Digital Health

AI Outperformed Emergency Doctors in Harvard Triage Study, Raising Questions About the Future of Clinical Decision-Making

Key Takeaways: 

• A Harvard-led study found that an AI reasoning model outperformed emergency physicians in diagnosing patients during hospital triage scenarios using text-based clinical information.
  
• Researchers said the findings represent a major advance in AI clinical reasoning, although they stressed that AI is not ready to replace human doctors.
  
• Experts warned that important concerns remain around accountability, bias, safety, and the risk of clinicians becoming overly reliant on AI systems.
  
  
  

AI shows strong performance in emergency medicine trial

From fictional emergency department heroes such as George Clooney in ER to Noah Wyle in The Pitt, emergency physicians have long been portrayed as the ultimate decision-makers in moments of medical crisis. However, a new Harvard study suggests artificial intelligence may increasingly play a major role in those same high-pressure situations.

Researchers from Harvard Medical School and Beth Israel Deaconess Medical Center found that advanced AI systems outperformed human doctors in emergency medicine triage scenarios, making more accurate diagnoses when presented with limited patient information during the critical early stages of hospital admission.

The findings, published in Science, were described by independent experts as representing “a genuine step forward” in AI clinical reasoning.

According to the study authors, large language models (LLMs) “have eclipsed most benchmarks of clinical reasoning”.

AI versus doctors in emergency room triage

One of the study’s central experiments examined 76 patients who presented to the emergency department of a Boston hospital.

Both the AI system and pairs of human physicians were given identical electronic health record information to assess. This included standard triage details such as:

• Vital signs
• Demographic information
• Brief nursing notes explaining why the patient attended hospital

Using only this text-based information, OpenAI’s o1 reasoning model identified the exact diagnosis or a very close diagnosis in 67% of cases.

By comparison, the human physicians achieved diagnostic accuracy rates of between 50% and 55%.

Researchers found the AI’s advantage was especially apparent in triage situations requiring rapid decision-making with minimal available information.

When additional clinical detail was provided, the AI’s diagnostic accuracy increased further to 82%. Human experts achieved accuracy rates between 70% and 79% under those circumstances, although researchers noted the difference was not statistically significant in that setting.

AI also performed better in treatment planning

The study also evaluated how AI performed in longer-term clinical planning tasks.

In this experiment, the AI system and a group of 46 doctors were asked to review five detailed clinical case studies and develop treatment strategies. These included decisions relating to:

• Antibiotic regimens
• Ongoing management plans
• End-of-life care processes

The AI significantly outperformed the doctors.

Researchers reported that the AI achieved a score of 89%, compared with 34% among physicians using conventional resources such as search engines.

AI detected a diagnosis human doctors missed

One example highlighted in the study involved a patient with a pulmonary embolism and worsening symptoms.

Human doctors believed the patient’s anticoagulant treatment was failing. However, the AI system identified something clinicians had overlooked – the patient had a history of lupus, which may have been responsible for inflammation in the lungs.

The AI’s interpretation was ultimately confirmed as correct.

Researchers say AI will reshape medicine – not replace doctors

Despite the strong performance shown by AI systems, researchers stressed that the technology is not ready to replace physicians.

The study only assessed AI systems using text-based patient information. It did not evaluate the AI’s ability to interpret non-verbal clinical signals that doctors routinely use during patient assessment, such as:

• Visible distress
• Facial appearance
• Behaviour
• Physical examination findings

As a result, researchers said the AI functioned more like a clinician reviewing paperwork and offering a second opinion.

“I don’t think our findings mean that AI replaces doctors,” said Arjun Manrai, one of the lead authors of the study who heads an AI lab at Harvard Medical School. “I think it does mean that we’re witnessing a really profound change in technology that will reshape medicine.”

Dr Adam Rodman, another lead author and physician at Beth Israel Deaconess Medical Center, described AI LLMs as among “the most impactful technologies in decades”.

He suggested healthcare may move towards what he described as a “triadic care model”.

“Over the next decade,” Rodman said, AI would not replace physicians but instead work alongside them in a new model involving “the doctor, the patient, and an artificial intelligence system”.

AI use in healthcare is already growing

The findings come amid rapidly increasing AI adoption within healthcare systems.

According to research published last month, nearly one in five physicians in the United States are already using AI to assist with diagnosis.

In the United Kingdom, a recent Royal College of Physicians survey found:

• 16% of doctors use AI daily
• A further 15% use AI weekly
• Clinical decision-making is among the most common applications

However, concerns around safety and accountability remain significant.

UK doctors surveyed identified AI errors and legal liability as among their biggest worries.

“There is not a formal framework right now for accountability,” said Rodman.

He also stressed the continuing importance of human clinicians in patient care.

“Patients ultimately want humans to guide them through life or death decisions [and] to guide them through challenging treatment decisions,” he said.

Experts warn against over-reliance on AI

Independent experts said the study highlights the rapidly improving capabilities of AI systems in medicine, but also demonstrates the need for caution.

Prof Ewen Harrison, co-director of the University of Edinburgh’s Centre for Medical Informatics, said the findings suggest AI systems are beginning to evolve beyond theoretical testing environments.

“These systems are no longer just passing medical exams or solving artificial test cases,” he said. “They are starting to look like useful second-opinion tools for clinicians, particularly when it is important to consider a wider range of possible diagnoses and avoid missing something important.”

However, Dr Wei Xing from the University of Sheffield warned that the study also raised concerns about how clinicians interact with AI recommendations.

He suggested some doctors may unconsciously defer to AI-generated answers instead of independently evaluating clinical information themselves.

“This tendency could grow more significant as AI becomes more routinely used in clinical settings,” he said.

Dr Xing also noted that the study provided limited information about where the AI may perform less effectively, including whether diagnostic accuracy differed among certain patient populations such as:

• Older adults
• Non-English speakers
• People with more complex communication needs

He cautioned against interpreting the findings as evidence that publicly available AI tools are ready for independent medical use.

“It does not demonstrate that AI is safe for routine clinical use, nor that the public should turn to freely available AI tools as a substitute for medical advice,” he said.

Source: The Guardian

Read More

AI and Ultrasound Data May Transform Detection of Advanced Heart Failure

Key Takeaways:

• A new artificial intelligence approach can estimate a key heart failure metric using routine ultrasound images and electronic health records
  
• The method may help identify people with advanced heart failure who are currently missed due to limited access to specialised testing
  
• Early results show approximately 85% accuracy, suggesting strong potential for real-world clinical use
  
  
  

A new approach to a persistent diagnostic challenge

Applying artificial intelligence to cardiac ultrasound data may offer a more accessible way to identify people with advanced heart failure, according to a new study led by researchers from Weill Cornell Medicine, Cornell Tech, Cornell Ann S. Bowers College of Computing and Information Science, Columbia University Vagelos College of Physicians and Surgeons, and NewYork-Presbyterian.

Advanced heart failure is typically diagnosed using cardiopulmonary exercise testing (CPET). While effective, this method requires specialised equipment and trained personnel and is usually limited to large medical centres. As a result, many people do not receive timely or appropriate diagnosis and care.

In the United States alone, an estimated 200,000 people are living with advanced heart failure. However, only a small proportion are properly identified each year, in part due to these diagnostic limitations.

The new study, published on 3 March in npj Digital Medicine, explored whether artificial intelligence could help overcome this bottleneck by using more widely available clinical data.

Predicting peak VO2 without specialised testing

The research team developed an artificial intelligence model capable of predicting peak oxygen consumption, known as peak VO2. This measure is a central output of CPET and a key indicator of heart failure severity and patient risk.

