concurrent SI in adolescents with MDD (all P < 0.05). Furthermore, receiver operating characteristic (ROC) curve analysis showed that the combination of these four factors had a good discriminatory ability for SI, with an area under the curve (AUC) of 0.851.ConclusionIn this cross-sectional study, TSH and FBG levels were associated with SI in adolescents with MDD. Nevertheless, whether these parameters can serve as clinically useful biomarkers requires further validation in larger prospective studies.
Effects of exercise interventions on executive function in autism spectrum disorder: a three-level meta-analytic review
ObjectiveThis study aimed to use a three-level meta-analysis to examine the effects of exercise interventions on executive function (EF) in children and adolescents with autism spectrum disorder (ASD), and to explore the relationship between exercise dose and intervention effects through subgroup analyses.MethodsA systematic search was conducted in Web of Science, PubMed, Embase, and the Cochrane Library from database inception to November 17, 2025. All included studies were randomized controlled trials. A three-level random-effects model was applied to integrate multiple dependent effect sizes. Hedges’ g was used to calculate effect sizes, and negative values indicated improvements in EF following exercise interventions. Study quality was assessed using the PEDro scale, and publication bias was evaluated using Egger’s regression and the trim-and-fill method.ResultsSeventeen studies involving 626 participants were included. Exercise interventions produced a moderate and significant improvement in EF (g = –0.34, p < 0.0001) with low heterogeneity (Q = 50.42, p =0.268). Subgroup analyses showed that only intervention duration significantly moderated the effects, with programs lasting ≥10 weeks yielding the greatest benefits.ConclusionExercise interventions significantly improve EF in children and adolescents with ASD, with the largest effect sizes observed in interventions lasting ≥10 weeks. Further large-scale studies are needed to clarify dose–response relationships and inform optimal exercise prescriptions. This meta-analysis was designed and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. Systematic review registrationhttps://www.crd.york.ac.uk/PROSPERO/, identifier CRD420251236112
Adolescent exposure to the psychedelic 25C-NBOMe in rats induces lasting competitive avoidance through disrupted hippocampal–prefrontal synchrony
Nature Neuroscience, Published online: 20 July 2026; doi:10.1038/s41593-026-02369-y
Yu, Zhang et al. show that repeated exposure to the psychedelic drug 25C-NBOMe in adolescent rats, but not in adult rats, reduces willingness to engaged in food resource competition in adulthood, due to reduced theta synchrony between ventral hippocampus and orbitofrontal cortex.
New €25 Million BioReliance Testing Facility Opens at Merck KGaA Global Headquarters in Germany
MilliporeSigma opened a new €25 million BioReliance® testing facility at the company’s global headquarters in Darmstadt, Germany. The facility expands access to commercial drug substance and drug product release testing as well as stability testing for biopharmaceutical companies developing and commercializing therapies in Europe, according to the company.
“As demand for biologics and novel therapies continues to grow, our customers need reliable, compliant testing capabilities closer to where their products are developed and commercialized,” said Paolo Carli, head of advanced solutions for the life science business of Merck KGaA. “Our new testing facility combines best-in-class analytical characterization services with more than 75 years of BioReliance expertise to help our European customers move critical therapies toward patients with greater speed and confidence.”
The 2,000-square-meter facility is designed to help customers meet European requirements for in-region drug substance and drug product release testing and to expand the company’s ability to support customers from drug development through commercialization. The site will also offer GMP-compliant stability studies for monoclonal antibodies and cell therapies, addressing the growing demand for biologics testing across Europe.
Located close to major clinical trial sites in Germany, France, Spain, the Netherlands, Belgium and Italy, the Darmstadt facility is strategically positioned to support biopharmaceutical companies seeking to release drug products into European markets, pointed out Carli. By adding these capabilities in the heart of Europe, the company is strengthening its support for customers managing increasingly complex development, quality and regulatory requirements, he added.
A MilliporeSigma spokesperson noted that the BioReliance sites form a global testing network that allows customers to scale across geographies and work with the company across continents. Among the company’s leading technologies is the Blazar® platform, which moves the biosafety testing paradigm from traditional methods to rapid molecular approaches to significantly reducing testing timelines for virus detection.
The Aptegra® CHO genetic stability testing streamlines a previously complex and time-intensive process into a single assay, continued the spokesperson.
