This study was case-control study with a herd as the unit of interest. A case-control study design was selected to find a clear difference in an outcome that was originally measured in a continuous scale (percentage of pleurisy in study farms). The target population included medium- to large-sized (more than 500 pigs per herd) Finnish herds rearing finishing pigs.
Sampling and data gathering
Purposive sampling was used. A study herd needed to fulfill the inclusion criteria of 1) at least 1000 finishing pigs sent for slaughter annually and 2) location in south-western Finland within a distance of 250 km from the University of Helsinki ambulatory clinic in Mäntsälä. Three major slaughterhouses were asked to compile a list of finishing pig herds fulfilling the above criteria. This list including 219 herds served as a sampling frame. At the same time, the slaughterhouses provided the research group with the pleurisy percentage of these herds during the year preceding enrolment in the study. The herds were then sorted in descending order according to their pleurisy percentage, separately for each slaughterhouse. Herds were tentatively divided into case and control herds based on their pleurisy values. Herds having a higher pleurisy value than slaughterhouse-specific mean pleurisy plus standard deviation value were considered to be tentative case herds. Similarly, herds having a lower pleurisy value than the slaughterhouse mean minus the standard deviation were considered to be tentative control herds. For slaughterhouse one, pleurisy percentage (±standard deviation) used in tentative classification was 1.9±1.4%, for slaughterhouse two 27.5±16.8% and for slaughterhouse three 4.0±2.9%. These tentative herd classifications were verified based on the post mortem findings of the study batch as described later in materials and methods.
Researchers then contacted the tentative case and control herds and requested their consent to participate in the study. During the first contact researchers ascertained that the herd had not suffered from acute respiratory disease outbreak during the last batch of finishing pigs. All voluntary herds were enrolled in the study until the desired number of herds (at least 60 herds representing all three slaughterhouses proportionally to their market share) was fulfilled. Of 93 herds contacted, 29 (31.2%) opted out or were unable to participate. Out of herds that opted out, 48% were not willing to participate, 48% no longer had animals or had changed slaughterhouse, making participation impossible, and 3% reported some other reason. Customers of slaughterhouse 1 and case herds (54% cases vs. 45% controls) were overrepresented in herds opting out. Opted-out herds had approximately twice as many animals as herds that were willing to participate.
Researchers visited all study herds once about ten weeks after the growers (body weight 25 kg) had arrived at the finishing pig herd or a finisher compartment in a farrow-to-finish herd. Visits were carried out during years 2012-2014. At the beginning of the visit, the researchers ascertained that the pig groups had not suffered from acute respiratory disease outbreak after entering the finishing herd or compartment. The herd owner or responsible caretaker was interviewed about basic information of the herd (herd type, number of animals), management and environmental factors using a herd visit questionnaire. The main categories of management-related information included vaccination routines, biosecurity and hygiene, animal flow, mean number of animals per compartment and per pen, mean size of pen, medication routines and feeding. The main categories of environmental factors were ventilation, temperature, heating and floor type.
In each herd, one finishing pig compartment containing at least 200 but less than 1000 finishing pigs was chosen for clinical inspection of pigs and environmental measurements. After carefully entering the compartment, the researchers counted the number of pens containing pigs lying on top of each other. After that, pen wall temperature (Testo 830-T1, Testo SE & Co, Germany), NH3 concentration (Dräger-Tube Pump Accuro, Drägerwerk AG & Co, Germany) and air temperature, humidity and airflow (Envic DM-101, Envic Oy, Finland) were measured in four different spots (two opposite corner pens and two pens in the middle of the row) of the compartment. A herd-level mean of these four measurements was used in further analyses. After environmental measurements, all pigs were gently forced to stand up and their coughing and sneezing episodes were counted for five minutes. A coughing/sneezing episode was defined as a single cough/sneezing or as a set of continuous coughs/sneezings of one animal. The number of soiled pens and the presence of various clinical signs (tear staining, conjunctivitis, neurological signs, lameness, acute and healed tail biting, diarrhoea, flank biting or skin scratches, runts and sitting pigs) were registered individually in approximately 100 pigs.
Altogether 15 pigs were selected evenly in different parts of the study compartment for blood sampling. These pigs were caught with a snout snare and a blood sample was taken from their vena jugularis with vacuum needles and serum tubes. The samples were centrifuged the next day in the laboratory, and the serum was stored at -20°C until analysed.
At the end of the visit, outdoor temperature was measured. Before leaving the herd, herd personnel were asked to inform researchers about any major unexpected events, e.g. acute disease outbreak or equipment malfunction, during the remaining rearing period.
After the study batch containing clinically examined animals was sent to slaughter, the slaughterhouses provided the researchers with the meat inspection (MI) findings of the batch, including mean carcass weight, kilograms of condemned meat, percentage of whole and partial carcass condemnations, lean meat percentage, abscesses, arthritis, milk spots in liver, organ condemnations, tail biting, pneumonia and pleurisy. Herds having pleurisy percentage higher than the slaughterhouse-specific mean pleurisy in the batches undergoing analysis were considered batch-specific case herds. Similarly, herds having pleurisy percentage lower than the slaughterhouse mean were considered batch-specific control herds. These batch-specific results were used to verify the herd classification. Batch-specific mean pleurisy percentages used for classification were 1.6%, 30.2% and 1.8% for slaughterhouses 1, 2 and 3, respectively. Finally, the group of confirmed case herds included only tentative case herds that were also batch-specific case herds. Similarly, confirmed control herds included only those tentative control herds that were also batch-specific control herds.