Instead of relying on exercise testing, the model uses routinely collected data, including cardiac ultrasound images and information from electronic health records. These sources are already embedded in standard clinical care, making the approach potentially scalable across a wide range of healthcare settings.

“This opens up a promising pathway for more efficient assessment of patients with advanced heart failure using data sources that are already embedded in routine care,” said study senior author Dr. Fei Wang, associate dean for AI and data science and the Frances and John L. Loeb Professor of Medical Informatics at Weill Cornell Medicine.

A collaborative effort across disciplines

The study represents a highly collaborative effort involving experts in artificial intelligence, informatics, and clinical cardiology. Alongside Dr. Wang’s team, key contributors included Dr. Deborah Estrin, associate dean for impact at Cornell Tech, and Dr. Nir Uriel, director of advanced heart failure and cardiac transplantation at NewYork-Presbyterian.

The work forms part of the broader Cardiovascular AI Initiative, a joint effort between Cornell, Columbia, and NewYork-Presbyterian aimed at advancing the use of artificial intelligence in heart failure diagnosis and management.

“Initially we put together a group of more than 40 heart failure specialists and asked them to tell us where they thought AI could best be applied,” said Dr. Uriel.

One of the most promising opportunities identified was the use of artificial intelligence to analyse cardiac ultrasound data in order to detect advanced heart failure earlier and more accurately.

How the AI model works

The research team developed a multi-modal, multi-instance machine learning model designed to process multiple types of clinical data simultaneously. These included:

• Moving ultrasound images of the heart
• Waveform imagery showing heart valve motion and blood flow
• Structured and unstructured data from electronic health records

“The close interaction between clinicians and AI researchers on this project ended up driving the development of new AI techniques that would not have been explored otherwise,” said Dr. Estrin. “So, this was a case of medicine shaping the future of AI – not just AI shaping the future of medicine.”

Training and validation

The model was trained using deidentified data from 1,000 people with heart failure treated at NewYork-Presbyterian/Columbia University Irving Medical Center.

Once trained, it was tested on a separate group of 127 people with heart failure from three additional NewYork-Presbyterian campuses. The goal was to assess how accurately the model could predict peak VO2 and identify individuals at high risk.

Strong early performance

The results demonstrated a level of accuracy that exceeds previous artificial intelligence approaches for predicting peak VO2.

Using a standard performance metric for risk prediction, the model achieved an overall accuracy of approximately 85%. This suggests that the tool could effectively distinguish between people at higher and lower risk of advanced heart failure in clinical settings.

Implications for clinical practice

If validated in further studies, this approach could significantly expand access to advanced heart failure assessment. By removing the need for specialised exercise testing in some cases, clinicians may be able to identify high-risk individuals earlier and initiate appropriate treatment sooner.

“If we can use this approach to identify many advanced heart failure patients who would not be identified otherwise, then this will change our clinical practice and significantly improve patient outcomes and quality of life,” Dr. Uriel said.

Next steps towards clinical adoption

The research team is now planning prospective clinical studies to further evaluate the model. These studies will be essential for regulatory approval, including review by the U.S. Food and Drug Administration, and for eventual integration into routine clinical workflows.

The work was partially supported by funding from NewYork-Presbyterian as part of the Cardiovascular AI Initiative. As with other research at Weill Cornell Medicine, relationships with external organisations are disclosed publicly to ensure transparency.

Source: Weill Cornell Medicine

Read More

Remote Culinary Coaching Shows Sustained Weight Loss Benefits in Adults with Overweight and Obesity

Key Takeaways:

• A fully remote culinary medicine programme combining cooking and health coaching led to sustained weight loss over 12 months
  
• Participants experienced significant fat mass reduction without loss of lean body mass
  
• Improvements in diet quality, calorie intake, and cooking confidence were observed alongside weight changes
  
  
  

Study overview

A recent randomised controlled trial has found that a fully remote culinary medicine intervention can support meaningful and sustained weight loss in people living with overweight and stage I obesity. The programme combined practical cooking education with health coaching, offering a patient-centred approach to improving dietary behaviours and long-term health outcomes.

Conducted across two hospitals between May 2019 and September 2022, the study examined the one-year impact of this combined intervention on weight, body composition, and dietary habits.

Methodology

Participant characteristics

The study included 50 adults with overweight or stage I obesity. Participants had a mean age of 47.5 years, and 70% were female. The average body mass index was 30.7, with a mean total fat mass of 40.37%. All participants reported cooking fewer than five meals at home per week at baseline.

Intervention design

All participants initially received two nutrition education sessions focused on the Mediterranean diet. Following this, they were randomly assigned to one of two groups:

• Intervention group: Participants took part in a structured culinary coaching programme consisting of 12 weekly one-to-one tele-sessions, each lasting 30 minutes. These sessions integrated culinary skills training with health coaching principles and provided access to culinary medicine resources.
• Control group: Participants were given access to the same culinary medicine resources but did not receive coaching sessions
  
  
  

Outcome measures

Researchers assessed a range of clinical and behavioural outcomes at baseline, and again at 3, 6, and 12 months within a hospital clinical research setting:

• Body weight and height were measured by a registered dietitian
• Body composition was analysed using dual-energy X-ray absorptiometry (DEXA)
• Dietary intake was calculated using 4-day food records reviewed by a registered dietitian
• Diet quality was evaluated using a 14-item Mediterranean diet assessment tool
• Culinary attitudes and self-efficacy were measured using a validated questionnaire

The primary outcome was change in body weight at 6 months, with secondary outcomes including dietary intake, body composition, and behavioural measures.

Weight loss outcomes

Participants in the culinary coaching group achieved significantly greater weight loss compared with the control group at all measured time points:

• 3 months: -3.23% vs -0.71% (between-group difference -2.52; P = .016)
• 6 months: -4.2% vs -1.22% (between-group difference -2.98; P = .027)
• 12 months: -4.02% vs a weight gain of 0.28% (between-group difference -4.30; P = .021)

These findings indicate that the intervention not only supported early weight loss but also helped sustain these changes over a full year.

Changes in body composition

At 6 months, participants receiving culinary coaching demonstrated favourable changes in body composition:

• Average fat mass decreased by 1.86% in the intervention group
• In contrast, the control group experienced a slight increase in fat mass of 0.11%
• The between-group difference was 1.96 (P = .039)

Importantly, these reductions in fat mass occurred without any significant changes in lean body mass, suggesting that weight loss was primarily driven by fat reduction rather than muscle loss.

Dietary improvements

The intervention also led to measurable improvements in diet quality and energy intake:

• At 3 months, Mediterranean diet scores increased by 2 points in the intervention group compared with 0.38 points in the control group (net difference 1.62; P = .020)
• At 6 months, daily calorie intake decreased by 452 calories in the intervention group compared with 62.4 calories in the control group (net difference 390 calories; P = .015)

These findings suggest that the programme successfully influenced both food choices and overall energy consumption.

Behavioural and skill-based outcomes

Participants who received culinary coaching reported significant improvements in their confidence and ability to prepare meals:

• Self-efficacy in cooking techniques and meal preparation improved significantly at 12 months in the intervention group compared with the control group (P = .040)

No serious adverse events were reported during the study, indicating that the intervention was safe and well tolerated.

Interpretation and clinical relevance

The study authors highlighted the broader significance of these findings, stating:

“This study is an important step in considering CM [culinary medicine] interventions as an effective patient-centered nutrition strategy for weight loss.”

This suggests that combining practical cooking skills with behavioural coaching may offer a scalable and effective approach to supporting people living with overweight and obesity, particularly in remote or resource-limited settings.

