You are on page 1of 10

This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:

Author's personal copy
Veterinary Parasitology 197 (2013) 614–622

Contents lists available at ScienceDirect

Veterinary Parasitology
journal homepage:

Hierarchical model for evaluating pyrantel efficacy against
strongyle parasites in horses
M.K. Nielsen a,∗ , A.N. Vidyashankar b , B.M. Hanlon c , G. Diao b ,
S.L. Petersen d , R.M. Kaplan e

M.H. Gluck Equine Research Center, Department of Veterinary Science, University of Kentucky, Lexington, KY, USA
Department of Statistics, George Mason University, Fairfax, VA, USA
Department of Statistics, University of Wisconsin, Madison, WI, USA
EquiLab Laboratory, Slangerup, Denmark
Department of Infectious Diseases, University of Georgia, Athens, GA, USA

a r t i c l e

i n f o

Article history:
Received 19 September 2012
Received in revised form 15 April 2013
Accepted 19 April 2013
Anthelmintic resistance
Hierarchical models

a b s t r a c t
Anthelmintic resistance is an increasing challenge for the control of equine parasites. The
fecal egg count reduction test (FECRT) is the practical gold standard method for evaluating
reduction in anthelmintic efficacy, but the interpretation is complicated due to high levels of variability. A hierarchical statistical model was described for analysis of FECRT data
from multiple farms to evaluate the role of biological factors in determining the strongyle
efficacy of pyrantel pamoate in a study performed in Denmark. The model was then used
to describe two notions of farm efficacy, namely conditional and marginal efficacy. The
median of the lower prediction limits was used to describe a robust classification rule. The
performance of the methodology was evaluated using Monte Carlo simulations. The field
study was performed on 64 Danish horse farms of different breeds. Of 1644 horses, 614
had egg counts ≥200 eggs per gram (EPG) and were treated. Individual coprocultures were
performed for identification of Strongylus vulgaris from all horses pre-treatment. Thirty-one
farms (48.4%) were positive for S. vulgaris, but pyrantel efficacy was unaffected by the presence of this parasite in the statistical model. Further, there were no significant effects of age,
gender, or interactions between these, while the pre-treatment egg count was negatively
associated with the egg count reduction. The statistical model classified 81.3%, 10.9%, and
7.8% of farms as no signs of resistance (NR), suspect resistance (SR), and resistance (RE),
respectively. In comparison, arithmetic calculations classified 68.8%, 17.2%, and 14.1% in
the same categories. Using 10,000 simulated data sets, the methodology provided a classification of farms into different efficacy categories with a false discovery of reduced farm
efficacy rate equaling 8.74%. In addition, model-classification was unaffected by presence
of single outlier horses in a separate simulation study.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction
Anthelmintic resistance has become a universal problem in gastrointestinal parasites of horses (Kaplan et al.,
2004), and it is considered an inevitable biological

∗ Corresponding author. Tel.: +1 859 218 1103; fax: +1 859 257 8542.
E-mail address: (M.K. Nielsen).
0304-4017/$ – see front matter © 2013 Elsevier B.V. All rights reserved.

consequence of anthelmintic treatment. Because these
reductions in anthelmintic efficacy are known to be a growing problem, it can no longer be assumed that a given drug
is working, and hence a routine monitoring of drug efficacy
is needed (Kaplan, 2004; Wolstenholme et al., 2004). For a
precise definition of efficacy, see Section 2.3.2.
However, in the absence of reliable genetic markers,
other biological measures can be used to evaluate if there
are reductions in efficacy. The ultimate gold standard for

