DRAFT: NOT FOR CITATION

Prepared for

The 20th Year Anniversary and 11th Biannual Public Management Research Association Conference, Syracuse, NY, June 2-4, 2011

Searching for Sweet Spots of Communication during an Emergency Situation:

Pandemic Influenza Outbreaks and Public Risk Communication

Wei Zhong Yushim Kim

Contact: School of Public Affairs Arizona State University Phone: (602) 496-1157

Fax: (602) 496-0950 wzhong1@asu.edu

ykim@asu.edu

This work was supported by the Arizona Department of Health Services (ADHS) through a Health and Human Services (HHS) preparedness grant, Arizona State University College of Public Program’s research seed grant, and the National Research Foundation of Korea Grant funded by the Korean Government (NRF-2010-330-B00262). We thank Megan Jehn and Tim Lant who helped conduct this research.

  1  

Abstract

This paper explores opportunities of public risk communication during an emergency situation such as a novel pandemic influenza outbreak. Using the 2009 H1N1/A influenza outbreak in Arizona as a research context, we build a systems dynamics (SD) model of the influenza virus spread and public risk communication. The SD model is used to conduct an experiment on the effects of the information transmission channel, coverage, and perceived credibility in mitigating pandemic impacts on the community. From a public survey conducted during October 2009 in Arizona, we find that local television and national newspapers are important communication channels that influence the public’s perceptions of H1N1/A influenza risk. Incorporating the role of key channels, risk perception, and avoidance behavior by the public, the simulation outputs demonstrate that, with the current level of local television coverage (95%) in Arizona, increasing the number of households who get pandemic information from national newspapers (4%) does not alleviate pandemic impacts. However, pandemic impacts to the community can be mitigated through increasing the perceived credibility of national newspapers and/or the proportion of the population following pandemic news and information closely. Therefore, efforts to communicate risk information to the public in pandemic situations in Arizona need to focus on increasing the credibility of national newspapers or the percent of the population who follows pandemic news closely rather than the coverage of national newspapers alone.

Keywords: Communication, Pandemic Emergency, Systems Dynamics

  2  

INTRODUCTION Natural and social emergencies and disasters have become more imminent over the past decade, and are creating unfamiliar consequences. For example, the tsunami that hit Japan in 2011 was a surprise largely because it hit so many seemingly unrelated parts of society, the costs of which are not easily calculable. During such disastrous events, it is critical to effectively communicate with the public in order to minimize potential damages to society. However, the public’s response to emergency risk communication has been “an Achilles heel to organized attempts at securing public safety in the face of disasters” (Donner, 2006, p. 1). There is a substantial body of literature on the importance of risk communication regarding disasters and hazards, but there is little research on how to best structure public risk communication during an emergency situation. In this paper we explored how an emergency situation can be mitigated through the effective use of information transmission channels for public risk communication. We first reviewed the literature to provide background information on public risk communication, including its key assumptions and challenges. We then built a computational model of a pandemic influenza spread, using it to simulate the second wave of the 2009 H1N1/A influenza outbreak in Arizona. This model incorporates the specific local conditions during that time period, including the communication strategy employed, the public’s perception of risk and information transmission channels, and the public’s protective behavior. Influenza-related morbidity was examined as a simulation output. We further conducted experiments to explore communication strategies that influence the dynamics of such a pandemic situation in the community. The paper ends with a discussion on improving the effectiveness of risk communication during an emergency.

PUBLIC RISK COMMUNICATION: ASSUMPTIONS AND CHALLENGES

Public risk communication is defined as the transmission of messages to individuals, groups, or populations to provide them with information about the existence of danger in addition to what can be done to prevent, avoid, or minimize the danger (Williams, 1964). It can be conceived as a general social system with the ultimate goal of reducing emergency impacts by initiating and motivating appropriate protective responses among those in danger. Public managers must design and issue risk messages in an appropriate way to facilitate timely protective behavior. Social scientists believe that the principal aspect of risk communication lies in its social and human component, particularly the responsive process of individuals (Balluz et al., 2000; Donner, Rodriguez, & Diaz, 2007; Gladwin et al., 2007). Over the past several decades, a significant number of studies have found that individuals respond to emergency risk information through a complex social process (Dynes & Quarantelli, 1973; Maxwell, 2003; Mileti & O’Brien, 1992; Quarantelli, 1983; Sorensen, 1993, 2000; Trainor & McNeil, 2008). The central component of this process is called risk perception, which refers to whether individuals believe they are personally endangered or not. In other words, risk perception is a crucial component of individuals’ spontaneous protective behavior, although it is not the only important component. Therefore, the fundamental assumption of efforts in social sciences regarding emergency risk

  3  

communication has been that “the actor acts toward his world on the basis of how he sees it and not on the basis of how that world appears to the outside observer” (Drabek & Boggs, 1968, p. 445). Risk perception plays a significant role in mediating the relationship between risk information and people’s protective behavior during an emergency situation. During decades of research, useful knowledge has been accumulated regarding risk information and perceptions. When emergency managers aim to influence individuals’ protective behaviors by influencing risk perceptions, they must consider the following key attributes of risk information: (1) the information source and frequency, (2) the message content, (3) the message style, and (4) the transmission channel. Individuals are more likely to personally perceive the danger when the information source is official and familiar (Donner, 2006; Mileti & Fitzpatrick, 1992; Perry, 1979, 1981) and when the risk message is repeated in a predicted way (Mileti & Sorensen, 1990). It is also important to consider whether the risk message encompasses answers to the key questions of risk information, such as what the risk and its characteristics are, what geographical area or location is threatened, what people can do to protect themselves, when the risk occurs and how much time is left before the impact, who the warning is issued from, and whether the information includes graphical information in addition to verbal messages (Donner, 2007; Mileti, 1995; Mileti & Darlington, 1997; Parker, Priest, & Tapsell, 2009; Pfister, 2002). The message style is also critical, including the consistency, continuity, certainty, urgency, sufficiency, specificity, clarity, and accuracy. When the style components related to answers to key questions included in the risk information are enhanced, people are more likely to perceive the risk of the danger and respond to the warning (Donner, 2007; Drabek, 1986; McLuckie, 1970; Mileti, 1995; Parker et al., 2009; Perry, Lindell, & Greene, 1981; Quarantelli, 1983; Reynolds, 2005). One underexplored dimension of public risk communication is the role of the transmission channel in risk perceptions and its impact within the emergency situation. A transmission channel is the medium through which a warning message is conveyed from its source to the target recipients. The same message delivered through different transmission channels can be perceived by individuals as having different degrees of credibility, therefore having varied influences on risk perceptions (Quarantelli, 1983). While researchers have found that information delivered through personal channels (e.g., fact-to-fact conversation) is more influential on risk perception (Drabek, 1969; Mileti & Sorensen, 1990; Quarantelli, 1983; Sorensen, 1991, 1992), the influences of different impersonal channels have not been sufficiently investigated. Most people in an emergency receive risk messages through impersonal transmission channels such as television, newspapers, or social media, which are convenient tools that can be easily utilized by public managers. A systematic knowledge of the influence of transmission channels on individual risk perception and the impacts of an emergency can help public managers develop effective risk communication strategies. Research Context To address this challenge in public risk communication, we focus on the 2009 H1N1/A influenza outbreak as the research context. H1N1/A influenza emerged as a new pandemic strain of influenza in April 2009. As the first global pandemic in over 40 years, it caused a substantial number of illnesses, hospitalizations, and deaths (CDC, 2010a). The United States experienced

