Journal Pre-proof Selective and deferred treatment of clinical mastitis in seven New Zealand dairy herds Andrew Bates, Richard Laven, Olaf Bork, Merlyn Hay, Jess McDowell, Bernardita Saldias
PII:
S0167-5877(19)30539-2
DOI:
https://doi.org/10.1016/j.prevetmed.2020.104915
Reference:
PREVET 104915
To appear in:
Preventive Veterinary Medicine
Received Date:
2 August 2019
Revised Date:
3 February 2020
Accepted Date:
4 February 2020
Please cite this article as: Bates A, Laven R, Bork O, Hay M, McDowell J, Saldias B, Selective and deferred treatment of clinical mastitis in seven New Zealand dairy herds, Preventive Veterinary Medicine (2020), doi: https://doi.org/10.1016/j.prevetmed.2020.104915
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier.
Selective and deferred treatment of clinical mastitis in seven New Zealand dairy herds
Andrew Batesa, Richard Lavenb, Olaf Borkc, Merlyn Hayd, Jess McDowelle, Bernardita Saldiasf
Vetlife NZ, Vetlife Scientific, 1, Waitohi-Temuka Road, Temuka, New Zealand
c
Mastaplex Ltd, Centre for Innovation, 87 St David Street, Dunedin 9016, New Zealand
d
e
Vetlife Oamaru, 281 Thames Street, Oamaru 9400, New Zealand
Vetlife Temuka, 1 Waitohi-Temuka Road, Temuka 7920, New Zealand
Centre for Dairy Excellence, 20, Wilson Street, Geraldine 7930, New Zealand
re
f
Institute of Veterinary, Animal and Biomedical Sciences, Massey University, New Zealand
ro of
b
-p
a
lP
Email addresses for correspondence
[email protected]
na
Abstract
Mastitis is the most frequent reason for antibiotic use in New Zealand dairy cattle and technologies
ur
reducing and targeting this use contribute to responsible product stewardship. Rapid identification of pathogen and antibiotic susceptibility facilitate targeted treatment but currently involve a minimum 24
Jo
hours delay. Studies from confinement systems where Gram-negative organisms are responsible for a significant proportion of mastitis, indicate selective treatment can reduce antibiotic use without reducing clinical or bacteriological cure. However, in New Zealand’s seasonal, pastoral dairy system, mastitis is dominated by Gram-positive organisms and if treatment is deferred, it is vital both shortand long-term clinical health outcomes are not compromised. Mastatest is a diagnostic system for bovine mastitis indicating the pathogen and its antibiotic sensitivity within 24 hours of sampling. This study focused on evaluating this system’s ability to 1
control antibiotic usage whilst achieving equivalent bacteriological and clinical cure rates alongside long term individual somatic cell count (ISCC) outcomes as conventional treatment choices. Mild to moderate mastitis cases in the 100 days after calving in 6,467 cows from 7 farms were milk sampled and randomly allocated to a positive control group (non-selective treatment) or a culturebased treatment. All milk samples were processed using Mastatest. For the positive control, the quarter was treated immediately with 3 treatments of procaine penicillin every 12 hours. For the selective treatment group, treatment was delayed for 24 hours and then informed by pathogen and
ro of
antibiotic sensitivity from the Mastatest result. Gram-negative and no-growth quarters were untreated. Gram-positive quarters were treated with the antibiotic for which the lowest in vitro antimicrobial sensitivity was reported.
Re-sampling was carried out from affected quarter(s) approximately 21 days after initial diagnosis and
-p
cultured for bacterial identification. Clinical recurrence within 60 days and ISCC data was recorded at herd tests over the duration of the lactation. Antimicrobial usage and days of milk withhold
re
pending clearance of antibiotic residues were also noted.
There was no difference in bacteriological or clinical cure rate between the two treatment groups.
lP
Final herd test ISCC and days of milk withhold from supply did not differ between groups. Antibiotic usage was 24% less (95% predictive interval = 12-47%) in the selective group.
na
This study suggests that on farm decisions about deferred treatment of mastitis using Mastatest to identify the intramammary pathogen can reduce the antimicrobial usage with no loss in bacterial or
ur
clinical cure and with no effect on ISCC over the lactation.
Jo
Key Words: Mastitis; deferred treatment; antibiotic sensitivity; Mastatest
Abbreviations
ISCC - Individual Somatic Cell Count NZ – New Zealand MIC – Minimum Inhibitory Concentration
2
Se – Sensitivity Sp – Specificity NS – Non-Selective S – Selective PI – Prediction Interval ROPE – Region of Probable Equivalence OR – Odds Ratio HPDI – Highest Probability Density Interval
ro of
IRR – Incidence Rate Ratio
Introduction
Across all dairy systems, mastitis is one of the most prevalent and costly diseases (Down et al., 2017;
-p
Hillerton et al., 2017). As such, it is a significant cause of antimicrobial use on farm (Bryan and Hea, 2017; Higham et al., 2018). Commonly, clinical mastitis is defined by visible changes in the udder or
re
milk (Roberson, 2003) and in many countries, treatment is initiated without bacteriological identification or antimicrobial sensitivity testing (McDougall et al., 2017; Vasquez et al., 2017).
lP
Bacterial culture by external laboratories is not routinely used because of cost and the time delay between submission of sample and reporting of results (Lago and Godden, 2018) means treatment
na
decisions are made before the results are known.
On farm diagnostic systems that identify the causative pathogen help farmers and their advisers in
ur
deciding whether to cull the animal or treat. It will also inform choice of antimicrobial therapy, likelihood of success and suitable control measures to prevent future cases (Steele et al., 2017).
Jo
Withholding antimicrobial treatment can sometimes be appropriate. In confinement systems, 10-40% of bacterial cultures from clinical mastitis yield no bacterial growth and so likely do not require antimicrobial therapy (Roberson, 2003). An average of 24% (range 7% to 40%) of clinical cases have been estimated to yield Gram-negative bacteria or yeasts (Lago and Godden, 2018; Ruegg, 2018). Approved intramammary drugs have no therapeutic activity against yeasts and many have little efficacy against Gram-negative infections. Moreover, a large proportion of Gram-negative infections
3
are quickly cleared by the body’s immune system (Lago et al., 2011a). Thus, identification of these cases at the point of diagnosis offers the opportunity to reduce antimicrobial use on farm (Lago and Godden, 2018; Ruegg, 2018). This will reduce costs associated with treatment and discarded milk, antimicrobial residues in milk and the potential risk of antimicrobial resistance in mastitis pathogens. Systems for mastitis diagnosis on farm, currently involve bacterial culture and require a minimum of 24 hours before treatment decisions can be made. Despite this interval before antimicrobial treatment, in confinement systems, deferred compared to immediate blanket antimicrobial treatment appears to give equivalent bacteriological and clinical cure, individual somatic cell count (ISCC), milk
(Lago et al., 2011; Lago et al., 2011b; Vasquez et al., 2017).
