Boar testing station

During our period of investigation between 2009 and 2012, the boar testing station in Norway [19] had a capacity of testing 1152 pigs in 16 separate rooms at the same time. Each room housed a cohort of 72 pigs grouped by breed (Landrace or Duroc) into six groups of 12 pigs and placed in pens of 14 m² in size. The station received weekly, 72 growing pigs (11–12 weeks old with a mean bodyweight of 33 kg) from 46 breeding herds in Norway to monitor their growth performances until they reached a bodyweight of 100 kg. Electronic feeding stations in all pig pens used FIRE (Feed Intake Recording Equipment, Osbourne Ltd, UK) to record daily feed intake and daily weight gain of each pig individually. Pigs fed one at a time ad libitum from one electronic feed dispenser in each pen on conventional concentrate containing 161 g and 136 g digestible protein, 9.68 MJ, and 9.50 MJ net energy/kg before and after 50 kg live weight, respectively, with 1 month of mixing the two feeds to facilitate the feed change.

Study sample

The study sample consisted of 1955 pigs (55% Landrace, 45% Duroc) from 43 breeding herds that were performance tested at the testing station between 2009 and 2012. All pigs were tested for antibodies against influenza A virus by cELISA before leaving the station. During a clinical outbreak of influenza at the boar station between 1^st April 2011 and 31^st July 2011, nasal swabs and blood samples were collected from a total of 375 of these pigs (three pigs per pen) to test for presence of virus and antibodies against the virus, respectively. The testing methods have been described previously [15]. Based on laboratory findings, the 1955 pigs were classified into five infection status groups (INFGP). The seronegative group (SERONEG) with 887 pigs was defined as pigs tested negative for antibodies against influenza A(H1N1)pdm09 virus at the end of their performance testing period. The seropositive group (SEROPOS) with 874 pigs was defined as pigs tested positive for antibodies against influenza A(H1N1)pdm09 virus but with unknown point of infection. Most (n = 859) of these pigs were tested at the end of the performance testing period when they were about 100 kg in weight thus ruling out maternal antibodies in cELISA test results. The virus positive group one (VIR1) with 123 pigs was defined as pigs tested positive for influenza A(H1N1)pdm09 virus by RT-PCR when their bodyweight was between 33 kg and 60 kg (growth phase one or GF1). Of these VIR1 pigs, twenty-two were in the upper weight range of between 51 kg and 60 kg when they were tested positive for the virus. The virus positive group two (VIR2), with 34 pigs, was defined as pigs tested positive for influenza A(H1N1)pdm09 virus by RT-PCR when their bodyweight was between 61 kg and 80 kg (growth phase two or GF2). The virus positive group three (VIR3), with 37 pigs was defined as pigs that tested positive for influenza A(H1N1)pdm09 virus by RT-PCR when their bodyweight was between 81 kg and 100 kg (growth phase three or GF3). Figure 1 is a histogram showing the distribution of the bodyweights of these 194 virus positive pigs when they were tested. These pigs were then classified into three groups to ensure that there were at least 30 pigs in each group.

During the period when the station experienced a clinical outbreak of influenza A(H1N1)09pdm virus, the staff intensified the clinical observation of the 2045 pigs that stayed at the station during this period. The source of the outbreak was not investigated as the infection has become widespread in the Norwegian pig population since its incursion into Norway in 2009 [15].