Author's personal copy
M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622

determining efficacy are controlled efficacy and critical test
type studies, where animals are euthanatized and total
worm counts are performed (Drudge and Lyons, 1977).
However, these methods are only practical in research
settings on a limited number of animals and cannot be routinely used at the farm level. In addition, ethical concerns
are growing with respect to studies involving euthanasia of
animals, and these may become difficult to obtain permission to perform in the future. Some in vitro assays have
been validated for usage in sheep, goats, and cattle, but
none have so far proven useful for horses (Matthews et al.,
2012). Furthermore, in vitro assays are technically difficult to perform, and have many other limitations, which
severely restrict their usefulness on a wide scale. Given
the lack of alternative methods, the only practical technique available for evaluating efficacy at the farm level is
the fecal egg count reduction test (FECRT). This method has
been applied routinely for measuring anthelmintic efficacy
in ascarid, trichostrongyle, and strongyle type parasites
in all livestock host species and for all drug classes; and
can be carried out by veterinary practitioners in the field.
The test is based on comparing fecal egg counts (FECs) in
groups of animals before and after treatment or by comparing FEC in anthelmintic treated and non-treated control
groups. FECRT data are straightforward to obtain, but the
interpretation of data is complicated due to a number of
factors, as recently outlined (Vidyashankar et al., 2012)
and mentioned in the following. Studying equine populations is more complex than ruminant or swine herds,
where animals typically are of the same breed and uniform
age groups. In contrast horse farms can house a variety
of different breeds, and age distributions are often wide
ranging from foals to geriatrics. It is known that younger
animals tend to have higher egg counts (Morris et al., 2004;
Fritzen et al., 2010), and anthelmintic efficacy can potentially vary with age groups. Additionally, a typical horse
farm is comprised of three genders: mares, geldings and
stallions, where stallions are often differently managed.
Indeed, variability in FECRT results between horse genders
has been reported previously (Drudge et al., 1982). This
suggests that on equine farms, it may be appropriate to
account for both gender and age effects when interpreting
FECRT data. Additional possible variables include breed and
use of the horses, and farm management. On the horse level,
over 50 strongyle species have been described infecting
horses (Lichtenfels et al., 2008), and the species composition of the parasite burden in the intestinal tract can vary
between horses. All these factors can in turn lead to heterogeneities in efficacies amongst different groups within
a farm making interpretations of farm efficacy complicated.
This paper describes statistical methodologies for interpreting and evaluating factors influencing drug efficacy
from a multiple farm study. A random effects model is
adopted to identify the role of biological factors affecting anthelmintic efficacy. The results of this model were
used to develop a robust classification method. The role
of outliers and their influence on anthelmintic efficacy is
described, and a mechanism is provided to address this
issue. The methodology is illustrated using data obtained
in a Danish field study aiming at evaluating the efficacy of
pyrantel pamoate.


2. Materials and methods
2.1. Field study
The study was carried out as an observational study
amongst farms submitting fecal material for routine
screening to a central laboratory (Equilab Laboratory,
Slangerup, Denmark). Data were collected between March
17 and May 1, 2008.
2.2. Farms
Farms participating in the study were selected among
farms following a parasite control strategy based on selective therapy for at least one year prior to the study. All farms
with at least six horses exceeding 200 strongyle eggs per
gram during the study period were included in the study. In
this parasite control regimen, all the farms submitted fecal
samples from all horses for analyses twice yearly; in the
spring (March–May) and in the fall (September–October).
On each occasion, horses exceeding the cutoff value for
treatment of 200 strongyle eggs per gram (EPG) were
treated. None of the horses in the study had been treated
for at least four months prior to the sampling for this study.
2.2.1. Regions
Farms represented all geographic regions (counties) of
Denmark. For the statistical analyses, farms were clustered according to their postal zip codes as follows: 1:
2000–2999: larger Copenhagen and suburbs (17 farms),
2: 3000–3999: Northern Zealand (9 farms), 3: 4000–4999:
Remainder of Zealand and adjacent islands (17 farms), 4:
5000–5999: Funen (14 farms) and 5: 6000 and above:
Jutland (7 farms). These regions are justified by Danish
geography, which is characterized by separate islands and
one peninsula (Jutland).
2.2.2. Fecal egg counts and treatments
Fecal egg counts were performed from all horses using
a modified McMaster technique with a detection limit of
20 EPG (Roepstorff and Nansen, 1998). All egg counts were
performed by the fifth author. Horses shedding 200 or more
strongyle EPG were orally treated with pyrantel embonate paste (Banminth vet., Orion Animal Health, Ballerup,
Denmark) at the labeled dosage of 19 mg/kg, which equals
6.6 mg pyrantel base/kg.
Pre- and post-treatment fecal samples were collected
at days 0 and 14, respectively, refrigerated and refrigerated
for up to five days until egg counts were performed according to general recommendations (Nielsen et al., 2010).
2.2.3. Coprocultures
Coprocultures were performed individually from all
1644 horses on the pre-treatment samples. Ten grams of
feces were moistened with tap water, mixed with vermiculite and set up for culture in a humidity chamber at room
temperature. After 14 days, cultures were baermannized
for 24 h at room temperature, and the whole sediment was
examined under the microscope at 100 and 400× magnification. Third stage larvae of Strongylus vulgaris were
identified according to described criteria (Russell, 1948),