ro of
production and cow survival outcomes, whilst reducing antimicrobial usage by approximately 50%
In pastoral, seasonal dairy systems such as those in New Zealand, Gram-positive organisms
-p
predominate (McDougall, 1998; McDougall et al., 2014). Under these conditions, the incidence of Gram-negative organisms is estimated to vary from <5% to 10% and samples yielding no growth
re
from 18-28% ( McDougall, 1998; Bryan et al., 2016; McDougall et al., 2018). Despite this, on-farm diagnosis of clinical mastitis in pastoral systems has shown a reduction in antimicrobial use of around
lP
28% and no difference in clinical outcome over 60 days of follow up (McDougall et al., 2018). However, long term effects on ISCC were not reported in this study.
na
Systems currently available for bacterial culture on-farm have been reviewed by Lago and Godden, (2018) and Ruegg, (2018) and rely on a decision tree approach to guide the operator to a diagnostic
ur
decision. None of them yield antimicrobial sensitivity information and choice of antimicrobial is based on expected sensitivity and knowledge of likely risk factors determining likelihood of a cure
Jo
(Higham et al., 2018; McDougall et al., 2017). Two recent surveys of antimicrobial sensitivities for common mastitis pathogens in New Zealand (NZ) suggest that 72-73.1% of isolates of Staph. aureus may be sensitive in vitro to penicillin (McDougall et al., 2014; Petrovski et al., 2011). However, for antimicrobial sensitivity to be useful in guiding individual treatment decisions, it needs to be validated against specific veterinary cut-points specific for the bacterial organism, host species, pathogen and disease process as these will all influence the outcome irrespective of the in vitro sensitivity. Thus, although knowing the pathogen’s in vitro antimicrobial sensitivity is useful, without validated 4
veterinary break points and knowledge of the likely differences between in vitro and in vivo performance it may not necessarily lead to improved cure rates through targeted antimicrobial use (Lago and Godden, 2018). However, at the regional and national level, the information may provide ad-hoc evidence for changes in antimicrobial sensitivity with time or location through evaluation of the distribution of minimum inhibitory concentrations (Watts et al., 2018). Mastatest (Mastaplex Ltd, Centre for Innovation, 87, St. David’s Street, Dunedin 9016, New Zealand: https://www.mastaplex.com/mastatest) is a diagnostic system for bovine mastitis indicating
of sampling and was the on-farm, diagnostic tool used in this study.
ro of
the pathogen and its in vitro antibiotic sensitivity using the microdilution technique, within 24 hours
The objective of this study was to evaluate this system’s ability to control antibiotic usage without
compromising bacteriological cure or the proportion of cases re-treated. Our primary null hypothesis
-p
was that there would be no difference in clinical outcome between the two treatment groups (deferred vs immediate treatment). Antimicrobial usage, milk discarded from cows because of antibiotic
re
residues in the milk and long term individual somatic cell count (ISCC) outcomes were also reported.
Materials and Methods
lP
Herd enrolment
Seven, spring calving pastoral dairy herds from the Canterbury and Otago regions of New Zealand
na
were convenience selected from farms serviced by Vetlife New Zealand Ltd. All farms were chosen based on the fact that they had expressed interest in on-farm diagnosis for mastitis, were prepared to
ur
follow the study protocol and contributed towards the funding of the project as part of their annual veterinary fee. Remaining funding was supplied by Vetlife Ltd and Mastaplex Ltd, the developers of
Jo
Mastatest. All procedures were pre-approved by the Animal Ethics Committee of Massey University, NZ, project number 16/75. On-farm training for farm staff was provided in the autumn preceding the trial and covered diagnosis of clinical mastitis using agreed criteria, aseptic milk sample collection, operation of the Mastatest system and the trial protocol and treatment decision trees. Over the dry period preceding the trial, internet connectivity was confirmed for all farms and a run through provided of the procedures involved for farm staff.
5
Case enrolment All cows were eligible for participation, from 14 days before calving to 100 days after calving. Farm staff were asked to enrol all cows with mild to moderate clinical mastitis that would normally be treated with antimicrobials. Clinical mastitis was diagnosed and recorded if milk from one or more quarters was abnormal in colour, viscosity or consistency with or without accompanying swelling, heat, redness or pain, or generalised illness (Lago et al., 2011a). Any cow with pyrexia (rectal temperature > 400C), dehydration, inappetence or other signs of systemic illness was excluded from the trial and treated with systemic antimicrobials and non-steroidal anti-inflammatory drugs under the
ro of
direction of the farm’s veterinarian. Any cow diagnosed with mastitis based on a ISCC or increased conductivity, but without visible signs of changes in the milk was excluded from the study. Sample collection and on-farm processing
-p
At detection, and before treatment, the teat end was scrubbed with 70% ethanol impregnated wipes
(Mediwipes; imported by Sulco Ltd, Manukau City, NZ) and a milk sample (>5 mL) collected into a
re
factory-clean polyethylene sample vial and then, using Mastatest, tested on-farm. Immediately after
lP
sample collection approximately 5mL of milk was poured into the Mastatest cartridge. This is a modified ELISA cartridge with 24 wells containing broth media that over a 24-hour period undergo a color change dependent on the bacterial species present. Six wells are used for bacterial
na
identification. The remaining 18 wells contain three antibiotics commonly used in New Zealand to treat mastitis: benzylpenicillin, cloxacillin and lincomycin/neomycin combination. Six wells are
ur
allocated to each antibiotic at concentrations of 4, 2, 1, 0.5, 0.1 and 0.05mg/L (drug as free form). Mastatest determines if there are bacteria present, distinguishes between Gram-positive and
Jo
coliform/Gram-negative bacteria and, in addition, identifies Streptococcus uberis, Staphylococcus aureus and cogagulase negative Staphyloccoccus bacteria. Other Gram-positive bacteria like Streptococcus dysagalactaie are detected and reported as Gram-positive bacteria. Once the cartridge has been loaded with suspect milk it is placed into an incubator (37 0C) on-farm which uploads a series of digital image of the Mastatest cartridge to the cloud. The colour changes detected are interpreted via an algorithm. Within 24 hours an email is sent to the farm and the farm’s
6
veterinarian, confirming the bacterial type and ranking of the minimum inhibitory concentration (MIC) values for the three antibiotics. In a recent Bayesian latent class analysis of test sensitivity (Se) and specificity (Sp), Jones et al., (2019) found for Strep. uberis, Mastatest had a Se of 0.88 (sd = 0.07) and Sp of 0.80 (sd = 0.10) and for Staph. aureus, Se was 0.85 (sd = 0.05) and Sp of 0.96 (sd = 0.02). No differences in test performance were detected between Mastatest and conventional culture for the range of organisms detected. Comparison of antibiotic susceptibility testing using Mastatest and the reference method
ro of
showed similar trends and, in some cases, identical MIC50 and MIC90 values, with at most one antibiotic dilution difference. Treatment groups
For a first clinical mastitis case, cows were randomly assigned to either a non-selective treatment
-p
(NS) or a selective treatment, (S) by consulting a farm specific randomised list of treatments. If more than 1 quarter was affected, all quarters from that cow were allocated to the same treatment system
re
group. Repeat cases were allocated as if they were a new case unless they had recurred in under
lP
seven days when they were re-treated with the original antimicrobial on the assumption that they were a recurrence of the initial infection. Treatment decision trees
na
In consultation with herd owners and farm veterinarians, a common treatment tree decision protocol was developed for all enrolled farms (Figure 1). All cows in the NS group were treated immediately
ur
after diagnosis with three, 12-hourly intramammary infusions of 1,000,000 i.u procaine penicillin (Intracillin 1000 Miliking Cow, Virbac NZ, Ltd). If more than one quarter was affected, three, 24-
Jo
hourly intramuscular injections of 5g i/m penethemate (Penethaject, Bayer Animal Health NZ, Ltd) were used in place of intramammary treatment. Cows in the selective treatment group were treated based on bacterial identification and antimicrobial sensitivity. Gram-negative and no-growth quarters were left untreated. Gram-positive quarters were treated with the antibiotic for which the lowest in vitro antimicrobial sensitivity was reported in the Mastaplex result. Treatment was allocated to one of: three infusions of 1,000,000 i.u procaine
7
penicillin (Intracillin 1000 Milking Cow, Virbac NZ, Ltd) every 12 hours; three infusions of 200mg cloxacillin (Nitroclox LA, Virbac, NZ, Ltd) every 24 hours; and three infusions of 330mg lincomycin, 100mg neomycin (Albiotic, AgriHealth NZ Ltd) every 12 hours. Where multi-quarters were affected, and penicillin was indicated as the drug of choice, three, 24-hourly intramuscular injections of 5g i/m penethemate (Penethaject, Bayer Animal Health NZ, Ltd) could be used. Milk from all cows diagnosed with clinical mastitis was withheld from sale until the milk was visually normal or for the duration of the prescribed withhold from the antimicrobial (if any) prescribed. Follow up samples
ro of
During the trial, a veterinary technician visited each farm once a week to aseptically re-sample the affected quarter(s) from all enrolled cows approximately 21 days after initial enrolment.
