We graphed the curves of H7N9 epidemic and severity by number of daily cases, first case(s) reported in each province/city, and cumulative case fatality rate over time. The daily posted and forwarded number and BAI were graphed by date, number of daily cases, and first case(s) in each province/city, respectively, to explore public reaction changing to the H7N9 epidemic situation over time. Median (P₅₀) and range were used to describe the distributions of daily posted and forwarded number and BAI indexes by province/city. Spearman’s rank correlation coefficient was used to check the relationship of epidemic trends with the daily posted and forwarded number and BAI, respectively, if the distribution of the variables is monotonic. Kruskal-Wallis test was used to test the differences in daily posted and forwarded number and BAI between provinces/cities with cases and without cases. Rumors of H7N9 outbreaks were qualitatively classified by date, province, rumor maker, media, and main content.

On March 31, 2013, the CNHFPC reported the first human infections with the H7N9 virus, including 2 cases in Shanghai and 1 in Anhui province [10]. Subsequently, cases were successively reported in Jiangsu, Zhejiang, Beijing, Henan, and other provinces. Until April 24, 2013, a total of 108 cases and 23 deaths (cumulative case fatality rate = 21.3%) of H7N9 infections were reported in 7 provinces/cities in China [10]. The cumulative case fatality rate was 33% (6/18) during the first week and then gradually declined to 21.3% (23/108) because of timely prevention and treatment. By April 12, 43 cases had been reported in 4 provinces in Eastern China and then the epidemic expanded to Beijing on April 13 and Henan province on April 14. The last case during the study period was reported in Shandong province on April 23, 2013 (Figure 1).

The total daily posted and forwarded number for the keyword H7N9 on Sina microblog peaked at 850,000 on April 3, 2013, from zero before March 31, increasing to 97,726 on April 1, and to 370,607 on April 2, and remaining above 500,000 from April 5 to 8 before declining to 208,524 on April 12. When the first case was reported in Capital Beijing on April 12 and the first case in Henan province on April 14, the total daily posted and forwarded number returned to a relatively high level of 368,500 on April 13 and 339,822 on April 14, declined to 262,667 on April 15 and 235,131 on April 16, and increased again to 273,752 on April 17 and 370,995 on April 18. The total daily posted and forwarded number then decreased and remained at a level of 100,000 (Figure 2). The median of the total daily posted and forwarded number was 289,325 (range 76,482-853,027). Spearman’s rank correlation coefficient analyses showed that the daily posted and forwarded number was positively associated with cumulative case fatality rate (Spearman’s rank correlation coefficient = 0.60, P=.002), that is to say, higher cumulative case fatality rate higher daily posted and forwarded number at the early stage of epidemic and lower cumulative case fatality rate lower daily posted and forwarded number later (Figure 3).

Since the first human H7N9 case reported on March 31, the observed daily posted and forwarded number in each province/city showed a similar trend to the total daily posted and forwarded number. However, the daily posted and forwarded number in Beijing on April 14 and Henan on April 15 rose to another peak when the local area first reported cases (the daily posted and forwarded number peaked at 163,512 in Beijing on April 13 and peaked at 72,512 in Henan province on April 14 as shown in Figure 4).

The observed median daily posted and forwarded number during the 25 days studied was 36,256 (range 1830-232,368) in Shanghai, 4944 (217-31,312) in Anhui, 21,424 (1521-143,376) in Jiangsu, 24,720 (897-131,840) in Zhejiang, 57,680 (7867-166,448) in Beijing, 11,536 (316-56,032) in Henan, 28,016 (513-56,032) in Shandong, 9888 (327-47,792) in Hubei, and 924 (229-26,368) in Shaanxi. The mean rank value of the daily posted and forwarded number in each province/city with human H7N9 cases reported except Anhui was significantly higher (P<.001) than that in the 2 provinces without cases (Table 1).

