Research articles

In a prospective multi-centre cohort in China, people living in areas of high soil selenium had a 35% increased risk of incident diabetes over a 3-year study period, with a stronger effect seen for men.

A transmission model, integrating trial-derived prophylaxis efficacy and behavioural factors such as screening frequency and intervention uptake and adherence, evaluates different enrolment programmes for doxycycline post-exposure prophylaxis in terms of their potential effects on syphilis incidence, using data and indications from Singapore and England.

A global analysis sheds light on the dynamics and features involved in influenza subtype composition in epidemic seasons between 2000 and 2023 and presents a series of tools to predict next year’s composition.

In a randomized preregistered study, 282 participants were asked to judge a hypothetical malpractice case in different conditions of artificial intelligence (AI) assistance to a radiologist. An AI-assisted second-read framework was found to reduce the fraction of participants who would side with the plaintiff.

In a retrospective analysis of 1,469 patient encounters, a large language model-based clinical decision support system deployed across 16 outpatient clinics provided recommendations generally aligned with local guidelines, but with contrasting results on safety and relevance.

A qualitative analysis of two decades of policy documents from 200 countries and interviews with 46 key informants found that adoption of policies to promote physical activity has increased since 2004, but implementation remains weak because physical activity is still a low, albeit gradually increasing, political priority in most countries.

A physical activity and climate change framework integrates climate mitigation, adaptation, health equity and Indigenous knowledge into public health campaigns for physical activity promotion.

Using data involving 24,984 clinicians and 5,210 trials, an AI model integrates information from medical claims, unstructured trial documents and historical enrolment relationships to suggest optimal trial-site configurations, to enhance cost-effectiveness and ethnicity representation.

Pregnant women exposed to cumulative days of high temperature saw a twofold-increased risk of gestational diabetes in a time-series study of 8,796 such cases in Barcelona between 2011 and 2022.

Including data for more than 125,000 compounds and biological agents, a dataset leverages a variety of sources including trial registries, news and stock prices and artificial intelligence to annotate outcomes and features automatically, as a resource to inform drug development from an organizational perspective and identify features that could be predictive of trial success.

Following a privacy-preserving framework, an open-source platform allows for continuous monitoring of a wide range of smartphone-based signals, including moment-by-moment capture of screenshots, application usage logs, interaction histories and phone sensor readings.

A generative model is used to simulate a digital twin population based on the records of 1,856 individuals in Singapore, on which different screening programmes for case finding in cognitive impairment are tested, balancing detection rate and cost.

Using dietary data from 8,931 participants in the WELL-China cohort, this study identifies a new diet rooted in traditional Chinese food elements associated with better adherence, favourable gut microbial profiles, lower rates of central obesity and reduced incidence of major cardiovascular events, and confirms these findings in an independent Achieving Better Omics validation cohort (n = 1,851).

Data from 12,608 pregnancies across 7 African countries and a simulation model suggest that the total number of malaria-exposed pregnancies across sub-Saharan Africa exceeded 13 million in 2023, and that current prevention measures avoided more than 2 million malaria-related anaemia cases.

This scoping review examines previous experience in performing silent evaluations of clinical AI applications, collecting evidence from 75 studies on implementation features and the sociotechnical context.

A systematic review including 574 studies extracts information about transmissibility, epidemiological delays and outbreaks for Zika virus disease at global scale.

A multi-source dataset including a survey on 2,364 Ukrainian citizens pinpoints the spatial and temporal features associated with stronger interactions between conflict-related and baseline stressors and health-related outcomes, in a Ukraine-wide analysis involving 461 cities.

In scenarios involving 1,000 acute-pain clinical vignettes, varied across 34 socio-demographic features, a panel of ten large language models is found to provide inconsistent recommendations, which also present disparities with respect to marginalized populations.

In a benchmark analysis comprising 5,609 clinical questions developed by 101 community health workers from Rwanda, a panel of 5 general large language models performed better than humans across all metrics.

In a randomized controlled study involving 58 physicians in Pakistan, assistance by a large language model in diagnostic reasoning resulted in a 27.5% increase in performance on 6 clinical vignettes.