Limitations

The study has several limitations, many of which were influenced by the COVID pandemic:

• High dropout rates after the first visit may have introduced attrition bias
• Some follow-up visits were conducted remotely, requiring participants to self-measure body weight
• Remote assessments limited the ability to collect body composition and other clinical data at certain time points
• Pandemic-related restrictions may have affected participants’ ability to cook at home

These factors should be considered when interpreting the findings.

Funding and disclosures

The study was led by Rani Polak at Harvard Medical School and Spaulding Rehabilitation Hospital in Boston and was published in Obesity.

Funding was provided by the US-Israel Binational Science Foundation and the National Institutes of Health Clinical Center. One author reported receiving royalties from a home cooking book and an honorarium from Wellcoaches.

Read More

AI Analysis of Health Records May Help Identify ADHD Risk Years Earlier

Key Takeaways: 

• Artificial intelligence can analyse routine electronic health records to estimate a child’s risk of developing ADHD years before diagnosis.
  
• The model demonstrated strong accuracy across diverse populations, using data from more than 140,000 children.
  
• The tool is designed to support earlier evaluation and intervention, not to replace clinical diagnosis.
  
  
  

AI and the challenge of delayed ADHD diagnosis

Attention-deficit/hyperactivity disorder (ADHD) affects millions of children worldwide. Despite its prevalence, many children experience significant delays before receiving a formal diagnosis. This can limit access to timely support, even when early signs are present.

New research from Duke Health suggests that artificial intelligence may help address this gap. By analysing routinely collected electronic health records, researchers have developed a tool capable of identifying patterns that indicate a higher likelihood of future ADHD diagnosis, potentially years in advance.

Unlocking insights from routine healthcare data

The study, published in Nature Mental Health on April 27, demonstrates how existing healthcare data can be used to support earlier clinical decision-making.

“We have this incredibly rich source of information sitting in electronic health records,” said Elliot Hill, lead author of the study and data scientist in the Department of Biostatistics & Bioinformatics at Duke University School of Medicine. “The idea was to see whether patterns hidden in that data could help us predict which children might later be diagnosed with ADHD, well before that diagnosis usually happens.”

Rather than relying on new or specialised testing, the approach draws on information already collected during standard healthcare visits. This includes developmental milestones, behavioural observations, and clinical events recorded from birth through early childhood.

How the AI model was developed

To build and test the model, researchers analysed electronic health records from more than 140,000 children, including both those diagnosed with ADHD and those without the condition.

The AI system was trained to identify combinations of factors that tend to appear before an ADHD diagnosis is made. Over time, it learned to detect subtle patterns across large datasets that may not be easily recognised through conventional clinical assessment alone.

The model showed strong performance in estimating ADHD risk in children aged 5 years and older. Notably, its accuracy remained consistent across different population groups, including variations in sex, race, ethnicity, and insurance status.

A support tool, not a diagnostic replacement

The researchers emphasise that the AI system is not intended to diagnose ADHD. Instead, it serves as a clinical support tool that can help identify children who may benefit from closer monitoring or earlier referral for specialist assessment.

“This is not an AI doctor,” said Matthew Engelhard, M.D., Ph.D., in Duke’s Department of Biostatistics & Bioinformatics, and senior author of the study. “It’s a tool to help clinicians focus their time and resources, so kids who need help don’t fall through the cracks or wait years for answers.”

By highlighting children who may be at higher risk, the tool could enable healthcare professionals to prioritise evaluation and initiate discussions with families sooner.

Potential benefits of earlier identification

Earlier identification of children at risk of ADHD could have meaningful implications for long-term outcomes. Research consistently shows that timely diagnosis and intervention are associated with improved academic performance, social development, and overall health.

“Children with ADHD can really struggle when their needs aren’t understood and adequate supports are not in place,” said study author Naomi Davis, Ph.D., associate professor in the Department of Psychiatry and Behavioral Sciences. “Connecting families with timely, evidence-based interventions is essential for helping them achieve their goals and laying a foundation for future success.”

The ability to flag potential concerns earlier may also help reduce the delays that many families face when seeking answers and support.

Next steps and ongoing research

While the findings are promising, the researchers stress that further validation is needed before such tools are implemented in routine clinical practice. Additional studies will be required to confirm effectiveness, assess real-world impact, and ensure safe integration into healthcare systems.

Hill and Engelhard have also explored the broader use of AI models in identifying risks and contributing factors for mental health conditions in adolescents, signalling a growing interest in predictive tools within this field.

Study authors and funding

In addition to Elliot Hill, Matthew Engelhard, and Naomi Davis, the study authors include De Rong Loh, Benjamin A. Goldstein, and Geraldine Dawson.

The research was supported by grants from the National Institute of Mental Health (K01-MH127309, UL1 TR002553) and the National Center for Advancing Translational Sciences.

Source: EurekAlert!

Read More

Mobile App May Help Prevent Excess Weight Gain in Pregnancy

Key Takeaways:

• A digital intervention combining self-monitoring and personalised feedback helped reduce excess weight gain during pregnancy among women with overweight or obesity
  
• Participants using the programme gained weight more slowly and were less likely to exceed recommended guidelines compared with standard care
  
• Greater engagement, particularly consistent self-weighing, was linked to better outcomes
  
  
  

Addressing a common challenge in pregnancy

Weight gain during pregnancy is both expected and necessary. However, gaining more than recommended levels is associated with increased health risks for both the pregnant person and the baby. These risks are particularly pronounced among individuals who are already living with overweight or obesity at the start of pregnancy.

Excess gestational weight gain has been linked to complications such as gestational diabetes and preeclampsia. It also increases the likelihood of pre-term birth and larger birthweight, both of which can complicate delivery and raise the child’s long-term risk of obesity.

In the United States, around half of pregnant individuals with overweight or obesity exceed national guidelines for weight gain, highlighting a persistent challenge in routine prenatal care.

The LEAP programme – a technology-enabled approach

To address this issue, researchers at Kaiser Permanente developed the Lifestyle, Eating, and Activity in Pregnancy (LEAP) programme. The intervention was designed to provide scalable, technology-driven support integrated into everyday clinical care.

The study, published in JAMA Network Open on 20 April, evaluated whether this digital approach could help individuals maintain healthier weight gain trajectories during pregnancy.

“Our results are especially exciting because we intentionally designed the LEAP program to be feasible for widespread implementation,” said first author Monique Hedderson, PhD, principal investigator with the Kaiser Permanente Division of Research (DOR). “This study represents the culmination of years of research on what works well for this high-risk population and how to leverage technology to make those benefits achievable across a health system.”

How the intervention worked

Participants in the LEAP group were provided with a digital scale for home use and a wearable activity tracker, either a Fitbit or their own Apple Watch. These devices were connected to a mobile application that delivered automated, personalised feedback on both weight and physical activity.

The programme also included weekly educational sessions delivered through the app, focusing on realistic and achievable goals related to diet and exercise. In parallel, clinicians received guidance on how to discuss gestational weight gain with patients more effectively.

A key feature of the programme was its adaptive design. Individuals whose weight gain began to accelerate received targeted support, including personalised chat messages from a lifestyle coach. If necessary, this support escalated to telephone consultations.

“About half of patients in the LEAP intervention group managed weight well on their own, but for those inching towards the upper limit of recommendations, LEAP provided extra, targeted support,” said Hedderson. “This conserves resources for patients who need help the most.”

Study design and participant outcomes

The effectiveness of the LEAP programme was assessed across 58 clinicians within Kaiser Permanente Northern California. All pregnant patients with overweight or obesity under their care were invited to participate.

