A review of estimated transmission parameters for the spread of avian influenza viruses

Abstract

Avian influenza poses an increasing problem in Europe and around the world. Simulation models are a useful tool to predict the spatiotemporal risk of avian influenza spread and evaluate appropriate control actions. To develop realistic simulation models, valid transmission parameters are critical. Here, we reviewed published estimates of the basic reproduction number (R ₀), the latent period and the infectious period by virus type, pathogenicity, species, study type and poultry flock unit. We found a large variation in the parameter estimates, with highest R ₀ estimates for H5N1 and H7N3 compared with other types; for low pathogenic avian influenza compared with high pathogenic avian influenza types; for ducks compared with other species; for estimates from field studies compared with experimental studies; and for within‐flock estimates compared with between‐flock estimates. Simulation models should reflect this observed variation so as to produce more reliable outputs and support decision‐making. How to incorporate this information into simulation models remains a challenge.

1. INTRODUCTION

Although highly pathogenic avian influenza (HPAI) has been recognized as a disease for more than a century, large‐scale HPAI outbreaks have only been observed since 1996, and recently outbreaks appear to be increasing (EFSA 2022, King et al. ²⁰²²). The 2005–2006 HPAI pandemic caused by H5N1 spread throughout Asia and into Europe and Africa, affecting more than 60 countries (Alexander et al. 2009). HPAI outbreaks have resulted in the death and culling of hundreds of millions of poultry worldwide. There are additional concerns regarding human infections with avian influenza viruses (AIVs); for example, of 903 human cases of H5N1 reported between 1997 and 2015, the overall case‐fatality risk was 53.5% (Lai et al. 2016). During the period 2015–2020, a total of 6412 HPAI outbreaks were reported worldwide; of these, 39% were due to H5N1 and 61% due to H5N8 (Amer et al. 2021), with H5N8 outbreaks increasing in occurrence from 2017 onwards. H5Nx HPAI variants (H5N1, H5N5, H5N8) continue to emerge within the Eurasian region in wild birds and poultry and threatens poultry production, food security and veterinary public health (Lewis et al. 2021). In Asia, since 2008, novel H5 subtypes have evolved from clade 2.3.4 into separate clade groups containing H5N2, H5N5, H5N6 and H5N8 types. Outbreaks have been reported in Asia, Europe, Africa and North America.

Prevention and control of HPAI outbreaks is a priority within the commercial poultry sector. For example, an outbreak of HPAI H7 in northern Italy in 1999, which apparently mutated from a LPAI H7N1 virus, affected more than 13 million poultry and caused huge economic losses to the Italian poultry industry with severe social and economic implications (Capua et al. 2002). The economic losses associated with this epidemic were estimated to be Euro 507 million (Sartore et al. 2010). An epidemic of HPAI H7N7 occurred in The Netherlands in 2003. It led to the culling of 30 million poultry (Stegeman et al. 2004). A 2004 outbreak of H5N1 HPAI in Thailand caused the death or slaughter of 60 million poultry and the disruption of poultry production and trade (Songserm et al. 2006).

HPAI viruses comprise the subtypes H5 and H7 and are thought to arise in poultry via mutation, following infection with low pathogenic avian influenza (LPAI) viruses (Alexander et al. 2009). However, since 2004, HPAI has been found to circulate in wild bird populations; this indicates a potential for direct transmission of HPAI to poultry flocks from wild birds (Smith et al. 2009). In 2016–2017, many European countries experienced a large epidemic of HPAI H5N8 (Scolamacchia et al. 2021), the largest HPAI epidemic ever recorded in Europe (Lycett et al. 2020). This might have resulted from an unusually high frequency of reassortments between H5 HPAI viruses and co‐circulating LPAI viruses, derived from both wild birds and domestic poultry. It suggests that there is an ongoing risk of emergence of reassortant viruses from wild bird populations that can be highly pathogenic for poultry (Lycett et al. 2020).

