Evolving epidemiology of human infections with avian influenza A(H7N9) virus across five epidemic waves in mainland China, 2013–17: an epidemiological study of laboratory-confirmed case series

INTRODUCTION

The emergence of novel avian influenza A(H7N9) virus in 2013 poses a pandemic threat to humans, which necessitates close monitoring of the virus and continuing assessment of pandemic risk.¹ Human cases of A(H7N9) virus infection have occurred during annual winter-spring epidemic waves in mainland China since 2013,¹ reaching 1,220 laboratory-confirmed cases and 494 deaths as of 23 February 2017. After peaking during 2013–14, the number of cases identified in subsequent epidemic waves was generally lower, until a surge in A(H7N9) cases was observed in December 2016, in the 5^th epidemic wave of human infections.² The 5^th epidemic wave that started on 1 October 2016 has included 447 laboratory-confirmed cases in mainland China (as of 23 February 2017), and is the largest epidemic wave to date. The earlier start and larger epidemic size in the 5^th A(H7N9) epidemic wave has prompted concerns of whether the epidemiology has changed to suggest an increasing pandemic threat.^2,3

Prior to 2016–17, the avian influenza A(H7N9) viruses circulating among poultry in China have been classified as low pathogenic, and infected poultry have been asymptomatic. Experimental studies have reported that some A(H7N9) virus strains have acquired tropism to bind to receptors that are present in the human upper respiratory tract that could facilitate increased transmissibility among humans.^4,5 On 19 February 2017, the Chinese Center for Disease Control and Prevention (China CDC) reported that A(H7N9) virus with a four basic amino-acid insertion in a host protease cleavage site in the HA protein was identified in two patients with A(H7N9) virus infection in Guangdong province,⁶ suggesting that some A(H7N9) virus strains have evolved to become highly pathogenic in poultry.

A(H7N9) virus has the highest risk score among the 12 novel influenza A viruses evaluated by the Influenza Risk Assessment Tool to date, and is characterized as posing moderate-high potential pandemic risk by the U.S. Centers for Disease Control and Prevention.³ In addition to the on-going monitoring of virological and molecular characteristics of A(H7N9) viruses recovered from infected poultry and humans, continuous epidemiological investigations and clinical severity assessments of A(H7N9) case-patients are critical to inform public health for pandemic preparedness.^7–12 The present study aims to assess changes of basic epidemiology, clinical severity and key epidemiological parameters of time-to-event distributions of A(H7N9) case-patients in the 5^th epidemic wave compared with previous epidemic waves in mainland China.

METHODS

RESULTS

The first epidemic wave of human infections with A(H7N9) virus occurred in the spring of 2013 in mainland China, followed by four successive winter-spring epidemic waves from 2013–14, 2014–15, 2015–16 and 2016–17 with laboratory-confirmed cases of 134, 306, 219, 114 and 447, respectively (Table 1, Figure 1A). The number of A(H7N9) cases identified in the 5^th epidemic wave was the highest among all epidemic waves and the epidemic wave is still on-going (Figure 1A). The heatmap of wave-specific weekly proportion of laboratory-confirmed A(H7N9) case-patients shows that the 5^th epidemic wave started earlier than previous epidemic waves (Figure 1B).

A total of 20/31 provinces have reported the identification of human cases of A(H7N9) virus infection, and cases were concentrated in the Yangtze River Delta in eastern China and Guangdong province in southern China (Figure 2 A–B; Appendix, page 4). Although Sichuan province was the only new province reporting laboratory-confirmed human case of A(H7N9) virus infection in the 5^th epidemic wave, there were 144 newly affected counties (Figure 2C; Appendix, page 5–6). Zhejiang, Guangdong and Jiangsu were the three provinces with the highest cumulative case numbers of 294, 247 and 232, respectively.

