bs_bs_banner
Multisite randomised controlled trial to evaluate polypropylene clips applied to the breech of lambs as an alternative to mulesing. II: multivariate analysis of relationships between clip treatment and operator, sheep, farm and environmental factors AR Rabiee,a MC Playford,a,b I Evans,c G Lindon,c M Stevensond and IJ Leana*
Objective A multivariate analysis approach was used to evaluate both the effects of application of occlusive polypropylene clips to the breech on bare area measurements and scores of lambs, and the influence of operator, region, sheep, farm and environmental factors on outcomes. Procedures A randomised controlled trial using 32,028 lambs was conducted on 208 commercial wool-growing properties across Australia. Differences in bare area measurements and scores between groups were estimated and analysed using a mixed model to investigate the effects of operator differences, farm and environmental factors and the interactions among these factors. Results Clip-treated lambs with higher body weight at visit 1 had higher bare area measures and scores, but lower changes in dag and urine scores. Lambs with tight skin showed improved response in bare area scores and measurements after clip treatment, but lambs with a high wrinkle score at visit 1 showed less response to the treatment in their urine, dag and wrinkle and bare area scores. These effects of the clip treatment were not significantly influenced by estimated fleece fibre diameter, operator or region, but were significantly influenced by farm. Conclusions The effect of occlusive clips on breech measurements and scores was significantly influenced by body weight, skin type and thickness, wrinkle score and sex of the lamb, but not by region, operator or estimated fibre diameter. The clip treatment significantly improved characteristics that influence the susceptibility of lambs to flystrike under most conditions. Keywords animal welfare; flystrike; non-surgical mulesing; randomised controlled clinical trial; sheep Abbreviation
V, Visit
Aust Vet J 2012;90:423–432
doi: 10.1111/j.1751-0813.2012.00992.x
S
urgical mulesing is an effective and widely adopted method of preventing cutaneous myiasis, specifically breech-strike in Merino sheep.1 Concerns about animal welfare and the productivity effects of mulesing2 have stimulated research into alternative procedures to mulesing.3,4 An occlusive polypropylene clip was devel*Corresponding author. a SBScibus, PO Box 660, Camden, New South Wales 2570, Australia;
[email protected] b Current address: Dawbuts Pty Ltd, PO Box 1118, Camden, New South Wales, Australia c Australian Wool Innovation, Sydney, New South Wales, Australia d Epicentre, Massey University, Palmerston North, New Zealand
© 2012 The Authors Australian Veterinary Journal © 2012 Australian Veterinary Association
PRODUCTION ANIMALS
PRODUCTION ANIMALS
oped by Australian Wool Innovation Ltd to modify the breech area of sheep using a closed method.5 Advantages of the clips over surgical mulesing are that lambs show lower levels of behavioural pain responses,6 growth retardation and mortality.7 After several pilot studies demonstrated that the clips were effective in increasing breech bare area,8 a large-scale, randomised controlled multisite trial across all wool-growing areas of Australia was conducted during the spring of 2007 to test the efficacy of the clips under field conditions. Details of the trial and results of descriptive analyses have been presented.9 Those results showed that lambs treated with clips had greater bare area measurements and lower scores for wrinkle, dag and urine stain than untreated controls.9 The study was conducted over a wide range of geographical regions, by many different operators, on different breeds and sexes of lambs with different fibre diameters, skin types and body sizes, and in a range of environmental conditions. In order to explore the optimal conditions for using the clips to improve breech measurements and scores, and