It is still unknown whether most AIV spread is via wild birds or within the poultry sector (Alexander et al. 2009). However, in an analysis of 52 individual H5N1 introduction events, the majority (20 out of 23) in Europe were found to be most likely via migratory birds, whereas this appeared to be minor in Asia (three out of 21) and of less importance in Africa (three out of eight) (Kilpatrick et al. 2006). In the case of such introductions, the interface between wild birds and domestic poultry is season, species and location specific (Hill et al. 2021).

The relative contribution of trade and the market chain versus wild birds in spreading AIV needs further investigation (Osmani et al. 2014). While the role of animal movements in AIV spread within national trade networks has been largely identified, the risk from human and fomite movements is still largely unknown. Although studies have identified several possible transmission pathways for between‐farm transmission, including human movements between poultry houses and other poultry farms (Ssematimba et al. 2013), empirical evidence from the field is lacking (Hautefeuille et al. 2020). In Romania, migratory birds appeared to have been responsible for the introduction of the H5N1 virus and its initial spread in the Danube Delta region in the autumn and winter of 2005, but the movement of poultry might have introduced the infection into central Romania during the spring and summer of 2006 (Ward et al. 2008). Identification of such a shift in the primary mechanism of disease spread is important for designing disease surveillance and control programs (Farnsworth and Ward 2009).

Simulation models can be used to predict the spread of disease outbreaks, and help implement the optimal control strategy (Kirkeby et al. 2021). In addition to predictive models for between‐flock transmission, within‐flock models are useful for decision makers when acting on specific outbreaks, as well as in combination with between‐flock simulation models (Central Veterinary Institute et al., 2017). Several models have been developed to predict the risk and spread of AIV within poultry populations, so as to inform decision‐makers (Stegeman et al. 2010, Stevens et al. ²⁰¹³). For example, Sharkey et al. (2008) concluded that there was significant potential for widespread infection throughout the British poultry industry, and that duck farms potentially play a key role in transmission. Likewise, Backer et al. (2015) concluded that culling only infected poultry flocks was not sufficient to stop a Dutch HPAI outbreak, and higher costs were associated with preemptive culling of nearby farms. However, the usefulness of such models depends on valid parameterization. One previous study collated parameter estimates for within‐flock transmission based on experimental studies, and found some differences in the basic reproduction number and infectious period between bird species (Central Veterinary Institute et al., 2017). However, in general, models have been calibrated using data collected from specific outbreaks. For example, in a simulation study of HPAI H5N8 in south‐west France 2016–2017, Andronico et al. (2019) inferred latent and incubation periods of 1 and 7 days, respectively, based on the best fit of a mathematical model to the observed outbreak data. The latent period is a critical parameter for simulation models, determining the speed of spread of the outbreak. Furthermore, as pathogenicity depends on the infected species and the virus strain, the latent period can potentially vary greatly between outbreaks. Thus, it is important to continuously gather and update knowledge of these key parameters.

The aim of this study was to review published literature estimates of the basic reproductive number, latency and infectiousness period, with regards to the spread of AIV within and between poultry flocks, and to assess the variation in these estimates by virus type, pathogenicity, species, study type and poultry flock unit. Since we are concerned with model parameterization, our focus was on describing the distributions of key parameters rather than estimating average values, so meta‐analysis methods were not used. Our goal is to better inform AIV disease spread models so as to improve decision‐making for the control and prevention of avian influenza.

2. METHODS

Peer‐reviewed publications that report estimates of AIV transmission in domestic poultry were identified via a search of the Web of Science database. The primary search was conducted using the search terms ‘Avian Influenza’ AND ‘Mathematical modelling’. No restrictions were applied to the date of publication or publication type, but the search was restricted to English language publications.

Each publication identified in the search was screened (MPW) on title and abstract for information on quantitative estimates of AIV transmission, specifically transmission rate, basic reproductive number, period of latency or period of infectiousness. For inclusion in further analysis, publications needed to report primary estimates of at least one of these four parameters. Those publications presenting only information on wild birds were excluded at this step.

Next, the reference lists of each publication were searched for additional publications which might meet the above criterion for inclusion. The titles and abstracts of these secondary publications were screened (MPW) in the same manner as the primary publications.