The median age of laboratory-confirmed human infections with A(H7N9) virus in the 5^th epidemic wave was 58 years, which was comparable to previous epidemic waves (Table 1). However, the proportion of cases under 16 years was reduced to 1% in the 5^th epidemic wave with an increase in the proportion of cases aged 16–59 years (Table 1, Figure 3A). The difference in age distribution across epidemic waves was statistically significant (Table 1), although the 0–15 years age group was sparse and the 16–59 years age group was the only group with a consistent increasing proportion over the waves. Men consistently accounted for around 70% of all cases during the five epidemic waves (Table 1), while the age and sex distribution varied slightly across epidemic waves (Figure 3 B–G). Over half of the cases were finally admitted into a provincial hospital in the 1^st epidemic wave, while cases were more likely to be admitted into prefecture level hospitals in subsequent epidemic waves (Table 1). Antiviral, antibiotic and corticosteroid treatments, oxygen treatment and mechanical ventilation were commonly administered for A(H7N9) case-patients, and less than 10% A(H7N9) case-patients received ECMO (Table 1). 34%–47% A(H7N9) case-patients died in the five epidemic waves (Table 1). Residence of A(H7N9) case-patients shifted gradually from urban to semi-urban and rural areas from the 1^st to the 5^th epidemic wave with a statistically significant difference in case residence across epidemic waves, although the situation varied by province (Table 1, Appendix page 7).

Around 70% of human cases of laboratory-confirmed A(H7N9) virus infection reported exposure to poultry within the 10 days before the onset of symptoms across five epidemic waves (Table 1). Visiting a live poultry market and exposure to backyard poultry were the two major sources of poultry exposure (Table 1). Although the chance of poultry exposure from visiting a live poultry market was lower for rural cases than urban and semi-urban cases, around 40% of rural A(H7N9) case-patients reported poultry exposure from visiting a live poultry market (Table 1, Appendix page 7). For cases with occupational poultry exposures, semi-urban cases reported a higher proportion of exposure than urban and rural cases in the 1^st, 2^nd, 5^th and total epidemic waves (Table 1, Appendix page 7).

Clinical severity for hospitalized A(H7N9) case-patients increased from the 1^st to subsequent epidemic waves (Figure 4A). A linear increase in fatality risk was observed from the younger age groups to older age groups (Figure 4B). Case-patients aged over 60 years consistently had higher risks for death, death/mechanical ventilation, death/mechanical ventilation/ICU admission than case-patients aged below 60 years (Figure 4 C–E). After adjusting for age, sex, underlying medical conditions, case residence and delay from illness onset to antiviral treatment, the aOR for fatality risk increased by epidemic waves except for the 4^th epidemic wave (Table 2), and trend test for wave was statistically significant (P=0·0342). Case-patients with older age had higher risk for death, and the aORs increased from 0·70 (95% CI, 0·03–6·38) for 0–15 years, 1 for 16–59 years (reference group), 1·84 (95% CI, 1·28–2·65) for 60–74 years to 2·28 (95% CI, 1·43–3·68) for ≥75 years, and trend test for age was statistically significant (P<0·0001). Case-patients with underlying medical conditions or that received antiviral treatment >48 hours after symptom onset consistently had higher risks for death, although the aORs were not statistically significant (Table 2). In the 5^th epidemic wave, case-patients from Guangdong province had a slightly lower risk for death, but this was not statistically significant (Table 2). The median days from illness onset to initiation of antiviral treatment were 5·5, 6·2, 5·4, 4·4 and 4·8 days from wave 1 to wave 5, respectively (Appendix, page 8).

DISCUSSION

Our study presented the temporal pattern and spatial dissemination of A(H7N9) epidemics in China, and provided a comprehensive description of the epidemiology of laboratory-confirmed cases of A(H7N9) virus infection, and assessed the clinical severity of A(H7N9) case-patients during the 5^th epidemic wave. The 5^th epidemic wave occurred earlier, spread to more counties in affected provinces and infected more people compared to previous epidemic waves. Although there was only one additional province (Sichuan province) reporting human infections with A(H7N9) virus in the 5^th epidemic wave as of February 2017, an analysis at the county level suggested much wider geographic spread with many newly affected counties, and the 5^th epidemic wave is on-going. Human infections with A(H7N9) virus shifted gradually from urban to semi-urban and rural areas, although the shift differed by provinces.