  4  

its first wave of 2009 H1N1/A influenza in the spring and summer months of 2009. A public health emergency was declared by the U.S. government on April 26, and by June 19, all states in the U.S. had reported cases of H1N1/A infection. The second wave occurred in the fall of 2009, with most parts of the country experiencing the influenza outbreak from October to early December 2009 (Ross et al., 2010). In Arizona, the first case of H1N1/A influenza was confirmed on April 29, 2009 (Shanks, 2009), and the Arizona Department of Health Service (ADHS) has been reporting the number of new infected and deceased cases each week since August 30, 2009. The 2009-2010 influenza season started on October 4, 2009 and continued through October 2, 2010 (ADHS, 2009a). By early October 2009, a total of 2,243 people had been infected and 30 people had died from the disease in Arizona (ADHS, 2009b). Newly infected cases continued emerging until May 2010. By the end of the 2009-2010 influenza season, 5,620 people had been infected with the virus and the total number of deceased cases was 122. Figure1 shows how the number of newly infected cases in Arizona changed each week during the 2009-2010 influenza season (solid curve).

Figure 1 Epidemiological curve for 2009 H1N1/A influenza newly infected cases in Arizona

0  

200  

400  

600  

800  

1,000  

1,200  

1,400  

1,600  

Num

ber  o

f  new

 infected

 cases  

Week    

  5  

MODELING PUBLIC RISK COMMUNICATION DURING A PANDEMIC INFLUENZA OUTBREAK

Due to the complexity of the outbreak and spread of pandemic influenza, computer simulation has been frequently utilized (Freimuth et al., 2008).We built a system dynamics model for two purposes: to model the dynamics of the flu epidemic and to test the hypothetical communication strategies with a focus on information transmission channels. In the following section we explain the structure of the simulation model. Modeling Flu Dynamics with the Classical SEIR Model The simulation model presented in Figure 2 utilizes the standard Susceptible-Exposed-Infected-Removed (SEIR) model in epidemic simulation (Li et al., 1999; Rost & Wu, 2008).

Figure 2 Schematic of a standard SEIR model In this model, there are four compartments representing four health statuses relative to an epidemic: SEIR. People can transit from one state to the next, and the transition rates are specified as follows:

!"!" = −!"#

! + !!

!"!" = !"#

! + !! − !"

!"!" = !" − ! + ! !

!"#!" = !"

!"!" = !"

- +

Latent Period Contact Rate Infection Rate Total Population

- + + + +

Susceptible (S)

Exposed (E)

+

+

Recovered (Re)

+

Infection Period

- Infected

(I)

Died (D)

Mortality Rate

+

  6  

The susceptible compartment (S) includes people who are susceptible to the disease. The infection rate is represented by α and is dependent upon the characteristics of the disease. The β is the normal contact (mixing) rate, which represents the average number of people each person contacts per day in a non-epidemic condition. Once infected, the susceptible transit into the latent status, and are then called the exposed (E). They are infectious, but do not show any disease symptoms. The

€

σ represents the progression rate from E to I. After the latent period (

€

σ−1) ends, the exposed become infected people (I) who are both infectious and symptomatic. The removed (R) compartment consists of people who either recover (Re) from the disease after the infection period (

€

γ −1) or die (D) when they are in the infected status. The γ is the recovery rate, and µ is the mortality rate among infected people. The recovered individuals are usually assumed to acquire full immunity to subsequent infection so that these individuals never re-enter the susceptible population (Larson & Nigmatulina, 2009; Yoo, Kasajima, & Bhattacharya, 2010). N is the total number of people in the system, excluding those who died. This standard SEIR model has become a prominent epidemiological model, but is also criticized in that it is too simple to provide insightful information for public managers (Larson & Nigmatulina, 2009; Epstein et al., 2008). It ignores human perceptional and behavioral responses to potential threats. Individuals are assumed to continue their regular activities as usual during an epidemic, while empirical studies have reported the opposite to occur in reality, especially in a pandemic situation (Ekberg et al., 2009; Lau et al., 2003; Lau et al., 2007). When informed of the threat of pandemic influenza, people change their perceptions and forego social contacts (Lau et al., 2003; Lau et al., 2007). This type of protective measure, where individuals reduce their overall social activities for self-protection, is also called avoidance behavior (Yoo et al., 2010). Previous researchers have reported that, once adopted, the avoidance behavior will continue until the pandemic ends (Leung et al., 2003; de Zwart et al., 2010). We incorporate this avoidance behavior and its links from risk perceptions into our modified SEIR model below. Incorporating Public Risk Communication to the SEIR Model The goal of the modified model is to incorporate public risk communication into the classical SEIR model. The first assumption is that the pandemic influenza epidemic is influenced by the public’s avoidance behavior, as discussed above. Thus, the population is divided into two subpopulations depending upon their avoidance behavior. Second, the public’s avoidance behavior is specified as a function of their risk perception regarding the particular disease and dynamic disease prevalence. Finally, public risk communication influences how people perceive the risk of getting such a disease during the epidemic. Figure 3 shows the schematic of the modified SEIR model. The total population (N) is divided into two subpopulations: the population of individuals who engage in avoidance behaviors (Na) and the population of individuals who do not take such measures (Nna). The standard four-compartment model is then modified into an eight-compartment model, including the susceptible who engage (Sa) and do not engage (Sna) in avoidance behavior, the exposed who engage (Ea) and do not engage (Ena) in avoidance behavior, the infected who engage (Ia) and do not engage (Ina) in avoidance behavior, the recovered (Re), and those who died (D).