The full‐length articles of all retained publications included were then read by both authors independently, and data was extracted on the following: year of publication, country in which the study was undertaken, AIV type, virus pathogenicity (high or low), poultry species, study unit (within or between flock), estimation method, rate of transmission (β, per day), basic reproduction number, period of latency (days) and period of infectiousness (α, days). For these last 4 parameters, 95% confidence/credible intervals were recorded, if reported. If a publication reported more than one estimate for any of these parameters, all estimates were included separately. Parameters were considered to be within‐flock parameters if only a single poultry production unit or one village (in the case of backyard poultry) were described. Between‐flock parameters were estimated in studies in which more than one population was described, and where the methods supported the estimation of between‐flock parameters.

In studies in which estimates of transmission coefficients (β) were presented and either estimated or assumed period of infectiousness (α) was provided, R ₀ was calculated as α × β and these estimates R ₀ were included in the study. In studies in which data was reported which allowed the period of latency to be estimated, such as serial daily testing in experimental studies, these inferred estimates were also included in the study.

In addition to exclusion of estimates of parameters made from wild bird populations, estimates based on recombinant virus strains, estimates from experimental studies in which birds had been vaccinated and field estimates from periods in which active disease control strategies had been implemented were excluded. No exclusions were applied regarding the diagnostic method used to determine infection status (such as virus isolation versus PCR).

The accuracy of data extracted was verified by comparison of each of the two authors’ independent reviews and data extraction. In cases in which differences were found, data extraction proceeded via discussion until a consensus was reached.

The data extracted was tabulated. The parameter estimates were described by box‐and‐whisker plots and frequency distributions. Univariable comparisons between key parameters and key study variables (virus type, pathogenicity, unit and species) were done using a Kruskal–Wallis test.

3. RESULTS

The primary literature search yielded 37 articles (36 journal articles [28 journals] and one book) published between 2006 and 2021. Applying exclusion criteria based on article titles and abstracts, nine articles were retained. On review of the reference lists of these nine articles, an additional 17 articles were identified that met the inclusion criteria. Therefore, a total of 26 articles (25 journal articles, one conference proceeding; 2003–2019; 11 journal titles) were included for data extraction. Half of these articles were published in Preventive Veterinary Medicine (five), Veterinary Microbiology (four) and Epidemiology and Infection (four). More than half (16) of the articles were published between 2007 and 2012. Articles reported data mostly on HPAI (18); types H5N1 (nine), H7N1 (seven) or H7N7 (five); within flock estimates (17); chickens (11) or mixed poultry (eight); and used field data to estimate parameters (15). The 26 articles included in analysis, and the data extracted from these articles, are shown in Supplementary Data.

Overall, the 26 articles generated a total of 100 R ₀ estimates. Similar estimates of R₀ were made between experimental (47) and field (53) studies. More estimates were based on high (63) than low (37) pathogenic AIVs, and on within‐flock transmission (70) than between flock transmission (30). The most common virus type for which R ₀ estimates were made was H5N1 (40), followed by H5N2 (16), H7N1 (15), H7N7 (11), H7N3 (seven) and H5N3 (one). Four estimates did not distinguish between two AIV types (H5N2 and H5N8 [three], H7N1 and H7N3 [two]) and six did not specify the AIV type. The majority of estimates were based on chickens (56), compared to poultry (28), turkeys (13) and ducks (three).

3.1. Basic reproduction number

The distribution of R ₀ estimates was right‐skewed (Figure 1). The median estimated R ₀ was 2.21 (IQR, 1.03–5.28), with estimates ranging from 0.17 to 208. The median (IQR) estimate for within‐flock transmission was 2.45 (1.2–5.55), and for between flock transmission was 1.30 (0.94–2.54) (Figure 2). For within‐flock transmission, 83% of the 70 estimates were >1, whereas for between‐flock transmission only 60% of the 30 estimates were >1.

FIGURE 1.

Distributions of estimates of the basic reproduction number (R ₀) for avian influenza virus transmission from a review of 26 published articles

FIGURE 2.

Distributions of estimates of the basic reproduction number (R ₀) for avian influenza virus transmission within (left) and between (right) flocks from a review of 26 published articles.