Clinicians were randomly assigned to one of two groups:

• Standard care plus the LEAP intervention
• Standard care alone

In total, 1,256 participants were included in the study. Of these, 677 received the LEAP programme, while 588 received standard care.

Standard care involved providing patients with a newsletter at their first prenatal visit outlining healthy weight gain targets and lifestyle advice. At around 20 weeks, clinicians reviewed weight gain against recommended guidelines.

The results showed clear differences between the groups. Participants in the LEAP programme experienced:

• Lower weekly rates of weight gain
• Lower total weight gain during pregnancy
• Reduced likelihood of exceeding recommended guidelines

In addition, fewer individuals in the LEAP group gave birth to infants classified as large for gestational age.

“This study demonstrates that combining wireless self-monitoring tools with an adaptive digital intervention can make a meaningful difference for patients with overweight or obesity at risk of excess weight gain during pregnancy,” said Kari Carlson, MD, director of women’s health for The Permanente Medical Group. “Importantly, it was tested in routine care, making the findings highly relevant for health systems looking for practical solutions.”

Engagement matters

Although all participants assigned to the LEAP group were included in the analysis, only around half actively engaged with the programme. This reflects real-world conditions, where not all individuals fully participate in digital health interventions.

However, among those who did engage, outcomes were notably stronger. Higher levels of interaction with the programme, particularly consistent self-weighing, were associated with lower overall weight gain.

“When we zoomed in on participants who actually engaged, we found that the more they engaged, especially by weighing themselves consistently, the less weight they gained,” Hedderson said.

Participants could view their weight trends in relation to Institute of Medicine guidelines within the app, a feature that appeared to play a particularly important role in supporting behaviour change.

“This seems to be a particularly effective component of the program,” she said.

Implications for clinical practice

The findings suggest that digital, scalable interventions such as LEAP could offer a practical solution to a longstanding challenge in prenatal care. By integrating self-monitoring tools, automated feedback, and targeted coaching into routine workflows, the programme demonstrates how technology can support both patients and clinicians without increasing burden.

“Excess gestational weight gain among these high-risk patients is a common and challenging issue in routine prenatal care,” said senior author Assiamira Ferrara, MD, PhD, a DOR senior research scientist. “What’s compelling about the LEAP study is that it shows we can support healthier weight gain using scalable, technology-enabled tools that fit within real-world clinical workflows, without adding burden for patients or clinicians.”

Looking ahead

The research team continues to follow participants to assess longer-term outcomes for both parents and children. There is optimism that the LEAP programme could be integrated into routine care in the near future.

“Bottom line, using technology to help patients manage their own health can be effective, and patients tend to like it, too,” Hedderson said.

Funding and research team

The study was funded by the National Institutes of Health.

Additional co-authors included Susan D. Brown, PhD; Charles P. Quesenberry, PhD; Fei Xu, MS; Emily Liu, MPH; Karen L. Li, MPH; Sneha B. Sridhar, MPH; Tali Sedgwick, RDN; Page Kissel, BA; and Hillary D. Serrato Bandera, BA, all from the Division of Research. Mibhali M. Bhalala, MD, contributed from Kaiser Permanente Northern California, and Cheryl Albright, PhD, MPH, contributed from the University of Hawaii at Manoa School of Nursing and Dental Hygiene.

About the Kaiser Permanente Division of Research

The Kaiser Permanente Division of Research conducts epidemiological and health services research aimed at improving health outcomes and healthcare delivery. With more than 720 staff and approximately 630 active research projects, the division focuses on understanding the determinants of illness and advancing the quality and cost-effectiveness of care.

Source: Kaiser Permanente Division of Research

Read More

Large Language Models Show Promise in Detecting Drug Safety Signals from Clinical Notes

Key Takeaways:

• Large language models can identify immune-related adverse events in clinical notes without task-specific training, offering a potential alternative to labour-intensive manual review
  
• Performance remains below the threshold required for clinical decision support, with models tending to overpredict adverse events
  
• Despite limitations, this approach may support large-scale safety monitoring and accelerate research into cancer immunotherapies
  
  
  

The challenge of detecting drug safety signals

Drug safety signals are often embedded within unstructured clinical text, particularly in electronic health records. Identifying these signals has traditionally required either manual chart abstraction, which is resource-intensive, or natural language processing systems tailored to specific drugs and healthcare settings.

This challenge is particularly evident in the case of immune checkpoint inhibitors. These cancer therapies, first introduced in 2011, are associated with a broad range of immune-related adverse events. These events can affect multiple organ systems, including the colon, liver, lungs, heart, nervous system, skin, and endocrine system, making systematic detection complex and time-consuming.

Exploring large language models as a solution

Large language models are increasingly being explored as a way to streamline the identification of drug safety signals within clinical text. A multicentre study, published in eBioMedicine, evaluated whether these models could detect immune-related adverse events associated with immune checkpoint inhibitors.

The study focused on a zero-shot learning approach. In this setting, the model receives a single, detailed prompt without prior examples. The prompt used by the researchers began: “You are a clinical expert in identifying immune-related adverse events caused by immune checkpoint inhibitors …” and included a list of six immune checkpoint inhibitors alongside numerous associated adverse events.

This prompt was applied to clinical notes from multiple sources. These included records from 100 people treated at Vanderbilt Health, 70 people from the University of California, San Francisco, and 272 people enrolled in seven Roche-sponsored clinical trials.

Study design and model performance

The research team evaluated three models: GPT-3.5, GPT-4, and GPT-4o, with GPT-4o demonstrating the strongest overall performance.

To assess accuracy, the investigators used F1 scores, a metric that balances false positives and false negatives. Scores range from zero to one, with values above 90 percent considered excellent. A score of 80 percent or higher may be sufficient for use in automated clinical decision support systems.

At the patient level, GPT-4o achieved average F1 scores of 56 percent for Vanderbilt Health data, 66 percent for University of California, San Francisco data, and 62 percent for Roche clinical trial data. The models showed a consistent tendency to overpredict the presence of immune-related adverse events.

When analysing individual clinical notes, the model achieved an average F1 score of 57 percent across 667 notes from Vanderbilt Health, evaluating 17 different adverse events.

Implications for clinical practice and research

The findings suggest that large language models can play a role in identifying drug safety signals, even without task-specific training data.

“Manual patient chart abstraction for monitoring the safety and efficacy of drugs already at market requires tremendous resources and puts a drag on the pace of discovery in precision medicine. And that’s especially true with immune checkpoint inhibitors, where the adverse events are so varied. If zero-shot learning with LLMs could help with these notes, it could significantly reduce time and costs for all concerned,” said the report’s corresponding author, Cosmin Bejan, PhD, assistant professor of Biomedical Informatics at Vanderbilt Health.

However, the current level of performance falls short of what would be required for clinical decision support.

“These results show that zero-shot learning with a powerful LLM is useful for detecting these adverse events,” Bejan said. “This performance does not rise to the level required for clinical decision support, but the method could be valuable for automated irAE extraction across multiple sites, potentially speeding discovery and enhancing the safety and effectiveness of cancer immunotherapies.”

Wider research context

The study involved collaboration among multiple researchers at Vanderbilt Health, including Yaomin Xu, PhD, Eric Mukherjee, MD, PhD, Matthew Krantz, MD, Douglas Johnson, MD, MSCI, Elizabeth Phillips, MD, and Justin Balko, PhD. Funding support was provided in part by the National Institutes of Health.