Statistical analysis

Growth and feeding performances were represented by five outcomes; average daily growth (ADG, weight gain in kg/day), feed conversion efficiency (FCE, kg feed/kg weight gain), average daily feed intake (ADFI, feed intake in kg/day), age of pig when they reached 100 kg (Age100kg) bodyweight; and the overall feed intake (OFI) of the pigs to grow from a mean starting weight of 33 kg to 100 kg. Besides the infection status groups (INFGP; SERONEG, SEROPOS, VIR1, VIR2 and VIR3) as the main predictor of interest, important control predictors were birthdate (BD), breed (Br), ADFI, and GF. For each of the three outcomes of ADG, FCE and ADFI, every pig had three measures aggregated for the three stipulated growth phases. The SERONEG pigs were the reference group for comparison with the SEROPOS group and the three virus positive groups. The data was structured longitudinally on daily measures of growth and feed intake [19] and was hierarchical with three levels. The observations for each outcome were nested in the 1955 pigs, which in turn were nested in the 43 herds. To estimate the negative effects of the virus infection on growth performance, we used multi-level random-intercept regression models to control for other important predictors, while accounting for clustering or random effects at the pig and herd levels [20]. We used analysis of covariance to determine the statistical significance (P value<0.05 for level of inclusion) of predictors for each of the five outcomes. Predictors BD (proxy for improvement over time from 2009 to 2012) and ADFI were continuous variables while INFGP, Br, GF and two interaction terms of GF*INFGP and Br*INFGP for stratification were categorical variables. The INFGP*GF interaction term was included in the regression model to investigate the modifying effect of age of infection on the virus [21]. The INFGP*Br interaction term was tested for statistical significance to see if the impact of the virus infection was different between Landrace and Duroc. The two outcomes, Age100 kg and OFI, were single measures for each pig’s overall performance when they reached 100 kg bodyweight. The predictors for them were the same as for FCE and ADG, but without GF since the measures were aggregated over all the 3 growth phases. To select the best multi-level models for the five outcomes with the selected predictors, we used Akaike Information Criterion (AIC). We selected the model with the lowest AIC value. To determine the significance of additional predictors for each of the five models (one outcome per model), a difference of ±2 of the AIC value was regarded as non-significant and the most parsimonious model was then chosen [22].

Multi-level random-intercept regression models

Where:

Y is one of the five outcomes in this study (ADG, FCE, ADFI, Age100kg, or OFI,). Y _ijk is the value of the response for ith (i = 1,2,3) observation for jth pig (n _j = 1955) nested within the kth (n _k = 43) herd.

β is a vector of coefficients for predictors and their interactions, and X _[i,j,k] is the vector of explanatory variables for the ith observation of the jth pig and kth herd.

u _jk is a vector of random intercepts unique to each pig in each herd, where u _jk ~ N(0, σ²_pig), and v _k is a vector of random intercepts unique to each herd, where v _k ~ N(0, σ ²_herd).

ε _ijk is the vector of error terms where ε _ijk ~ N(μ, σ²).

Predictors

Apart from BD and ADFI that are continuous predictors, the following are categorical predictors:

INFGP: SERONEG; SEROPOS; VIR1; VIR2; VIR3.

Br: Landrace; Duroc

GF: GF1 (33 kg to 60 kg); GF2 (61 kg to 80 kg); GF3 (81 kg to 100 kg).

Quantitative bias analysis by Episens [23] was used to estimate the magnitude of misclassification bias for adverse effects on FCE given that the cELISA test had a respective sensitivity and specificity of 93.7% and 99.1% (manufacturer’s data sheet). A small number of misclassification of the 887 seronegative pigs and 874 seropositive pigs may have been possible. This required dichotomizing the continuous outcome FCE into high and low with median FCE value of 2.54 for the seronegative pigs chosen for the dichotomy.

We used software SAS Enterprise Guide 4.3 (SAS Institute Inc., Cary, NC, USA) and STATA version 12.0 (StataCorp LP, College Station, TX, USA) for data handling and statistical analysis. We plotted the predicted marginal effects of the five infection status groups of pigs (INFGP) for each of our five outcomes based on our regression models by keeping the other covariates at the sample mean values.

Ethics

This was a field study that involved pigs at a commercial pig testing station. No pigs were harmed during the collection of blood samples from the jugular vein or nasal swabs to ascertain their infection status.