Significant differences (Kruskal–Wallis test statistic, p value) in R ₀ estimates were identified for virus type (21.177, p = .0070), species (19.045, p < .001) and unit (5.316, p = .021) (Table 1). The estimated median R ₀ was significantly higher for H7N3 (5.6) than H5N1 (2.18), H7N1 (1.9) and H5N2 (1.03). The estimated median R ₀ was significantly higher for both ducks (14.9) and turkeys (5.5) than chickens (2.19) and poultry (1.08). The estimated median R ₀ was significantly higher for within‐flock transmission (2.45) than between‐flock transmission (1.3). No significant difference was found for pathogenicity (0.888, p = .346) or study type (0.996, p = .318).

TABLE 1.

Estimates of the basic reproduction number (R ₀) of avian influenza virus identified in a review of 26 published articles, categorized by key variables

Variable	Number of estimates	Median R 0	Inter‐quartile range	Kruskal–Wallis statistic	p value
Virus type				21.177	.0070
H5N1	40	2.18	0.99−3.40
H5N2	16	1.03	0.62−1.17
H7N1	15	1.9	1.50−3.01
H7N7	11	1.9	1.10−7.80
H7N3	7	5.6	4.70−9.10
H5N3	1	−	−
H5N2, H5N8	3	0.94	−
H7N1, H7N3	1	−	−
Not specified	6	5.45	3.95−6.80
Pathogenicity				0.888	.346
Low	37	3.70	1.06−5.60
High	63	2.18	0.97−3.40
Species				19.045	<.001
Chickens	56	2.19	1.06−4.0
Poultry	28	1.08	0.94−2.40
Turkeys	13	5.5	3.01−7.80
Ducks	3	14.9	−
Study type				0.996	.318
Experimental	67	1.77	1.0−5.30
Field	33	2.40	1.03−5.50
Unit				5.316	.021
Within flock	70	2.45	1.20−5.55
Between flock	30	1.3	0.94−2.54

3.2. Latency period

In eight articles, the period of latency was either estimated (seven) or could be inferred from the data reported. In these articles, the assumption regarding period of latency was derived either from previous publications, the authors’ opinion, or the source was not stated. A total of 29 estimates of R ₀ were made in these studies. Of these estimates, the study type was mostly experimental (25), and all were focused on within‐flock transmission. The median latency period was 1.0 days (IQR, 1.0–1.9 days) (Figure 3). Of the 20 R ₀ estimates made, in most cases (18) the assumed latency period was made in field studies (18) rather than experimental studies (two). Also, in most cases when the latency period was assumed this was for the purposes of estimating between‐flock transmission (14) rather than within‐flock transmission. For the former, latency periods of either 2 (four estimates) or 7 (10) days was assumed, whereas for the latter the latency period was assumed to be either 0 (one) or 1 (five) day.

FIGURE 3.

Distribution of estimates of the latency period (days) of avian influenza viruses for transmission within flocks from a review of 26 published articles. No literature estimates for between‐flock latency were found.

3.3. Period of infectiousness

A total of 45 estimates of period of infectiousness were reported, 29 based on experimental studies and 16 based on field studies. There were 35 estimates for within‐flock transmission and 10 for between‐flock transmission; median (IQR) estimates of period of infectiousness were 6.35 days (1.76–7.73) and 9.6 days (8.05–12.15), respectively (Figure 4). The mean infectious period for LPAI studies was 6.2, while it was 7.7 for HPAI studies. In eight articles, a period of infectiousness was assumed. All of the 25 R ₀ estimates made were from field studies, and assumptions regarding period of infectiousness were made equally for within‐flock estimates of transmission (12) and between‐flock estimates (13). For within‐flock estimates, assumed period of infectiousness was either 1 (two estimates), 2 (four), 3 (two) or 4 (four) days. For the between‐flock estimates it was either 1 (two), 2 (four), 3 (two) or 4 (four) days.

FIGURE 4.

Distributions of estimates of the period of infectiousness (days) of avian influenza viruses for transmission within (left) and between (right) flocks from a review of 26 published articles.

