Incidence of asymptomatic human influenza A(H5N1) virus infection

BangladeshWhen virologists Fouchier and Kawaoka were isolating avian influenza H5N1 viruses that could transmit among ferrets by aerosol, there was consternation from some quarters that such viruses might escape from the laboratory and cause a pandemic in humans. Part of the fear came from the fact that the case fatality ratio for human infections with the H5N1 virus exceeds 50%. This number could be substantially higher than the lethality ratio, which is the number of symptomatic cases divided by the total number of infections. Divining the latter number has been difficult. Results of a meta-analysis published in 2012 suggest that H5N1 seropositivity approaches 1-2% in certain populations. Others have concluded that these studies are flawed, clouded by false positives and cross-reacting antigens. Recently two additional studies have been published that contribute to this discussion.

The first paper is a case report of subclinical avian influenza H5N1 virus infection in a Vietnamese household in which family members were involved in slaughtering, preparing, and consuming chickens, and birds were permitted to roam freely in the sleeping area. Four chickens from this household were found to be positive for H5N1 virus by polymerase chain reaction (PCR) of throat and cloacal swab specimens. The 40-year old father died after a severe four day respiratory illness requiring hospitalization; H5N1 viral RNA was detected by PCR of a throat swab on day 3 of illness. A throat swab from his daughter, taken 6 days after she had killed a chicken, was positive by PCR, and H5N1 virus was recovered by inoculation of cell cultures. Her hemagglutination-inhibition (HI) titer, a measure of anti-viral antibodies, increased from <20 to 160, but she showed no signs of illness, perhaps because she was treated with oseltamivir from day 5 of her father’s illness.

The authors note the difficulty in detecting subclinical H5N1 infections:

…it is unclear whether serologic testing reliably detects subclinical cases. According to the World Health Organization,MN (microneutralization) titers >80 are indicative of infection but must be confirmed by a second serologic test because of the possibility of cross-reactivity. The interpretation of results from a single serum sample is limited by the specificity or sensitivity of serologic tests, and viral shedding times may mean that infected cases may be missed.

The second study examined seroprevalence of anti-H5N1 virus antibodies in poultry workers in Bangladesh. Sera were collected in 2009 from poultry workers on farms (212 from 87 farms) and live bird markets (210 from 3 markets). Some of the farm workers (91%) reported handling sick animals during laboratory-confirmed H5N1 outbreaks. Sera were screened for antibodies to H5N1 virus by two methods: microneutralization and hemagglutination-inhibition. None of the individuals were seropositive for anti-H5N1 virus antibodies.

I have several reservations about this study. Although H5N1 virus was identified on the poultry farms whose workers were sampled, the sera were drawn from 22 to 543 days after the onset of poultry deaths. If any of the workers had been infected with H5N1 virus, anti-viral antibody titers might have already declined by this time. Although the sera were examined for anti-viral antibodies by two different tests, paired sera were not used, as recommended by the authors of the first paper discussed above.

Therefore the answer to the question ‘what is the fatality rate of influenza H5n1 virus infections in humans?’ still cannot be answered. As the authors of the first paper conclude:

Estimating the incidence of asymptomatic influenza A(H5N1) virus infection in humans exposed to sick poultry or human case-patients requires further careful study using early collection of swab samples and paired acute and convalescent serum samples.

Human infections with influenza H5N1 virus: How many?

The lethality of avian influenza H5N1 infections in humans has been a matter of extensive debate. The >50% case fatality rate established by WHO is high, but the lethality of the virus might be lower if there are many infections accompanied by mild or no disease. One way to answer this question is to determine how many individuals carry antibodies to the virus in populations that are at risk for infection. A number of such studies have been done, and some have concluded that the results imply a low but substantial level of infection (even less than one percent of millions of people is a lot of infections). The conclusion of a new meta-analysis of H5N1 serosurveys is that most of the studies are flawed, and that the frequency of H5 infections appears to be low.

Twenty-nine different H5N1 serological studies were included in this meta-analysis. None of these are particularly satisfactory according to the authors:

None of the 29 serostudies included what we would consider to be optimal, blinded unexposed controls in their published methodologies, i.e., including in the serology runs blinded samples from individuals with essentially no chance of H5N1 infection. Serological assays can easily produce misleading results, especially when paired sera are not available.

Some of the problems identified in the serological surveys include the possibility that many H5N1 positive sera are the result of false positives, that is, cross reaction with antigens from other influenza virus strains. In addition, many studies utilized H5N1 strains that are no longer circulating.

