(This is part of the Health Rounds newsletter, where we feature the latest medical studies on Tuesdays and Thursdays)
Dec 17 (Reuters) – Insecticide treatment used in soldiers’ uniforms on clothing used to carry children significantly reduced the incidence of malaria in children, researchers found.
The six-month study, conducted in areas of Uganda with a malaria epidemic, included 400 mothers and their babies aged 6 months to 18 months. Half were randomly assigned to use cotton cloth wraps treated with Sawyer Products’ permethrin, while the others received plain water-treated cloths as the control group. The wraps were retreated every 4 weeks.
All pairs received insecticide-treated sleeping nets.
According to a study report published in The New England Journal of Medicine, permethrin-treated baby wraps reduced malaria cases in infants by 66%.
Adverse events were mild, and the frequency was low and similar in the treatment group and the sham group, the report said.
“Given the expected duration and frequency of use… extended follow-up of children, particularly for neurodevelopmental effects of permethrin exposure, is warranted,” the researchers acknowledge.
“Yet malaria, both severe and uncomplicated forms, can cause long-term cognitive impairment, and careful weighing of potential risks and benefits will be necessary.”
AI needs special training to detect cancer in low-risk groups
Two recent studies highlight the potential for artificial intelligence tools, if the tools are not properly trained, to be less accurate in some patients than others.
It is well known that if AI tools are trained on data collected from a disproportionate proportion of patients from different demographic groups, they have difficulty making accurate diagnoses in minority groups that are not well represented in the training set.
But in the current analysis, the models sometimes performed worse in one demographic group even though the sample sizes were comparable, the researchers reported in Cell Reports Medicine.
The reason may be that some cancers are more common in certain groups, so the models are better at diagnosing those groups. As a result, the models may have difficulty diagnosing cancers in populations where they are not common, the researchers found.
Additionally, subtle molecular differences may exist in biopsy samples from different demographic groups, and AI may detect those differences and use them as a proxy for cancer type, potentially making it less effective at diagnosing populations in which these mutations are rare.
“We found that because AI is so powerful, it can differentiate many obscure biological signals that cannot be detected by standard human assessment,” study leader Kun-Sing Yu of Harvard Medical School said in a statement.
As a result, models may pick up signals that are more related to demographics than to disease—and inferring demographic information from pathology slides may affect their diagnostic ability across groups.
Together, Yu said, these explanations suggest that bias in pathology AI comes not just from the variable quality of training data but from how researchers train the models.
When his team applied the new framework to the models they tested, it reduced diagnostic disparities by 88%, they said.
“We show that by making this small adjustment, the models can learn robust features that make them more general and better across different populations,” Yu said.
The finding is encouraging, he added, because it suggests that bias can be reduced even without training models on completely unbiased, representative data.
In a separate study published in PLOS Biology, the researchers found that even with broad samples of bacterial populations, bias may be limiting AI’s ability to predict and combat antibiotic resistance.
(Sign up for free here to receive the full newsletter in your inbox)
(Reporting by Nancy Lapid; Editing by Bill Burkrot)
