London

A database is set to help doctors treat HIV. It attempts to identify the best drug combinations to keep a patient's infection in check, on the basis of genetic sequences of drug-resistant strains of the virus.

The international collaboration that is developing the system — the HIV Resistance Response Database Initiative (RDI) — unveiled its latest results on 12 June at the 12th International HIV Drug Resistance Workshop in Los Cabos, Mexico. And although the system isn't the first to deduce information about the drug resistance of particular HIV strains, it is the first to attempt to do so on such a scale, and to be based on clinical results.

“We're aiming to produce tools that can predict treatment response in actual patients,” says Brendan Larder, a virologist who chairs the initiative's core group. The system could eventually be available to all physicians online, enabling them to use a patient's HIV gene sequence to help to predict what drug treatment options would be most successful.

Rapidly mutating HIV strains readily develop resistance to drugs. Up to half of all patients living with HIV must switch drug combinations at least once to control their viral load. Studies have identified about 200 mutations associated with resistance to the 16 drugs commonly given to control the virus.

Doctors already use computer models to establish the optimal combination of two or three drugs that should be given to treat a particular, resistant strain of HIV. But that is an inexact business, given that a patient's virus may harbour as many as 40 drug-resistance mutations. Also, input to the existing computer models usually derives from the ability of different strains to survive in test-tube experiments — which may not reflect the outcomes in patients.

“What's missing is a link to clinical response,” says Joep Lange, director of the National AIDS Therapy Evaluation Center in Amsterdam. “At the moment, a lot of mistakes are being made.” Because the RDI software 'learns' from real patient data, it could surmount these problems, he says.

The latest test of the software showed that the system could correctly predict which drug combinations would bring HIV under control, and by how much, 78% of the time.

This isn't reliable enough for clinical use, Larder admits. But the latest results are from the 750 cases in the original database. And data from about 4,000 patients at AIDS treatment centres in Europe and North America are currently being added. Once that happens, “we hope to achieve 90% accuracy”, says Larder.