Samples
were immediately frozen and cultured by an independent laboratory operating in accordance with
-p
National Mastitis Council guidelines (Hogan et al., 1999) to determine bacterial identification at the species level.
re
Treatment records
Electronic treatment records were used on all farms via the national Livestock Improvement
lP
Corporation (LIC, Hamilton, NZ) MINDA database. Third party access was granted by herd owners for the study. Records included cow age, breed, calving date, date of diagnosis, intended treatment
days before dry-off.
ur
Statistics
na
allocation, Mastatest result, treatment administered and ISCC at the last herd test, recorded 30-40
Power analysis
Jo
Power analysis suggested that 250 quarters in each group would retain power above 0.8 at a type I error risk of 0.05 to detect a difference in (1) the proportion of bacterial cures of from 75% to 85%; (2) a difference in clinical recurrence from 15% to 5%; (3) a difference in the ISCC from 200,000 to 180,000; (4) a difference in the number of daily doses of 3 to 6 and (5) a difference in the number of days of milk withhold from 5 to 6. This is very similar to the number of cows recruited across eight herds by Lago et al., (2011b, 2011a). Independent variables 8
The primary independent variables were treatment group (categorical, NS vs S) with parity, breed, length of dry period (days), days from planned start of calving for the herd to individual cow calving date, days calved at enrolment and pathogen isolated at diagnosis and herd as potential confounders or covariates. Dependent variables The dependent variables were the proportion of clinical quarters with a bacterial cure (no evidence of the original pathogen) at 21 days from first diagnosis, the proportion of cases with clinical recurrence within 60 days from initial diagnosis, the ISCC recorded at the last herd test of the season, the average
was withheld from supply because of antimicrobial residues.
ro of
quantity of antimicrobial used in each group per case and the average number of days for which milk
Antimicrobial use was defined as the sum of daily treatment doses for each course, calculated using
-p
the method of McDougall et al., (2018) but applied at the quarter level. In brief, the sum of daily
doses was calculated by dividing the total number of treatments given by the dose frequency. For
re
example, if a quarter was treated with three intramammary infusions of penicillin every 12 hours the sum of daily doses was 3/2 = 1.5. If three treatments of cloxacillin were infused every 24 hours, the
lP
daily dose was 3/1=3. If three doses of penethamate at 24-hour intervals were used as parenteral treatment for multi-quarter cases the daily dose recorded for each quarter was 3/1=3. Quarters
na
allocated to the selective group that yielded no-growth or Gram-negative bacteria and subsequently not treated, were recorded as zero daily doses.
ur
The period for which milk was withheld from sale was calculated from the group allocation plus treatment given plus the label specified milk withhold. The milk from animals in the selective group
Jo
that gave Gram-negative or no growth samples was returned to sale (if visually normal) on receipt of this result and so 24 hours after initial clinical diagnosis. Quarters were excluded if no pre-treatment sample was collected, if the cow’s identity could not be established, if treatment allocation had been non-compliant and if there was no follow-up sample collected 21 days after initial diagnosis (Figure 2). In all cases, the experimental unit was the quarter. Proportions were binomially distributed, ISCC, quantity of antimicrobial and days of milk withhold were numeric but right skewed. 9
Models To investigate the distribution of the independent variables between the treatment allocation groups, a univariable Bayesian comparison was made. For the effect of the numeric independent variables (parity, length of dry period, days from herd planned start of calving to calving date and days calved at enrolment between treatment groups which were visually assessed to be right tailed and nonnormally distributed), the Bayes equivalent of a robust t-test was used with a log-normal distribution (Kruschke 2016). To compare the distribution of categorical variables (farm and breed) between treatment groups we used a modified version of the bayes.prop test in R (Bååth, 2014).
ro of
These models were written in JAGS (“Just Another Gibbs Sampler” a computer programme for the construction of Bayesian models with Markov chain Monte Carlo sampling (Plummer, 2012)) and
implemented in R (R 2013). We used mildly informative priors for the numeric variables, where for
-p
the sth subject in the cth treatment group, the distribution of the independent variable(Y) was: Ys|c ~ lnorm (μlogyc,1/ σlogyc2)
re
μlogyc ~ norm(mean of log(Y), 0.001 x 1/standard deviation of log(Y)2)) σlogyc~ uniform(0, (standard deviation of log(Y)+ 6.9)
lP
We anticipated that each level of categorical variables would be evenly distributed between NS and S groups. Consequently, we used a Beta (10,10) prior to reflect a distribution centred at 0.5 and bound
na
between 0 and 1.
