Predicting swine flu deaths

In a paper published in the British Medical Journal, researchers have called for improved data to map the spread of swine flu and to make accurate estimates of the number of people likely to die from the virus.

Key points

The researchers say that current estimates of the projected number of deaths may be innaccurate for several reasons:

• Death rates are overestimated because only more severe cases are counted in the total number affected, while mild cases do not appear because they do not present to medical care.
• Death rates are underestimated because deaths are attributed to other seemingly unrelated causes besides swine flu, or because of the delay between symptom onset and death (cases who are counted as living at time of assessment may later die).

The researchers suggest several ways to minimise these biases:

• If information on hospitalisation rates from cases confirmed early in the epidemic is combined with sampling of hospitalised cases later in the epidemic, this can indicate fatality rates among severe cases.
• Adjusting total cases of H1N1 for time delay between symptoms and death/recovery can minimise the underestimation of the fatality rate.
• Studies involving sampling of selected population groups and H1N1 screening are important to gain accurate numbers of those with asymptomatic or mild infection.
• Age-specific analysis to determine whether the trend of a higher rate of infection in young people continues.

Where was the article published?

This research was carried out by Dr Tini Garske and colleagues from the MRC Centre for Outbreak Analysis and Modelling, Department of Infectious Disease Epidemiology, Imperial College London. The study was published in the British Medical Journal, and supported by the Medical Research Council.

What does the research say?

This article discusses the methods used to estimate the proportion of deaths caused by infection from the pandemic (H1N1) 2009 virus, known as the case-fatality ratio. The authors say that early data suggests that the new virus appears to be fairly mild, and the case-fatality ratio is similar to seasonal flu (around 0.5%). However, they say that this ratio seems to vary considerably between countries and, notably, a younger population appears to be affected compared with seasonal flu.

The authors say that the current method of calculating the case-fatality ratio could result in inaccurate estimates. They say that this standard calculation - dividing the number of deaths by the total number of cases – could be inaccurate for a number of reasons:

• The fatality rate is overestimated because people with milder symptoms or no symptoms are not visiting their doctor. Therefore only the most severe cases are reported and taken into account, i.e. there are more actual cases than those confirmed, so the ratio of deaths to cases is less than estimated. (They cite Mexico as a possible example where fatality rates have been overestimated due to an underestimation of the total number of people being infected).
• The present calculation does not take into account the time delay between infection and death, i.e. cases alive at the time of assessment may go on to die, making the death rate higher than estimated.
• That the number of deaths attributable to swine flu are being underestimated because the person has died from an apparently unrelated cause, e.g. cardiovascular death, when in fact this complication may have been precipitated by swine flu.

What do the researchers suggest?

A new way of calculating the case-fatality ratio. They suggest that data from the first few hundred cases confirmed in the UK (when cases were more closely followed) can be used to estimate the early hospitalisation ratio. This can be combined with an estimate of the case-fatality ratio in selected cases that were admitted later during the epidemic.

The researchers point out that it is important to obtain data on the reasons for hospital admission in order to gain an accurate measure of disease severity. Large-scale testing for the virus on a selected population group would also give a better indication of the number of people with clinical symptoms who are actually infected with the virus. They say that such studies need to be set up alongside household studies to assess the extent of asymptomatic infection, so that changing patterns of virulence are detected rapidly.

To counter the bias introduced by the time delay between onset of symptoms and death, the researchers propose either dividing the number of deaths by the total number of cases for whom the outcome was known (both deaths and recoveries), or, more reliably, by adjusting the total number of cases for the delay from symptom onset to death (using information taken from existing data or past epidemics).

What is the implication and importance of this?

This is timely and important research. Accurately estimating the severity of the pandemic (H1N1) 2009 virus is important for planning the most effective healthcare and social measures (such as school closures) to reduce the number of deaths caused by the virus.

The researchers have highlighted areas in which current methods of estimation of case fatality and hospitalisation ratios are likely to involve some inaccuracies. Reliable population level estimates of the prevalence and case-fatality ratio will help to identify populations at risk and to determine which groups are given priority for vaccination when a vaccine becomes available. The proposed methods to get more reliable estimates seem plausible.

At this early stage in the epidemic, many confirmed cases have been in young people, and so it is important to collect age-specific data to determine whether this trend will continue with the spread of the virus. As the researchers say, carefully implemented systems of data collection such as these will be of great value in improving estimates of case-fatality ratio. It will also ensure that any changes in H1N1 virulence are rapidly detected.