This review looked at data from over 100,000 patients to see if computer programs could guess who might stop working well on tuberculosis medicine. The results suggest these tools can work, but they are not ready for doctors to use right now in regular care.
Researchers found that the computer models were fairly good at spotting risk, scoring about 0.84 out of 1.0 overall. However, performance dropped when looking at people with HIV, and results varied widely between different studies. Only a small number of studies checked if the models worked in new groups of people.
There are important gaps in the data. Many studies did not include children, people with HIV, or those from countries with high tuberculosis rates. Most studies also had high risks of bias, meaning the results might not be fully reliable.
While the technology shows potential, patients should not rely on these models yet. More research is needed to make sure they work safely and fairly for everyone before they become part of standard treatment plans.