The daily BAI for keyword H7N9 increased sharply in the first 3 days of the H7N9 outbreak, with a peak at 126,825 on April 3. The daily BAI then declined with fluctuations, remaining at a high level above 100,000 from April 4 - 8. Following the cases reported in Beijing and Henan, the daily BAI increased again and peaked at 116,991 on April 15. Afterwards, it steadily decreased to a lower level of approximately 40,000. The median daily BAI was 79,687 (range 18,007-79,687) (Figure 5). Spearman’s rank correlation coefficient analyses showed that the daily BAI was positively associated with cumulative case fatality rate (Spearman’s rank correlation coefficient = 0.43, P=.04), that is to say, higher cumulative case fatality rate higher daily BAI at the early stage of epidemic and lower cumulative case fatality rate lower daily BAI later (Figure 6).

The daily BAI for H7N9 in each province/city increased substantially after the first human cases were reported regardless of where human cases were reported. The trend of daily BAI change in each province/city was similar to the total daily BAI. Moreover, there was also an obvious increase of the daily BAI in Beijing and Henan from April 14-15 because of the first case report in Beijing on April 13 and in Henan on April 14 and in Shandong province when the government first announced a case on April 23.

Overall, the median daily BAI was 4227 (range 0-9014) in Shanghai, 1723 (0-2765) in Anhui, 6592 ( 0-11,914) in Jiangsu, 6432 (0-15,011) in Zhejiang, 4790 (0-9488) in Beijing, 2731 (0-7592) in Henan, 3065 (0-6763) in Shandong, 1420 (0-2707) in Hubei, and 1301 (0-2198) in Shaanxi. The mean rank values of daily BAI in the provinces with reported cases were significantly higher than that in the provinces without cases (P<.001; Table 1).

All 32 rumors were disseminated by 58 people and distributed in 19 provinces in China. In all, 84% (27/32) rumors were disseminated through Sina microblog, QQ (the most popular Chinese communication software), and Wechat (an instant mobile communication software), and 16% (5/32) were spread via bulletin board system (BBS), cell phone message, and telephone calls. Of these 27 rumors, 72% (23/32) disseminated information that human infections with H7N9 virus occurred in the local areas, and 16% declared that eating chicken feet with pickled peppers (a popular Chinese food) would cause H7N9 infection (Figure 8). Among 15 rumor investigations that released the characteristics of the rumor makers, 5 were students; 5 were migrant workers; and 5 were farmers and company employees. Four rumor investigations announced the aims for making and disseminating rumors were to obtain more attention or get more visits to their individual forum or microblog. Some rumors had very strong transmissibility. The rumor that “there was a suspected human case with H7N9 virus infection in a local school” disseminated by a student from Gansu province, Western China, was forwarded by 34,000 people. The rumor that “eating chicken feet with pickled peppers would cause H7N9 infection” was spread across 5 provinces in China.

Our study used Internet surveillance to investigate the public reaction to the H7N9 outbreaks over time during the first 25 days of the H7N9 epidemic in China in the spring of 2013. The indexes of public reaction, daily posted and forwarded number and BAI, to H7N9 outbreaks were significantly higher in 7 provinces/cities with human H7N9 cases than those in 2 provinces without cases. Both daily posted and forwarded number and BAI were positively associated with the cumulative case fatality rate of human H7N9 infection. Our findings indicate that the first 3 days of an epidemic is a critical period for the authorities to take appropriate responsible action through Internet surveillance to control and prevent the epidemic, including preparation of personnel, technology, and other resources; information release; collection of public opinion and reaction; and clarification, prevention, and control of rumors. Internet surveillance can be used as an efficient and economical tool to control and prevent public health emergencies, such as H7N9 outbreaks.

This is the first study that investigated the importance of infoveillance [7,14-17] methods in monitoring public reaction to the H7N9 avian influenza outbreak on Sina microblog and Baidu search engines and to have collected rumors through the Internet in China. When an emerging infectious disease outbreak occurs, it is important for risk communicators in the government to clearly know the public concern [18]. In the past, to obtain information on the public, community, and media reactions or concern, most researchers used questionnaires and telephone interviews after the outbreak [19-23]. In contrast, network monitoring can be timelier to know public reactions to the outbreak because it is a real-time surveillance during the emergencies and also it is very economical without consuming too many resources [6,7,13,16,24,25]. During this human H7N9 avian influenza outbreak, we obtained the information by Internet monitoring and provided timely policy suggestions to the governments and related shareholders who play a role in disease outbreak control and prevention.