Statistical models

ADG and FCE

Tables 2, 3 and 4 show that with SERONEG pigs as the reference, there were statistically significant marginal adverse effects on ADG and FCE in all infected groups (SEROPOS and VIRs 1–3). The marginal plots in Figures 2 and 3 show the predicted means of ADG and FCE while the covariates of BD, ADFI and Br were kept at the sample means. Only GF and INFGP were allowed to vary so that the effect of the virus infection on ADG and INFGP could be studied in the three strata of growth phases. The differences in the predicted FCE and ADG between the five groups of pigs in each of the three growth phases can then be attributed to infection status of the pig.

Average Daily Growth (ADG)
Predictors	Coefficients	SE	P values	95% CI
Breed
Landrace	Reference
Duroc	−0.028	0.006	<0.001	−0.039	−0.016
Growth phase
GF1 (33_60 kg)	Reference
GF2 (61_80 kg)	−0.062	0.0071	<0.001	−0.08	−0.05
GF3 (81_100 kg)	−0.134	0.0083	<0.001	−0.15	−0.12
Birthdate	0.00006	0.000006	<0.001	0.00005	0.00007
Average daily feed intake	0.31	0.0053	<0.001	0.30	0.32
Interaction term: INFGP*GF
SERONEG*GF(1–3)	Reference
SEROPOS*GF1	0.006	0.0064	0.37	−0.01	0.02
SEROPOS*GF2	−0.004	0.0064	0.54	−0.02	0.01
SEROPOS*GF3	−0.015	0.0064	0.02	−0.03	0.00
VIR1*GF1	0.008	0.0127	0.51	−0.02	0.03
VIR1*GF2	−0.033	0.0127	0.01	−0.06	−0.01
VIR1*GF3	−0.058	0.0127	<0.001	−0.08	−0.03
VIR2*GF1	0.002	0.0230	0.94	−0.04	0.05
VIR2*GF2	−0.053	0.0230	0.02	−0.10	−0.01
VIR2*GF3	−0.015	0.0230	0.51	−0.06	0.03
VIR3*GF1	0.014	0.0221	0.52	−0.03	0.06
VIR3*GF2	−0.020	0.0221	0.37	−0.06	0.02
VIR3*GF3	−0.045	0.0221	0.04	−0.09	0.00
Constant (β 0 )	−0.746	0.1113	<0.001	−0.96	−0.53

Abbreviations: SERONEG Seronegative pigs; SEROPOS Seropositive pigs (not tested for virus), VIR1 pigs viral positive between bodyweight 33 kg and 60 kg, VIR2 pigs viral positive between bodyweight 61 kg and 80 kg, VIR3 pigs viral positive between bodyweight 81 kg and 100 kg.

The model is hierarchical with three levels; observations (n = 5865); pig (n = 1955) and herd (n = 43). *denotes interaction term.

Feed Conversion Efficiency (FCE)
Predictors	Coefficients	SE	P value	95% CI
Breed
Landrace	Reference
Duroc	0.0426	0.010	<0.001	0.0230	0.0622
Growth phase
GF1(33_60 kg)	Reference
GF2(61_80 kg)	0.2865	0.011	<0.001	0.26	0.31
GF3(81_100 kg)	0.5442	0.014	<0.001	0.52	0.57
Birthdate	-0.0002	0.00001	<0.001	-0.0002	-0.0001
Average daily feed intake	0.0238	0.010	0.019	0.004	0.044
Interaction term: INGP*GF
SERONEG*GF(1-3)	Reference
SEROPOS*GF1	-0.015	0.011	0.159	-0.037	0.006
SEROPOS*GF2	0.002	0.010	0.849	-0.018	0.022
SEROPOS*GF3	0.029	0.010	0.004	0.010	0.049
VIR1*GF1	-0.001	0.023	0.951	-0.046	0.043
VIR1*GF2	0.058	0.020	0.004	0.018	0.097
VIR1*GF3	0.125	0.020	<0.001	0.086	0.165
VIR2*GF1	-0.049	0.037	0.189	-0.121	0.024
VIR2*GF2	0.122	0.036	0.001	0.051	0.192
VIR2*GF3	0.033	0.036	0.361	-0.038	0.103
VIR3*GF1	-0.016	0.037	0.659	-0.089	0.056
VIR3*GF2	0.045	0.035	0.196	-0.023	0.112
VIR3*GF3	0.091	0.035	0.008	0.023	0.159
Constant (β 0 )	4.6609	0.1934	<0.001	4.282	5.040

