Mind reading machines … and other research (2024)

1. Tom Nolan, clinical editor; sessional GP, Surrey

1. The BMJ, London

A lesson in TACT

There was no plausible mechanism to suggest that removing heavy metals such as cadmium and lead using edetate disodium (EDTA) infusions might reduce cardiovascular events in people who have had a myocardial infarction, but the TACT randomised control trial was funded nonetheless. In 2013 the findings surprised the academic world when those in the EDTA chelation arm of the trial were found to have lower rates of cardiovascular events than those in the placebo group. Eleven years on, the TACT2 study has failed to reproduce these results, finding no cardiovascular benefits from EDTA chelation compared with placebo among the nearly 1000 participants with type 2 diabetes and a history of myocardial infarction. An aptly tactful editorial note accompanying the study in JAMA calls for us to learn from this story and reminds us that “common sense dictates that not all hypotheses need be subjected to a clinical trial.”

JAMA doi:10.1001/jama.2024.11463

Looking for indications

A study of 4752 nursing home residents in the US found that 582 (12.2%) were receiving an oral anticoagulant and an antiplatelet, and 45% of those receiving an antiplatelet (mostly aspirin) had no indication for the prescription recorded. We’ve all been there, hunting through consultation notes that give a Pulitzer-worthy account of the patient’s narrative but nothing about why a new medication was prescribed, clinic letters that joyfully declare cancer has been excluded but refuse to offer any opinion on symptoms or treatment, and medication history audit trails that have an uncanny tendency to reveal that the medication was first prescribed on one of the rare days when nothing else was recorded in the patient’s notes. Ten minutes later, with the time allocated to the consultation over, you turn back to the patient: “What was it you were saying?”

JAMA Intern Med doi:10.1001/jamainternmed.2024.3819

Mind reading machines

In these frustrating situations, it's just as well we don’t wear devices that can translate the words in our head into speech. That might sound a bit sci-fi, but a study in the New England Journal of Medicine shows that it’s possible. A brain-computer interface—known as a neuroprosthesis—was able to help a person with severe dysarthria to speak with a 97.5% accuracy at a rate of 32 words per minute, and to continue this over an eight month follow-up period. It was done by placing 256 electrodes into the left precentral gyrus, which controls the lips, jaw, tongue, and larynx. Signals from the electrodes were processed—with the aid of AI—to work out what the person was trying to say, then turned into speech using software that emulates the sound of the person’s voice.

N Engl J Med doi:10.1056/NEJMoa2314132

Uncertainties in diagnosing pneumonia

So we live in a world where we can make machines that literally speak your mind, and yet finding relevant clinical details in someone’s medical notes can be a needle in a haystack task that sends a whole clinic spiralling out of control. As well as needing a good edit, electronic patient records would be much improved if we could better express uncertainty in them. Diagnosis is often uncertain, and pragmatic treatment decisions are made all the time, as shown by a retrospective cohort study of people hospitalised with pneumonia in the US.

From a database of over two million hospitalisations, 13.3% were diagnosed with pneumonia at either initial assessment or discharge. However, a third (33%) of those diagnosed with pneumonia on discharge weren’t diagnosed on initial assessment, and a third (36%) of those diagnosed initially did not have pneumonia as a diagnosis on discharge. Many patients with pneumonia received treatments for other possible diagnoses—with 10% receiving antibiotics, corticosteroids, and diuretics within 24 hours of admission. The authors conclude that this demonstrates the “need to recognize diagnostic uncertainty and treatment ambiguity in research and practice of pneumonia-related care.”

Ann Intern Med doi:10.7326/M23-2505

Adapting to climate change

Globally, 2023 was the warmest year on record—and the second warmest in Europe. A modelling study in Nature Medicine estimates that there were 47 690 heat related deaths in Europe in 2023 (95% confidence interval 28 853 to 66 525). The study also estimates that the number would be 80% higher if it weren’t for adaptations made since the turn of the century. What these adaptations are, and how they’ve had such a positive effect, isn’t clear, however, as they’re described in the paper only as “a broad range of factors encompassing infrastructure, technological and behavioral adaptation.” As the world heats up, understanding how we can adapt in an environmentally sustainable way will be needed to ensure that heat related deaths don’t rise even further.

Nat Med doi:10.1038/s41591-024-03186-1

Footnotes

• Competing interests: None declared.

• Provenance and peer review: Not commissioned; not peer reviewed.