Author's personal copy

M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622

and horses were recorded as either being S. vulgaris positive or negative. Farms were classified as S. vulgaris present
if any of the horses tested positive for the presence of S.
vulgaris on the coproculture.
2.3. Statistical analyses
All statistical analyses were performed using the glimmix procedure with SAS version 9.3 (SAS Institute Inc., Cary,
NC, USA). Six horses with post-treatment egg count larger
than the pre-treatment egg count were removed from the
data set prior to the analysis. Biologically, anthelmintic
treatment is expected to reduce egg counts, even if some
efficacy is lost, and thus it was assumed that these six
horses were either incorrectly treated or accidentally not
treated at all. All remaining horses had egg count reductions of 40% or higher.
Data were first fitted to a hierarchical generalized linear mixed logistic regression model where pre-treatment
egg count, age, gender, and the presence of S. vulgaris on
farms were fixed effects and horses, farms and regions
were random effects. The overall variability in observed
efficacy can be described to be due to intra-farm and interfarm variability. The intra-farm variability was captured
using a horse specific random effect and inter-farm variability was captured using a farm specific random effect.
The horse specific random effect was introduced to account
for possible differences in the species composition of the
parasitic burden between horses. To account for common
characteristics between farms within a region, a region specific random effect was incorporated. All the explanatory
variables and random effects were combined additively to
model the efficacy on a logit scale. All possible interaction
terms were evaluated during model construction, as well as
residual analyses and goodness of fit tests were performed
to check the adequacy of the model fit.
Using the fitted statistical model (1.2), the lower limits
of the 95% predicted intervals for pijk were used to assess
reduction in pyrantel efficacy and classify the farms into
signs of resistance (RE), suspect resistance (SR), and no
resistance (NR) categories. Since the prediction intervals
depend on the covariates, either the mean or the median of
these lower limits could be used as a summary measure of
efficacy for each farm. The median was chosen as this summary measure since it is a robust estimator of the center of
the distribution. Using the proposed methods, farms were
classified with three different methods: The model-based
classification, the same model without random effects and
simple arithmetic calculations without statistical modeling.
2.3.1. Statistical model
Let the indices i, j and k represent the horse, farm, and
region respectively. The farms were grouped into the five
regions defined in Section 2.2.2 and the possible values for
k were 1, 2, 3, 4, and 5.
The number of farms within a region changed with the
region and number of horses within a farm changed with
the farm. nk denoted the number of farms in region k, while
njk denoted the number of horses on farm j of region k. Let
(Xijk , Yijk ) represent the pre-treatment and post-treatment

egg count of the ith horse, in the jth farm which belonged
to the kth region. It was assumed that,
Yijk |pijk , Xijk ∼Bin(Xijk , 1 − pijk ),


where pijk represented the pyrantel efficacy on horse i, on
farm j, and belonging to region k. Let Aijk denote the age of
the ith horse in the jth farm which belongs to the kth region.
The regression model to assess reduction in efficacy was 


1 − pijk 

= ˇ0 + ˇ1 Xijk + ˇ2 Aijk + ˇ3 Vjk



ˇ3+l Gijkl + Hijk + Fjk + Rk



where ˇ0 was the overall mean efficacy, ˇ1 was the slope
associated with pre-treatment egg count, and ˇ2 was the
slope associated with the age. The variable Vjk was used as
an indicator variable representing the presence or absence
of S. vulgaris in each farm. Thus, the effect ˇ3 represented
the mean effect of S. vulgaris presence on a farm on the
logit of the efficacy of the pyrantel paste. In general, Gijk
represented the gender variable and took values 1, 2, 3 and
4, where 1 represented male horses, 2 represents female
horses, 3 represented neutered males and 4 for horses
whose gender information was unknown. Thus,
Gijkl = I[Gijk=l ], for l = 1, 2, 3, 4.


The parameters ˇ4 , ˇ5 , and ˇ6 represented mean effects
of each gender group. Hijk represented the horse effect, Fjk
represented the farm effect, and Rk represented the region
effect. It was assumed that Hijk , Fjk , and Rk were latent
effects. These latent effects are assumed to be mutually
independent normal random variables with expectations
0 and variances H2 , F2 , and R2 respectively. This formulation implies that the, variances of farms were all the same
across regions and the correlations between farms within
a region were all the same.
We notice that, conditioned on pre-treatment egg count
values, the post-treatment values have a binomial distribution, whose success probabilities are random and drawn
from a normal distribution on the logit scale. This is biologically justified, since the group of equine strongyle
parasites comprise more than fifty species, existing in complex mixed-species infections (Lichtenfels et al., 2008), but
the eggs of these species cannot be differentiated using traditional methods. The composition of the strongyle fauna
could potentially be different between individual horses.
This justifies the use of binomial distribution with random
success probabilities.
It has been observed that pre-treatment egg count levels cause bias in the calculated values for drug efficacy
(Levecke et al., 2012a,b). Animals with low to moderate pre-treatment egg count levels tend to have higher
observed efficacies than horses with high pre-treatment
egg counts, and this appears to be explained by the detection limit of the egg counting technique (Levecke et al.,
2012a). Thus, a high measurement of efficacy is biased
toward horses with lower pre-treatment egg count levels.
We note that a typical farm contains a mix of horses belonging to three basic categories: low egg shedders, moderate

Author's personal copy
M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622


and 0.92, with LPLs between the two values falling in the SR
category. The principles behind this are outlined in Fig. 1.
Where classification discrepancies occurred between
the model-based classification and the arithmetic calculations, data were revisited by removing the one horse with
the lowest arithmetic FECRT on each farm to evaluate the
possible influence of single outlier horses on classification
using the two methods.