Related research further highlights safety concerns associated with immune checkpoint inhibitors. In a research letter published in JAMA Oncology, Mukherjee, Phillips, and colleagues used logistic regression analysis of adverse event reports from the Food and Drug Administration. They confirmed that these therapies are independently associated with an increased risk of Stevens-Johnson syndrome and toxic epidermal necrolysis, which are severe and potentially life-threatening skin reactions. The study also found that this risk may be linked to exposure to human leukocyte antigen–restricted drugs.

Conclusion

Large language models represent a promising tool for extracting clinically meaningful insights from unstructured health data. While their current performance limits direct clinical application, their ability to operate across multiple datasets without task-specific training suggests potential for supporting large-scale pharmacovigilance efforts. As these models continue to improve, they may contribute to more efficient and comprehensive monitoring of drug safety in clinical practice.

Read More

AI Scribes Deliver Modest Time Savings in Clinical Documentation, Large Study Finds

Key Takeaways:

• AI scribes were associated with small but measurable reductions in electronic health record use and documentation time
• Greater benefits were seen among clinicians who used the tools more frequently
• The reductions observed do not fully explain previously reported improvements in clinician burnout
  
  
  

The burden of clinical documentation

Documenting patient encounters within the electronic health record is a core component of modern healthcare delivery. However, it remains one of the most time-intensive aspects of clinical practice and is widely recognised as a contributor to clinician burnout.

In response, artificial intelligence-enabled ambient documentation tools, commonly referred to as AI scribes, have emerged. These systems automatically generate draft clinical notes based on conversations during patient appointments, allowing clinicians to review and edit them afterwards. While earlier research has suggested these tools may reduce burnout, there has been limited large-scale evidence examining how they affect day-to-day clinical workflows.

A large, real-world study across multiple hospitals

A new study co-led by researchers from Mass General Brigham and the University of California, San Francisco provides insight into this question. The study tracked the use of ambient documentation tools across five hospitals in the United States over a period exceeding two years.

More than 1,800 clinicians using AI scribes were compared with 6,770 clinicians who did not use the technology within the same institutions. This work forms part of the Ambient Clinical Documentation Collaborative, a multi-organisational research initiative.

Modest reductions in time spent on documentation

The findings, published in JAMA, indicate that AI scribes were associated with modest efficiency gains. On average, clinicians using these tools spent 13 fewer minutes per day on the electronic health record and 16 fewer minutes on documentation tasks.

These reductions correspond to relative decreases of 3% in overall EHR usage and 10% in documentation time.

The study also identified a small increase in productivity. Clinicians using AI scribes completed approximately 0.5 additional patient visits per week compared with those who did not use the technology.

Frequency of use influences impact

The benefits of AI scribes were not evenly distributed. The most notable improvements were observed among primary care physicians, advanced practice providers, female clinicians, and those who used the tools in at least half of their patient encounters.

Clinicians who used AI scribes for more than 50% of visits experienced roughly twice the reduction in total EHR time and three times the reduction in documentation time compared with less frequent users. Despite this, only 32% of clinicians adopted the technology at this level of regular use.

Financial impact remains limited

Although the increase in patient visits translated into higher revenue, the financial gains were modest. On average, clinicians using AI scribes generated an additional $167 per month.

This suggests that while the tools may offer efficiency benefits, their economic impact at an individual clinician level remains relatively small.

No change in after-hours workload

One notable finding was that time spent using the electronic health record outside of standard working hours did not differ significantly between clinicians using AI scribes and those who were not.

This raises important questions about how time savings during the working day are being redistributed and whether they meaningfully reduce workload burden or are absorbed by other clinical or administrative tasks.

Understanding the link to burnout

Despite prior evidence suggesting that ambient documentation tools may reduce clinician burnout, the mechanisms behind this effect remain unclear.

“Previous studies link ambient documentation to a significant decrease in burnout, but the underlying drivers of this reduction have been unclear,” said senior author Rebecca G. Mishuris, MD, MS, MPH, Chief Health Information Officer at Mass General Brigham.

“The modest reductions in documentation time we observed are unlikely to fully account for changes in burnout, underscoring the need to understand how these tools change how clinicians approach care delivery while using them.”

Adoption and real-world implementation

The study highlights both the promise and the limitations of AI scribes in real-world clinical settings. While measurable improvements were observed, their magnitude was relatively small and depended heavily on consistent use.

“Ambient documentation use is expanding rapidly across U.S. health care, making it essential to study how these technologies are impacting clinicians in real time,” said lead and corresponding study author Lisa Rotenstein, MD, MBA, an associate professor of medicine at the UCSF School of Medicine, and director of The Center for Physician Experience and Practice Excellence at Brigham and Women’s Hospital.

“Our study demonstrates the impact of AI scribes in diverse real-world implementations at multiple sites. It also emphasizes the value of helping clinicians become comfortable with the technology so that they are reaping its full benefits via frequent use.”

The need for further research

The findings suggest that while AI scribes can improve efficiency, they are not a complete solution to the challenges associated with clinical documentation or clinician burnout.

Further research is needed to understand how these tools influence clinician behaviour, how saved time is reallocated, and whether broader system-level changes are required to fully realise their potential benefits.

Read More

Automated Weight Loss Programme Shows Promise for People Living with Cancer in Landmark Trial

Key Takeaways: 

• A fully automated, web-based programme delivered clinically meaningful weight loss in people living with and beyond cancer, without any in-person support
  
• More than 43 percent of participants achieved at least 3 percent weight loss, with nearly one in three reaching 5 percent or more
  
• The intervention also improved a range of health outcomes, including diet quality, physical functioning, and cardiometabolic markers
  
  
  

A new model for post-cancer care

A large national randomised clinical trial has demonstrated that a fully automated, web-based weight loss intervention can deliver substantial health benefits for people living with and beyond cancer. The programme, developed by researchers at the University of Alabama at Birmingham, represents a significant shift in how post-cancer care may be delivered in the future.

Published in the Journal of the National Comprehensive Cancer Network, the study reported the highest level of weight loss ever achieved through a fully automated intervention in this population. The programme, known as the AMPLIFY Diet (AiM, PLan and act on LIFestYles), was designed to provide structured, evidence-based lifestyle support without requiring direct clinician involvement.

Addressing a major unmet need

A substantial proportion of people living with and beyond cancer are also living with overweight or obesity. In the United States, this figure is estimated to be around 70 percent. This places individuals at increased risk of cardiovascular disease, type 2 diabetes, functional decline, cancer recurrence, and the development of second primary cancers.

Despite this, access to specialist oncology dietitians remains limited. Traditional weight management programmes often rely on in-person consultations or regular coaching, which can be difficult to scale and may not be accessible to all patients.

The AMPLIFY Diet intervention was developed specifically to address these barriers by delivering personalised nutrition and behavioural support entirely online.

A fully automated intervention

The programme operates without live coaching, counselling calls, or face-to-face appointments. Instead, it uses a structured digital platform that includes weekly interactive sessions, goal-setting tools, progress monitoring, and automated personalised feedback.

Participants engage with the system independently, receiving guidance that is grounded in established behavioural and nutritional science. This approach allows for scalability while maintaining a consistent standard of care.

“This is a game changer for cancer survivorship care,” said Wendy Demark-Wahnefried, Ph.D., R.D., senior author and professor at UAB’s School of Health Professions and O’Neal Comprehensive Cancer Center. “We showed that a completely automated online program grounded in decades of behavioral and nutrition science can safely and effectively help cancer survivors lose weight and improve their health at scale.”

Study design and participant profile

Between 2020 and 2024, the study enrolled 349 participants aged between 50 and 82 years from 31 states across the United States. All participants were living with and beyond cancers associated with obesity.