4. DISCUSSION

We found a large variation in the estimated R ₀, latency and infectiousness between the studies, supporting a previous study of estimates restricted to experimental data (Central Veterinary Institute et al., 2017). These parameters are crucial when planning the response to an outbreak, especially if the response is based on simulation studies. Among the virus types included, H5N1 and H7N3 had the highest estimated median R ₀ (Table 1), which might highlight the importance of the virus type in an outbreak situation. However, for some types the number of estimates were sparse, and therefore more studies should be conducted to investigate this variation. Whilst many studies on H5N1 transmission were found and 40 estimates were identified, estimates of H5N8 transmission were scarce. This HPAI type caused the largest epidemic on record in Europe in 2016–2017 (Scolamacchia et al. 2021; Lycett et al. ²⁰²⁰), and so there is a research gap in terms of modelling H5N8 spread in Europe. Other estimates, for example an R ₀ of 208 for H7N7, appear as outliers (Goot et al. 2005). That estimate might be biased by the experimental study design and a small sample size (experiment using 10 infected chickens and 10 contact chickens that all acquired H7N7). The same might be the case for estimates by Goot et al. (2005) who used five inoculated chickens and five contact chickens in different experiments and estimated an R ₀ of up to 31.7. Likewise, Marquetoux et al. (2012) estimated an R ₀ of up to 22.4 for between poultry farm spread of H5N1 in Thailand. These estimates were based on an exponential growth model with assumed latency and infectiousness periods; other analytical approaches might generate lower R₀ estimates. Also, in field studies such as that of Marquetoux et al. (2012), estimates likely can be biased by suboptimal surveillance, reporting bias and diagnostic bias. Another source of bias in field studies is the expected stochasticity in the context of outbreaks; some subtypes might have had better conditions for spread than others, by chance. For instance, housing style has previously been considered to be a risk factor, and likely differs between studies, countries and species (Sergeant et al. 2022, Scott et al. ²⁰¹⁸). An experimental study by Niqueux et al. (2014), which was included in the present study, identified large differences between strains and subtypes (Niqueux et al. 2014), which was also found in field data from Garske et al. (2007). This indicates that there might be some real differences between subtypes. In addition, for experimental studies included in the current review in which conditions are controlled and estimates are less affected by bias, there is also an apparent difference in median R ₀ estimates: 2.20 for H5N1 and 2.04 for H7N1 compared with 1.35 for H7N7 and 1.0 for H5N2. Estimating and comparing differences in transmissibility between AIV subtypes is challenging. Regardless, such variability in parameter estimates, if it exists, should be captured by simulation models of AIV spread so as to produce realistic model outputs. Therefore, more epidemiological and virological studies are still needed to investigate if the previous estimates are representative of the underlying properties of each subtype.

We also found a lot of variation between poultry species. Ducks had a much higher R₀ than other species, supporting other studies that found ducks in the wild are generally more affected than other species (Sharkey et al. 2008, Kjær et al ²⁰²², Vergne et al. ²⁰²¹). This is also supported by the previous findings of Iqbal et al. (2013), who conducted an experimental study and found considerable variation in the titres between bird species, with highest titres estimated in quail. Similarly, Criado et al. (2021) found differences in the infectiousness of AIV between turkeys and chickens, with viremia higher in turkeys than in chickens in the case of LPAI. In contrast, a previous review of experimental data found that the range of R ₀ for chickens, turkeys and ducks overlapped considerably, indicating that there might be no real difference (Central Veterinary Institute et al., 2017). The different R ₀ estimates between species should be utilized when designing the response to outbreaks: biosecurity could be focused on high risk species such as ducks and turkeys, while chickens and poultry in general might exhibit a reduced risk of transmission. It should be noted that in some studies reviewed, species‐specific details and results were not reported and for these we classified the species as ‘poultry’. Better reporting of this information in published field studies, in particular, would facilitate parameterization of disease spread models.