Fig. 1. Illustration of the model-based method for classifying farms as
no resistance (NR), suspect resistance (SR), or resistance (RE). The figure
shows the means and prediction intervals (error bars) for three theoretical
farms. The two lower prediction limit (LPL) cutoff values used here are
92% and 88% (bold lines). For the first farm, the whole confidence interval
stays above the 92% cutoff, and it is determined having NR. The second
farm moves the LPL into the suspect range of 88–92% and is determined
SR. The third farm extends the LPL well below the 88% cutoff value and is
determined RE.

egg shedders, and high egg shedders. For these reasons,
we include pre-treatment egg count as a covariate in the
2.3.2. Definitions of farm efficacy
Since pijk defined above is a random variable influenced
by horse level, farm level, and region level random effect
distributions, the farm efficacy for the jth farm in the kth
region can be defined in two ways: (1) marginal farm
efficacy and (2) conditional farm 
njkefficacy. The marginal
farm efficacy is defined to be
E(p )/njk .The condii=1n ijk
p /njk . Ejk is the
tional efficacy is defined to be Ejk =
i=1 ijk
farm efficacy given the random effects as opposed to the
marginal efficacy where we average over all the random
effect distributions.
Assuming that the covariates are centered at the expectations, the marginal efficacy, if defined as an average of
logit of the efficacies, will correspond to the intercept 

pij (k) =

exp ˇ0 + ˇ1 Xij + ˇ2 Aij + ˇ3 Vj + 

2.3.4. False discovery of reduced farm efficacy
The accuracy of the classification method was evaluated by identifying false discovery of reduced farm efficacy
(FDRFE) using Monte Carlo simulations. FDRFE rate is
defined as the percentage of truly no resistance farms
which are incorrectly classified as resistant by the methodology adopted for classification.

2.4. Simulation studies
For the simulation study, we fitted the model (1.2) without region effect and generated data as described below. To
the marginal efficacy for the jth farm, recall that
E(p·j ) =
E(pij )/mj , where mj is the number of horses on
the jth farm. Since E(p.j ) does not have a closed form expression, a Monte Carlo approach was adopted. Specifically, for
every farm 10,000 farm specific normal random variables
and 10,000 horse specific normal random variables were
generated for each horse on that farm. For the kth simulation, we denote the realization of pij as pij (k). For example, if
a farm had six horses, 60,000 horse specific normal random
variables and 10,000 farm specific random variables were
generated. Finally E(p.j ) was calculated using the formula

E(pj ) =
mj × 10, 000

2.3.3. Farm classification
Cutoff lower prediction limit (LPL) values for the classification of farms into RE, SR, and NR categories were 0.88 


pij (k),




ˇ I(Gij = l) + Hij (k) + Fj (k)
l=1 3+l

1 + exp ˇ0 + ˇ1 Xij + ˇ2 Aij + ˇ3 Vj +

term of the statistical model (1.2). However, in the probability metric studied in this manuscript, this is not the case.
These two notions of efficacy are alternatives to the arithmetic mean. When the horses are homogeneous and the
variability in efficacy between horses is negligible, these
two notions reduce to the arithmetic mean and hence they
can be viewed as generalizations of arithmetic mean to the
non-homogeneous case. Due to multiple sources of variability and outliers in various farms, the conditional farm
efficacy was used for the classifications made in this study.

10,000 mj 




ˇ I(Gij = l) + Hij (k) + Fj (k)
l=1 3+l

The true status of the farm was then determined using the
0.95 cutoff: if E(p.j ) ≥ 0.95, the farm’s status is NR, otherwise
the farm’s status is RE. The parameter values were chosen
to be similar to the values obtained from the analysis of the
equine data set.
Now each simulated data set was analyzed using a variant of model (1.2). This fitting facilitated the model-based
classification, and each farm was then classified as RE, SR,
or NR. The FDRFE rate for this study was then computed by
comparing the farms true resistance status and the classified status. Specifically, 10,000 independent data sets were
generated, and the FDRFE was computed as the proportion
of non-resistant farms (in these 10,000 data sets) wrongly
classified as resistant.