7  

 

 

Figure 3 Schematic of the modified SEIR model

Ena Ina Na + Nna

Ea + Ena Ia + Ina

Re

D

Removed (Re)

Died (D)

% of people taking avoidance behavior

Ia Ea

Latent Period Contact Rate Infection Rate

Total Population

Avoidance Behavior Effect

D

Ea + Ena

Re

Na + Nna

Ia + Ina

Infection Period

Mortality Rate

Exposed (Ea)

Infected (Ia)

Susceptible (Sa)

Infected (Ina)

Susceptible (Sna)

Exposed (Ena)

  8  

Since the key to divide the population in the model is the proportion of the susceptible population engaging in avoidance behavior during each time unit, Figure 4 presents how that proportion is determined in the current model. While not every high-risk perceiver takes avoidance actions, the proportion engaging in avoidance behaviors is primarily determined by the proportion of the susceptible population perceiving a high risk of disease in the model. These people do not change their avoidance behavior despite changes in their health status throughout the simulation, as previously shown in the literature (Lau et al., 2007; Leung et al., 2003).

Figure 4 Determining the subpopulations in the modified SEIR model In this modified model, public risk perception is largely formed by risk communication components that focus on the characteristics of the information channel. We considered the types of information transmission channels people use to get information about pandemic flu (channel type), how many people use each channel for risk information (channel users), and whether people believe that the channel they use is important in providing updates on the pandemic flu (perceived credibility). Disease prevalence, which indicates the severity of the disease spread condition in the system, is another important component in the model. The prevalence influences the number of disease news followers and is also influenced by the percent of people engaging in avoidance behaviors. Adopting Larson and Nigmatulina’s (2009) measurement, disease prevalence at time t can be captured as

!"#$%#$  !"#$%&#'(#   ! =!!"#$%& 0!!"#$%& !

Nactive represents the population at a single time point, excluding those who have been infected or died. It is assumed that more people will follow risk information as the number of people who are infected or die increases (Aakko, 2004). When the disease becomes more severe at time t, more people will follow the risk information, perceive a high risk of the disease, and engage in avoidance behaviors at time (t + 1). This change then reduces the size of the infectious and died

Public Risk Communication

Risk Perception

Avoidance Behavior - Channel Types

- % Channel Users

- % Perceived credibility on

each channel

% Covered % Followers

Disease Prevalence

 

Ea Ia Ra

Ena Ina Rna

Sa

Sna

   

  9  

population, which further reduces disease prevalence at time (t + 1), and again the percentage of people who engage in avoidance behaviors among the susceptible population at time (t + 2). Incorporating risk communication dynamics in avoidance behavior influences a key parameter in the SEIR model: the contact rate (β). Avoidance behavior reduces the contact rate among people engaging in it, and the avoidance behavior effect (φ) is set to represent the reduction in the contact rate due to such behavior. Previous researchers estimate that the effective contact rate, which is the product of the infection rate and the contact rate, could be reduced by 30-90% through the early implementation of non-pharmaceutical measures against a pandemic influenza outbreak, such as social distancing (Jefferson et al., 2008; Larson & Nigmatulina, 2009). The infection rate could be considered a constant representing the biological features of the disease, so we can infer that avoidance behavior could reduce people’s contact rate by 30-90%. Here we set that people who engage in avoidance behavior reduce their contact rate by 60% on average. Avoidance behavior does not influence the values of other parameters. The latent and infectious periods remain the same regardless of whether exposed or infected people engage in avoidance behavior. New infectious cases occur when a susceptible individual is infected by an infectious individual. Given the different contact rates resulting from individuals’ choices regarding avoidance behavior, four influenza transmission mechanisms can be structured. A matrix β is used to present the virus transmission mechanisms between those engaging in avoidance behavior and those not engaging in such behavior:

! =!!,! !!,!"!!",! !!",!"

= ∅!!! ∅!!∅!! !!

βna,na represents influenza virus transmission to susceptible individuals who are not engaging in avoidance behavior from those who are infectious and are not engaging in avoidance behavior. The contact rate in a normal situation (β0) is used to structure the interaction between the two. βa,na and βna,a capture, respectively, the interaction between susceptible people engaging in avoidance behavior and infectious people not engaging in such a behavior, and the interaction between susceptible people not engaging in avoidance behavior and infectious people engaging in such a behavior. Assuming that only one side takes avoidance actions, the contact rates are of the same value and assigned as φβ0. Finally, βa,a represents the virus transmission to the susceptible who are engaging in avoidance behaviors from the infectious who are also engaging in avoidance behaviors. It has the smallest magnitude (φ2β0) since people in both interaction compartments reduce their contact rate. Using this matrix, the transition rate from the susceptible to the exposed compartment within avoidance behavior takers and non-takers is specified as below. Other transition rates remain as in the standard SEIR model.

!!!!" = !!!,!!!

!! + !!! + !!!,!"!!

!!" + !!"! − !!!

!!!"!" = !!!",!!!"

!! + !!! + !!!",!"!!"

!!" + !!"! − !!!"

  10  

In summation, we modeled the relationship between public risk communication and risk perceptions, between risk perceptions and avoidance behavior, and between avoidance behavior and disease dynamics. We examine how the model performs in a specific context in the following section.

CASE: 2009 H1N1/A INFLUENZA OUTBREAK IN ARIZONA The above computational model is used in the context of the second wave of the 2009 H1N1A influenza outbreak in Arizona. We simulate the disease spread dynamic in Arizona during the 2009-2010 influenza season, as shown in Figure 1 by the solid curve. The simulation results are partially evaluated against empirical data from October 2009 (or the first four weeks of the season: October 4, 2009 to October 31, 2009). A vaccine against 2009 H1N1A influenza has been widely available in Arizona, at least to high-risk priority groups, since November 2009 (ADHS, 2009c). The introduction of vaccination can greatly change the spread dynamic (CDC, 2010b).Our model does not include any public intervention other than public risk communication, so simulated and empirical data are most comparable in October 2009, before the vaccination was available. Data Table1 summarizes the parameters used in the simulation, their (or initial) values, and the sources of these values. There are approximately 2 million households in Arizona, with an average household size of 2.8. Studies on the number of interpersonal contacts and the contact rate have generally been based on convenience samples or conducted in European countries (Edmunds et al., 2006; Mossong et al., 2008). Few have studied the contact rate using the U.S. population (DeStefano et al., 2010). We start the simulation using the assumption of Larson and Nigmatulina (2009) which is an average daily contact rate of 21 in the U.S.. Table 1 Key model parameters and data sources