The cohort included individuals with a range of cancer types, including breast, colorectal, prostate, endometrial, ovarian, thyroid, renal, and haematologic cancers. Participants were randomly assigned to either the AMPLIFY Diet programme or a control group receiving standard survivorship information.

Clinically meaningful weight loss outcomes

After six months, the results showed clear differences between the intervention and control groups.

More than 43 percent of participants in the AMPLIFY Diet group achieved weight loss of at least 3 percent of their body weight. In comparison, only 13 percent of those receiving usual care reached this threshold.

In addition, nearly one in three participants in the intervention group lost at least 5 percent of their body weight. This level of weight loss is widely associated with reductions in cardiovascular risk and improvements in cancer-related outcomes.

On average, weight loss in the intervention group was nearly five times greater than that observed in the control group.

Broader health improvements

The benefits of the programme extended beyond weight loss alone. Participants in the AMPLIFY Diet group experienced improvements across multiple domains of health and wellbeing.

These included reductions in waist circumference and overall caloric intake, as well as improvements in diet quality. Biochemical markers also shifted in a favourable direction, with lower circulating levels of leptin, a hormone associated with cancer progression and cardiometabolic disease.

Further gains were observed in blood pressure, physical functioning, and cognitive performance. Participants also reported improvements in depression and their ability to engage in social roles, suggesting a broader impact on quality of life.

Strong engagement without human support

One notable finding from the study was the level of participant engagement. Individuals completed an average of 60 percent of the weekly sessions, which is considerably higher than engagement rates typically reported in other digital lifestyle interventions.

This suggests that a well-designed automated system can maintain user engagement even in the absence of direct human interaction.

Implications for scalable care

Unlike many conventional weight management programmes, the AMPLIFY Diet intervention does not require ongoing staff involvement. This makes it particularly well suited for integration into healthcare systems, cancer centres, and community-based services.

The ability to deliver consistent, evidence-based care at scale may help address longstanding gaps in survivorship support, particularly in settings where specialist resources are limited.

The role of behavioural and nutritional care

The researchers emphasise that lifestyle-based interventions remain a cornerstone of care for people living with and beyond cancer, particularly as pharmacological approaches continue to evolve.

“Behavioral and nutritional interventions are essential,” Demark-Wahnefried said. “Diet quality, muscle preservation, cognition, and long-term sustainability of a healthful lifestyle and body weight are critical for cancer survivors, and even if weight loss medications eventually receive broadscale endorsement, they alone do not address all of these needs.”

Future directions

The research team is now focusing on expanding the reach of the AMPLIFY Diet programme across both clinical and non-clinical settings. The aim is to improve access to effective survivorship care while also contributing to broader cancer prevention efforts.

The study was funded by the National Institutes of Health and the American Cancer Society.

Read More

Growing Use of Emojis in Electronic Health Records Raises Safety Questions

Key Takeaways:

• Emoji use in electronic health records has increased steadily between 2020 and 2025, appearing in thousands of clinical notes.
  
• While emojis may offer a quicker, more expressive way to communicate, they introduce risks of misinterpretation across clinicians and patients.
  
• Experts suggest that clearer governance and guidance may be needed to ensure safe and consistent use in clinical documentation.
  
  
  

Study reveals growing presence of emojis in clinical records

A recent study from Michigan Medicine, published on 14 January, examined 218.1 million clinical notes from the electronic health records of 1.6 million people receiving care. The findings revealed a notable rise in emoji usage by both healthcare professionals and patients between January 2020 and September 2025.

Across all records analysed, researchers identified 372 unique emojis appearing in 4,162 notes. While this represents a small proportion of total documentation, the upward trend signals a shift in how digital communication is entering clinical environments.

David Hanauer, clinical associate professor of paediatrics and learning health science at Michigan Medicine, explained that the study was initially driven by simple curiosity.

“It was mostly out of interest, just trying to explore if anything was there at all,” Hanauer said. “Our understanding had been that emojis and other symbols are actually not supposed to be used in a medical record, so we were wondering: Were there any there at all, and how often were they being used, and which ones?”

Concerns around clarity and misinterpretation

Despite their increasing use, emojis raise important concerns about clarity in clinical communication. Hanauer highlighted that the ambiguity of many emojis could lead to misunderstanding.

“Most of the concerns that people have is that it’s hard to understand from an emoji what is being conveyed,” Hanauer said. “Maybe a smiley face is pretty obvious to most people, but there’s a lot of different faces with nuances and other symbols. I think there can be a lot of miscommunication, misinterpretation.”

The issue becomes more complex when considering variation in interpretation across different groups of people receiving care and healthcare professionals.

Kim Ford, a health information business systems analyst lead at Michigan Medicine, emphasised how generational differences may influence understanding.

“If you have older patients who may not be familiar with emojis, it’s almost like a foreign language to them,” Ford said. “(For) our younger generation – or those people that have grown up with technology – it’s a second language for them that they understand very well. That’s my biggest concern.”

Accessibility challenges for some people receiving care

Beyond interpretation, accessibility presents another potential barrier. Hanauer noted that small visual symbols may be difficult for some individuals to distinguish clearly, particularly those with visual impairments.

“For older people, having small emojis might actually be hard for them to see and make out, so they might see its face but they can’t tell what the specific expression is,” Hanauer said. “I think we found over 300 different kinds of emojis being used. That’s a lot of different symbols that people would have to understand what they mean.”

This highlights a broader concern that even seemingly simple visual cues may not be universally interpretable or accessible.

Potential implications for patient care

A key concern raised by the study is whether emoji misinterpretation could affect clinical outcomes. While there is currently no direct evidence linking emoji use to adverse outcomes, the possibility remains.

“We hope that doesn’t happen, but I think because of that concern, there’s probably going to be a little bit more oversight,” Hanauer said. “I don’t think we would easily be able to find a circumstance in which there was actually some sort of better or negative outcome from an emoji being misinterpreted.”

The absence of clear evidence does not eliminate the risk, particularly in high-stakes environments where precise communication is essential.

Balancing efficiency with professionalism

Some healthcare professionals recognise potential benefits in using emojis, particularly in reducing communication burden within electronic systems. However, concerns remain about maintaining professionalism and objectivity.

Leah Beel, a medical assistant at American Family Care in Ann Arbor, expressed reservations about their place in formal documentation.

“From my experience, EHRs are used to get quick information and try to communicate with each other in a fast and reliable way,” Beel said. “The only thing I would use is an exclamation point, which, even then, is kind of out there. It’s a good thing that emojis can show enthusiasm or certain reactions, but I also think to a degree – it’s not unprofessional but just someone might take it the wrong way. My perspective on EHR is that you write very objectively.”

In contrast, Elizabeth Rossmann Beel, a paediatric anaesthesiologist at Texas Children’s Hospital, noted that emojis may offer a more efficient way to communicate in certain contexts.

“It’s a way to react to something without putting as much effort into it, or into making that person who’s reading it feel like they need to reply,” Rossmann Beel said. “I think it can cut down a little bit on the burden of replying to and responding to messages in the EHR, which is nice. However, it’s definitely more casual, and so sometimes that’s not the best tone to be setting in a medical record.”

The case for governance and standardisation

Given the growing use of emojis, there is increasing interest in whether formal guidance or regulation should be introduced.

Ford suggested that healthcare organisations may need to consider structured governance around emoji use.

“Maybe emojis are an acceptable means of communication,” Ford said. “The other piece is, should there be a governance process around what emojis can be used? And in what situations? I need to think a little bit about what their structure might look like – what department should be involved in reviewing and approving those, what should be the process to submit an emoji for consideration for use? There’s a lot of pieces to the governance process that need to be figured out there.”