Overall, we found no significant difference between the estimated R ₀ for pathogenicity (0.888, p = .346) or study type (0.996, p = .318). If a comparison is made for pathogenicity only estimated within experimental studies, for HPAI the median R ₀ is 2.20 versus 2.04 for LPAI. In an experimental study conducted by Criado et al. (2021) using turkeys and chickens, the amount and duration of H7N3 viral shedding was greater for HPAI than LPAI strains. Whether HPAI strains have a higher R ₀ than LPAI strains in the field is uncertain. Potential study design bias in HPAI studies (for example, culling as a control method in HPAI field outbreaks and underestimation of the period of infectiousness) needs to be considered. Nearly two‐thirds of the studies reviewed focused on HPAI, so it appears that investigation of the transmission of LPAI is relatively under‐researched. Given the propensity for H5 and H7 LPAI viruses to mutate into HPAI, this is a research gap. In addition, uncertainly regarding transmission rates for HPAI versus LPAI has implications for the parameterization of disease spread models.

As expected, we observed higher within‐flock transmission than between‐flock transmission (Figure 2 and Table 1). Transmission between flocks is dependent on a large number of external factors including climate, wild bird species, poultry density and human‐mediated transmission between flocks

Statistical comparison of the estimated parameters may be biased by confounding factors in each study, such as housing systems and densities. Some of these confounding factors are specific for, for example, bird species, and therefore, the estimates represent a combination of the species properties and the facilities they are kept under. This is still useful for simulation models that aim to provide a realistic estimate of species‐specific scenarios. However, when simulating alternative scenarios such as implementing control actions, this must be taken into account. It should also be noted that identifying risk factors for AIV spread is a different goal to estimating parameters to incorporate into disease spread models. For the latter, unadjusted parameter values are needed because such models (if well specified) will already include all important factors that influence disease spread. To accurately parameterize such a model, unadjusted parameter estimates are preferred. Furthermore, many of the estimates in this study were inferred from data that included culling, reflecting realistic situations where control actions are put into action after discovery of an outbreak (see Supplementary Data).

In this review, we identified (or inferred) 100 estimates of R ₀, 28 estimates of the latency period and 45 estimates of the period of infectiousness. Most (117 out of 173) of these reported point estimates were accompanied by 95% confidence or credible intervals (Supplementary Data). In the case of R ₀, 95% confidence or credible intervals were reported for even more of the point estimates (76 out of 100). For LPAI, the mean infectious period was 6.2 and for HPAI it was 7.7, possibly indicating a small difference between these two groups. While it is practically impossible to measure the duration of infectiousness for HPAI in field studies given the high mortality and often mandatory culling of infected poultry flocks (see Supplementary Data), it is more feasible for LPAI. However, it is still challenging to estimate the period of infectiousness when observing at the flock level, highlighting the need for more experimental studies with each circulating strain. Such estimates of variability are useful when parameterizing disease spread models (Kirkeby et al., 2021). For example, probability functions fit to distributions of parameters that include both point and confidence/credible limits could be used to provide more realistic model outputs. So for example, a total of 252 estimates of R ₀ could be used rather than just the 100 point estimates. Alternatively, these confidence/credible limits could be used to define distributions that can be used in model sensitivity analyses. Enriching models based on the published literature provides better information on which to make decisions. How to maximize the value of the information collected within simulation models needs to be considered.

5. CONCLUSIONS

The estimated transmission parameters for AIV differ greatly between published studies. This highlights the importance of including the context – such as the poultry species and virus type involved – when considering the response to an outbreak. Moreover, it is important to include the estimated variation in simulation models for the spread of avian influenza, so as to develop appropriate policy and respond to AIV outbreaks properly.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

FUNDING

M. P. W. and C. K. were funded by Erasmus+ Staff Mobility program, grant number KA107‐2019‐007. C. K. was also funded by the ENIGMA project, a veterinary contingency project funded by the Danish Veterinary and Food Administration.

Supporting information

Supporting Material

Kirkeby, C. , & Ward, M. P. (2022). A review of estimated transmission parameters for the spread of avian influenza viruses. Transboundary and Emerging Diseases, 69, 3238–3246. 10.1111/tbed.14675

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available in the supplementary material of this article