Key Variables Value Data Sources Population characteristics Number of households 2,276,865 ACS (2009) Average household size 2.8 Same as above Contact (mixing) rate (β) 21 Larson & Nigmatulina (2009) H1N1/A news/information followers 73.1% ASU/ADHS Influenza Survey (2009) Local TV coverage 95.6% Same as above Local TV is a credible source 95.7% Same as above National newspaper coverage 3.8% Same as above National newspaper is a credible source 73.7% Same as above People took avoidance actions among high-risk perceivers

92.6% Same as above

Disease characteristics Infection rate (α) 1.4% Coburn et al. (2009), Yang et al. (2009) Latent period (σ-1) 2 days CDC (2009a) Infectious period (γ-1) 5 days Same as above Period of exposed being infectious (ω) 1 day Same as above

  11  

Mortality rate (µ) 0.3% Donaldson et al. (2009), Tuite et al. (2010) Avoidance behavior effect (φ) 60% Jefferson et al. (2008),

Larson & Nigmatulina (2009) Model initialization Initial number of people exposed 328 Assumption Initial number of people infected 328 ADHS (2009b) Initial number of people recovered 115,758 ACS (2009) , ADHS (2009b), CDC (2009b) Natural immune population 7% Assumption Population characteristics related to public risk communication came from the 2009 ASU/ADHS Influenza Survey (Jehn et al., forthcoming). This survey was conducted using a random-digit telephone survey of representative households in Arizona in October 2009. It included questions about risk perception, information search behavior, and protective behaviors along with other influenza-related questions. Note that the survey was conducted during the H1N1/A influenza pandemic outbreak. The survey asked two risk perception questions: the likelihood and concern of getting H1N1/A flu for family members in the household. About 37% of respondents answered either “very likely” or “very concerned”, and these individuals are grouped as high-risk perceivers. Around 73% of respondents reported that they closely or somewhat closely followed H1N1/A news, and these people are called H1N1/A flu information followers. Regarding risk communication, items in the survey were related to channel type, its coverage, and perceived credibility. Survey respondents were asked to identify the type of channel through which they obtained H1N1/A flu information during the month of the survey. The choices were local TV, national TV, local newspaper, national newspaper, Internet, radio, magazine, personal channel, and other. The personal channel included family, friends, neighbor, college, school, and doctor. Respondents could choose multiple channels and are defined as a channel user of the specific transmission channel they used. Finally, those who indicated that the channel they used to provide updated information on the flu was “very important” or “somewhat important” were considered to perceive that type of channel credible. The analysis of the survey showed that two risk communication channels were significantly related to high risk perceivers: local TV and national newspapers. The perception of getting H1N1/A flu was dependent on whether individuals use either channel for pandemic information and also whether they perceived each channel as credible, which is stated as follows:

! = 0.36− 0.25!!"# − 0.44!!!"# + 0.29!!"#!!"#$ + 0.51!!!"#!!!"#$ where ltv is local TV, nnew refers to national newspapers, and c implies credibility. In the simulation, the percent of people perceiving a high risk of getting H1N1/A flu was calculated based on the percent of channel users (here it is defined as the product of the percent of people covered by each channel (channel coverage) and the percent of information followers), and the percent of people who believe the channel is credible. In Figure 4, we assumed a relationship between risk perception and avoidance actions among the susceptible. However, we do not know what percent of the susceptible perceiving a high risk of

  12  

H1N1/A flu take avoidance behavior. So we used the information collected for the total population from the survey. In other words, 92.6% of the high-risk perceivers took at least one precautionary behavior among the six precautionary behavior measures included in the survey. Therefore, we assumed that 92.6% of high-risk perceivers among the susceptible are decided as the subpopulation engaging in avoidance behaviors at each simulation step. Parameters related to the epidemic characteristics of 2009 H1N1/A influenza were collected from CDC reports and earlier studies. According to the CDC, “the incubation period for influenza is estimated to range from 1 to 4 days with an average of 2 days” (CDC, 2009a) and “influenza virus shedding (the time during which a person might be infectious to another person) begins the day before illness onset and can persist for 5 to 7 days” (CDC, 2009a). In the simulation model, we assume a latent period of 2 days and an infection period of 5 days. Another implication from the CDC’s statement is that exposed individuals are not infectious at all times; they only begin to transmit the virus from the last day of their latent period (CDC, 2009a). The virus transmission rate can be estimated based on previous findings on the basic reproduction number.1 We use a value of 1.4%, which is based on the medium value of the estimated basic reproduction number (Coburn et al., 2009; Yang et al., 2009). The mortality rate for this H1N1/A influenza is 0.3% (Donaldson et al., 2009; Tuite et al., 2010). The model is initialized with 328 exposed, 328 infected, and 115,758 recovered cases. The estimated initial number of infected people is based on the Arizona Department of Health Services weekly activity report (ADHS, 2009b). No data are available for the estimation of exposed cases, so we assume the number is at least the same as that of the infected cases at the beginning of the simulation. The recovered population initially consists of two groups of people: some portion of the elderly population who may have pre-existing immunity to the 2009 H1N1/A influenza virus (CDC, 2009b) and those who recovered from the disease by October 3, 2009 (1,915 cases; ADHS, 2009b). We assume that 7% of the population has pre-existing immunity. Simulation Outputs The simulation output is measured using two indicators: the number of people in infected status on each day (morbidity), and the number of people in infected status by each day since the beginning of the influenza season (cumulative morbidity). Note that in the model one individual actually represents one household. Previous researchers have found that self-protective action is a family-level decision, and the whole household usually acts as one respondent (Ekberg et al., 2009; Vaughan & Tinker, 2009). Simulation outputs are therefore calculated based on the average household size in Arizona. We run the model with the parameters set to the values shown in Table 1. As discussed earlier, the first step of the simulation represents October 4, 2009, and one day is used as one time step in the simulation. Experiments are run 364 time steps to cover the whole 2009-2010 influenza                                                                                                                          1 The reproduction number (R0) is the number of secondary infections caused by a single infectious case

introduced into the susceptible population. Considering the assumption that people can only die when they become infected, R0is equal to the product of the effective contact rate and the infectious period (Keeling & Rohani, 2008).  