It is clear that most of the H5N1 serosurveys have not been done as well as they should have been. The authors conclude that “it is essential that future serological studies adhere to WHO criteria and include unexposed control groups in their laboratory assays to limit the likelihood of misinterpreting false positive results.”

Let’s not forget that a completely different way of assessing H5N1 infection – by looking for virus-specific T cells – has been reported. The results provide further evidence for subclinical H5N1 infection and are not subject to the caveats noted here for antibody surveys.

I come away from this meta-analysis with an uneasy sense that the authors are not being sufficiently objective, and that they firmly believe that there are no mild or asymptomatic H5N1 infections. One reason is the authors’ use of ‘only’ to describe their findings. For example: “Of studies that used WHO criteria, only [italics mine] 4 found any seropositive results to clades/genotypes of H5N1 that are currently circulating”. The use of ‘only’ in this context implies a judgement, rather than an objective statement of fact. Furthermore, despite the authors stated problems with all H5N1 serosurveys, they nonetheless conclude that there is little evidence for asymptomatic H5N1 infection. If the studies are flawed, how can this conclusion be drawn?

My concern about the authors’ objectivity is further heightened by the fact that they are members of the Center for Biosecurity at the University of Pittsburgh. These are individuals whose job it is to find dangerous viruses that could be used as weapons. On the front page of the website for the Center for Biosecurity is a summary of the meta-analyis article which concludes that “In the article, Assessment of Serosurveys for H5N1, Eric Toner and colleagues discuss their extensive review of past studies and conclude that there is little evidence to suggest that the 60% rate is too high.”

I would argue that if the H5N1 serosurveys are flawed, then do them properly; it is incorrect to simply assume that the H5N1 virus is as lethal as WHO suggests. The World Health Organization should call for and coordinate a study that satisfies criteria established by virologists and epidemiologists for a robust analysis of human H5N1 exposure.

How lethal is rabies virus?

Desmodus rotundusWhen I am asked to name the most lethal human virus, I never hesitate to name rabies virus. Infection with this virus is almost invariably fatal; just three unvaccinated individuals have been known to survive. New evidence from humans in the Peruvian Amazon suggests that the virus might be less lethal than previously believed.

Rabies virus is typically transmitted to humans by the bite of an infected mammal, often a carnivore or a bat. Recently there have been numerous outbreaks of rabies in Peru that have been linked to bites of vampire bats. A study of two communities at risk for vampire bat bites was undertaken to determine whether subclinical infection with rabies virus might occur. Over half of 92 individuals interviewed reported having been bitten by bats. Neutralizing antibodies against rabies virus were detected in 7 of 63 serum samples obtained from this population. Antibodies against the viral nucleoprotein were found in three individuals, two of whom were also positive for viral neutralizing antibodies. All 9 seropositive individuals indicated that they had previously had contact with a bat (a bite, scratch, or direct contact with unprotected skin). One of these individuals had previously received rabies vaccine.

The finding of neutralizing antibodies against rabies virus suggests that these individuals were likely infected, but did not develop fatal disease. It is also possible that they received a sufficiently large dose of virus to induce antibodies, but that viral replication did not occur. Another explanation for the findings is that these individuals were infected with an unknown virus that is highly related to rabies virus, but which is not pathogenic for humans.

There have been numerous seroprevalence studies of rabies infection in wildlife. For example, foxes and other canids have low (0-5%) seroprevalence rates, while 5-50% of bats can harbor rabies neutralizing antibodies, indicating that these animals are less susceptible to fatal rabies. In contrast, there have been few studies on rabies seroprevalence in humans. In one study of 30 raccoon hunters in Florida, low levels of rabies virus neutralizing antibodies were found in 2 samples. Low neutralizing antibody titers were also detected in 9 of 31 Canadian Inuit hunters; in a separate study, high rabies antibody titers were detected in the serum of 1 of 26 Alaskan fox trappers. All of these individuals had not been immunized with rabies virus vaccine.

Rabies virus causes 55,000 human deaths each year, so even if the results of the Peruvian study indicate subclinical infection, they would have little impact on the nearly 100% fatality rate associated with infection. More extensive studies are needed to determine if nonfatal human rabies infection is more common than believed. Understanding why some individuals do not die after infection might reveal immunological and genetic factors that protect against the disease.

Amy T. Gilbert, Brett W. Petersen, Sergio Recuenco, Michael Niezgoda, Jorge Gómez, V. Alberto Laguna-Torres and Charles Rupprecht. Evidence of Rabies Virus Exposure among Humans in the Peruvian Amazon. Am. J. Trop. Med. Hyg. 87:206 (2012).