Laboratory analysis
APP antibodies were measured using two commercial test kits: IDEXX APP-ApxIV ELISA (IDEXX, Liebefeld-Bern, Swizerland) to detect antibodies against ApxIV toxin, which is produced by all known APP serotypes (19), and IDvet ID Screen APP 2 indirect ELISA (IDvet, Grabels, France) to detect antibodies against lipopolysaccharides (LPS) specific to APP serotype 2 (APP2), with a sensitivity of 82.9% and a specificity of 99.6% for IDEXX APP ApxIV ELISA and a specificity of 99.68% for IDVet APP2 ELISA. A pig was considered positive if the test used detected any antibodies in the serum sample.
All blood samples were tested with influenza A antibody ELISA (ID Screen® Influenza A Antibody Competition, IdVet, Grabels, France) according to the manufacturer’s instructions. A sample was considered unclear when the competition percentage (S/N%) was 46–49% and positive when the competition percentage was ≤ 45%. If a herd had at least one unclear or positive blood sample (pig) in the ELISA test, blood samples of that herd were further analysed using a haemagglutination inhibition (HI) test according to the European Surveillance Network for Influenza in Pigs. This was done with the antigens H1N1 (SW/Best/96), H1N2 (SW/Gent/7625/99) and H3N2 (SW/St. Oedenrode/96). All antigens were provided by GD Animal Health Service (Deventer, the Netherlands). A sample was considered HI positive if the HI titre was ≥ 1:20.
Statistical analysis
A required sample size of 24 herds in both groups (control and case) was calculated assuming the proportion of exposed as 40% for controls and as 80% for cases in presumably the most influential variable (herd type). Alpha 0.05 and power 0.8 were used. The least extreme odds ratio to be detected is 6.0 [14].
All gathered data were scrutinized, and all unreliable answers were either checked and corrected, or if this was impossible, removed from the dataset. Most of the variables describing management-related and environmental factors were transformed into meaningful categories. Most of the count variables and measurements were handled as continuous variables.
The outcome variable was a categorical variable “confirmed case or control”. Regarding the predictors, the herd-level prevalence (%) of different clinical signs was calculated. Furthermore, the percentage of soiled pens or pens where pigs were lying on top of each other was calculated. The prevalence of diarrhoea, neurological signs or skin scratches and soiled pens was almost zero and these variables were not included in further analyses. For modelling serological results, herd-level APP2, APPIV and SIV antibody prevalences (%) were calculated. Both APP2 and APPIV seroprevalences were categorized in two categories (low/high prevalence within herd, median used as a cut-off point). Descriptive statistics of all predictor variables at herd level (management-related factors, clinical signs, environmental measurements, and serology) were compiled containing all herds, and a comparison between case and control herds was carried out.
Univariate associations between predictor variables and outcome were evaluated using logistic regression. The liberal p-value of 0.2 was used as a keep-in or drop-out threshold. The correlations between predictor values were scrutinized. The herd type (farrow-to-finish or finishing only) was detected to correlate strongly with many predictor variables (e.g. room-level all-in all-out production). The decision to force the variable “herd type” in further models and drop correlating, intervening variables was made. None of the variables related to environmental measurements showed to be statistically significant and therefore were not included in the multivariable model. Furthermore, SIV serology univariate association with herd pleurisy status was clearly insignificant and this variable was not included in the further modeling.
Finally, a multivariable logistic regression model was built. The initial model contained predictor variables: herd type (fattening or farrow-to-finish), number of finishing pigs in the herd, compartment disinfection (always between batches, sometimes, never), littering frequency (once or twice per day or continuously available), proportion of slatted flooring (≤50%/>50%), airspace per pig, feeding type (liquid/dry), piggery temperature when weaners entered the compartment, piggery temperature when finishing pigs left the compartment, heating (yes/no), ventilation system service (yes/no), ventilation adjustment difference in winter, loading corridor (yes/no), handwashing facility for visitors (yes/no), sitting pigs (%), flank biting (%), conjunctivitis (%), herd-level APP2 prevalence and APPIV antibody prevalence (high/low). Backward elimination model building strategy was then utilized. The final model contained only the predictor variables herd type and number of finishing pigs per herd.
For model diagnostics, the basic assumptions of logistic models were inspected with regard to data structure and nature of the predictor variables. Observations were independent from each other and the continuous variable in the final model (no. of finishing pigs) had a sufficiently linear relationship with the outcome. The Hosmer-Lemeshow goodness of fit test with seven groups collapsed on quantiles of estimated probabilities was used, and it gave a non-significant result (p=0.5) as the null hypothesis is “model fits the data”. In addition, residuals, deltabeta and leverage values were scrutinized to observe potential outliers. Furthermore, the area under the ROC curve (0.92) was evaluated to assess predictive ability of model.