Author's personal copy

M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622

Evaluation of the conditional efficacy was performed
similarly to the above, but used the conditional efficacy for
farm j, namely Ej , given by
Ej =



Fixed effects

pij ,


and mj is the number of horses in farm j. Note that the
predicted values of the random effects were used in the
calculation. The true status of the farm (RE or NR) was
again defined using the 0.95 cutoff. Specifically, a farm’s
true status was NR if Ej ≥ 0.95; otherwise it was defined as
2.5. Outlier effects
The effect of single outliers on farm classification was
evaluated using simulated data from a single farm with
12 horses. The data were generated using a regression
model with age, gender, and pre-treatment egg count as
covariates. Let Xi and Yi denote the pre-treatment and posttreatment egg count of the ith horse. The gender of the ith
horse was coded as Gi = 1, 2, and 3 for stallion, mare and
gelding, respectively. Then,
Yi |Xi , pi ∼Bin(Xi , 1 − pi ),
where pi represented the anthelmintic efficacy of the ith
horse. To complete the model specification, we modeled 



1 − pij

= ˇ0 + ˇ1 Xi + ˇ2 Ai + 


I(Gi = l) + ˛i

˛i ∼N(0, 12 ),
where ˛i was the horse specific random effect, Ai is the
age of the ith horse. The values of the covariates were
randomly chosen from a farm with 12 horses. The regression coefficients were set as ˇ = (ˇ0 , ˇ1 , ˇ2 , ˇ3 , ˇ4 ), where
ˇ0 = 3.5 (intercept), ˇ1 = −0.001 (pre-treatment egg count),
ˇ2 = 0.01 (age), ˇ3 = −0.01 (gender = stallion), and ˇ4 = 0.01
(gender = mare). An outlier effect was induced by altering
the post-treatment egg count for horse 1. The observed
post-treatment egg count for horse 1 was replaced with
the altered value Y1∗ , given by 

Table 1
Summary of parameter estimates for fixed and random effects from the
analysis of the observational farm data.

Y1∗ = Y1 ,
for different values of  in the range [1,X1 /Y1 ]. Increasing 
decreased the observed efficacy of horse 1;  = 1 corresponded to the true data, where  = X1 /Y1 corresponded to
the case of zero efficacy. Note that the symbol x denoted
the integer part of x.
3. Results
3.1. Farm data
Sixty-four farms with at least six horses exceeding
200 EPG were included in the study. On each farm, all
horses exceeding this threshold were included in the data
set. Farm size ranged from six to 68 horses (mean = 25.7,
median = 24), with 1644 horses being included in the


Pre-treatment egg count
Gender (no information)
Gender (mare)
Gender (gelding)
Gender (stallion)
Presence of Strongylus vulgaris 0
Presence of Strongylus vulgaris 1


Standard error

Random effects


Standard error




The residual variance was 3.29.

study. A total of 614 out of these horses were treated
during the time period of this study. On the individual
farms, the number of treated horses ranged from 6 to 22
(mean = 9.6, median = 9). Horses aged between 1 and 33
years (mean = 6.9, median = 5) with 49% being geldings,
48% mares and 3% stallions. Horses were primarily mixed
type riding horses, such as Danish Warmbloods, Icelandic
horses, Norwegian Fjordhorse, and other pony breeds. S.
vulgaris was detected in 78 horses (4.75%), with 31 farms
(48.4%) having at least one horse positive.
All statistical estimates provided below are given on the
logit scale, unless otherwise stated. The variance estimates
and standard errors for random effects (horse, farm, and
region) in the model are presented in Table 1. The analyses did not reveal an effect of region. An analysis without
the region as a random effect did not change the conclusions (results not shown). Therefore, region was left in the
model to emphasize the usefulness of this strategy. The
estimate of the variance (standard error) of the horse, the
farm, and the region were estimated to be 6.4137(0.5244),
0.5154(0.3451) and 0.1705(0.2137), respectively. The proportion of variance explained by the horse effect was
61.73% while that by the farm was 4.96%. The proportion
of variance explained by the region was 1.64%.
The estimate of the slope of the pre-treatment egg
count was determined to be −0.0010 (SE = 0.0002), which
was statistically different from 0 (P < 0.0001). The slope for
the age was determined to 0.0113 (SE = 0.0214). Age was
not a significant factor to describe variations in efficacies
(P = 0.5981), and there were no significant differences in
efficacies of horses between genders (P = 0.6838). The data
was also analyzed by removing all horses with missing gender (16%), and results were similar (not shown). Finally,
there were no significant differences in efficacy between
farms with and without S. vulgaris detected (P = 0.9403). No
apparent interactions were found between the parameters
mentioned above. Summaries of parameter estimates of
fixed effects are presented in Tables 1 and 2. Residual analyses and goodness of fit tests (p-values > 0.10) suggested
that the model fit was adequate.
The overall arithmetic mean pyrantel efficacy, obtained
by averaging the FECR from all horses on all farms, was

Author's personal copy
M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622


Table 2
Summary of type III tests of fixed effects from analyzing the observational
farm data set.

Num DF

Den DF

F value


Pre-treatment egg
Presence of Strongylus









Num DF: degrees of freedom of the numerator; Den DF: degrees of freedom of the denominator.