A shift in digital communication within healthcare

The findings from this study reflect a broader evolution in digital communication, where informal elements are beginning to intersect with traditionally formal systems such as electronic health records.

While emojis may offer efficiency and emotional nuance, their integration into clinical documentation raises important questions about clarity, accessibility, professionalism, and patient safety. As their use continues to grow, healthcare systems may need to balance flexibility with standardisation to ensure communication remains precise, inclusive, and clinically appropriate.

Read More

Algorithm-Guided Insulin Dosing Improves Blood Sugar Control in Type 2 Diabetes

Key Takeaways:

• An algorithm paired with continuous glucose monitoring significantly increased time in target glucose range compared with standard self-monitoring approaches
  
• The tool provides personalised weekly insulin dose recommendations based on recent glucose data, helping to simplify titration
  
• Early findings suggest strong patient acceptability and potential to enhance diabetes management at scale, though larger trials are needed
  
  
  

A data-driven approach to insulin adjustment

A novel algorithm developed by researchers at the University of Virginia Center for Diabetes Technology has demonstrated encouraging results in supporting people living with Type 2 Diabetes to better manage their blood glucose levels.

The system works in combination with a continuous glucose monitor and provides tailored recommendations for insulin dose adjustments. Rather than relying solely on manual interpretation of glucose readings, the algorithm analyses patterns over time and offers structured, data-informed guidance.

In a clinical trial involving 30 participants, individuals were randomly assigned to one of two approaches over a 16-week period:

• Algorithm-guided insulin adjustment using continuous glucose monitoring data
  
• Traditional self-monitoring of blood glucose with independent dose adjustment
  

The results showed a marked improvement in glycaemic control among those using the algorithm. Participants in this group increased their average time spent within a safe blood glucose range from 54.1% to 75.3%. By contrast, those relying on self-monitoring alone saw a more modest increase from 50.2% to 55.3%.

Moving beyond traditional insulin management

The findings highlight the growing role of digital health tools in diabetes care. According to Marc D. Breton, the study’s lead author:

“These results clearly show that diabetes technology and advanced algorithms can be leveraged to great effects, well beyond the classical paradigm of automated insulin delivery. As continuous glucose monitoring and connected medical devices become ubiquitous, we have the opportunity to provide highly personalized advice and monitoring to people with diabetes and guide their use of insulin and medications. Showing the impact of these technologies in early insulin therapy (only one dose a day) opens the door to helping the vast majority of people using insulin, well beyond what we were able to achieve with automated insulin delivery.”

This perspective reflects a broader shift towards personalised, technology-enabled care. Rather than fully automated systems alone, there is increasing interest in decision-support tools that augment clinical judgement and patient self-management.

Addressing the challenges of insulin titration

For many people living with type 2 diabetes, treatment often begins with oral or non-insulin therapies. However, as the condition progresses, insulin may become necessary to maintain adequate glycaemic control.

Adjusting insulin doses – a process known as titration – can be complex and burdensome. It typically requires frequent monitoring, interpretation of glucose patterns, and iterative dose changes. Importantly, there is no universally standardised titration protocol, which can create variability in care and outcomes.

To address this, Anas El Fathi developed the algorithm with the aim of streamlining and improving this process. The system evaluates two weeks of continuous glucose monitoring data and generates weekly recommendations for insulin dose adjustments, offering a structured and personalised approach.

Strong acceptance and clinical potential

The study also explored how well the technology was received by participants. According to Ralf Nass:

“From a medical point of view, it was fascinating to see that the algorithm was not only better than the standardized insulin titration recommendations, but also how well the technology was accepted by the participants with type 2 diabetes. This type of technology has the potential to help physicians enable their patients to achieve better glycemic control faster by using a personalized approach.”

This combination of improved outcomes and user acceptability is particularly important, as adherence and engagement remain key challenges in long-term diabetes management.

Future directions – towards more personalised diabetes care

While the results are promising, the researchers emphasise that further validation is required. Larger and longer clinical trials will be needed to confirm the effectiveness of the algorithm across more diverse populations.

Looking ahead, the integration of more advanced data-driven approaches may further enhance personalisation. Breton noted:

“It is only the very beginning of these efforts. With early demonstration behind us, we can focus on robust approaches that will be effective with more varied populations. Integrating recently developed data-driven methodologies, especially digital twins, to further improve our capacity to tailor diabetes managements to individuals is likely to once more revolutionize diabetes care.”

Such developments could represent a significant step forward in precision medicine for people living with diabetes.

Study publication and funding

The findings have been published in the peer-reviewed journal Diabetes Technology & Therapeutics, with the article available as open access.

The research team included El Fathi, Nass, Carol J. Levy, Camilla Levister, Grenye O’Malley, Nirali A. Shah, Shaziah Hassan, Cheryl Quainoo, Chaitanya L.K. Koravi, Taylor N. Nguyen, Giulio Matteo Santini, Emma Emory, Carlene Alix, Dillon K. Flanagan, David Fulkerson, Mary Clancy Oliveri, Christian Laugesen, Jonas K. Lineolov, Peter W. Hansen and Breton.

The clinical trial was supported by a grant from Novo Nordisk.

Read More

AI Diet Recommendations for Adolescents Show Significant Nutritional Gaps, Study Finds

Key Takeaways:

• AI-generated diet plans consistently underestimated energy and key macronutrients required by adolescents
  
• Macronutrient balance was frequently misaligned with clinical guidelines, with lower carbohydrates and higher fat and protein levels
  
• Researchers caution that AI tools should not replace dietitians for adolescent nutrition without professional oversight
  
  
  

Growing demand for accessible nutrition support

Artificial intelligence is increasingly being used to support dietary planning, particularly in areas where access to qualified professionals is limited. However, a new study published in Frontiers in Nutrition raises important concerns about the reliability of these tools when applied to adolescents living with overweight or obesity.

Globally, adolescent overweight and obesity are rising at pace, affecting an estimated 390 million young people in 2022. In many regions, this now represents the most common form of malnutrition. Excess body weight in adolescence is associated with a range of adverse health outcomes, including type 2 diabetes, dyslipidaemia, hypertension, and sleep apnoea. It also increases the likelihood of obesity in adulthood and is linked to reduced quality of life.

Alongside physical health risks, adolescents may experience body image concerns and engage in harmful weight control behaviours such as self-induced vomiting or misuse of laxatives.

Dietary modification remains central to improving outcomes. Dietitians play a key role in delivering tailored, evidence-based nutrition plans aligned with established guidelines. However, limited access and workforce pressures can restrict the availability of personalised support.

AI tools, including chatbots and large language models, are increasingly being explored as a way to bridge this gap. While they can provide general dietary guidance, concerns remain about their accuracy, safety, and ability to replicate the individualised care provided by trained professionals.

Study design – comparing AI models with dietitian plans

To better understand the role of AI in adolescent nutrition, researchers conducted a direct comparison between AI-generated diet plans and those created by a dietitian.

Five AI systems were evaluated: ChatGPT-4o, Gemini 2.5 Pro, Claude 4.1, Bing Chat-5GPT, and Perplexity. Across two sessions, these models generated a total of 60 diet plans. Each plan covered three days and was based on four standardised adolescent profiles, including boys and girls living with overweight or obesity.

These AI-generated plans were compared with dietitian-designed one-day plans developed in line with established nutritional recommendations. The reference plans followed a macronutrient distribution of:

• 45–50 % carbohydrates
• 30–35 % fat
• 15–20 % protein

The researchers then analysed energy intake, macronutrient composition, micronutrient content, safety, and feasibility.