To estimate the probability that the effects we observed were real, we calculated the 95% predictive
ur
interval (PI) for each parameter value estimated by the models. Values inside the 95%PI are “ the 95% most credible values of the parameter” (Kruschke, 2018). To confirm a difference between
Jo
treatment groups we looked at the predicted difference in numerical variables between the treatment groups, calculated in terms of the effect size relative to the standard deviation of the population. We set our null value for this difference at zero but with a range of values either side corresponding to the region of practical equivalence (ROPE, Kruschke, 2018). To compare parity, length of dry period, days from herd planned start of calving to calving date and days calved at enrolment between treatment groups, we used an effect size defined as +/- 1 standard deviation of the population. By
10
deliberately using a wide ROPE, this gave us an indication of whether the values were broadly similar between treatment groups. To compare the distribution of categorical variables between treatment groups we used a ROPE of +/10%. To investigate the effect of treatment allocation on the proportion of quarters with a bacteriological cure, clinical recurrence, ISCC at the last herd test, quantity of antimicrobial used and days of milk discard, we used mixed models, where quarter was nested within cow and cow within farm. A logistic distribution with a logit link function was used to assess proportions of quarters with a
ro of
bacteriological cure and the proportion of quarters with a clinical recurrence. A negative binomial
distribution was used to assess ISCC at the last herd test, antimicrobial usage and days for which milk from treated animals was withheld from sale. The sum of daily doses was multiplied by two to ensure
-p
integer values (and representing the sum of half-day doses). These models were fit with the following
Intercept ~ Student’s t (3,0,2.5) Coefficients ~ Student’s t (3,0,2.5)
lP
Sigma ~ Exponential (0.32)
re
priors:
Covariance ~ Decovariance (1,1,1,1)
na
All distributions are reported after the variables have been centred. For linear models, sigma represents the variability with which the outcomes deviate from the predictions of the model.
ur
Predictive variables were added in a forward and backward step-wise manner and retained if their 95% PI excluded zero. Two-way interactions between treatment group and farm and between
Jo
treatment group and pathogen were assessed in a similar manner. Treatment group was forced into all models.
For all models, an effective sample size of 10,000 was used and convergence of the Monte Carlo Markov Chains, autocorrelation and comparison of the observed data with a random sample of replicates was assessed visually. Values of the Gelman and Rubin potential scale statistic (maximum tolerated value 1.1) and the standard error of the mean of the posterior draws (maximum tolerated
11
value 10% of the posterior standard deviation) were assessed using ShinyStan (Stan Development Team, 2018).
Results Across the seven participating farms, out of the 6,467 cows calving, 608 cows and 648 quarters were diagnosed with clinical mastitis by farm staff. Overall, complete data records were available for 259 quarters in the NS group and 276 quarters in the selective group. These quarters were from 471 cows with a single quarter affected, 29 cows with two quarters affected and 2 cows with three quarters affected. The distribution of enrolled quarters between farms is shown in Table 1.
effect of deferred and selective treatment for bovine mastitis
-p
re
ur
Non-selective Selective Total
lP
Non-selective Selective Total
Farm 1 2 3 4 5 6 7 760 769 1680 840 834 813 771 Quarters enrolled 36 18 64 32 76 56 27 42 22 67 34 77 66 31 78 40 131 66 153 122 58 Quarters for diagnosis and treatment allocation 32 14 57 26 70 54 24 39 20 60 31 69 62 25 71 34 117 57 139 116 49 Quarters for diagnosis, treatment allocation clinical and bacteriological cure 30 12 55 24 61 54 23 36 18 53 30 64 53 22 66 30 108 54 125 107 45
na
Herd size Group Non-selective Selective Total
ro of
Table 1. Distribution of enrolled quarters by farm and treatment group for a study looking at the
Jo
Figure 2. Losses to follow-up for quarters enrolled in a study looking at the effect of deferred and selective antimicrobial treatment for mastitis.
12
ro of
Summary descriptive statistics for the numerical independent variables for each treatment group are given in Table 2.
Table 2. Summary statistics for the independent numerical variables between selective and non-
-p
selective treatment groups
Median 2.5th percentile 97.5th percentile Difference and Effect size and 95% HPDIa 95% HDIa Parity Non-selective 4 1 10 0.26 0.07 Selective 5 1 9 (-0.33-0.89 (-.0.9-0.23) Length of dry period (days) Non-selective 77 47 121 2.6 0.13 Selective 79 49 123 (-0.74-6.15) (-0.04-0.30) Days from planned start calving to calving date Non-selective 27 7 76 2.8 0.12 Selective 30 6 74 (-1.30-6.94) (-0.005-0.28) Days from calving date to diagnosis of mastitis Non-selective 34 0 121 -2.4 -0.02 Selective 28 -1 125 (-23.67-18.94) (-0.14-0.10) a 95% Highest posterior density interval is the narrowest interval containing the specified probability
ur
na
lP
re
Group
mass (95%)
Jo
There were small differences between the treatment groups in parity, length of dry period, days from planned start of calving to calving date and days from calving to date of diagnosis of mastitis. However, in all cases the effect size was very small (<0.2). The distribution of the categorical variables for each treatment group is given in Supplementary Material. No significant differences were detected between treatment groups in the distribution of breed or in the distribution of treatment groups between farms. The distribution of pathogens diagnosed at enrolment is shown in Table 3. 13
Table 3. Number and distribution by proportion of pathogens isolated at enrolment between selective and non-selective treatment groups Pathogen at enrolment
Number NonSelective selective
Proportion NonSelective selective
Difference in proportion
5
8
0.38
0.62
-0.1
9
8
0.53
0.47
0.03
Contaminated2
3
5
0.38
0.62
-0.08
Gram-positive3
37
33
0.53
0.47
0.05
Mixed4
26
34
0.43
0.57
-0.1
No growth
26
20
0.57
0.43
0.09
Staph. aureus5
24
27
0.47
0.53
-0.04
Strep. uberis6
129
141
0.48
0.52
-0.04
ro of
Non aureus staphylococci1 Coliforms
Difference 95% Probability that HPDIa absolute difference is ≤10% (-0.4338% 0.25) (-0.2945% 0.34) (-0.4337% 0.29) (-0.1661% 0.25) (-0.3147% 0.12) (-0.1646% 0.33) (-0.2856% 0.19) (-0.1584% 0.07)
Also known as coagulase negative staphylococci (CNS)
2
Three or more different minor pathogens detected
3
Strep sp other than Strep. uberis
4
Three or more different organisms, one of which was a major pathogen or two or more organisms if
re
lP
neither included a major pathogen
-p
1
Staph. aureus alone or in combination with one other pathogen
6
Strep. uberis alone or in combination with one other pathogen excluding Staph. aureus
na
5
There was some evidence for a difference in the distribution of pathogens between the treatment
ur
groups with the HPDI exceeding the bounds of a 10% ROPE for all isolated pathogens. Pathogen at
Jo
diagnosis was forced into all models. Proportion of clinical quarters with a bacterial cure Out of 535 quarter cases, 451 (84%) were bacteriologically cured at re-sampling. For the prediction of bacterial cure, breed was retained in the model in addition to treatment group and pathogen, with farm and cow as random effects. The model predicted no difference in the cure proportion by treatment group but the probability of a cure for cows infected with Staph.aureus was significantly less than that for Strep. uberis infected cows in both groups (Table 4). There were numerical 14
differences in the median bacteriological cure proportion by farm, but the coefficients spanned zero and overlapped. Including farm as a random effect changed the value of the remaining coefficients by over 10%. There was no evidence for a significant interaction between treatment group and farm, nor between treatment group and pathogen. The distribution of predicted values for the odds ratio (OR) for the effect of treatment group on bacteriological cure is shown in Figure3 along with the 95%PI. Eighty-four percent of values exceeded 1.0 but the 95% HPDI exceeded the ROPE of +/-10% difference in the probability of a cure
ro of
(equivalent to a change in OR from 0.59 to 2.32, calculated at 84% cure proportion). Figure 3. Distribution of the predicted values for the odds ratio for the effect of selective treatment on the likelihood of a bacteriological cure. The 95% highest probability density interval is marked by
vertical red lines while the region of probable equivalence is marked by the thickened horizontal line
Jo
ur
na
lP
re
-p
on the y axis
15
Predicted proportion of bacteriological cures, adjusted for breed, farm and cow for each pathogen by treatment group are given in Table 4. The distribution of the predicted proportion of bacteriological cures for Staph. aureus by treatment group is shown in Figure 4a and Strep. uberis in Figure 4b. Table 4. Predicted proportion bacteriological cures, adjusted for breed, farm and cow for each pathogen by treatment group Selective Median probability and 95%PI 97 (87-100) 100 (97-100) 99 (88-100) 91 (78-97) 79 (57-91) 100 (99-100) 44 (24-69) 96 (91-98)
ro of
Non-selective Median probability and 95%PI Non aureus staphylococci 97 (86-100) Coliforms 100 (97-100) Contaminated 99 (86-100) Gram-positive 88 (72-96) Mixed 73 (48-88) No growth 100 (99-100) Staph. aureus 37 (18-62) Strep. uberis 96 (91-98) Pathogen
-p
Figure 4a. Predicted distribution of bacteriological cure for quarters from which Staph aureus was
isolated. Quarters were subjected to non-selective or selective antibacterial treatment and resampled
Jo
ur
na
lP
dotted vertical lines the predicted median.