Abbreviations: SERONEG Seronegative pigs, SEROPOS Seropositive pigs (not tested for virus), VIR1 pigs viral positive between bodyweight 33 kg and 60 kg, VIR2 pigs viral positive between bodyweight 61 kg and 80 kg, VIR3 pigs viral positive between bodyweight 81 kg and 100 kg. The model is hierarchical with three levels; observations (n = 5865); pig (n = 1955) and herd (n = 43). *denotes interaction term.

For SEROPOS pigs, the negative effect on growth performance was seen during GF3 where FCE was reduced by +0.029 kg feed/kg weight gain. Correspondingly, ADG also decreased by −0.015 kg/day.

For VIR1 pigs, adverse growth performance effects were not seen until GF2 even though these pigs were positive for virus during GF1. During GF2, FCE was reduced by +0.058 kg feed/kg weight gain, which led to a lower ADG by −0.033 kg/day. These negative effects extended into GF3 where FCE was reduced by +0.125 kg feed/kg weight gain and a corresponding lower ADG by −0.058 kg/day.

For VIR2 and VIR3 pigs, the negative effects were confined to the same growth phase that they were positive for virus. For VIR2 pigs during GF2, FCE was reduced by +0.122 kg feed/kg weight and ADG was lower by −0.053 kg/day correspondingly. The FCE and ADG of VIR2 pigs returned to the same levels as seronegative pigs during GF3. For VIR3 pigs during GF3, the FCE was reduced by +0.091 kg feed/kg weight gain and ADG was lower by −0.045 kg/day correspondingly.

Average daily feed intake

Age of pigs at 100 kg

Consequent to the reduced FCE and hence lower ADG, the virus positive pigs required longer time to reach the bodyweight of 100 kg (Figure 4 and Table 6). The VIR1, VIR2 and VIR3 pigs were slower by 1.6 days, 1.8 days and 2.4 days, respectively, in reaching 100 kg bodyweight. Although VIR2 was statistically insignificant (P value = 0.12), it was on similar trajectory as VIR1 and VIR3.

Overall feed intake

Consequent to the reduced FCE experienced by the four infected groups, Figure 5 and Table 7 show that all four infected groups needed additional feed to reach 100 kg bodyweight. The SEROPOS, VIR1, VIR2 and VIR3 pigs needed 2.3 kg, 8.0 kg, 5.9 kg and 7.2 kg additional feed, respectively.

Other predictors (covariates or control variables) in the models

The predictors birthdate, breed and growth phase were highly significant (P value<0.001) in our regression models for all five outcomes). For birthdate as a predictor, the outcomes of ADG, FCE, ADFI, Age100kg and OFI of pigs improved each year or every 365 days by 0.022 kg bodyweight/day, −0.055 kg feed/kg bodyweight, −0.007 kg feed/day, −2.2 days and −6.2 kg feed respectively.

On breed differences, Landrace pigs grew faster (ADG of +0.028 kg/day), had better efficiency in feed conversion (FCE was improved by −0.043 kg feed/kg bodyweight), reached the targeted weight seven days earlier, and required less feed (−5.9 kg feed) to reach 100 kg bodyweight.