92.93%. Twelve farms (19%) had arithmetic means below
90%, while another 18 farms (28%) were within the 90–95%
range. The LPL-based classification of farms into the categories no resistance (NR), suspect resistance (SR) and
resistance (RE) is presented in Fig. 2. For comparison, the
classification is presented using LPLs or lower confidence
limits (LCLs) obtained from the model-based classification,
the same model without random effects and simple arithmetic calculations without statistical modeling.
Classification of each farm with these three methods
is presented as supplementary files. Fourteen farms (22%)
were differently classified using the model-based classification compared to using arithmetic calculations. On ten of
these farms, the model classified no resistance (NR), while
the arithmetic approach classified either suspect resistance
(SR) or resistance (RE). Removing the horse with the lowest
individual FECR changed the classification for the arithmetic method in all ten cases. For farms CC, E, L, HH, MM,
MMM, P, WW, and Y the arithmetic classification changed
from RE to NR, while farms LLL and MMM both changed
from RE to SR. In each case, the model-based classification
was unaffected by removing the horses with the lowest
As mentioned in Section 2.3, we initially removed six
horses from the data set. It is worthwhile to mention here
that if these horses were included, then (pre-post)/pre
would become negative. As the logarithm of any negative

Fig. 2. Classification of farms into the three categories no resistance (NR),
suspect resistance (SR) and resistance (RE) to pyrantel embonate. The
outcomes are presented comparing three different methods for analyzing FECRT data: full model-based classification (black), model without
random effects (gray), arithmetic classification (white). The arithmetic
classification is based on lower 95% confidence limits.

Fig. 3. Single farm simulation evaluating the influence of single outlier
horses on the classification of the farm into anthelmintic efficacy categories. The farm has 12 horses and one of these is introduced as an outlier
by changing its anthelmintic efficacy as measured by the fecal egg count
reduction test. In the simulation, this horse represented efficacies in the
range of 0.1–1.0. The model-based classification is unaffected by this outlier horse, whereas the arithmetic classification changes for all efficacy
values of this horse in the range of 0.1–0.9. The horizontal stippled lines
indicate the threshold values for pyrantel efficacy categories used in the
present study.

number is not defined, this would make the model undefined. Alternatively, we performed the same analyses, with
the entire data set by setting the post-treatment values to
equal the pre-treatment values. This did not change any of
the conclusions described in Tables 1–3 and the numerical values were very close to the values presented in these
tables. Further, the classifications of the 64 farms remained
unchanged (results not included).
3.2. Simulation data: results for false discovery of
reduced farm efficacy rate
The simulations evaluating the marginal efficacy
revealed that amongst 61 farms, 56 farms had an expected
mean efficacy greater than 92%, and 8.74% of them were
wrongly classified as resistant or belonging to the suspect
resistant category. This then determined the FDRFE rate for
this study.
Table 3 presents the results of the simulations study
where efficacy is evaluated using the conditional definition.
The results are displayed for each of the twelve farms and
then aggregated across all of the farms. The table compares
the performance of the model with the arithmetic classification. In this study, of the 12,000 total simulated farms,
3161 were resistant and 8839 non-resistant. Both methods
had low rates of FDRFE; the model-based classification had
an FDRFE of 0 compared to an FDRFE of 0.0547 for the arithmetic classification. The model outperformed arithmetic
classification in terms of power in detecting truly resistant
farms: 0.7062 versus 0.5135.
3.2.1. Outlier effects
The influence of a single outlier horse on a single
farm is illustrated in Fig. 3. The figure clearly shows that

Author's personal copy

M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622

Table 3
Classification of farms according to their anthelmintic efficacy status based on 1000 simulated data sets, each based on 12 farms. Farms are classified using
the model-based and the arithmetic classifications. Classification results are presented both separated out on each farm, and aggregated across farms.

True status





Data aggregated across farms


















NR = no resistance; SR = suspect resistance; RE = resistance.

model-based classification using the median was unaffected by the outlier, whereas the arithmetic classification
is highly driven by the FECR value of the outlier.
4. Discussion
This study provides a flexible model for combining
horse-specific covariates and random effects on multiple
farms within different regions, and a statistical methodology for evaluating anthelmintic efficacy, which is robust to
the presence of outliers in farms. As a summary statistic
following the model fit, we introduce two new notions of
efficacy, namely, marginal and conditional efficacies. These
arise due to the introduction of random effects into the statistical model. Finally, we validated our methodology using
The choice of cutoff values for determining anthelmintic
resistance deserves more critical attention. Although
guidelines have been published, these are primarily
focused on sheep (Coles et al., 1992), and no recommendations exist for horses. As a result, there is no consensus
for the choice of any of these cutoff values, and each study
tends to use its own definitions (Ihler, 1995; Craven et al.,
1998; von Samson-Himmelstjerna et al., 2007; Lind et al.,
2007; Traversa et al., 2009). As the performance of a diagnostic test critically depends on the choice of cutoff, this
constitutes an underlying problem in assessing reduction
in efficacy of a drug using FECRT data. Clearly, the cutoffs
chosen in this study were arbitrary as well. However, we
based these on historical data available, where pyrantel

