Clustering

Helen Branswell of Canadian Press shows once again why she is the best flu reporter on the planet (hat tip crofsblog). Branswell alerts us abut a paper to be published in the journal Emerging Infectious Diseases about clustering of H5N1 infections. The paper presents evidence for at least 15 instances since the disease reappeared in 2003. Human-to-human transmission is strongly suspected in some or all of them, although data are insufficient to show this beyond doubt. But beyond doubt is an unrealistic and probably inappropriate standard.

The authors combed published reports and consulted regional contacts looking for clusters of cases within families that occurred from January 2004 to July 2005. Since their report was submitted another family cluster occurred in Indonesia involving a woman and her young nephew, both of whom tested positive for H5N1 virus.

[snip]

A cluster was considered two or more cases, where at least one member tested positive for the virus and other members of the cluster experienced severe pneumonia or death from respiratory disease.
Olsen said where human-to-human transmission may have taken place, the dates of onset of illness suggest transmission stopping after one generation. In other words, if a person passed the disease on to someone else, the newly infected individual did not appear to have spread it further.

"It's not to say there weren't tertiary cases, but we're not clearly seeing that in these data," she said from Bangkok.

Most clusters comprised two or three cases, though one stretched to five. The researchers could not determine what happened to the members of that cluster, which occurred in Vietnam in early March.
In one grouping of three family members in Vietnam, two nurses who looked after someone with the virus were subsequently hospitalized with severe pneumonia. But infection with H5N1 was confirmed in only one of the health care workers.

Branswell quotes CIDRAP's Michael Osterholm as saying the start of a pandemic will have multiple clusters (ten or twenty) in a given area and they will be larger. While this is plausible, the evolution to pandemicity may not happen all at once or so suddenly. As suggested by one of our readers last February, it may not be the number of clusters but the average size of each cluster that is the signal. Cluster size is related both to basic reproductive number (R₀) and to network topology. Larger clusters are the more highly connected nodes.

The lead author of the EID paper, Sonja Olsen, is acting director of CDC's International Emerging Infections Program in Bangkok. One suspects her scientific contribution is not entirely welcomed by WHO:

A spokesperson for the World Health Organization said the data, while helpful, are retrospective.

"If transmission from birds to humans had become more efficient, then we would be seeing more cases," Maria Cheng said from Geneva.

"But it's certainly very useful to have this information in terms of tracking the disease in the future."

This is not only faint praise, it is incomprehensible. What does it mean for the data to be "retrospective"? They want her to report clusters before they happen? Unfortunately, WHO's stance seems to be just as expressed. They won't believe it until they see cases filling the hospitals. One wonders if they'll "see" it even then.

We should say that this evidence of clustering has been noted persistently by Henry Niman at Recombinomics. WHO never liked it there, either. But credit where credit is due.