re
21 days later. Red shading indicates non-selective treatment, blue shading selective treatment and
16
Figure 4b. Predicted distribution of bacteriological cure for quarters from which Strep. uberis was isolated. Quarters were subjected to non-selective or selective antibacterial treatment and resampled 21 days later. Red shading indicates non-selective treatment, blue shading selective treatment and
Jo
ur
na
lP
re
-p
ro of
dotted vertical lines the predicted median.
Proportion of quarters with clinical recurrence
17
Out of 535 quarter cases, 43 (8%) were re-diagnosed with clinical mastitis within 60 days of the original diagnosis. For the prediction of clinical recurrence, breed was retained in the model in addition to treatment group and pathogen, with farm and cow as random effects. There were numeric differences in the median predicted proportion of cases recurring by pathogen, but the coefficients spanned zero and overlapped. Including farm as a random effect changed the value of the remaining coefficients by over 10%. The model did not predict differences in the clinical recurrence proportion by treatment group. There was no evidence for a significant interaction between treatment group and farm, nor between treatment group and pathogen.
ro of
The distribution of predicted values for the OR for the effect of treatment group on clinical recurrence is shown in Figure 5 along with the 95%PI. Eighty-four percent of values were below 1.0 but the 95% HPDI exceeded the ROPE of +/-10% difference in the probability of clinical recurrence
-p
(equivalent to a change in OR from 0.89 to 1.12, calculated at 8% clinical recurrence).
Figure 5. Distribution of the predicted values for the odds ratio for the effect of selective treatment on
re
the likelihood of clinical recurrence of mastitis. The 95% highest probability density interval is
Jo
ur
na
horizontal line on the y axis
lP
marked by vertical red lines while the region of probable equivalence is marked by the thickened
18
Predicted proportion of cases with clinical recurrence, adjusted for breed, farm and cow for each pathogen by treatment group are given in Table 5. Table 5. Predicted proportion of cases with clinical recurrence, adjusted for breed, farm and cow for each pathogen by treatment group Selective Median probability and 95%PI 0.9 (0.1-5.4) 1.4 (0.1-10.5) 0.3 (0.0-7.6) 2.4 (0.5-8.7) 0.1 (0.0-1.1) 2.6 (0.5-10.7) 3.1 (0.6-11.7) 3.0 (0.8-8.9)
ro of
Non-selective Pathogen Median probability and 95%PI Non aureusstaphylococci 1.2 (0.1-7.2) Coliforms 1.9 (0.1-13.3) Contaminated 0.3 (0.0-9.6) Gram-positive 3.3 (0.7-12.0) Mixed 0.1 (0.0-1.7) No growth 3.5 (0.7-13.5) Staph. aureus 4.2 (0.8-15.6) Strep. uberis 4.1 (1.1-11.6)
The distribution of the predicted proportion of cases with clinical recurrence for Staph. aureus by
-p
treatment group is shown in Figure 6a and Strep. uberis in Figure 6b.
Figure 6a. Predicted distribution of proportion of cases with clinical recurrence for quarters from
re
which Staph aureus was isolated. Quarters were subjected to non-selective or selective antibacterial treatment and resampled 21 days later. Red shading indicates non-selective treatment, blue shading
Jo
ur
na
lP
selective treatment and dotted vertical lines the predicted median.
19
ro of -p re lP
na
Figure 6b. Predicted distribution of proportion of cases with clinical recurrence for quarters from which Strep. uberis was isolated. Quarters were subjected to non-selective or selective antibacterial treatment and resampled 21 days later. Red shading indicates non-selective treatment, blue shading
Jo
ur
selective treatment and dotted vertical lines the predicted median.
20
ro of -p re lP na
ur
ISCC recorded at the last herd test of the season Out of 502 cow cases, 226 cows in the NS group and 209 cows in the S group had an ISCC reading
Jo
30-35 days before dry-off. None of the cows with a contaminated test result were present for the final herd test but there was no evidence for any difference in losses to follow up between the other pathogens isolated. The geometric mean for ISCC was 225,000 cells/mL (95%PI=25,000-4,543,145). For the prediction of ISCC recorded at the last herd tests, breed was retained in the model in addition to treatment group and pathogen, with farm and cow as random effects. The model predicted no differences in ISCC by treatment group. There were numeric differences in the median predicted ISCC by pathogen and farm, but the coefficients spanned zero and overlapped. There was no 21
evidence for a significant interaction between treatment group and farm, nor between treatment group and pathogen. The distribution of predicted values for the incidence rate ratio (IRR) for the effect of treatment group on ISCC is shown in Figure 7 along with the 95%PI. Eighty-four percent of values were below 1.0 but the lower limit of the 95% HPDI exceeded the ROPE of +/-10% difference in the IRR (equivalent to a change in ISCC of 22,500 cells/mL at a mean of 225,000 cells/mL). Figure 7. Distribution of the predicted values for the incidence rate ratio for the effect of selective treatment on the likelihood of clinical recurrence of mastitis. The 95% highest probability density
ro of
interval is marked by vertical red lines while the region of probable equivalence is marked by the
Jo
ur
na
lP
re
-p
thickened horizontal line on the y axis
22
ro of -p re lP na ur
The predicted ISCC 30-35 days before dry-off, adjusted for breed, farm and cow for each pathogen by
Jo
treatment group is given in Table 6. Table 6. Predicted individual somatic cell count (ISCC) 30-35 days before dry-off, adjusted for breed, farm and cow for each pathogen by treatment group Non-selective Median ISCC x 103 and 95%PI Non aureus staphylococci 240 (120-482) Coliforms 180 (85-387) Gram-positive 323 (208-513) Mixed 80 (19-336) No growth 248 (150-411) Pathogen
Selective Median ISCC x 103 and 95%PI 214 (110-423) 162 (75-346) 287 (184-457) 72 (16-309) 220 (133-365) 23
Staph. aureus Strep. uberis
365 (222-588) 252 (172-371)
325 (189-532) 224 (152-327)
The distribution of the predicted individual somatic cell count (ISCC) 30-35 days before dry-off for cows from which Staph. aureus was isolated by treatment group is shown in Figure 8a and Strep. uberis in Figure 8b. Figure 8a. Predicted distribution of individual somatic cell count 30-35 days before dry-off for cows from which Staph aureus was isolated. Quarters were subjected to non-selective or selective
ro of
antibacterial treatment. Red shading indicates non-selective treatment, blue shading selective
Jo
ur
na
lP
re
-p
treatment and dotted vertical lines the predicted median.