to study associations between treatment and other factors such as environmental conditions (temperature, rainfall), study sites, sheep and management factors,10 a multivariate approach was used to evaluate the data. Interactions among these factors in relation to the efficacy of clip application were investigated. Materials and methods Study design The multisite randomised controlled trial was conducted in six states on 208 purposively-selected trial sites from August 2007 to January 2008 (Table 1). Clip application procedure Occlusive polypropylene clips were applied by 35 trained operators at lamb marking. At each site, 160 lambs (range 90–168) were weighed, measured for breech bare area height and width, and scored for bare area, wrinkle, faecal (dag) and urine staining, skin type and skin thickness. Lambs were allocated to control (no clips) or treatment (two tail and two breech clips) groups. Lambs were assessed at 14 days (range 10–19) for wound healing and at 55 days (range 34–129) for body weight, breech bare area measurements and scores. Skin movement (loose or tight) and skin type (thick or thin) were scored as 1 or 2. The temperature and average rainfall of the trial sites during the period of study were obtained from the bureau of meteorology website (www.bom.gov.au). The cut-points for temperature
Australian Veterinary Journal Volume 90, No 11, November 2012
423
PRODUCTION ANIMALS
PRODUCTION ANIMALS
Table 1. Wool regions as defined by Australian Wool Innovation Ltd for the 2007–08 clip trial
Wool region
South-west coast, WA WA wheatbelt Northern pastoral, SA Eyre and York Peninsulas, SA Southern and eastern VIC Wimmera/Mallee/northern VIC Tasmania NSW Tablelands (northern, central, southern) Far-west NSW Riverina, NSW North- and central-west slopes and plains, NSW Darling Downs and central highlands, QLD Charleville Longreach, western QLD Region not recorded
Zonea
No. of sites
High rainfall Wheat–sheep Pastoral zone Wheat–sheep High rainfall Wheat–sheep High rainfall High rainfall Pastoral Wheat–sheep Wheat–sheep
4 26 2 19 34 14 15 39 5 15 19
Wheat–sheep Pastoral
4 4 3
a
Zones defined as per the Australian Bureau of Agricultural and Resource Economics. NSW, New South Wales; QLD, Queensland; SA, South Australia; TAS, Tasmania; VIC, Victoria; WA, Western Australia.
were 30°C and for the average rain fall were 0 mm (no rain), ⱕ10 mm/day (light rain) and >10 mm/day (heavy rain). An estimate of fibre diameter of the lambs at hogget shearing was provided by flock owners or managers. Statistical analysis
Data preparation. Before analysis, data were screened for errors by checking the electronic data against the hard copies that were completed by the operators at the trial sites. The process was conducted on two levels: (i) property or farm information, which included region, site, operators, breeds, estimated fibre diameter of wool, vaccination history, fly treatments, rainfall and temperature at each of the three visits; (ii) animal data, which included body weight, horizontal (A1) and vertical (A3) bare area measurements, and wrinkle, urine stain, dag and bare area scores at visits (V) 1 and 3. Tail and breech healing scores at V2 were also checked for errors and missing values. The electronic database and hard copies of each participant were used to cross-check and find errors and missing values. Following the auditing of 1.43% of all data, selected at random, the error rate was estimated to be 0.18% and not systematic in nature. This was accepted as a satisfactory level of precision for data collection and entry. Any identified errors or missing data were corrected. Of the 208 trial sites, 203 returned full data sets suitable for analysis. Five sites were excluded because of incomplete data (n = 4) or nonrandomised allocation of treatment (n = 1). Of the 203 trial sites, the 197 Merino and 1 Dohne flocks were considered as one breed group and the other flocks (4 Merino-cross and 1 Corriedale) were categorised as a separate group.
