Effective surveillance, containment response, and field evaluation are essential to contain potential pandemic strains

Highly pathogenic avian influenza A (HPAI) subtype H5N1 has caused family case clusters, mostly in Southeast Asia, that could be due to human-to-human transmission. Should this virus, or another zoonotic influenza virus, gain the ability of sustained human-to-human transmission, an influenza pandemic could result. We used statistical methods to test whether observed clusters of HPAI (H5N1) illnesses in families in northern Sumatra, Indonesia, and eastern Turkey were due to human-to-human transmission. Given that human-to-human transmission occurs, we estimate the infection secondary attack rates (SARs) and the local basic reproductive number, _{0}_{0}

Highly pathogenic avian influenza A (HPAI) subtype H5N1 is repeatedly crossing the species barrier to humans. Since December 2003, a total of 291 cases of HPAI (H5N1) have been reported in humans, resulting in 172 deaths (i.e

During late April and early May 2006, a cluster of 8 cases of HPAI (H5N1) was detected and investigated by the Indonesian public health surveillance system in northern Sumatra (

The index patient was a 37-year-old woman, thought to have been exposed to dead poultry and chicken fecal material before onset of illness. She also reportedly maintained a market stall that sold live chickens. Although her illness was not confirmed to have been caused by the (H5N1) avian influenza virus, her death on May 5, 2006, is suspected to be the result of HPAI (H5N1) infection because of her reported symptoms, illness progression, and prior contact with diseased or dead poultry.

Twenty members of her extended family are suspected to have been in contact with her, many during a family gathering on April 29, 2006 (

Of the remaining 11 family members, 4 became ill and died. The 29-year-old sister of the index patient, who lived in an adjacent house, became ill after she provided direct personal care to her ill sister (

From December 18, 2005, (

Before onset of symptoms, 4 children from 1 household, 3 of whom had confirmed cases (including the index patient), were reported to have had close contact with the dead bodies of sick chickens (

We used a previously developed statistical transmission model (

We define _{1} as the probability that an infectious household member infects another household member in 1 day. If the distribution of the infectious period is known, we can obtain the household secondary attack rate (SAR_{1}) from _{1}, defined as the probability that an infectious household member infects another household member over his or her infectious period. Similarly, we define the daily transmission probability (_{2}) and the community SAR (SAR_{2}) for between household spread. Finally, we define the daily probability (

Schematic of estimation method. An infectious person (in red) infects a susceptible person (in green) in the same household with probability of household secondary attack rate (SAR_{1}) and infects a susceptible person in a different household with probability SAR_{2}. The common infectious source (i.e., avian hosts) infects a susceptible person with probability

We establish the likelihood function for each person and then for the whole population for statistical inference. The likelihood function for a person is equivalent to the probability of observing the realized data on that person throughout the outbreak. The likelihood function for a person labeled _{1}, for the same household, or _{2} for exposure in the community, multiplied by the probability of person

The likelihood function for the whole population is the product of all the individual likelihood functions. In the event that human-to-human transmission occurs, SAR estimates are used to estimate the local basic reproductive number (_{0}_{0}

We set up a statistical test with the null hypothesis being that no human-to-human transmission occurs, that is, _{1} = _{2} = 0. The alternative hypothesis is either _{1} or _{2} is not equal to 0, or both are not equal to zero. The test statistic we use is proportional to the ratio of the maximum value of the likelihood function assuming the null hypothesis is true (null likelihood) and the maximum value of the likelihood function at the estimated parameter values (full likelihood).

Specifically, we define the likelihood ratio test statistic as –2 log (the null likelihood function divided by the full likelihood function). If no human-to-human transmission occurs, the 2 likelihood functions would be roughly equal, and we expect to see a likelihood ratio close to 1, and, thus, a likelihood ratio statistic close to 0. A large value of the likelihood ratio statistic is evidence of deviation from the null hypothesis. The question is how to obtain a reference set of the likelihood ratio statistic values that we would see under the null hypothesis. Given no human-to-human transmission, all the observed case-patients must have been infected by the zoonotic source. Since the exposure to the zoonotic source is assumed constant for each person on each day, the null likelihood function will not change if we reassign the infection and symptom status of the observed case-patients to a different group of people in the population. By performing such reassignment many times, we obtained a collection of datasets that were each equally likely to have been observed had there been no human-to-human transmission. The values of the likelihood ratio statistic calculated from these datasets form the null distribution for statistical testing. This method is referred to as a permutation test. The p value is given by the proportion of the reference values that are equal to or larger than the observed likelihood ratio statistic value. More technical details are given in the online appendix.