  13  

season. Table 2 summarizes the simulation results on morbidity and cumulative morbidity in the standard and modified SEIR models (called risk communication model here). Table 2 Simulation outputs

Peak morbidity

(step) No morbidity

(step) Cumulative morbidity

at season end SEIR Model 419,814

(92) (275) 3,927,736

Risk Communication Model 51,407 (270)

/ 1,654,599

Note: (step) indicates time step in the simulation Figures 5 and 6 show how morbidity and cumulative morbidity change over time in each model. In the SEIR model, where there is no public risk communication, morbidity peaks on January 3, 2010 (92th time step), and 419, 814 people are in infected status on that day. Although the morbidity keeps decreasing after that day, the disease continues to spread until July 5, 2010 (275th time step). By the end of the influenza season, the total number of people infected is 3,927,736, which is 61.6% of the total Arizona population.

Figure 5 Simulated morbidity during the 2009-2010 influenza season

0  

50,000  

100,000  

150,000  

200,000  

250,000  

300,000  

350,000  

400,000  

450,000  

0   14  

28  

42  

56  

70  

84  

98  

112  

126  

140  

154  

168  

182  

196  

210  

224  

238  

252  

266  

280  

294  

308  

322  

336  

350  

364  

Morbidity  

Day  

SEIR  model   Risk  communica=on  model  

  14  

Figure 6 Simulated cumulative morbidity during the 2009-2010 influenza season

When considering public risk communication in the modified model, the impacts of the pandemic are mitigated and the spread of the disease slows. The peak morbidity is reduced to 51,407 on June 30, 2010, which is almost 6 months after the peak morbidity day without risk communication. By the end of the influenza season, the cumulative morbidity is 1,654,599, with a reduction of 57.9% as compared to the SEIR model. Another important characteristic of the pandemic spread found in this condition is that the disease continues spreading during the entire season, and the morbidity and cumulative morbidity are still increasing at the end of the influenza season. However, in this case there is more time to prepare for the arrival of a peak. Note that the model did not include other policy interventions such as vaccination, which could drastically change the dynamics of morbidity. To evaluate the model, we compare the cumulative morbidity value from the models against the empirical data for Arizona. The comparison is made from October 4 to 31. The numbers of cases are grouped on a weekly basis because the empirical data available were updated by ADHS by the end of each week. As shown in Figure 7, the simulation results from the risk communication model are close to the actual number of cases in Arizona while the SEIR model produced a much higher number of H1N1/A cases. We are confident, at least in the short run, that the model provides some important information on the dynamic relationship between public risk communication and the disease spread process in Arizona.

0  

500,000  

1,000,000  

1,500,000  

2,000,000  

2,500,000  

3,000,000  

3,500,000  

4,000,000  

4,500,000  

0   14  

28  

42  

56  

70  

84  

98  

112  

126  

140  

154  

168  

182  

196  

210  

224  

238  

252  

266  

280  

294  

308  

322  

336  

350  

364  

Cumula0

ve  m

orbidity  

Day  

SEIR  model   Risk  communica=on  model  

  15  

Figure 7 Simulated and empirical cumulative morbidity by week in October 2009

Computer Experiments Here we conduct an experiment on the communication components by varying the coverage of national newspapers and credibility of local TV and national newspapers as well as the percent of information followers. Table 1 showed that local TV covered 95.6% of households in Arizona as a source of H1N1/A news and information, but national newspaper coverage was 3.8% in October 2009. This indicates that, in Arizona, local TV had almost complete coverage, but national newspapers reached only a small proportion of the population. According to the simulation output, the peak morbidity was 51,407 at step 270 under the condition, as shown in Table 2. Therefore, the first question is whether increased coverage of national newspapers would mitigate the dynamics of H1N1 flu morbidity under the current local TV coverage in Arizona.

• At the current coverage of local TV, increasing the household who get H1N1/A information from national newspapers does not alleviate pandemic impacts on the community.

Three national newspaper coverage scenarios (10%, 50%, and 90%) were experimented with to answer the question above. The assumptions here are that the local TV coverage is 90%, the percent of information followers does not change, and the percent of people who think that local TV and national newspapers are credible sources does not change.

0  

2,000  

4,000  

6,000  

8,000  

10,000  

12,000  

14,000  

16,000  

10/10/2009   10/17/2009   10/24/2009   10/31/2009  

Cumula0

ve  m

orbidity  

Day  

Empirial  data   SEIR  model   Risk  communica=on  model  

  16  

Table 3 shows that the set level of 90% local TV coverage and 10% national newspaper coverage performs very closely to the results using the empirical value set (96% local TV coverage and 4% national newspaper coverage). The peak morbidity under this scenario of media coverage is 54,540 at step 264. However, by increasing the coverage of national newspapers to 50%, the peak morbidity actually increased to 69,901 from 54,540, occurring at step 238. The situation worsened when the coverage of national newspapers reached 90%. The morbidity increased to 86,426 at peak time of 217 steps, even though the coverage of national newspapers increased to 90% of the household in Arizona for H1N1/A news. Table 3 National newspaper coverage scenarios

Peak Morbidity (Step) Local TV Coverage

90% National Newspaper Coverage

10% 54,540 (264) 50% 69,901 (238) 90% 86,426 (217)

Figure 8 presents the simulation results on morbidity from the three experimental scenarios in a graphical format, comparing them with the results of the risk communication model as a baseline. When other conditions are set to the values found in the empirical survey, the increase in the coverage of national newspapers may actually aggravate the situation in terms of peak morbidity and when the peak occurs.

Figure 8 Simulated morbidity when increasing the coverage of national newspapers

0  

10,000  

20,000  

30,000  

40,000  

50,000  

60,000  

70,000  

80,000  

90,000  

100,000  

0   14  

28  

42  

56  

70  

84  

98  

112  

126  

140  

154  

168  

182  

196  

210  

224  

238  

252  

266  

280  

294  

308  

322  

336  

350  

364  

Morbidity  

Day  

ltv:  90%  &  nnews:  10%  ltv:  90%  &  nnews:  50%  ltv:  90%  &  nnews:  90%  

  17  

We speculated that this result occurred because the percent of H1N1/A news followers and the percent of people who believe the channel are fixed at the empirical level in the experiment. Below, we explore the effect of these two parameters under the assumption of 90% coverage of local TV and 10% national newspaper coverage (i.e., the closest to the empirical condition).

• With the current channel coverage levels of local TV and national newspapers, increasing the perceived credibility of national newspapers and the percent of information followers can mitigate pandemic impacts.