was originally reported to be in the range of 94–100% with
substantial levels of variability within and between farms
(Lyons et al., 1975; Drudge et al., 1982). This is much different compared to other anthelmintics such as ivermectin,
where the expected efficacy is 99.9% (Lyons et al., 1980).
Given this information, a lower prediction limit of 92%
appears as a reasonable choice, while it should be higher
for a drug like ivermectin. As pointed out in Vidyashankar
et al. (2012), there are several technical, biological, and
sampling factors which are typically not well-controlled
leading to biased estimates of empirical farm efficacies. To
address this issue, we found it prudent to introduce a category called suspect resistance (SR) category to encourage
people to collect more data before a conclusion is reached.
We choose to classify farms to belong to this category if the
LPL was between 88% and 92%, but other cutoffs may apply.
The concept of using lower prediction limits is new. We
chose this as it reflects both the mean efficacy as well as represents the variability of the data in just one number. This
could be seen as an alternative to the traditional approach
with evaluating a combination of means and lower confidence limits (Coles et al., 1992). However, further work is
needed for identifying useful guideline cutoff criteria for
classifying anthelmintic resistance.
Fig. 2 illustrates the need for including the random
effects in the model. Only accounting for covariates is not
sufficient, since in many cases the sample size will be too
small to make an inference. Thus, combining the information through random effects in individual animals and their
farms is a useful approach. Other authors have addressed

Author's personal copy
M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622

sources of variability by assuming a Poisson distribution of
eggs in a sample and incorporating this into a statistical
approach (Morgan et al., 2005). However, variation arising
from the FEC methodology is just one of the many sources of
variation that impede interpretation of FECRT. By considering some of the important biological sources of variability,
significant progress can be made.
Understanding the slope estimates provides useful
information about the covariates used in the model. For
instance, a negative slope for the pre-treatment egg count
means that as the pre-treatment egg count increases, the
pyrantel efficacy decreases. There are several possible biological explanations for this relationship. In fact, we believe
the explanation is technical, not biological, for the following reasons. Using a technique with a detection limit of 20
EPG, a horse with a 200 EPG count treated with a drug with
a 90% efficacy will be likely to exhibit 0 EPG post-treatment,
because the low level egg count is likely to go undetected.
If a 1000 EPG horse is treated with the same drug, however,
eggs are likely to be detected post-treatment leading to a
lower observed efficacy. Since the level of pre-treatment
egg count is highly variable within and between farms, one
way to account for this relationship is to include it in the
model. Similar observations regarding pre-treatment egg
count levels and detection limit of egg counting techniques
have recently been made by other authors (Levecke et al.,
The slope for age was positive, although not significantly, meaning that pyrantel efficacy tends to increase
with increasing age. This makes sense, because pretreatment egg counts decrease with age, and young horses
tend to exhibit higher egg counts. Understanding the role
of the pre-treatment egg count level outlined above, it
can be deduced that the observed efficacy will tend to
increase with age. There are experimental data suggesting
that anthelmintic efficacy is lower in younger horses (Herd
and Gabel, 1990). It remains plausible that such observations could be fully or partially explained by the higher
pre-treatment egg counts in young horses.
An important question is whether a single observation
on any of these farms can affect the conclusions of the
data analysis if arithmetic methods are used. On ten of the
farms with classification discrepancies between the modelbased classification and the simple arithmetic calculations,
it was clear that removing the horse with the lowest FECR
strongly affected the arithmetic classification, while the
model-based classification was unaffected. This suggests
that any anthelmintic efficacy classification using arithmetic data may be unstable because of the large variability
between horses on each farm and the small group sizes.
Removing just one horse can change the results substantially, which makes the inference concerning anthelmintic
efficacy questionable. In Fig. 3, we show that using farm
median LPLs the outlier effect can be reduced considerably. Altogether, these discrepancies and our simulation
results illustrate that the model classification provides a
reliable and stable classification by taking into account
pre-treatment egg counts and the different sources of variability in FECRT data.
The experience with this data set clearly illustrates that
data sets evaluating pyrantel efficacy should be handled


