How much can we trust statistical models of the HIV epidemic?

A Wednesday session at the International AIDS

Conference in Barcelona helped demystify the role that

statisticians play in helping us to understand the HIV

epidemic.

Dr Geoff Garnett of the UNAIDS Epidemiology Group,

explained how six teams of mathematicians and

statisticians brainstormed to create a universal,

mathematical model that was used by UNAIDS in their

latest country-specific estimates of HIV prevalence

and projection of future infections.

Using a relatively simple equation that calculated low

and high risk populations with the crude adult death

rate and the HIV+ survival rate, along with variables

including the start date of HIV for that country and

the demand for risky sex, Garnett and his team claim

to have created the most sophisticated model yet.

Garnett admits, however, that there are still too many

individual variables for these statistics to be taken

at face value. The UNAIDS Global total estimate of

people living with HIV is currently 40 million, but

the low estimate is 30 million and the high estimate

is 50 million: a wide margin of error. He also

concedes that since the numbers cannot represent

actual human behaviour, and the success of

interventions cannot be predicted, “the model relies

on observed patterns of prevalence and can only

predict in the short-term.”

Another presentation by French researcher Bertran

Auvert, showed how statistical models can help

discover which specific factors contribute to the

spread of the epidemic.

By comparing the rate of HIV infection in four very

different cities of sub-Saharan Africa, Auvert and his

team was able to identify that the absence of male

circumcision was the most significant factor in the

spread of HIV, something which has only recently begun

to be recognised in HIV prevention work.