In Table 1 we reported that approximately 73% of respondents followed H1N1/A flu news in Arizona during the month of the survey. In terms of perceived credibility, 96% of respondents stated that local TV is a credible source of H1N1/A flu news and 74% indicated that national newspapers are a credible news source. Since the percent of respondents indicating that local TV is a credible source reached 96%, we examine the scenarios that consider the increase in the percent of people who believe that national newspapers are a credible source of information and the increase in the percent of information followers as variables. Table 4 presents the peak morbidity and the number of steps taken to reach the peak under the different scenarios of the two variables. The first cell presents the result of the risk communication model with the empirical value set for both variables. Regardless of the scenario, the increases in the two parameters reduce the peak morbidity. Under the assumption that the proportion of people who believe that national newspapers are a credible source is not changed (73.7%), the increases in the percent of people who follow H1N1 flu news and information reduces peak morbidity and delays the time to the peak. When the percent of H1N1/A news/information followers is set to the empirical level (73.1%), the increase in the percent of national newspaper believers to 80% or 90% also reduced peak morbidity to 52,707 and 49,844, correspondingly. Table 4 Experimental results under the risk communication variable scenarios

Peak morbidity (Step) % of H1N1/A flu information follower 73.1% Scenario: 80% Scenario: 90%

% of national newspaper believers

73.7% 54,540 (264) 53,549 (266) 52,124 (269) Scenario: 80% 52,707 (268) 51,558 (270) 49,906 (273) Scenario: 90% 49,844 (274) 48,452 (277) 46,458 (281)

Note: The bold is the simulated peak morbidity (occurrence step) under the empirical survey values for the two variables. It also appears that the increase in the percent of national newspaper believers influences the peak morbidity more severely than the increase in the percent of H1N1/A flu information followers. For example, when the percent of national newspaper believers is set to 80%, the increase in the percent of news followers from 80% to 90% reduced peak morbidity by 1,625 (difference between 51,558 and 49,906). When the percent of information followers was set to 80%, the increase in the percent of national newspaper believers from 80% to 90% reduced peak morbidity by 3,106 (difference between 51,558 and 48,451).

  18  

DISCUSSION

The outbreak of pandemic influenza can cause serious social and economic consequences in communities. In 2009, the H1N1/A influenza epidemic occurred, corroborating the expectation the CDC director stated in April 2007: “we know that a pandemic will eventually occur. We always say it’s not a question of if; it’s a question of when” (Ulene, 2007). This recent pandemic outbreak does not exclude the possibility of another novel virus-caused influenza outbreak. To minimize the impact of pandemic influenza, both CDC and HHS have made it a priority for public managers at all levels to engage in pandemic planning and response efforts (Das, Savachkin, & Zhu, 2008; Ferguson et al., 2005). Public risk communication is an integral component of emergency management. Understanding its dynamics is crucial for effectively managing public emergencies in communities. This study makes several theoretical and practical contributions. First, the simulation results suggest that effective public risk communication can slow the pandemic spread and therefore help buy time to introduce other public interventions, particularly the production and distribution of enough vaccines for the general public. Although public managers cannot solely rely on risk communication and people’s avoidance responses to manage adverse social outcomes, effective risk communication makes the impact less devastating. Second, we compared the effectiveness of different information transmission channels in addition to their coverage and perceived credibility on impact mitigation. In the empirical context used in this study, the public survey revealed that the coverage and perceived credibility of local TV and national newspapers are associated with individual risk perceptions. Using computational simulation, we further modeled the relationship between risk perception and avoidance behavior by the public, and observed that local public managers may need to increase the number of people who believe that national newspapers are a credible source of information regarding disease outbreaks rather than increasing the coverage of national newspapers in order to mitigate pandemic impacts. The increase in the percent of people who follow H1N1 flu information can also mitigate influenza morbidity. Third, the computational model can serve as an exploration instrument for researchers as well as a decision support tool for local public managers. Because it includes individual risk perceptions and responsive behaviors, which commonly occur during an epidemic but are ignored by most epidemic models, it can be used to more effectively anticipate the spread of a pandemic. As it includes the component of public risk communication, public managers can use it to systematically evaluate and compare the effectiveness of different strategies related to public risk communication. The risk communication model can be applied to other types of infectious disease or public health threats, such as bioterrorism, or in different community settings. There are several limitations in the present study. No other public intervention effort, particularly vaccination, is considered in this model. However, different types of public interventions can be included in this model in the future to test their effectiveness. In addition, exposed and infected individuals can, in reality, be medically treated to shorten their latent and infectious period, which is not taken into consideration in the present models. However, the influence of those

  19  

medical treatments on the simulation results may be limited because we use the average length of the latent and infectious period from the empirical data, which already included the effect of medical treatments. The computational model we developed is constrained by some limitations inherent in the SEIR compartment model. One major concern is the homogeneous population assumption. Standard compartment models assume a group of identical people in terms of their contact patterns and the impact of an epidemic. We attempted to include some heterogeneity in the model by splitting the population into eight compartments. However, such a change introduces only a limited degree of heterogeneity in people’s contact patterns, and the biological characteristics of the disease are still the same for all individuals. While it can be resource demanding, one way to address this problem is to develop an individual-based model and include important types of heterogeneity for epidemic simulation (Mniszewki et al., 2008; Bobashev et al., 2007; Rahmandad & Sterman, 2008). The value for some parameters in the model cannot be easily determined empirically, particularly the population characteristics related to public risk communication, and there are few studies that can be referenced regarding this topic. Also, the contexts in existing studies generally differ from one another and the number from one context cannot be simply applied to another. For example, the perceived credibility of the same media usually varies greatly in different communities, time periods, and epidemics. To accurately anticipate the disease dynamic and the effectiveness of certain communication strategies, public managers should parameterize the model with reasonable values for those parameters. Despite these detractors, the current study addressed critical challenges in studying public risk communication during an emergency situation such as a pandemic influenza outbreak. The flexibility of the approach presented can be used to conduct further experiments and test social interventions in specific contexts without the issues of real experiments or observational studies. This can be a useful decision support tool to inform decision-makers.

  20  

REFERENCE

Aakko, E. (2004) Risk communication, risk perception, and public health. Wisconsin Medical Journal, 103(1), 25-27.

Arizona Department of Health Service (ADHS) (2009a). Arizona weekly influenza summary: MMWR Week 40. Retrieved February 3, 2011, from http://www.azdhs.gov/phs/oids/epi/flu/pdf/h1n1_report_october14.pdf

Arizona Department of Health Service (ADHS) (2009b). Arizona weekly influenza summary: MMWR Week 39. Retrieved February 3, 2011 from http://www.azdhs.gov/phs/oids/epi/flu/pdf/h1n1_report_october7.pdf

Arizona Department of Health Service (ADHS) (2009c). Arizona 2009 H1N1 influenza vaccine distribution program 2009-2010 background document. Retrieved February 3, 2011, from http://www.azdhs.gov/flu/h1n1/pdfs/vaccine/VaccineDistributionBackgroundDocument.pdf

Balluz, L., Schieve, L., Holmes, T., Kiezak, S., & Malilay, J. (2000). Predictors for people's response to a tornado warning: Arkansas, 1 March 1997. Disasters, 24(1), 71-77.