Clinical observations between April and July 2011

During a clinical outbreak at the station from April 2011 to July 2011, a group 2045 pigs present at the station were monitored closely for clinical signs. These pigs entered the station at different time since the station received 72 new pigs on a weekly basis. Influenza-like illness was observed in 137 pigs, giving a crude morbidity of 7%. Clinical signs observed were transient anorexia or lethargy (45%), respiratory signs (coughing, laboured breathing or nasal discharge, 39%), and pyrexia (above 39°C, 27%). The prevalence of these clinical signs recorded within each of the 16 rooms ranged from 0% to 17%. For group specific morbidity, clinical signs were not detected in VIR1 pigs, while one (3%) out of 34 VIR2 pigs and five (14%) out of 37 VIR3 pigs had clinical signs reported. In the seropositive group, 39 (10%) out of 381 pigs had clinical signs of influenza-like illness during the testing period. No clinical signs were reported in the SERONEG pigs.

Bias analysis on misclassification

The cELISA test had a sensitivity and specificity of 93.7% and 99.1% (Manufacterers data sheets). Although these values are considered high for test performance, given the largely lack of clinical picture in this disease for corroboration, there could nevertheless be a small number of SERONEG and SEROPOS pigs that were misclassified and hence biased the adverse effects towards the null. Our quantitative bias analysis using Episens [23] for adverse effects on a dichotomized FCE showed that the odds ratio for a poorer FCE if the pig was SEROPOS versus SERONEG was 1.13. The odds ratio after adjusting for misclassification bias was 1.3, a change of 2 percent. The small bias of 2 percent in our study was towards the null (OR= 1).

The literature [7] states that SIVs in general, cause near if not 100% morbidity when they infect naïve pigs. An experimental study with influenza A(H1N1)pdm09 virus by Brookes et al. in 2010 [9] reported 100% morbidity involving 19 pigs in their experimental study, where 2 pigs were infected by contact. In contrast, we found in our field study that influenza A(H1N1)pdm09 virus infection of Norwegian pigs was largely subclinical, with only six (<4%) in a sample of 194 virus positive pigs reported to have clinical signs. Apart from coughing, clinical signs, like nasal discharge, were so mild that only through closer observation or handling of the pig, for example nasal swabbing, could the signs be detected. The mild clinical signs and low morbidity recorded during the observation period were similar to the morbidity experienced by other Norwegian pig farms infected with influenza A(H1N1)pdm09 virus for the first time [15],[16]. This shows that despite Norwegian pigs having no cross-protective immunity [27] to other strains of SIVs, the A(H1N1)pdm09 virus experienced in Norwegian pigs appeared to be of low pathogenicity that caused no or only mild clinical signs.

In recording the clinical signs at the pig testing station, possible bias of focusing on pigs with anorexia (shown on the computer records) may have led to underestimating the morbidity of the disease because we found no statistically significant decrease in appetite in the infected pigs in the stipulated three growth phases. A transient drop in appetite for one or two days would be masked by the three growth phases since the intervals in each growth phase were longer than two days and if there were compensatory increases in feed intake following one or two days of depressed feed intake.

Favourable conditions in our study

Pigs in our study did not have co-infections of other subtypes of influenza A viruses, M. hyopneumoniae, PRRSV, Aujesky’s disease virus and porcine respiratory coronavirus, given Norway’s disease free status for these pathogens [12]. Secondly, the daily recordings of feed intake and bodyweight for each pig were computerised without the presence of human interference. This ruled out human bias or error in making measurements to provide accurate calculations of the performance parameters. Thirdly, the repeated measures allowed the study of progressive effects of the virus in pigs infected at different ages and also the duration of the negative effects on growth performance.

Validity of study design and statistical models

With 34 or more pigs in each of the five groups of pigs, the sample sizes are considered large for multi-level models [28]. The five statistical models were based on maximum likelihood in estimating the predictors that allow for inference to the Norwegian pig population.