How lethal is ebolavirus?

Should we fear avian H5N1 influenza?

N.Y. Times: H5N1 ferret research should not have been done

ferretThe prominent lead editorial in the New York Times of Sunday, 8 January 2012 is entitled ‘An Engineered Doomsday’. It concerns recent avian influenza H5N1 research, in which scientists in the Netherlands and at the University of Wisconsin found that by passaging the virus in ferrets it could acquire aerosol transmissibility. Let’s determine if the scientific facts warrant the frightening title.

The editorial begins by rebuking the scientists who carried out the experiments on H5N1:

…the research should never have been undertaken because the potential harm is so catastrophic and the potential benefits from studying the virus so speculative.…they created a virus that could kill tens or hundreds of millions of people if it escaped confinement or was stolen by terrorists. …the new virus…ought to be destroyed.

The intent of the experiments was not to create a doomsday virus, but to answer questions about why the H5N1 virus transmits well among birds but not humans. This experiment cannot of course be done in humans, so it was carried out in ferrets, a model for influenza. The results show that aerosol transmissibility in ferrets can be achieved with just five amino acid changes, with no reduction in the virulence of the virus. That result does not mean that the same amino acid changes would have the same effect in humans – it just tells us that achieving aerosol transmissibility in an animal model is relatively easy.

Whether tens or hundreds of millions of people would be killed depends on the ability of a virus to not only transmit among humans, but to retain virulence. There is no evidence that the ferret-passaged H5N1 virus has these properties. In the unlikely event that the virus somehow escaped and began to infect people, its spread could be controlled by vaccines (candidates are under development) and antivirals (existing neuraminidase inhibitors are active against influenza H5N1).

The heart of the H5N1 controversy is encapsulated by the next passage:

Thus far the virus has infected close to 600 humans and killed more than half of them, a fatality rate that far exceeds the 2 percent rate in the 1918 influenza pandemic that killed as many as 100 million people.

This statement refers to the fact that nearly 60% of the 573 WHO-confirmed H5N1 cases have died. This death rate appears staggering until one considers how it is calculated. The WHO case definition for H5N1 influenza states that an individual must have a febrile respiratory illness, known exposure to H5N1 virus in the previous 7 days, and confirmation of infection by virus culture, polymerase chain reaction, or tests for antibodies. These conditions are highly unlikely to be fulfilled in rural populations where most H5N1 infections probably occur. The case fatality ratio can only be calculated by dividing the number of deaths by the total number of infections – and we do not know the latter number. Of ten large studies that have tested for H5N1 antibodies in rural populations, two were negative and 5 reported the presence of H5 antibodies in 0.2 – 5.6% of indiviudals. Much more work needs to be done to determine the actual fatality rate of influenza H5N1, but the WHO estimate is orders of magnitude too high.

Next the Editors weigh in on publication of the ferret results:

The Erasmus team believes that more than 100 laboratories and perhaps 1,000 scientists around the world need to know the precise mutations to look for. That would spread the information far too widely. It should suffice to have a few of the most sophisticated laboratories do the analyses.

As I have argued before, limiting the dissemination of scientific information only serves to impede progress. It is impossible to predict which laboratory is going to do the breakthrough experiment, and picking ‘sophisticated’ laboratories is meaningless.

Then the Editors argue that the research has no value:

Defenders of the research in Rotterdam claim it will provide two major benefits for protecting global health. But it is highly uncertain, even improbable, that the virus would mutate in nature along the pathways prodded in a laboratory environment, so the benefit of looking for these five mutations seems marginal.

I would like to see the studies on which this statement is based. It is well known in virology that mutations selected in laboratory experiments can be been identified in nature. For example, the mutations identified in the H5N1 influenza viruses that transmit among ferrets have indeed been observed naturally in animals. The fact is that no particular viral mutation is improbable, given the enormity of  viral diversity in nature.

The Editors write that the results of the H5N1 studies do not help determine if existing antiviral drugs and vaccines would be effective:

But genetic changes that affect transmissibility do not necessarily change the properties that make a virus susceptible to drugs or to the antibodies produced by a vaccine, so that approach may not yield much useful new information.