Bobashev, G. V., Goedecke, D. M., Yu, F., & Epstein, J. (2007). A hybrid epidemic model: Combining the advantages of agent-based and equation-based approaches. In S. G. Henderson, M. H. Hsieh, J. Shortle, J. D. Tew & R. R. Barton (Eds.), Proceedings of the 2007 Winter Simulation Conference. Washington, D. C.: The Society for Computer Simulation International (SCS).

Center for disease control and prevention (CDC) (2009a). Updated interim recommendations for the use of antiviral medications in the treatment and prevention of influenza for the 2009-2010 season. Retrieved February 3, 2011, from http://www.cdc.gov/h1n1flu/recommendations.htm

Center for disease control and prevention (CDC) (2009b). 2009 H1N1 early outbreak and disease characteristics. Retrieved February 3, 2011, from http://www.cdc.gov/h1n1flu/surveillanceqa.htm

Center for disease control and prevention (CDC) (2010a). 2009 H1N1 flu. Retrieved February 3, 2011, from http://www.cdc.gov/h1n1flu/

Center for disease control and prevention (CDC) (2010b). Key facts about 2009 H1N1 flu vaccine. Retrieved February 13, 2010, from http://www.cdc.gov/h1n1flu/vaccination/vaccine_keyfacts.htm

Coburn, B. J., Wagner, B. G., & Blower, S. (2009). Modeling influenza epidemics and pandemics: Insights into the future of swine flu (H1N1). BMC Medicine, 7, 30.

Das, T., Savachkin, A., & Zhu, Y. (2008). A large scale simulation model of pandemic influenza outbreaks for development of dynamic mitigation strategies. IIE Transactions, 40(9), 893-905.

de Zwart, O., Veldhuijzen, I. K., Richardus, J. H., & Brug, J. (2010). Monitoring of risk perceptions and correlates of precautionary behaviour related to human avian influenza during 2006 - 2007 in the Netherlands: Results of seven consecutive surveys. BMC Infectious Diseases, 10, 114.

Destefano, F., Haber, M., Currivan, D., Farris, T., Burrus, B., Stone-Wiggins, B., et al. (2010). Factors associated with social contacts in four communities during the 2007-2008 influenza season. Epidemiology and Infection, 10, 1-10.

Donaldson, L. J., Rutter, P. D., Ellis, B. M., Greaves, F. E. C., Mytton, O. T., Pebody, R. G., et al. (2009). Mortality from pandemic A/H1N1 2009 influenza in England: public health survelliance study. BMJ, 339, b5213.

  21  

Donner, B. (2006). Public warning response to Hurricane Katrina: A preliminary analysis. Newark, DE: Disaster Research Center, University of Delaware.

Donner, W. R. (2007). An integrated model of risk perception and protective action: Public response to tornado warnings. Doctoral dissertation. Newark, DE: University of Delaware.

Donner, W. R., Rodriguez, H., & Diaz, W. (2007). Public warning response following tornadoes in New Orleans, LA, and Springfield, MO: A sociological analysis. Paper presented at the Second Symposium on Policy and Socio-economic Research, 87th Annual Meeting of the American Meteorological Society, San Antonio, Texas.

Drabek, T. E. (1969). Social processes in disaster: Family evacuation. Social Problems, 16(3), 336-349.

Drabek, T. E. (1986). Human system responses to disaster: An inventory of sociological findings. New York, NY: Springer-Verlag.

Drabek, T. E., & Boggs, K. S. (1968). Families in disaster: Reactions and relatives. Journal of Marriage and Family, 30(3), 443-451.

Dynes, R. R., & Quarantelli, E. L. (1973). The family and community context of individual reactions to disaster. Paper presented at the NIMH Continuing Education Seminar on Emergency Mental Health Services, Washington, D.C.

Edmunds, W. J., Kafatos, G., Wallinga, J., & Mossong, J. R. (2006). Mixing patterns and the spread of close-contact infectious diseases. Emerging Themes in Epidemiology, 3, 10.

Ekberg, J., Eriksson, H., Morin, M., Holm, E., Strömgren, M., & Timpka, T. (2009). Impact of precautionary behaviors during outbreaks of pandemic influenza: Modeling of regional differences. American Medical Informatics Association Annual Symposium Proceedings, 2009, 163-167.

Epstein, J. M., Parker, J., Cummings, D., & Hammond, R. A. (2008). Coupled contagion dynamics of fear and disease: Mathematical and computational explorations. PLoS One, 3(12), e3955.

Freimuth, V. S., Hilyard, K. M., Barge, J. K., & Sokler, L. A. (2008) Action, not talk: A simulation of risk communication during the first hours of a pandemic. Health Promotion Practice, 9(s4), 35s-44s.

Ferguson, N. M., Cummings, D. A. T., Cauchemez, S., Fraser, C., Riley, S., Meeyai, A., et al. (2005). Strategies for containing an emerging influenza pandemic in Southeast Asia. Nature, 437(8), 209-214.

Gladwin, H., Lazo, J. K., Morrow, B. H., Peacock, W. G., & Willoughby, H. E. (2007). Social science research needs for the hurricane forecast and warning system. Natural Hazards Review, 8(3), 87-95.

Jefferson, T., Foxlee, R., Mar, C. D., Dooley, L., Ferroni, E., Hewak, B., et al. (2008). Physical interventions to interrupt or reduce the spread of respiratory viruses: Systematic review. BMJ, 336, 77.

Jehn, M., Kim, Y., Bradley, B., & Lant, T. (forthcoming). Community knowledge, risk perception and preparedness for the 2009 influenza A (H1N1) pandemic. Journal of Public Health Management and Practice.

Keeling, M. J., & Rohani, P. (2008) Modeling infectious disease in humans and animals. Princeton, NJ: Princeton University Press.

Lau, J. T., Kim, J. H., Tsui, H. Y., & Griffiths, S. (2007). Anticipated and current preventative behaviors in response to an anticipated human-to-human H5N1 epidemic in Hong Kong Chinese general population. BMC Infectious Diseases, 7, 18.

  22  

Lau, J. T., Yang, X., Tsui, H., & Kim, J. H. (2003). Monitoring community psychological responses to the SARS epidemic in Hong Kong: From day 10 to day 62. Journal of Epidemiology and Community Health, 57(11), 864-870.