It was appropriate to use multi-level analysis because of the hierarchical nature of the data. The data, having 5865 observations nested in the 1955 pigs, which in turn were nested in the 43 herds, were handled to account for the variations between individual pigs and between the various herds including unmeasured confounders like other infections at the herd or individual level. Pigs from the study sample of 43 herds were represented in the reference group of 887 seronegative pigs. Consequently, the effects of virus infection (primary predictor of interest) and the known covariates (predictors we wanted to control) were more accurately estimated. Such hierarchical models solved some of the problems mentioned in a similar study by Straw et al. [18], in that this study was designed to control heterogeneities due to the environment, herd health status, host characteristics and management conditions inherent at the pig and herd level to reduce if not eliminate confounding [29],[30]. Furthermore, keeping the pigs at one location in a uniform environment and husbandry eliminated these factors as potential confounders in our model. Our models also proved to have a relatively high explanatory ability on the variance as the achieved adjusted R² were 51% (ADG), 59% (FCE), 51% (Age100 kg), 66% (ADFI) and 20% (OFI), which were proportions of variation that were explained by the predictors in the models. The longitudinal nature of the data for each pig allowed the statistical models to account for changes to FCE, ADFI and hence ADG with respect to the stage in the pig’s growth phase by including growth phase (GF) as a dummy variable in the statistical models thus controlling for confounding due to normal variation of feed conversion efficiency and daily feed intake with stage in growth phase.

Our samples of 1955 pigs included pigs tested at the station over four years from 2009 and 2012. We saw in our models that birthdate was a significant covariate because pigs born later had better growth performance as a result of improvement over time due to genetic selection, improved feed, and management improvement. Pigs belonging to the five infection status groups were disproportionately distributed over these four years since all 194 virus positive pigs were sampled in a single year (2011) while 560 pigs from seronegative and seropositive group were sampled in 2009 and 2010. Despite these disproportions, we were able to account for the marginal effects attributed to improvement over time by including birthdate as a covariate in our multi-level regression models. This allowed us to increase the study sample and hence the power of our study.

As normal occurrence and also seen in our study, the feed intake increases and FCE declines as a pig grows. Despite the reduced FCE in an older pig, its ADG is still higher because it consumes a higher amount of feed than the younger pig. Hypothetically, if older pigs ate the same amount of feed as the younger pigs, their ADG would be lower because of a reduced FCE as depicted by the coefficients of GF for outcome ADG and FCE in Tables 2, 3 and 4. This is also depicted in our marginal plots at Figures 2 and 3. This again underlines the importance of having growth phase as a dummy variable in our models to ensure comparisons of our outcomes between the 5 infection-status groups were valid because comparisons between groups were made in the same growth phase.

The coefficients of covariates breed, birthdate, growth phase and average daily feed intake in all five models were useful for the validation of our statistical models by comparing their values with other sources of pig performance data. An improbable coefficient would have raised a red flag on the models.

All five outcomes in our models were correlated. The calculation of feed conversion efficiency was based on average daily feed intake and average daily growth recordings. They in turn determined the remaining 2 outcomes on overall growth performance, which were age of pig at 100 kg bodyweight and overall feed intake. These latter 2 outcomes on overall growth performance are especially useful in evaluating economic consequences of the infection for farmers. Cost of extra feed and a delay in getting the pig to the market will lead to higher overheads (feed and veterinary costs) and lower income for the farmer since fewer pigs are sold in the fixed time period. We found that pigs infected when they were young (33 kg - 60 kg) required an additional 8 kg feed and were 1.6 days slower in reaching 100 kg bodyweight. Farmers can estimate the added operational costs if they know that their pigs were infected at a young or older age.

In other parts of the world, SIVs seldom act alone, but with concurrent infections to cause porcine respiratory disease complex where SIVs are the most common primary pathogens [1],[2],[7],[31],[32]. Hypothetically, the severity of influenza A(H1N1)pdm09 virus infection in terms of growth performance would be aggravated by concurrent infection of these other respiratory pathogens [7]. On the other hand, these pigs could also be protected from influenza A(H1N1)pdm09 virus infection because of presence of protective immunity against other strains of SIVs [27].