Emerging infectious diseases usually have the features of unknown pathogens and high mortality, no vaccine or effective treatment available, and quick and wide dissemination that trigger high public and media attention [18,26]. As soon as the government announced the outbreak, the public would have concern about the event and be eager to find relevant information to obtain a certain sense of safety [22,25]. Our study suggested that the first 3 days of the outbreak was the period when Internet attentions rose rapidly. At this point, the government needs to timely and transparently release the information, including personal protective methods of the disease, which can ease public tension quickly [18]. During this human H7N9 outbreak, when the first human case was announced, the authorities timely released case numbers and disease prevention knowledge online, which may be one of the key reasons for the early decline of public attention. However, the high attention to this outbreak lasted 3 - 5 days, which is a good time for the Department of Emergency Control and Prevention to disseminate health education knowledge and skills through network and social media.

Some studies have shown that the public concern with the H1N1 influenza infection was associated with the number of reported cases and hospitalizations by the governments [22,23]. However, other studies reported that although the number of flu cases rose, the tweets about the H1N1 influenza showed a downward trend [16]. This fact reflected that public concern or interests for the outbreak might be weakened along with the progress of the epidemic as information saturation sets in [16,27]. Our study showed public reaction was positively associated with the cumulative case fatality rate but not daily number of cases. When case fatality rate was high, the public attention was high. The public attention decreased with a decrease in case fatality rate. This indicates that the public attention decreases when the threat from the epidemic disease decreases. Our study showed a decreasing trend of Internet user’s attention during the middle and later period. However, when there were new cases reported in a new province/city, the daily posted and forwarded number and BAI increased significantly, especially in the local province/city. This phenomenon indicates that the extension of epidemic area triggered more public attention than the increase of reported cases in the original outbreak areas. It is necessary to monitor the public reaction and take specific responses at that moment.

Rumors generate and occur early during the period of public health emergency [28-30]. The rumors of human H7N9 outbreaks collected in this study covered most provinces in China, and also spread quickly. Most of the rumors claimed that H7N9 avian influenza cases occurred in the local areas. The uncertainty of emerging infectious diseases often generates the rumors, which also suggests the importance of early information release [19,25,31-32]. In this new age of social media, microblogs are easy to generate and spread rumors, yet they could also be used to clarify wrong information and stop the dissemination.

There were some limitations in this study. The daily posted and forwarded number and BAI were collected from the Internet and could not be standardized because of lack of exact numbers of daily Internet users. Thus, the mean rank of daily posted and forwarded number and BAI only provide relative comparisons between provinces/cities with and without H7N9 cases, but there is not much meaning to comparisons among individual provinces/cities. Our study was limited to the first 25 days of the outbreak because the government stopped daily case reports, instead using weekly case reports; therefore, this might not thoroughly reflect public reaction after the first 25 days. Our study primarily focused on Chinese websites, which limits the representativeness of the public reaction from Internet users on English and other language websites. Our study is an exploratory study that was designed to use a quick way to obtain public reaction during the study period of a public health emergency. The qualitative research on rumors was simple and did not use frameworks to analyze because of the diversity of the information released on the Internet. The results of our study could be applicable to public health emergency response, but may not be directly applied to public health practice yet. Further studies with advanced methods or study designs are needed.

This exploratory digital epidemiologic study provides evidence that Internet surveillance is a rapid and efficient way to evaluate public reactions, which will help prepare the government, health authorities, and the public to respond to public health emergencies. Our findings showed the public reaction increased quickly during the early stage of the outbreak. The first 3 days could be a “golden 3 days” for the government and public health authorities to release information in time and make the information transparent and open. The high public attention would last for a week; thus, this week could be a critical period for health education. Expansion of the disease into other areas has more impact on public attention than the number of cases reported in the same outbreak area. This implies that extra methods should be taken when the epidemic area is expanded. Rumors are always hard to avoid, but using media, such as releasing information via an official microblog, could be effective to quickly clarify rumors and prevent their further spread.