MilliporeSigma lists the opening of the Darmstadt facility as one of several significant investments the company has made to grow its global contract testing footprint. In 2024, the company opened a €290 million biosafety testing facility in Rockville, MD, and expanded biosafety testing capacity by 40% across its Glasgow and Stirling sites through a €22 million investment. The company also cites the new BioReliance facility as reflecting the firm’s continued commitment to its global headquarters in Darmstadt, where €2.5 billion has been invested since 2015.
The post New €25 Million BioReliance Testing Facility Opens at Merck KGaA Global Headquarters in Germany appeared first on GEN – Genetic Engineering and Biotechnology News.
STAT+: Bristol Myers Squibb becomes latest company to claim it’s building pharma’s largest NVIDIA AI supercomputer
It’s officially a trend. For the third time in nine months, a pharma company has announced that it is assembling the largest AI supercomputer in the life sciences industry. This time, it is Bristol Myers Squibb.
When the company began its partnership with NVIDIA three years ago with a smaller computing cluster, it was focusing on simpler problems with individual AI tools, like protein structure prediction. But “we actually consumed all the space we had,” said Greg Meyers, chief digital & technology officer at BMS.
Adding extra computing power is necessary for the company now that it’s “become more convinced” that computationally hungry foundation models can give the company valuable insight into how its drug candidates interact with both the body and with disease, Meyers said in an interview with STAT. He mentioned oncology and neurodegeneration as examples of areas where BMS has developed such models.
AI is more likely than humans to form biases when hiring
The next time you apply for a job, AI may screen your résumé before any human sees it. But there’s good reason to question whether AI will judge you fairly. Researchers already know that LLMs pick up human biases from their training data. New research suggests that LLMs can also develop their own biases from experience—and stereotype job applicants more than humans do. As AI companies race to build agentic models that remember the tiniest details about users, they may be handing them ammunition for forming those biases.
Researchers at Princeton University and the University of Chicago ran LLMs, including ChatGPT, Claude, and Gemini, through a simulated hiring game, adapted from a psychology study that explored how humans can form stereotypes. Each model was told it had been hired as a consultant by the mayor of a fictional city and was then asked to help hire people for 20 jobs, including doctors, lawyers, child-care aides, and janitors. Candidates came from four fictional ethnic groups: Tufa, Aima, Reku, and Weki.
In each round, there was a new job opening and four candidates, one from each group. After the model hired a candidate, it learned whether they succeeded at their job and moved onto the next round. The model was told to make as many successful hires as possible over 40 rounds. Unbeknownst to the models, all candidates were equally likely to succeed at every job.
The models quickly started segregating candidates from different groups into different jobs on the basis of early observations of hiring outcomes. For example, when a model was told an Aima had failed as a doctor, a job considered to require high levels of warmth and competence, it veered away from hiring all Aimas as doctors. Instead, it started hiring Aimas as janitors, which the model classified as being less warm and competent than doctors.
The models were even more likely to stereotype people by demographic group than the human participants in the original study. On the study’s segregation scale, where 2 means every group has been completely confined to its own job niche, human participants scored 0.84. The models scored roughly 65% higher, with OpenAI’s reasoning model o3 scoring 1.83, close to the maximum possible.
That’s because LLMs “really are eager to create generalizations from limited data,” says Ryan Liu, a PhD student at Princeton University and a coauthor of the study, which was published in a paper at ICML in Seoul in July. “That’s literally a lot of what they’re optimized for.”
Every decision-maker, human or machine, faces a trade-off between sticking with what worked before and trying something new that might work better—a phenomenon psychologists call the “exploration-exploitation dilemma.” It’s like choosing between a new restaurant and your reliable favorite.
Because LLMs are trained on math, coding, and science problems—tasks that reward generalizing from just a few examples—they can settle on a hunch too early. And the same instinct that helps LLMs crack logic puzzles also makes them quick to stereotype.
In the experiment, newer models with higher reasoning capabilities, such as OpenAI’s o3 and DeepSeek’s R1, showed even stronger biases. When LLMs rush to generalize in social settings, “that’s when things tend to go wrong,” says Liu. OpenAI and Anthropic did not respond to requests for comment.
The finding is especially relevant now that chatbots are gaining improved memory and personalization features, says Angelina Wang, a computer scientist at Cornell University who did not work on the study. When a chatbot draws on its previous conversation history, it can “over-index on the same kinds of behaviors it’s experienced before” and form biases, she says.