Larson, R. C., & Nigmatulina, K. R. (2009). Engineering responses to pandemics. Information Knowledge Systems Management, 8(1-4), 311-339.

Leung, G. M., Lam, T.-H., Ho, L.-M., Ho, S.-Y., Chan, B. H. Y., Wong, I. O. L., et al. (2003). The impact of community psychological responses on outbreak control for severe acute respiratory syndrome in Hong Kong. Journal of Epidemiology and Community Health, 57(11), 857-863.

Li, M. Y., Graef, J. R., Wang, L., & Karsai, J. (1999). Global dynamics of a SEIR model with varying total population size. Mathematical Biosciences, 160(2), 191-213.

Maxwell, T. A. (2003). The public need to know: emergencies, government organizations, and public information policies. Government Information Quarterly, 20(3), 233-258.

McLuckie, B. F. (1970). The warning system in disaster situations: A selective analysis. Columbus, OH: Disaster Research Center, Ohio State University.

Mileti, D. S. (1995). Factors related to flood warning response. Paper presented at the U.S.- Italy Research Workshop on the Hydrometeorology, Impacts, and Management of Extreme Floods, Perugia, Italy.

Mileti, D. S., & Darlington, J. D. (1997). Society for the study of social problems. Social Problems, 44(1), 89-103.

Mileti, D. S., & Fitzpatrick, C. (1992). The causal sequence of risk communication in the parkfield earthquake prediction experiment. Risk Analysis, 12(3), 393-400.

Mileti, D. S., & O' Brien, P. W. (1992). Warnings during disaster: Normalizing communicated risk. Social Problems, 39(1), 40-57.

Mileti, D. S., & Sorensen, J. H. (1990).Communication of emergency public warnings: A social science perspective and state-of-the-art assessment. Oak Ridge, TN: Oak Ridge National Laboratory.

Mniszewski, S. M., Valle, S. Y. D., Stroud, P. D., Riese, J. M., & Sydoriak, S. J. (2008). Pandemic simulation of antivirals + school closures: Buying time until strain-specific vaccine is available. Computational and Mathematical Organization Theory, 14(3), 209-221.

Mossong, J., Hens, N., Jit, M., Beutels, P., Auranen, K., Mikolajczyk, R., et al. (2008). Social contacts and mixing patterns relevant to the spread of infectious diseases. PLos Medicine, 5, 381-198.

Parker, D. J., Priest, S. J., & Tapsell, S. M. (2009). Understanding and enhancing the public's behavioral response to flood warning information. Meteorological Applications, 16(1), 103-114.

Perry, R. W. (1979). Evacuation decision-making in natural disasters. Mass Emergencies, 4(1), 25-38.

Perry, R. W. (1981). Citizen evacuation in response to nuclear and non-nuclear threats. Seattle, WA: Battelle Human Affairs Research Center.

Perry, R. W., Lindell, M. K., & Greene, M. (1980). The implications of natural hazard evacuation warning studies for crisis relocation planning. Seattle, WA: Battelle Human Affairs Research Center.

Pfister, N. (2002). Community response to flood warnings: The case of an evacuation from Grafton, March 2001. The Australian Journal of Emergency Management, 17(2), 19-29.

  23  

Quarantelli, E. L. (1983). People's reactions to emergency warnings. Newark, DE: Disaster Research Center, University of Delaware.

Rahmandad, H., & Sterman, J. (2008). Heterogeneity and network structure in the dynamics of diffusion: Comparing agent-based and differential equation models. Management Science, 54(5), 998-1014.

Reynolds, B. (2005). Crisis and emergency risk communication. Applied Biosafety, 10(1), 47-56. Ross, T., Zimmer, S., Burke, D., Crevar, C., Carter, D., Stark, J., et al. (2010). Seroprevalence

following the second wave of pandemic 2009 H1N1 influenza. PLoS Currents, 2, RRN1148. Rost, G., & Wu, J. (2008). SEIR epidemiological model with varying infectivity and infinite

delay. Mathematical Biosciences and Engineering, 5(2), 389-402. Shanks, J. (2009). Arizona swine flu: Maricopa county gets first case. Retrieved February 3,

2011, from http://www.nationalledger.com/cgibin/artman/exec/view.cgi?archive=36&num=25842

Sorensen, J. H. (1991). When shall we leave? Factors affecting the timing of evacuation departures. International Journal of Mass Emergencies and Disasters, 9(2), 153-165.

Sorensen, J. H. (1992). Assessment of the need for dual indoor/outdoor warning systems and enhanced tone alert technologies in the chemical stockpile emergency preparedness program. Oak Ridge, TN: Oak Ridge National Laboratory.

Sorensen, J. H. (1993). Warning system and public warning response. Paper presented at the Socioeconomic Aspects of Disaster in Latin America, San Jose, Costa Rica.

Sorensen, J. H. (2000). Hazard warning systems: Review of 20 years of progress. Natural Hazards Review, 1(2), 119-125.

Trainor, J. E., & McNeil, S. (2008). A brief summary of social science warning and response literature: A report to COT Netherlands. Newark, DE: University of Delaware, Disaster Research Center.

Tuite, A. R., Greer, A. L., Whelan, M., Winter, A.-L., Lee, B., Yan, P., et al. (2010). Estimated epidemiologic parameters and morbidity associated with pandemic H1N1 influenza. CMAJ, 182(2), 131-136.

Ulene, V. (2007). Bracing for a flu pandemic. Los Angeles Times, June 4, 2007. Retrieved February 3, 2011, from http://www.latimes.com/features/health/la-he-themd4jun04,1,3259716.column

Vaughan, E., & Tinker, T. (2009). Effective health risk communication about pandemic influenza for vulnerable populations. American Journal of Public Health, 99(s2), S324-S332.

Williams, H. B. (1964). Human factors in warning-and-response systems. In G. H. Grosser, H. Wechsler & M. Greenblatt (Eds.), The threat of impending disaster: Contributions to the psychology of stress. Cambridge, Mass: MIT Press.

Yang, Y., Sugimoto, J. D., Halloran, M. E., Basta, N. E., Chao, D. L., Matrajt, L., et al. (2009). The transmissibility and control of pandemic influenza A (H1N1) virus. Science, 326(5953), 729-733.

Yoo, B. K., Kasajima, M., & Bhattacharya, J. (2010). Public avoidance and the epidemiology of novel H1N1 influenza A. Cambridge, MA: National Bureau of Economic Research.