Consistent underestimation of energy and macronutrients

The findings revealed a clear and consistent pattern across all AI models. Diet plans generated by AI underestimated both total energy intake and key macronutrients when compared with dietitian-designed plans.

On average:

• Energy intake was lower by 695 kcal
• Protein intake was reduced by 20 g
• Fat intake was reduced by 16 g
• Carbohydrate intake was reduced by 115 g

Given the high energy demands of adolescence, such deficits could have meaningful clinical implications, particularly for growth, development, and overall health.

Macronutrient imbalance – a shift away from guidelines

Beyond total intake, the balance of macronutrients was also significantly altered in AI-generated plans.

Some AI models recommended:

• Protein intake up to 23.7 %
• Fat intake up to 44.5 %

Both values exceeded recommended levels. In contrast, carbohydrate intake accounted for no more than 36.3 %, falling below guideline recommendations.

Dietitian-designed plans, by comparison, remained closely aligned with clinical standards:

• Carbohydrates: 44 %–46 %
• Protein: 18 %–20 %
• Fat: 36 %–37 %

The authors noted:

“This pattern illustrates a systematic shift across all AI models to lower CHO, higher protein, and higher lipid meal structures, indicating that the macronutrient balance, not just the amount of gram-based nutrients, is significantly disrupted in AI-generated plans.”

Researchers suggest that AI models may be influenced by popular dietary trends, such as low-carbohydrate or ketogenic approaches, rather than evidence-based adolescent nutrition guidelines. This shift may pose risks during a critical period of physical and cognitive development.

Micronutrient variability raises additional concerns

In addition to macronutrient discrepancies, the study identified significant variability in micronutrient composition across AI-generated plans.

No model consistently matched the dietitian-designed reference diet across all nutrients. This inconsistency raises concerns about potential micronutrient deficiencies, which could further compromise adolescent health.

The findings suggest that AI tools currently lack the technical precision required to accurately estimate both macro- and micronutrient needs in personalised dietary plans for adolescents.

Strengths and limitations of the study

The study offers several notable strengths. It evaluated multiple AI models, allowing for robust comparison across systems. The use of three-day diet plans enabled identification of consistent patterns rather than isolated outputs. Dietitian-designed plans provided a credible clinical benchmark, and the inclusion of both macro- and micronutrient analysis allowed for a comprehensive assessment of dietary quality.

However, there are limitations to consider. The findings are specific to the models tested, which are rapidly evolving. Standardised adolescent profiles may not fully capture real-world complexity, limiting personalisation. The use of simulated scenarios rather than real-life behaviours may reduce ecological validity. Additionally, prompts were standardised and delivered in a single language, which may limit generalisability across populations.

Implications for clinical practice and AI use

The study highlights important risks associated with the unsupervised use of AI for adolescent dietary planning.

As the authors conclude:
“AI models have exhibited clinically significant deviations in diet plans for adolescents at both macro and micro levels.”

These deviations include consistently lower energy and carbohydrate recommendations compared with dietitian-designed plans.

Until these limitations are addressed, AI-generated diet plans should be used with caution. They may serve as a supplementary tool under professional supervision, but they are not currently a safe or reliable substitute for qualified dietary guidance in adolescents.

Read More

Virtual Care Could Reduce Hospital Admissions in Severe Eating Disorders, Study Finds

Key Takeaways:

• A fully virtual, multidisciplinary treatment programme demonstrated strong engagement and positive clinical outcomes for adults living with severe eating disorders
  
• Structured online support during high-risk transition periods may help reduce hospital admissions and sustain recovery following discharge
  
• The model highlights the potential for digitally delivered, evidence-based care to bridge gaps between inpatient and community services
  
  
  

Evaluating a new approach to severe eating disorder care

An evaluation conducted by Oxford Health NHS Foundation Trust has examined whether intensive, fully virtual treatment can effectively support people living with severe eating disorders. The study, published in the Journal of Eating Disorders, focused on a service known as Step Care, developed through the HOPE Provider Collaborative.

Researchers describe this as the first prospective study to investigate a completely virtual, intensive treatment model using multidisciplinary enhanced cognitive behavioural therapy (CBT-E). The programme is designed to support individuals as they transition between inpatient treatment and community-based care.

This period of transition is widely recognised as a critical phase in recovery. People living with severe eating disorders face a particularly high risk of relapse shortly after discharge from hospital, especially within the first two months. The study therefore explored whether structured virtual support during this vulnerable period could help maintain recovery and reduce the likelihood of readmission.

Addressing gaps in existing services

Step Care was developed in response to well-documented challenges within eating disorder services. These include fragmented transitions between inpatient and community care, repeated hospital admissions, and limited access to intensive day treatment.

The service is delivered entirely online and brings together a multidisciplinary team, including professionals from psychology, nursing, dietetics, and art therapy. This integrated approach aims to provide consistent and coordinated care across different stages of recovery.

Step Care operates through three distinct pathways:

• Starting Well – for individuals at risk of requiring hospital admission, with a focus on prevention
  
• Staying Well – for those recently discharged from inpatient care, supporting early recovery
  
• Working towards Recovery – for individuals who have begun restoring weight and are focusing on longer-term recovery
  

Lucy Gardner, professional lead dietitian within the Step Care service, highlighted the importance of integrating nutritional support within a broader therapeutic framework:

“Nutrition plays a crucial role in mental health, yet access to the right level of dietetic support is often inconsistent,” she said. “Our model offers a clear, evidence-informed way to tailor dietetic input to individual need, delivering CBT-E virtually as part of a multidisciplinary team.”

Positive outcomes across key measures

The evaluation reported high levels of engagement and programme completion, including among individuals who had been living with eating disorders for an extended period.

Participants within the Starting Well pathway experienced significant improvements across several clinical and psychological measures, including:

• Body mass index (BMI)
  
• Eating disorder symptoms
  
• Psychosocial impairment
  
• Mood
  

Importantly, most individuals in this group were able to avoid hospital admission during the course of the programme.

For those in the Staying Well pathway, outcomes were also encouraging. Participants maintained their weight and experienced a reduction in the overall impact of their illness during a period typically associated with high relapse risk. Unplanned hospital admissions were reported to be rare, and many individuals were successfully supported in transitioning to community-based care.

Sharon Ryan, nurse lead within the Step Care service, emphasised the importance of this post-discharge phase:

“The weeks after leaving hospital are often the most fragile,” she said. “Step Care provides consistent multi-disciplinary support at that point, helping people maintain their recovery with support to feel safe and confident out of hospital.”

Implications for future care models

The findings suggest that intensive, evidence-based treatment for severe eating disorders can be delivered effectively in a virtual format. This approach may offer a valuable additional option for supporting individuals at home, particularly during critical transition periods.

Agnes Ayton, clinical lead for the HOPE Provider Collaborative, explained the underlying aim of the service:

“Step Care was designed to bridge the gap between inpatient and community services,” she said. “The findings show that intensive, evidence-based treatment can be delivered safely online, providing continuity of care at a time when people are most vulnerable.”

She also noted that both engagement and clinical outcomes were encouraging, including among individuals who had experienced long-term illness.

A complement to existing services

The authors conclude that virtual programmes such as Step Care may serve as an important complement to traditional inpatient and community services. By providing structured, multidisciplinary support during high-risk periods, these models have the potential to enhance continuity of care and support sustained recovery for people living with severe eating disorders.

As healthcare systems continue to explore digital and hybrid models of care, this study adds to a growing body of evidence suggesting that virtual interventions can play a meaningful role in complex, long-term conditions.

Read More