(This is an excerpt from the Health Rounds newsletter, where we publish the latest medical research on Tuesdays and Thursdays)
Dec 17 (Reuters) – Treating clothes used to carry babies with an insecticide used on soldiers’ uniforms could significantly reduce the incidence of malaria in children, researchers have found.
The 6-month study was conducted in a malaria-endemic area of ​​Uganda and included 400 mothers and their infants aged 6 to 18 months. Half were randomly assigned to use cotton wraps treated with Sawyer Products permethrin, while the others used water-treated cloth wraps as a control group. ‌Packs are re-treated every 4 weeks.
All pairs received insecticide-treated sleeping nets.
According to a study published in the New England Journal of Medicine, permethrin-treated baby wraps reduced infant malaria cases by 66%.
Adverse events were mild, less frequent and similar in the treatment and sham groups, the report said.
The researchers acknowledged that “long-term follow-up of children is needed, particularly regarding the neurodevelopmental effects of permethrin exposure, given the expected duration and frequency of use.”
“However, malaria, whether severe or uncomplicated forms, can cause long-term cognitive impairment, so the potential risks and benefits need to be carefully weighed.”
Artificial intelligence requires nuanced training to spot cancer in low-risk groups
Two recent studies highlight that AI tools may be less accurate for some patients than others if they are not properly trained.
It is well known that if AI tools are trained on data collected from unequal proportions of patients from different demographic groups, they have difficulty making accurate diagnoses of minority groups that are underrepresented in the training set.
But in the current analysis, the models sometimes performed worse within one population group, even with comparable sample sizes, the researchers report in the journal Cell Reports Medicine.
The reason may be that certain cancers are more common in certain groups, so the model is better able to diagnose in those groups. As a result, the models may have trouble diagnosing people where cancer is less common, the researchers found.
Additionally, there may be subtle molecular differences in biopsy samples from different population groups that AI can detect and use as a proxy for cancer types, which may make it less effective at diagnosis in populations where these mutations are less common.
“We found that because AI is so powerful, it can distinguish many biological signals that are difficult to detect through standard human assessment,” study leader Kun-Hsing Yu of Harvard Medical School said in a statement.
Therefore, these models may pick up signals more relevant to demographics rather than disease, and inferring demographic information from pathology slides may affect their diagnostic capabilities across groups.
These explanations suggest that bias in pathological AI stems not just from the varying quality of training data, but also from the way researchers train their models, Yu said.
When his team applied the new framework to the models they tested, diagnostic differences were reduced by about 88%, they said.
“We show that by making such small adjustments, models can learn powerful features that make them more general and fair across different populations,” Yu said.
He added that this finding is encouraging because it shows that bias can be reduced even without training models using completely fair, representative data.
In a separate study published in PLOS Biology, researchers found that even with a broad sample of bacterial populations, bias can hinder the potential of artificial intelligence to predict and combat antibiotic resistance.
(To receive the full newsletter in your inbox for free, sign up for free here)
(Reporting by Nancy Rapid; Editing by Bill Burkrot)
