Details of computer models that can predict the chances of a patient responding to their HIV drugs with 80% accuracy are published today in the journal AIDS. The models were developed by the HIV Resistance Response Database Initiative (RDI) using almost half a million pieces of data from approximately 6,000 clinical cases from hundreds of clinics around the world. The models are now available online as part of an experimental treatment support tool, HIV-TRePS.
The random forest models were trained to predict the probability of any combination of HIV drugs reducing the virus in the patient’s blood to an undetectably low level (<50 copies/ml). They use the genetic code of the virus, the patient’s immune status, their treatment history and a measure of the level of HIV in the blood, to make their predictions.
“The publication of these results is an important milestone in the development of expert computer systems to aid clinical practice”, commented Professor Julio Montaner, etc. “The models harness the experience of hundreds of physicians treating thousands of patients and puts this distilled expertise in the hands of the individual physician via the click of a mouse.”
Currently, when a patient’s treatment fails and the levels of the virus increase, physicians usually run a genotype test, which detects mutations in the genetic code of the virus that can make it resistant to certain drugs. The physician then selects a combination of drugs that the test indicates will still be effective against the mutated virus. When the results of this test were compared with those of the RDI models they proved significantly less accurate as a predictor of response.