The earlier occurrence of the 5^th epidemic wave may be associated with changes in environmental factors such as weather conditions, and more research is needed to elucidate the mechanisms for seasonality. The geographic expansion of human infections with A(H7N9) virus suggests that A(H7N9) viruses circulating among poultry have spread more widely in China since the emergence of human cases in 2013. The majority of human infections with A(H7N9) virus to date have been with a low pathogenic avian influenza (LPAI) A virus until recently as some virus strains have evolved to be highly pathogenic in the 5^th epidemic wave. Since LPAI A(H7N9) virus does not cause illness in infected poultry, prevention and containment can be very challenging. LPAI A(H7N9) virus may be undetected in poultry farms, and wholesale and retail markets, unless active surveillance is conducted.²¹ Moreover, as increasing numbers of live poultry markets were closed in counties where laboratory-confirmed A(H7N9) cases were reported, it might have resulted in movement of ready-for-market poultry infected with LPAI A(H7N9) virus to wider geographic areas without or with less strict implementation of market closure, including previously unaffected counties. Hence, we suspect that LPAI A(H7N9) virus could spread silently along the poultry trade and transportation routes, although we do not have data on trans-regional poultry trade in China.

Laboratory-confirmed cases with A(H7N9) virus infection reported similar low frequencies of exposure to sick/dead poultry across the five epidemic waves. However, the exposure pattern may change in the future depending upon the proportion of low and highly pathogenic avian influenza A(H7N9) viruses circulating among poultry in China. On-going assessment of poultry exposures is essential to study the changes in exposure risk, especially to sick/dead poultry. Although the pattern of poultry exposures for reported human infections with A(H7N9) virus has not changed substantially, the prevalence of poultry infected with A(H7N9) viruses is probably higher than before, as more human infections were reported during the 5^th epidemic wave.

Although the risk for death showed an increasing trend across waves, no statistically significant risk difference was observed when comparing the 5^th epidemic wave to the 4^th epidemic wave. Avian influenza A(H7N9) viruses have undergone some genetic changes,⁶ but our findings suggest that the clinical severity of A(H7N9) case-patients in the 5^th epidemic wave remained similar to the previous epidemic wave. However, if we set the 1^st epidemic wave as the reference group, we could observe the adjusted ORs from the logistic regression model were statistically significant for the 2^nd, 3^rd and 5^th epidemic waves. The increased clinical severity after the first epidemic wave could result from case ascertainment bias. After announcement of the first human case of A(H7N9) virus infection in the 1^st epidemic wave, both the government and the public were extremely concerned about this novel avian influenza A virus. For example, people became more worried than usual if they developed influenza-like illness (ILI), especially in urban areas.²² Laboratory testing for A(H7N9) virus was conducted intensively among suspected cases in the 1^st epidemic wave. It is possible that after 2013, an improved public perception of a lower risk of A(H7N9) virus infection may have changed people’s health seeking behavior. We found that the initiation of antiviral treatment within 48 hours after illness onset was associated with a lower risk for death than the delayed treatment beyond 48 hours since symptom onset. Although antiviral treatment started slightly earlier for A(H7N9) case-patients identified in the latest two epidemic waves than the first three epidemic waves, a reduced fatality risk was not observed in the hospitalized patients detected in recent waves.