24
ro of -p re lP na ur
Jo
Figure 8b. Predicted distribution of individual somatic cell count 30-35 days before dry-off for cows from which Strep. uberis was isolated. Quarters were subjected to non-selective or selective antibacterial treatment. Red shading indicates non-selective treatment, blue shading selective treatment and dotted vertical lines the predicted median.
25
ro of -p
re
Quantity of antimicrobial used in each group
The mean antimicrobial usage per quarter case was 1.46 (95%PI=0.00-1.50) daily doses. This is
lP
consistent with the expected range of usage from the treatment protocols (0.0-3.0 daily doses per quarter case, with an expectation closer to 0.0-1.50, given that most treatments involve 3 x 12 hourly
na
treatments). For the prediction of antimicrobial use, group was forced into the model with farm and cow as random effects. Including farm as a random effect changed the value of the remaining
ur
coefficients in the mode by more than 10% and the coefficients for farm spanned zero. The model predicted differences in antimicrobial usage by treatment group. There was no evidence for a
Jo
significant interaction between treatment group and farm. The distribution of predicted values for the incidence rate ratio for the effect of treatment group on antimicrobial use is shown in Figure 9 along with the 95%PI. Ninety-eight percent of values were below 1.0 but the lower limit of the 95% HPDI exceeded the ROPE of +/-10% difference in the IRR (equivalent to a change in antimicrobial use of 0.15 at a mean of 1.46 daily doses/case). Figure 9. Distribution of the predicted values for the incidence rate ratio of the effect of selective treatment on antimicrobial use for cows treated with mild to moderate clinical mastitis. The 95% 26
highest probability density interval is marked by vertical red lines while the region of probable
Jo
ur
na
lP
re
-p
ro of
equivalence is marked by the thickened horizontal line on the y axis
The model predicted that the there was a 98% chance that antibiotic usage in the S group (1.3 daily doses per case, 95%PI=1.1-1.6) was less than in the NS group (1.7, 95%PI=1.4-1.9). The predicted distribution of antimicrobial use for NS and S groups is shown in Figure 10. Figure 10. Predicted distribution of antimicrobial use for cows treated for mild to moderate clinical mastitis. Quarters were subjected to non-selective or selective antibacterial treatment. Red shading indicates non-selective treatment, blue shading selective 27
Days for which milk was withheld from supply The mean period for which milk was withheld from sale was 5.7 days per quarter case (95%PI=1.06.5). For the prediction of milk withhold, group was forced into the model with farm and cow as
ro of
random effects. Including farm as a random effect changed the value of the remaining coefficients in the mode by more than 10% and the coefficients for farm spanned zero. The model predicted no
differences in the period for which milk was discarded by treatment group. There was no evidence for
-p
a significant interaction between treatment group and farm.
re
The distribution of predicted values for the incidence rate ratio for the effect of treatment group on days for which milk was withheld shown in Figure 11 along with the 95%PI. The null value of 1.0
lP
was the most probable, but 22% of values were above the upper limit of the ROPE of +/-10% difference in the IRR (equivalent to a change in antimicrobial use of 0.57 at a mean of 5.7 days/case).
na
Figure 11. Distribution of the predicted values for the incidence rate ratio of the effect of selective treatment on milk withhold for cows treated with mild to moderate clinical mastitis. The 95% highest
ur
probability density interval is marked by vertical red lines while the region of probable equivalence is
Jo
marked by the thickened horizontal line on the y axis
28
ro of -p re lP na ur Jo The model predicted that the there was a 63% chance that milk withhold in the S group (5.6 days/ per quarter case, 95%PI=4.0-8.6) was more than in the NS group (5.2, 95%PI=3.7-7.4). 29
The predicted distribution of milk withhold for NS and S groups is shown in Figure 12. Figure 12. Predicted distribution of milk withhold for cows treated for mild to moderate clinical mastitis. Quarters were subjected to non-selective or selective antibacterial treatment. Red shading
ur
na
lP
re
-p
ro of
indicates non-selective treatment, blue shading selective
Jo
Discussion
We found the use of an on-farm diagnostic system to inform therapeutic choice for mild to moderate clinical mastitis reduced the use of antimicrobials by 24% with no evidence detected for a difference in bacteriological cure at 21 days, clinical recurrence within 60 days, ISCC measured 30-35 days before dry-off or days milk was withheld from sale, The reduction in use of antimicrobials reported here is less than the 56% reported by Lago et al., (2011) using on-farm culture in confinement systems but very similar to the 25% reduction from an 30
on-farm diagnostic study in a NZ pasture based systems McDougall et al., (2018). Neither of these studies used antimicrobial sensitivity to guide treatment choices. The proportion of no-growth and Gram-negative isolates will be a key driver of differences in outcome metrics for on-farm diagnostic systems as will the difference in cure rate between deferred and blanket treated cows (Down et al., 2017). At 8.6%, the proportion of no-growths that we report is similar to the 8% reported by McDougall et al., (2018) for samples analysed on-farm in a NZ pasture based study. This is in contrast to other NZ studies using laboratory based diagnosis where nogrowths have made up 26% (McDougall, 1998); 18% (Bryan et al., 2016) and 24% (McDougall et al.,
ro of
2018). In this last study, these were duplicates to the samples processed on-farm. These workers
suggest that the increase in no-growths from on-farm to laboratory analysis may have been due to
freeze/thawing of samples collected on-farm and diagnosed off-site. In our study, an additional factor
-p
may have been the greater volume of milk processed through the Mastatest cassette (circa 5mL) compared to a standard inoculum of 10-20 μL.
