Difficulties in estimating early incidence of influenza infection in a pandemic – study with multiple approaches in Singapore and a review of other studiesArchived

Lee V, Chen M,  Yap J, Ong J, Lim W-Y, Lin R et al  Comparability of different methods for estimating influenza infection rates over a single epidemic wave.  Am. J. Epidemiol. (2011) 174 (4): 468-478. doi: 10.1093/aje/kwr113

This paper and review from a Singapore-based group compared results obtained for estimating rates of new infections during the 2009 pandemic. They derived rates from paired specimens from the same patient (serum cohort approach), cross-sectional serological surveys, rates of unconfirmed syndromic influenza-like-illness (ILI) obtained from primary care physicians in sentinel general practices, and combined clinical repos with laboratory confirmed samples.

Estimates from paired seroconversion were 17% (95% credible intervals 14% to 20%) and from cross-sectional serology  were 12%, (95% credible intervals 9% to 17%). Though the two results do not differ significantly the authors suggests that cross-sectional serological data will usually tend to underestimate true seroconversion rates because of a tendency to over-compensate for pre-existing immunity. Adjusting the data from general practices using the serological information the authors derived estimates of influenza of 15% (95% credible intervals of 10% to 25%) while adjusted estimates from combined clinical presentation and laboratory data were 12% (95% credible intervals 8% to 18%).

A useful comparison of the advantages and disadvantages of the different methods is made in a table  There is another table of fifteen early studies that were undertaken in the pandemic to estimate infection using a variety of serological and non-serological approaches with very large differences in their results. Though with hindsight the studies are not comparable as they represent the results that were the best to hand early on in the 2009 pandemic.

ECDC Comment:  (August 15th 2011)

The difficulty of making early estimates of incidence of influenza in a pandemic (or seasonal influenza) is demonstrated by the review table in this article.(1)  Looking at the problems in more detail it is considered that serological data are optimal, especially serial serological data. Unless there are plans to get the data through early the results from serological studies are usually delayed, especially in a crisis. Developing diagnostic test is usually given priority over serologial approaches.  Also the patients from whom the sera are derived are usually a biased sample in some way or another. Syndromic data or combined clinical and virological information are available much more quickly. However here another group of biases apply, mostly reflecting care seeking behaviours and care availabilities, across age-groups, social groups, settings etc. Finally there will be true geographical and population differences. All these issues may not present such a problem with seasonal influenza since at least the biases in a certain local may be relatively constant. However in a pandemic it is very unlikely that the biases will be the same as during a normal influenza season.  They are likely to change in a pandemic over time as care is sought more and less vigorously by the public and mechanisms of delivery of care can change.(2) Some of the best data on this are still awaited from the European Flu-Watch programme which sadly are only available in abstract form.(3)  

1. Lee V, Chen M,  Yap J, Ong J, Lim W-Y, Lin R et al  Comparability of different methods for estimating influenza infection rates over a single epidemic wave.  Am. J. Epidemiol. (2011) 174 (4): 468-478. doi: 10.1093/aje/kwr113
2. ECDC scientific advance - Interpreting trends in consultations for influenza disease – varying ‘multipliers’, February 2011.
3. Lim M, Bermingham A, Edmunds J, Fragaszy E,  Harvey G, Johnson A et al. Flu Watch. Community burden of Influenza during three influenza seasons and the summer wave of the 2009 H1N1 pandemic in England – implications for interpretation of surveillance data. Poster No P-132 Options Conference, Hong Kong, China, September 3-7th 2010.