Although clusters of human cases have been reported, with epidemiological evidence suggesting some instances of limited human-to-human transmission,¹⁴ there has been no evidence of sustained human-to-human A(H7N9) virus transmission to date. Poultry-to-human transmission remains the major transmission route, and high proportions of laboratory-confirmed cases of A(H7N9) virus infection reported poultry exposure by visiting live poultry markets. Environmental surveillance conducted in Zhejiang province during 2015–16 revealed that 17·3% of the environmental samples collected from live poultry markets were contaminated by H7 subtype viruses.²³ Closure of live poultry markets has been demonstrated to be highly effective in reducing the risk of human infection with A(H7N9) virus by reducing poultry exposures.^24,25 Although some cities have temporarily or permanently closed live poultry markets, there are still reported A(H7N9) cases from these cities. Although closure of large live poultry markets may have been implemented in central urban areas, it is possible that smaller live poultry markets in semi-urban and rural areas may still be operating. Epidemiological investigations revealed that more than 40% of semi-urban and rural cases reported that they had visited a live poultry market before illness onset. Closure of live poultry markets in many areas may have been implemented too late in the 5^th epidemic wave, and only after many A(H7N9) cases had been identified. Regular closure of live poultry markets before the annual winter A(H7N9) epidemic period could be a more effective prevention measure than closure after A(H7N9) cases have been identified.²¹ The expansion of areas with permanent closure of live poultry markets might also help to further reduce A(H7N9) cases in the future.

Zhejiang, Jiangsu and Guangdong were the three provinces with the highest cumulative numbers of reported human infections with A(H7N9) virus since early 2013. We observed a weak association between the incidence of A(H7N9) human infections and chicken/duck density,²⁶ which may be confounded by chicken breeds. Yellow chicken is produced and consumed much more commonly then white chicken in the southern part of the country. Further investigation of factors associated with the persistence of A(H7N9) virus among poultry in these three provinces is necessary to inform prevention and control measures to reduce the risk of human cases. Risk mapping of A(H7N9) virus in the Yangtze River Delta and Pearl River Delta may provide evidence for making future preventive strategies.

Sentinel surveillance for ILI contributes to identification of clinically mild A(H7N9) cases. The ILI surveillance network consists of 554 sentinel hospitals across 31 provinces, covering 2·5% of all hospitals in China. For each sentinel hospital, 10–15 respiratory specimens were collected each week by convenience samples of patients with ILI symptoms. Considering the sampling method currently used for sentinel surveillance, a large number of undetected A(H7N9) cases with mild symptoms could be occuring.¹⁵ Identification of clinically mild cases through ILI surveillance reveals the “iceberg under the sea”, with implications for the estimation of the infection-fatality risk (which may be much lower than the hospitalization-fatality risk estimated in our study) and the total number of A(H7N9) virus infections that have occurred.^15,27 However, the true denominator of human A(H7N9) virus infections to date, including asymptomatic and clinically mild illness in China, to better inform severity assessments, is unknown.

Our study had several limitations. First, our study analyzed only laboratory-confirmed human infections with A(H7N9) virus that were hospitalized and excluded clinically mild cases. Therefore, the full clinical spectrum of human infection with A(H7N9) virus was not assessed. Second, our estimates for clinical severity could be biased if there were undetected hospitalized patients with A(H7N9) virus infection that were not suspected by clinicians and therefore not tested for A(H7N9) virus infection, or due to lack of access to laboratory testing in some areas. Third, it is possible that the cases or their family members may not have accurately recalled and reported exposure to poultry or to live poultry markets, which might affect estimation of the incubation period. Fourth, some cases did not have an exact date of hospital admission reported in the China CDC A(H7N9) case database, which could result in a slightly biased estimation of the clinical severity.

In conclusion, the 5^th epidemic wave of human cases of A(H7N9) virus infection occurred earlier, spread to more areas and infected more people compared to previous epidemic waves. Cases shifted from the elderly to middle-aged adults and from urban locations to semi-urban and rural areas. The clinical severity remains unchanged to date. Continuous monitoring and regular assessments of the epidemiology characteristics and clinical severity of human infections with avian influenza A(H7N9) virus are necessary for pandemic risk assessment.

Supplementary Material