The tool uses routine flu prevalence data to predict in advance the dominant strain and extent of an epidemic.
Researchers led by Dr Edward Goldstein from Harvard School of Public Health in Boston, Massachusetts, said the model could allow decision makers to adjust their response during a flu epidemic in real time.
The team examined patient records for cases of influenza-like illness and virological data from 1997 to 2009.
This information was used to estimate the rate of new infection rates for each major flu strain. These included influenza types B and A/H1N1 ('swine flu') and A/H3N2 ('bird flu').
This estimate was then used to create an algorithm, which was found to accurately predict real flu outbreaks.
The early circulation of a particular flu strain is linked with a reduction in the total incidence of other strains, the researchers found.
Significantly, the findings suggest early season flu surveillance data can predict the relative size of epidemics by each flu strain. In some cases, the model could predict the size of the flu outbreak several weeks before its peak.
The authors said: 'Such predictive methods may be useful to decision makers when they are trying to determine in real time which measures to recommend for an influenza season.'