424
Australian Veterinary Journal Volume 90, No 11, November 2012
Multivariable mixed modelling. Mixed model procedures were used to assess data and examine interactions between clip treatments and V1 observations, farms and operators. Horizontal and vertical bare area measurements (A1, A3, respectively) and scores (wrinkle, dag, urine stain and bare area) were analysed using a mixed model regression procedure. Statistical analyses were conducted using Stata v.10.1 (StataCorp, College Station, TX, USA) and the outputs were validated with R v.2.8.1 outputs (The R Foundation for Statistical Computing, Vienna, Austria). The direction and magnitude of treatment effects were explored across various subsets of sheep. A univariable evaluation of associations between variables was undertaken using graphical and statistical methods.9 A screening process was conducted for each outcome to identify variables needed in the mixed models. A multivariable regression model was used to screen variables for inclusion as covariates. This was conducted by first screening individual variables (e.g. region, skin type and movement, temperature, rainfall, fibre diameter and days between visits). Variables with P ⱕ 0.15 were included in subsequent multivariable models. Sex, interactions of sex and treatment, effects of weight and wrinkle score at V1 were a priori considered to be factors that might influence the response to treatment. The screening model that was developed for each outcome variable included treatment group + sex + sex*group + V1 weight + V1 observation for each variable + V1 wrinkle score. The number of days between V1 and V3, estimated fibre diameter, bare area measurements (A1, A3) and the scores (wrinkle, dag, stain and bare) at V1 were treated as continuous variables. Other variables (rainfall, temperature, days, estimated fibre diameter, skin type and movement) were categorised according to predetermined cut-points. All variables meeting the first screening criteria were included in a backward exclusion modelling process until all variables that remained in the model were significant at P ⱕ 0.05. Once variables were excluded from the model, they were not re-included in the subsequent steps. Wrinkle, urine stain and dag scores at V3 were not normally distributed because of the large number of ‘1’ scores. Transformation procedures were not effective in providing a normal distribution. Differences between the V3 and V1 observations were found to meet normality assumptions and used as the outcome variables, treating them as continuous (i.e. interval) data. Because the wrinkle, dag, urine stain and bare area scores were ordinal data, simulation procedures using different underlying assumptions were used to assess the effects of deviations from the properties of interval data on the results, as previously described.8 Briefly, the estimated coefficients from the regression models were compared after modifying the intervals between successive categories of these score variables at V1 and V3. We repeated this, based on different assumptions about the underlying biological values of scores. The results of these simulations showed that estimated magnitude of the effects changed under different scalar assumptions, but the estimated directions of the effects were almost always unchanged. These results and the finding that flystrike risk increased linearly with wrinkle and dag © 2012 The Authors Australian Veterinary Journal © 2012 Australian Veterinary Association
scores (Smith, 2010; CSIRO,Armidale; pers. comm.) supported the use of the score data as continuous variables. The mixed model of each outcome variable included the fixed effect terms and covariates that met the selection criteria, random effects terms for ‘property’ or farm of trial, and operator at V1 and V3. Variables that were significant following the screening process were retained in the model. The non-significant variables were removed from the model and each model was assessed and compared with the previous model using the probability values, Wald values and Likelihood ratios collectively.11 The factor of ‘property’ or farm of trial was included in the model as a random effects term, and the operators present at V1 and V3 were used as fixed and random effects as follows: (i) operators at V1 and V3 were both included as fixed effect terms; (ii) operator at V1 was included as a fixed effect and operator at V3 as a random effects term; (iii) operator at V3 was included as a fixed effect and operator at V1 as a random effects term; and (iv) operators at V1 and V3 were both included as random effects terms. This process allowed the effect of operator to be examined in detail before the final models outlined above were developed. After the final model main effects were evaluated, two-way interaction terms between treatment and other variables that remained in the model were generated and included in the model.12 Significant twoway interaction terms were then assessed for biologically plausible effects on treatment. Those with plausible effects on treatment responses were retained in the final model. Results Descriptive statistics The descriptive statistics from this study have been presented previously by Playford et al.9 Mixed models Of the 208 trial sites, 203 returned full data sets suitable for analysis. The results of the mixed models are presented in Table 2 for all outcome variables.
Body weight. Body weight at the time of treatment (V1) ranged from 4.5 to 44 kg, with means of 17.35 ⫾ 0.06 (SE) kg for ewes and 18.46 ⫾ 0.06 kg for wethers. Treated lambs had slightly, but significantly, lower body weight (26.02 ⫾ 0.05 kg) at V3 compared with the control group (26.40 ⫾ 0.05 kg). The effect of treatment on weight depended on an interaction with body weight at V1. Heavier, treated lambs gained less weight than lighter, treated lambs by V3, although this effect was not of great magnitude. Body weight at V3 was significantly greater in wether lambs and lower in lambs with higher wrinkle scores and tight skin movement at V1. Body weight of lambs was also significantly greater at sites with lower environmental temperature (