re
In the present study, days of milk withhold was numerically greater in the selective group. This contrasts to the 12% reduction in discard time reported for selective therapy by Lago et al., (2011a)
lP
and 34% by Vasquez et al., (2017) and probably reflects the lower proportion of no-growth and Gram-negative isolates in pastoral systems. The treatment choice used for each group will also affect
na
differences in milk discard time. For example, McDougall et al., (2018) used extended therapy for the treatment of some Staph. aureus cows in the selective group while others have recommended
ur
combined cloxacillin-penicillin therapy for Staph. aureus infections (Clews et al., 2018). In their simulation study looking at factors affecting the cost effectiveness of on-farm diagnosis,
Jo
Down et al., (2017) used a difference in predicted cure rate between selective and blanket treatment estimated at -11% (0-22%) from Lago et al., (2011). The simulation suggested that even if the difference in cure rate was -5 to 0%, on-farm diagnosis was only cost effective where the proportion of gram-positive bacteria was less than 47%. In our study, 86.6% of isolates were Gram-positive but numerically, the probability of bacterial cure was greater in the selective group with 84% of the estimated values for the OR for the effect of deferred treatment exceeding 1.0. However, the HPDI interval was wide and included some negative values. 31
Given the necessary caveats about the value of the antimicrobial sensitivity testing in regard to the in vivo clinical response of individual cows (Watts et al., 2018) it is not clear from the present work how the addition of antimicrobial sensitivity may have influenced the bacterial cure rate of the selective group. If true in vivo antimicrobial performance could be predicted, a high proportion of penicillin resistant infections would be expected to lead to more treatment failure in the blanket group (Barkema et al., 2006) and increase the benefit of selective therapy. Antimicrobial sensitivity varies between countries and across time (Erskine et al., 2002) and needs to be defined under particular conditions for each pathogen, host species and disease process (www. https://clsi.org/). In NZ, 73-74% Staph.
ro of
aureus infections are sensitive in vitro to penicillin (McDougall et al., 2014; Petrovski et al., 2011)
compared to 79% of isolates in the present study. In the absence of validated veterinary break-points for the pathogen, drug, disease process, and host combination in vitro antimicrobial sensitivity may
-p
not reflect in vivo performance. Although Mastatest appears to accurately predict pathogen type and whether antimicrobials are indicated, in the absence of robust, randomised field data as to likely in
re
vivo antimicrobial performance, pathogen type in conjunction with advice from the herd’s veterinarian should remain the basis for choice of individual antimicrobial type.
lP
Most other plate based on-farm diagnostic systems also provide a readable result within 24 hours except for a proportion of no-growths which require a further 24 hours of culture (Lago and Godden,
na
2018). Test performance will also be a key determinant of benefit. For the detection of Gram-positive organisms, Lago and Godden, (2018) reported a Se of 78% and Sp of 83% for on-farm culture,
ur
compared to Se of 94.6% and Sp of 72.1% against all target species for Mastatest reported by Jones et al., (2019). Issues with farmer interpretation of culture results (Lago and Godden, 2018; Ruegg,
Jo
2018) are also avoided with the automated interpretation of Mastatest results. We cannot rule out the effect of unintentional biases in our study. The farms involved were potentially atypical in that they had all expressed interest in on-farm diagnosis and were prepared to invest time and money in the study. The level of clinical mastitis reported in this study is similar to that reported in other similar NZ studies (McDougall et al., 2018) but it is possible that the enrolled farms are atypical and so care needs to be taken with extrapolation of the results. The study design
32
was designed in consultation with farm staff and veterinarians, but more effort was required for treatment of mastitis than normal. Analysis based on farmer diagnosis of clinical mastitis will lead to inconsistencies in application of inclusion and exclusion criteria (Bates and Saldias, 2017) and farmers may have ignored treatment allocation rules if they felt cases required immediate treatment (McDougall et al., 2018). Regular farm visits by veterinary staff during the study were designed to offset some of these risks. To be a confounder, farm bias would have to be causally associated with the outcome and associated (causally or not) with the testing method (Aly et al., 2010). This would mean that farmers would have to deliberately skew the balance of cases that they put into each
ro of
diagnostic path way (so ignoring their prior agreement to follow the treatment schedule) even though they would have had no prior knowledge of what a particular case would yield on culture. This seems unlikely. Moreover, the distribution of cases and the other independent variables was not different
the model to statistically adjust for any confounding.
-p
between the two treatment groups and where there was any doubt we forced the relevant variable into
re
Although adequately powered to detect the major effects of our study, we had insufficient power to detect differences in outcome for the proportion of bacteriological cure and clinical recurrence for
lP
quarters infected with Staph. aureus. With an expected bacteriological cure rate of 35% we would have needed > 375 quarters in each group. In general, the 95% HPDI was wide and at one end,
na
exceeded the ROPE that we had used. This implies a non-zero probability that the observed difference between the treatment groups could have been greater than the null. There was
ur
considerable variation between farms in the effects measured and although the HPDI for individual farms overlapped, the overall effect of farm was significant in the model. Although we found no
Jo
evidence of an interaction between group and farm or group and treatment, the likelihood of cure particularly for Staph. aureus infections may potentially be influenced by unmeasured cow and farm factors (Barkema et al., 2006; Roberson, 2003; Ruegg, 2018). Our study adds to the evidence that there are potential gains in on-farm diagnosis in terms of reduction in antimicrobial use without a decrease in short and medium term health outcomes (Lago et al., 2011a, 2011b; McDougall et al., 2018; Vasquez et al., 2017). In deciding whether this will be an attractive option for the majority of NZ dairy farmers, we acknowledge that potential economic 33
benefits will be influenced by the proportion of Gram-negative and no-growth isolates and the extent to which our findings from 7 convenience selected farms will be replicated elsewhere. However, our work and that of McDougall et al., (2018) appear to suggest that for at least on these farms and for some pathogen groups the delay in treatment is beneficial in aggregate to decreasing antibiotic use without detriment to animal health or production. As such, the economic benefits of deferred treatment could be calculated using an approach similar to (Down et al., 2017) and this will form part of a follow up study to the present paper.
ro of
Acknowledgements We would like to thank the farm owners and operatives who took part in this trial and our colleagues throughout Vetlife who collected samples, made time available and facilitated the smooth running of
-p
this trial. We would like to acknowledge the support of Vetlife and Mastatest management in
Jo
ur
na
lP
re
investing in clinical trial work
34
References
Aly, S., Anderson, R., Adaska, J., Jiang, J., Gardner, I., 2010. Association between Mycobacterium avium subspecies paratuberculosis infection and milk production in two California dairies. J. Dairy Sci. 93, 1030–1040. doi:10.3168/jds.2009-2611
Bååth, R., 2014. Bayesian First Aid: A Package that Implements Bayesian Alternatives to the
ro of
Classical *.test Functions in R., in: Proceedings of UseR! 2014 - the International R User Conference. p. 1.
Barkema, H.W., Schukken, Y.H., Zadoks, R.N., 2006. Invited Review: The Role of Cow, Pathogen,
-p
and Treatment Regimen in the Therapeutic Success of Bovine Staphylococcus aureus Mastitis. J.
re
Dairy Sci. 89, 1877–1895.