Two of the currently used drugs for controlling influenza, Tamiflu and Relenza, act by inhibiting the viral neuraminidase enzyme. Its function is to allow viruses to spread from cell to cell, and could very likely be involved in ferret transmission. If changes in this protein lead to aerosol transmission among ferrets, they could alter sensitivity to the drugs. Other changes in the H5N1 virus might alter its protein profile, making it less sensitive to currently proposed vaccines. These are only two of many reasons why studying this H5N1 virus would yield a great deal of useful information. As noted in A flu virus risk worth taking by Anthony Fauci, Gary Nabel, and Francis Collins in the Washington Post, “…new data provide valuable insights that can inform influenza preparedness and help delineate the principles of virus transmission between species.”

I think it is a good idea to have a public dialogue to understand the goals of influenza H5N1 research. But the discussion should be based on scientific fact, not doomsday scenarios. The Times does everyone a disservice by basing their opinion on science fiction.

How many people die from influenza?

WHO reports that as of 15 June 2009, 76 countries have officially reported 35, 928 cases of influenza A(H1N1) infection, including 163 deaths. These numbers can be used to calculate a case fatality ratio (CFR) of 0.45%. Is this number an accurate indication of the lethality of influenza?

Determining how many people die from influenza is a tricky business. The main problem is that not every influenza virus infection is confirmed by laboratory testing. For example, early in the Mexico H1N1 outbreak, the apparent CFR was much higher because the total number of infections had not been established. Even with the intense surveillance being conducted at the onset of this pandemic, many infections are still not diagnosed. Virologic surveillance is likely become even more incomplete as health systems become overburdened:

The size of Victoria’s outbreak is now so great that only those most at risk – the elderly, pregnant women and those with other underlying medical conditions – are being tested, resulting in 199 new cases last week. “At the moment cases confirmed in the laboratory signify only a small fraction of the cases,” Dr Lester said. “It could be three or four times the laboratory confirmed number, but it’s very hard to estimate, given the mild nature of the virus. It is not anywhere near the one in three some have suggested.

So how do we determine how many people are killed by influenza virus?

In fact, the Centers for Disease Control and Prevention of the US does not know exactly how many people die from flu each year. The number has to be estimated using statistical procedures.

There are several reasons why influenza mortality in the US is estimated. States are not required to report to the CDC individual influenza cases, or deaths of people older than the age of 18. Influenza is rarely listed as a cause of death on death certificates, even when people die from influenza-related complications. Many flu-related deaths occur one or two weeks after the initial infection, when influenza can no longer be detected from respiratory samples. Most people who die from influenza-related complications are not given diagnostic tests to detect influenza.

To determine the level of influenza-related mortality, each week, from October to mid-May, the vital statistics offices of 122 cities report the number of death certificates which list pneumonia or influenza as the underlying or cause of death. The percentage of deaths due to pneumonia and influenza are compared with a seasonal baseline and epidemic threshold value determined each week. The seasonal baseline is calculated using statistical procedures using data from the previous five years, and the epidemic threshold is calculated as 1.645 standard deviations above the seasonal baseline. This is the point at which the observed proportion of deaths attributed to pneumonia or influenza becomes significantly higher than would be expected without substantial influenza-related mortality.

For the 2007–08 influenza season, the percentage of deaths attributed to pneumonia and influenza exceeded the epidemic threshold for 8 consecutive weeks from January 12–May 17, 2008, with a peak at 9.1% at week 11, as shown below. In contrast, pneumonia and influenza deaths remained below the epidemic threshold in the relatively mild 2008-2009 season:


This method clearly is not perfect. The rationale is that the ‘excess mortality’ (over the epidemic theshold) is likely to be caused by influenza, but so could at least some of the deaths between the baseline and excess threshold. For example, the pneumonia and influenza deaths are below the epidemic threshold this season, yet we know that people have died from influenza. It also misses deaths caused by influenza, but for one reason or another influenza or pneumonia were not entered on the death certificate.

The answer to this dilemma is more statistics – methods that use the CDC data to estimate the number of deaths caused by influenza. In the paper cited below, the authors calculated an average of 41,400 deaths each year , for the years 1979 – 2001, in the US due to influenza. Remember that this is an average, and the actual numbers may vary substantially each year.

To answer the question posed at the beginning of this post: except in well-contained outbreaks in which the number of infected individuals can be determined with precision, the case-fatality ratio is bound to be inaccurate. The use of serological assays to determine the extent of infection, coupled with statistical estimates of influenza mortality, are likely to provide more reliable data.

Dushoff, J. (2005). Mortality due to Influenza in the United States–An Annualized Regression Approach Using Multiple-Cause Mortality Data American Journal of Epidemiology, 163 (2), 181-187 DOI: 10.1093/aje/kwj024