Researchers have developed a device that can diagnose 17 diseases from a single breath.
Geege Schuman stashed this in Medical Breakthroughs
Breakthrough: they have identified a unique “breathprint” for 17 different diseases, such as kidney cancer or Parkinson’s disease, and have designed a device that screens breath samples with 86% accuracy using an array of nanoscale sensors and analyzing the results with artificial intelligence techniques.
Reddit comment is that this is a good start.
So that 86% is very misleading. What the journalist did was take the biggest sounding number and reprint it.
They had an average of 86% accuracy when they created a binary classifier using a machine learning technique called discrimination function analysis. So this means they grabbed a small group of two diseases and tried to predict which of the two that breath signature had. Sometimes they got 100%, sometimes 56%.
For their healthy control group and picking out a specific disease vs no disease, it was as low as thirty percent.
What wasn't clear on first read is if this was cross validation accuracy or the test set accuracy, which would further lower its power.
But 86% sensitivity for a breath test would be phenomenal. You don't need super high specificity to justify a screening test. However, this is GCMS, a
million dollar60-250k machine, which would probably make it rather expensive and impractical. You aren't going up to the drug store and blowing into the cancer breathalyzer any time soon.