Bates, A., Saldias, B., 2017. Effect of treatment with an internal teat sealant at drying-off in cows
lP
wintered on forage crops in New Zealand on clinical mastitis and somatic cell counts. N. Z. Vet. J. 1–25. doi:10.1080/00480169.2017.1401494
na
Bryan, M., Hea, S.Y., 2017. A survey of antimicrobial use in dairy cows from farms in four regions of New Zealand. N. Z. Vet. J. doi:10.1080/00480169.2016.1256794
ur
Bryan, M.A., Hea, S.Y., Mannering, S.A., Booker, R., 2016. Demonstration of non-inferiority of a novel combination intramammary antimicrobial in the treatment of clinical mastitis. N. Z. Vet. J.
Jo
doi:10.1080/00480169.2016.1210044 Clews, M., Kenyon, A., Johnston, K., Oehley, S., Pike, H., Taylor, K., Wyatt, K., Durel, L., 2018. Treatment outcomes of clinical mastitis treated with penicillin or cloxacillin on either a case by case basis after milk culture or by prescription based on numbers of days post partum., in: Proceedings of the 2018 International Bovine Mastitis Conference. Milan, pp. 1–2. Down, P.M., Bradley, A.J., Breen, J.E., Green, M.J., 2017. Factors affecting the cost-effectiveness of 35
on-farm culture prior to the treatment of clinical mastitis in dairy cows. Prev. Vet. Med. doi:10.1016/j.prevetmed.2017.07.006 Erskine, R.J., Walker, R.D., Bolin, C.A., Bartlett, P.C., White, D.G., 2002. Trends in antibacterial susceptibility of mastitis pathogens during a seven-year period. J. Dairy Sci. 85, 1111–1118. Higham, L.E., Deakin, A., Tivey, E., Porteus, V., Ridgway, S., Rayner, A.C., 2018. A survey of dairy cow farmers in the United Kingdom: knowledge, attitudes and practices surrounding antimicrobial use and resistance. Vet. Rec. doi:10.1136/vr.104986
ro of
Hillerton, J.E., Irvine, C.R., Bryan, M.A., Scott, D., Merchant, S.C., 2017. Use of antimicrobials for animals in New Zealand, and in comparison with other countries. N. Z. Vet. J. doi:10.1080/00480169.2016.1171736
-p
Jones, G., Bork, O., Ferguson, S.A., Bates, A., 2019. Comparison of an on-farm point-of-care
doi:10.1017/S0022029919000177
re
diagnostic with conventional culture in analysing bovine mastitis samples. J. Dairy Res. 1–4.
lP
Kruschke, J., 2016. Bayesian estimation of log-normal parameters - Update [WWW Document]. URL http://doingbayesiandataanalysis.blogspot.com/2016/04/bayesian-estimation-of-log-normal.html
na
Kruschke, J.K., 2018. Rejecting or Accepting Parameter Values in Bayesian Estimation. Adv. Methods Pract. Psychol. Sci. 1, 270.
ur
Lago, A., Godden, S., Bey, R., Ruegg, P., Leslie, K., 2011a. The selective treatment of clinical mastitis based on on-farm culture results: I. Effects on antibiotic use, milk withholding time, and
Jo
short-term clinical and bacteriological outcomes. J. Dairy Sci. 94, 4441–4456. doi:10.3168/jds.2010-4046
Lago, A., Godden, S.M., 2018. Use of Rapid Culture Systems to Guide Clinical Mastitis Treatment Decisions. Vet. Clin. North Am. Anim. Pract. doi:10.1016/j.cvfa.2018.06.001 Lago, A., Godden, S.M., Bey, R., Ruegg, P.L., Leslie, K., 2011b. The selective treatment of clinical mastitis based on on-farm culture results: II. Effects on lactation performance, including clinical 36
mastitis recurrence, somatic cell count, milk production, and cow survival. J. Dairy Sci. 94, 4457–4467. McDougall, S., 1998. Efficacy of two antibiotic treatments in curing clinical and subclinical mastitis in lactating dairy cows. N. Z. Vet. J. 46, 226–232. doi:10.1080/00480169.1998.36094 McDougall, S., Compton, C.W.R., Botha, N., 2017. Factors influencing antimicrobial prescribing by veterinarians and usage by dairy farmers in New Zealand. N. Z. Vet. J. 1–9. doi:10.1080/00480169.2016.1246214
ro of
McDougall, S., Hussein, H., Petrovski, K.R., 2014. Antimicrobial resistance in Staphylococcus
aureus, Streptococcus uberis and Streptococcus dysgalactiae from dairy cows with mastitis. N. Z. Vet. J.
-p
McDougall, S., Niethammer, J., Graham, E.M., 2018. Antimicrobial usage and risk of retreatment for mild to moderate clinical mastitis cases on dairy farms following on-farm bacterial culture and
re
selective therapy. N. Z. Vet. J. 66, 98–107. doi:10.1080/00480169.2017.1416692
lP
Petrovski, K.R., Laven, R.A., Lopez-Villalobos, N., 2011. A descriptive analysis of the antimicrobial susceptibility of mastitis-causing bacteria isolated from samples submitted to commercial
na
diagnostic laboratories in New Zealand (2003-2006). N. Z. Vet. J. doi:10.1080/00480169.2011.552853
ur
Plummer, M., 2012. JAGS: Just Another Gibbs Sampler. Comput. Softw. Man. R Core Team (2013). R: A language and environment for statistical computing., 2013.
Jo
Roberson, J.R., 2003. Establishing treatment protocols for clinical mastitis. Vet. Clin. North Am. Food Anim. Pract. 19, 223–234.
Ruegg, P.L., 2018. Making Antibiotic Treatment Decisions for Clinical Mastitis. Vet. Clin. North. Am. Anim. Pract. doi:10.1016/j.cvfa.2018.06.002 Stan Development Team, 2018. ShinyStan: Interactive Visual and Numerical Diagnostics and
37
Posterior Analysis for Bayesian Models. Steele, N., Williamson, J., Thresher, R., Laven, R., Hillerton, J., 2017. Evaluating a commercial PCR assay against bacterial culture for diagnosing Streptococcus uberis and Staphylococcus aureus throughout lactation. J. Dairy Sci. 100, 3816–3824. doi:https://doi.org/ 10.3168/jds.2016-11752 Vasquez, A.K., Nydam, D. V, Capel, M.B., Eicker, S., Virkler, P.D., 2017. Research: Clinical outcome comparison of immediate blanket treatment versus a delayed pathogen-based treatment
ro of
protocol for clinical mastitis in a New York dairy herd. J. Dairy Sci. 100, 2992–3003.
Watts, J.L., Sweeney, M.T., Lubbers, B. V, 2018. Antimicrobial Susceptibility Testing of Bacteria of
Jo
ur
na
lP
re
-p
Veterinary Origin. Microbiol. Spectr. doi:10.1128/microbiolspec.ARBA-0001-2017
38
Jo
ur
na
lP
re
-p
ro of
Figure 1. Treatment decision tree used on-farm for cows diagnosed with mild to moderate mastitis in a study looking at the effect of deferred treatment for clinical mastitis on 7 New Zealand dairy farms
39