The Pharmaceutical Journal
Vol 271 No 7257 p42
12 July 2003

This article
Reprint
Photocopy

News summary

"Artificial intelligence" picks best AIDS therapy

"Artifical intelligence" could be used to find effective treatments for HIV-infected patients whose drug therapy is failing, say researchers.

Dr Brendan Larder, RDI Ltd, described the system at a workshop on HIV drug resistance in Mexico recently. RDI Ltd is a not-for-profit company that is building databases of clinical data relating to HIV drug resistance in practice.

RDI has developed a "neural network" using data from 350 heavily treated patients to predict how well patients will respond to different combinations of drugs. The neural network "learns" as it analyses the relationships between HIV genotype, viral load and the patient's previous response to drugs.

RDI chief executive Andrew Revell told The Journal that data from around 10 per cent of patients were available to test the neural network. In these patients, the model predicted viral load change with 79 per cent accuracy. "These are spectacular results," he said. The model was then used to predict how patients would respond to different combinations of drugs when fed information from 139 failing patients. In every case, it identified an alternative drug combination that it predicted would be effective. On average, the alternatives were predicted to reduce the amount of virus in the patient's bloodstream by over 99 per cent. Based on its accuracy, had the system been used to select treatment, 110 out of 139 patients might have responded to treatment instead of failing. The company is now following up these patients to see what happens in practice.

Mr Revell commented that the tool was not yet developed fully enough to allow it to be in general clinical use. But he hoped that 20,000 patients would eventually be included in the database and that the neural network would be available free to predict the results of alternative HIV drug combinations in failing patients.