A rapid and accurate estimate of the prevalence and severity of influenza in an emerging pandemic can be made by using a population-based telephone survey together with information on the background rates of influenza-like illness (ILI), say researchers from the New York City Department of Health and Mental Hygiene this month in PLoS ONE.¹

In the early stages of a pandemic, health authorities need to know how severe the illness caused by the virus is in order to determine how best to deal with it. One important measure of this is the case-fatality rate (CFR), often defined as the number of deaths per 1000 people infected. But it is difficult to estimate accurately without good information on the number of people infected. In the early stages of the 2009 H1N1 pandemic, estimates for the CFR varied by as much as 150-fold.

“Determining severity of illness as soon as possible after a potentially pandemic influenza strain is recognized as one of the highest priorities for public health authorities,” write James Hadler and colleagues. “We believe that it is possible to get such estimates during the peak of a first wave if not sooner using population-based survey methods.”

Official estimates of influenza prevalence in the USA and UK during the early stages of the 2009 H1N1 pandemic were based on the number of reports of infection obtained from healthcare providers and laboratories. However, this only captures the number of people who seek medical attention, and many people who have flu do not seek treatment, the authors say. In addition, collecting data in this way is time consuming.

To overcome these limitations, the researchers conducted two telephone surveys of around 1000 randomly selected households in New York City in May and June 2009. With each survey they collected data on self-reported influenza-like symptoms during the previous 30 days and demographic information for each member of the household. The data were then weighted by age, sex, household size, and ethnicity, to estimate the prevalence of symptoms in the city population as a whole.

The data were initially released to the public and shared with other researchers to help model the spread of the virus. However, to make an accurate estimate of the level of ILI attributable to the pandemic H1N1 virus, rather than to other respiratory viruses, the researchers needed to determine the background rate of ILI that would be expected in the absence of the pandemic virus.

To do this, the authors used a combination of historical data on ILI prevalence from Emergency Department surveillance systems, and data on self-reported ILI symptoms from areas of the city that had not reported a significant number of pandemic flu cases at the time the survey was carried out.

These methods suggested that the prevalence of infections with the H1N1 virus was between 7% and 12%. This allowed the authors to estimate a case-fatality rate between 0.05 and 0.09 per 1000 — just lower than that of seasonal flu, and lower than that suggested by health authorities in the US and in the UK in the early stages of the pandemic.

“This survey is important,” says John Edmunds, of the London School of Hygiene and Tropical Medicine, UK. “How well you measure the case-fatality ratio depends on how well you count your cases, and a telephone survey captures far more case s [than traditional surveillance].”

“All our surveillance systems [in the UK] give a very partial view,” says Edmunds. “It’s essential to measure cases at the population level.”

In the UK, flu surveillance is mainly based on reports from doctors. These reports are biased, Edmunds says, because they only measure the number of people seeking treatment — and the data became more biased over the course of the 2009 pandemic because people’s health-seeking behaviour changed.

To try to obtain a better estimate of the number of cases in the UK, Edmunds helped set up FluSurvey — a population-based survey, which used a weekly online questionnaire, rather than a telephone survey, to gather data.

This online survey allows much more data to be collected on each individual compared with a telephone survey, as it is faster to fill in a questionnaire than to answer questions over the telephone, Edmunds points out. It also allows researchers to track changes over time, rather than just taking a snapshot of influenza prevalence. However, people tended to respond to the survey only during the weeks when they were ill, potentially skewing the data by undercounting the number of people with no symptoms, he says. “We thought we could get a good measure of incidence, but I don’t think we did.”

Both telephone and internet-based methods also have the advantage of being much more flexible than surveillance based on reports by healthcare providers, Edmunds says. If the illnesses caused by the next pandemic don’t have classical symptoms, such as fever, then doctors will miss it. But it is easy to adapt a questionnaire as necessary.

The best way to establish the number of people that have been infected, though, is serological testing, where blood samples are tested for antibodies against the virus to confirm infection, Edmunds says. This is expensive and time-consuming to carry out — in New York City, Hadler et al. point out, there are still no serological data available to validate their estimate of pandemic flu prevalence.

Edmunds argues that serological testing should have been carried out in Mexico in the early stages of the pandemic. One thousand blood samples, taken in an area where the pandemic has finished and no further cases are being imported, could have established the proportion of people infected by the virus, he says. Instead, house-to-house surveys conducted in La Gloria, the town where the pandemic is thought to have originated, put the proportion at 60% in April last year, when the real figure was more like 10%, according to Edmunds. This explains why the early estimates for the upper bound of the number of deaths from the pandemic (65,000 in the UK) were so high, he argues.

However, Edmunds believes that case-fatality rate estimates are ultimately not very useful, in part because they come with high uncertainty. “What you can look at, and what you should look at and people didn’t, is the age distribution of deaths,” he says. “What was really worrying about the 1918 [flu] pandemic was that it was people outside risk groups dying, younger people. From the very first days in Mexico the age distribution of deaths looked normal [comparable to seasonal flu].”