The probability of infection by the zoonotic source may not be estimable together with SAR_{1} or SAR_{2} from an observed cluster. In such a situation, a statistical test of the occurrence of human-to-human transmission is still meaningful because the likelihood ratio test statistic is still estimable from the permuted datasets.

A list of the inputs that are required for estimation and statistical testing are listed in the _{0}

Category | Parameter/data | Required* |
---|---|---|

Entire outbreak | Outbreak begin date | X |

Outbreak end date | X | |

Latent/incubation period, d† | X | |

Infectious period, d† | X | |

All persons | Neighborhood of residence | X |

Household of residence | X | |

Sex | X | |

Age, y | X | |

Case status (yes or no) | X | |

Case-patients | Whether outbreak index case-patient (yes or no) | X |

Date of illness onset | X | |

Outcome (recovered, died, or don’t know/still ill) | X | |

Date of outcome | X | |

Dates of hospitalization | O | |

Period of receiving treatment (dates) | O | |

Non–case-patients | Dates of hospitalization | O |

Period of prophylactic treatment (dates) | O | |

Inter-residence visits | Identifier for visiting person | X |

Neighborhood visited | X | |

Household visited | X | |

Dates of the visit | X | |

Analysis parameters | End of exposure to the common source of infection (date) | X |

Final day of observation (date) | X | |

R_{0} estimation | Mean no. residents per household | X‡ |

Mean no. community contacts per person/d | X‡ |

*X, required; O, optional; R_{0}, basic reproduction number.
†The user defines the distribution of this period, including the minimum and maximum length of the period.
‡Required to estimate R_{0}.

For the outbreak in Indonesia, _{1}) can be estimated. We determined that human-to-human spread did occur by rejecting the null hypothesis of no human-to-human transmission (p = 0.009). The estimated household SAR is 0.29 (95% confidence interval [CI] 0.15–0.51). Thus, a single infected person in a household infected another household member with the probability of 0.29. The average household size for rural Indonesia is ≈5 people. Because we do not have an estimate of the community SAR, we have an estimate of the lower limit of the local _{0}_{,}_{1} and _{0}

For the outbreak in Turkey, all the parameters are estimable, but we do not reject the null hypothesis of no human-to-human transmission (p = 0.114). Our estimate of the daily probability of infection from the common source is 0.011 (95% CI 0.005–0.025).

We have presented statistical evidence that the strain of HPAI (H5N1) that caused the family cluster of human cases in northern Sumatra was spread from human to human and that the household SAR was 29%. This household SAR is similar to statistical estimates for interpandemic influenza A in the United States (12.7%–30.6%) (_{0}_{0}

We did not consider the role of heterogeneity—such as age, sex, treatment status, or quarantine—in transmission. The parameters could be made to be functions of time-dependent covariates, as we have done with similar models (

Computer simulations have shown that the targeted use of influenza antiviral agents could be effective in containing a potential pandemic strain of influenza at the source (_{0}

Ascertaining whether a potential pandemic strain of influenza is capable of sustained human-to-human transmission and estimating key transmission parameters are important. To estimate more than the household SAR, more detailed community data need to be collected. This would include a complete census of potentially exposed households and persons in the area where immediate transmission could occur from both potential zoonotic and human sources. Such data would enable estimation of important parameters and a more complete estimate of the _{0}

We have developed a software application, TRANSTAT, for implementing these analyses. This application provides a stand-alone environment for the entry, storage, and analysis of data from outbreaks of acute infectious diseases. A partial list of the input information is given in the Table. The statistical methods presented here can be applied to the data along with several standard epidemiologic tools. This information system would allow for real-time analysis and evaluation of control measures for an outbreak. We would encourage outbreak investigators to use this tool, taking care to input data on the exposed nonpatients as well as case-patients. The authors will provide a link to this software upon request.

Detecting Human-to-Human Transmission of Avian Influenza A (H5N1)

Exposure and disease events for each member of the family cluster in northern Sumatra, Indonesia. Dark boxes, duration of illness; white boxes without text, recovery period; thick dark vertical line, death; dark triangles, known contacts between members; shaded triangles, suspected contacts. *Unknown location of residence.

Exposure and disease events for each member of the family cluster in Eastern Turkey. Dark boxes, period of illness; white boxes without text, recovery period; thick dark vertical line, death; *Exposed to corpses of potentially diseased poultry. **Most of the members of houses 2 and 3 attended.

This work was supported by the National Institute of General Medical Sciences MIDAS grant U01-GM070749 and National Institute of Allergy and Infectious Diseases grant R01-AI32042.

Dr Yang is a staff scientist in the Biostatistics and Biomathematics Program in the Division of Public Health Sciences at the Fred Hutchinson Cancer Research Center, Seattle, Washington. His primary research interest is in the statistical and mathematical analysis of infectious disease data and intervention studies.