Simply having chatbots remember less isn’t a fix, though, because users want chatbots to remember what they say. “We still are trying to figure out just the right amount that isn’t too much or too little,” says Wang.
Telling the model to be fair didn’t change its behavior much. “Either it can’t put these values into action or that process is being submerged under the tendency to try to optimize for the goal of getting the most correct hires,” says Liu. But promising the models an additional bonus for diverse hiring made them far less biased. The trick, then, is to design goals that “incorporate desirable social values in order to make the large language model act in socially desirable ways,” says Liu.
The models also became less biased when they were told more personal information about individuals. In another experiment in the same study, the researchers asked the models to resettle members of different ethnic groups in cities across Canada. When the models were told personal information relevant to the ability to adapt to a new city, such as age and education, they were less likely to segregate people by their ethnicity. But when they were given irrelevant information, such as hair color and tattoo shape, the models largely fell back to sorting people by their ethnicity again.
To what extent AI systems will stereotype job applicants in the real world is still an open question. While the models in the experiment immediately learned whether they’d made successful hires, a model screening résumés in the real world doesn’t get an instant report card. Companies can take a long time to find out whether a new hire is any good.
But when feedback does trickle in, a model could still read too much into those results when making future hires. As companies increasingly deploy LLMs to screen résumés and even conduct interviews, the finding that models can form biases from their hiring experience “is a really serious implication that they should grapple with,” says Wang.
As LLMs learn from experience to make decisions about who gets hired, who gets a loan, or who gets parole, the biases we should worry about may include ones no human ever taught them. “These novel biases—they’re sort of ever present,” says Liu.
STAT+: Telemedicine company touted by Novo Nordisk stressed profits over patient safety, ex-workers say
Novo Nordisk, maker of Ozempic and Wegovy, lists the telehealth company LifeMD on its website as a provider that offers “legitimate medicine sourcing and patient support” for people seeking GLP-1 drugs. But some former employees describe LifeMD differently, as a company that has sought to maximize the volume of prescriptions it doles out at the expense of patient safety.
Former workers told STAT that providers were pressed to expedite their work to a pace that was not clinically responsible, with two of them saying providers at times were expected to review the cases of 25 people per hour based only on electronic forms the patients filled out — the equivalent of spending about two minutes on each case.
The company also discouraged providers from asking what they felt were medically relevant questions to patients, so that they don’t “delay care,” former employees said.
STAT+: State audit of Medicaid records points to methods used by PBMs to obscure drug costs
A recent audit of state Medicaid records revealed how pharmacy benefit managers are using complicated and sophisticated approaches for handling prescription drug claims that ultimately overcharge taxpayers, a finding that underscores controversy surrounding these crucial middlemen in the pharmaceutical supply chain.
The audit of Iowa’s state Medicaid program found that one large pharmacy benefit manager appeared to have made more than $100 million by adjusting the amount of money paid to pharmacies without passing some of the savings back to managed care plans working on behalf of the state, according to Iowa officials. The review scrutinized records from 2019 through 2021.
The overall conclusion was similar to audit results conducted in a few other states, but in this instance, the auditors identified what amounted to creative accounting maneuvers, which not only made it possible to obscure the flow of money but evade prohibitions on a controversial pricing practice that is now outlawed in Iowa and some other states.
Opinion: It’s time to treat pandemics as threats to national security
Pandemics like the great influenza of 1918-1919 and Covid-19 are often viewed and treated as health care challenges. That view, however, is far too narrow. They should be viewed and combated as threats to national security, too.
Why treat pandemics as national security threats? Because that’s what they are. Framing the danger this way would inform and motivate reforms that strengthen our pandemic defenses and reduce the devastation pandemics inflict.
Opinion: My husband’s suicide shows there’s something very wrong with the U.S. insurance industry
I remember standing outside an exam room, phone vibrating in my breast pocket, when the coroner’s call came through. I hadn’t wanted to believe the neighbor’s frantic text about a body bag removed from my husband’s condo, but the medical examiner’s office on my caller ID was unmistakable. I answered, and sounds fell out of my mouth into the phone: “Hello, this is Dr. Hardison.”
I am an emergency physician living in the U.S., and we had one of the top commercial health insurance plans. When my husband, Randy, became suicidal for the first time in his life, I thought I could find him the best care, our insurance would pay for it, and he would get better. Only one of those three things actually happened.