with care. The overall arithmetic mean efficacy was high
(92.93%), but the large variability complicated the interpretation of data. Overall, it appears that pyrantel would be
useful on most of the studied farms. A direct comparison
with other studies reporting pyrantel efficacy is not possible since different methods were used. Given the higher
variability of pyrantel data, interpretation is further complicated. For instance, the study by Ihler (1995) is often
being reported as the first documentation of pyrantel paste
resistance in cyathostomins based on a 94% mean FECRT
found on one farm. Similarly, a Danish study suggested
early signs of pyrantel resistant cyathostomins on two
farms based on a 95% mean FECRT cutoff value for determining reduced efficacy (Craven et al., 1998). In Germany,
von Samson-Himmelstjerna et al. (2007) found pyrantel
efficacies in the range of 92–100% and concluded four farms
to be suspect resistant due to mean efficacies below 95%.
A Swedish study reported pyrantel efficacies above 95% for
all 23 farms, but a large variability on nine farms led the
authors to conclude suspect resistance (Lind et al., 2007).
Similarly, a recent multinational European study reported
pyrantel efficacies below 90% on farms in Italy, United Kingdom and Germany (Traversa et al., 2009). Altogether, farm
pyrantel efficacy levels reported in other European studies
were often in the 80–100% range, and it is a challenge to
delineate between reduced efficacy and chance variations
of FECRT data.
It is often assumed that presence of large strongyle
species can affect the efficacy data measured for cyathostomins, since Strongylus spp. have not yet shown signs of
developing anthelmintic resistance (Craven et al., 1998;
von Samson-Himmelstjerna et al., 2007). However, this
could not be supported in our analyses. There may be several reasons for this, but in practice large strongyles, when
present, will never dominate the egg output, and typically
more than 95% of the eggs will be of cyathostomin origin. In addition, there are 50 different cyathostomin species
described as infecting equids, and hence this group cannot
be treated as uniform either. With coprocultures, it may be
possible to account for only a handful of the many strongyle
species potentially present. Using the molecular tools that
were recently developed for detection of 21 of the most
common cyathostomin species (Hodgkinson et al., 2003;
Traversa et al., 2007; Cwiklinski et al., 2012.) there is now
a possibility to further account for the species composition
in strongyle burdens and to connect this information with
the efficacy data. This information could then be built into
the statistical model, which would be worthwhile in future
It should be noted that in this data set, FECRTs were
only performed on horses exceeding 200 EPG. This threshold was chosen because of the detection limit of the
egg counting method used (20 EPG), making the method
unsuitable for detecting minor reductions in pyrantel efficacies. Given the clear role of the pre-treatment egg count
level in the FECRT illustrated with this study, it is realistic
to assume that if horses with egg counts below 200 EPG
had been included, the efficacy estimates would have been
biased toward a higher level. However, further studies are
needed to confirm this notion. Further, other factors possibly affecting parasite populations could be recorded in

Author's personal copy

M.K. Nielsen et al. / Veterinary Parasitology 197 (2013) 614–622

future studies and subsequently included in the statistical
model. These could include farm size, stocking rate, breed,
use, level of horse traffic, and anthelmintic treatment regimen employed on the farm. This might allow for accounting
for more of the variability observed in the data set, which
could potentially improve the classification model further.
Ultimately, different egg counting techniques are likely to
perform differently, and more sensitive methods appear to
be preferable (Levecke et al., 2012b).
The proposed statistical methodology is useful to identify beginning stages of reduction in efficacy when data
from multiple farms are available. However, this approach
may not be easily applicable for routine use especially by
veterinary practitioners when dealing with a specific farm.
The methodology used here is not host, worm species, or
drug specific. Therefore, a similar approach can be adapted
to analyze data from other host species. For instance, in
cattle the efficacy of avermectin/milbemycin anthelmintics
against Cooperia spp. is variable even in the absence of
reduced anthelmintic efficacy (Vermunt et al., 1996), and
hence FECRT data are subject to high levels of variability.
In conclusion, we developed a hierarchical model for
evaluating occurrence of reduced anthelmintic efficacy
using observational data from multiple farms. Using the
random effects in the statistical model, we combined information among farms to distinguish between variability and
genuine reduction in efficacy, and thus provide a robust
method to determine which farms are likely to be resistant,
suspect resistant, or not resistant to a given anthelmintic
drug. Alternatively, count models can be used to model the
egg counts and these can be extended to include the random effects. It will be interesting to compare our results
with these alternative approaches.
Appendix A. Supplementary data
Supplementary data associated with this article can be
found, in the online version, at
Coles, G.C., Bauer, C., Borgsteede, F.H., Geerts, S., Klei, T.R., Taylor, M.A.,
Waller, P.J., 1992. World association for the advancement of veterinary parasitology (WAAVP) methods for the detection of anthelmintic
resistance in nematodes of veterinary importance. Vet. Parasitol. 44,
Craven, J., Bjørn, H., Henriksen, S.A., Nansen, P., Larsen, M., Lendal, S., 1998.
Survey of anthelmintic resistance on Danish horse farms, using 5 different methods of calculating faecal egg count reduction. Equine Vet.
J. 30, 289–293.
Cwiklinski, K., Kooyman, F.N.J., van Doorn, D.C.K., Matthews, J.B., Hodgkinson, J.E., 2012. New insights into sequence variation in the IGS region
of 21 cyathostomin species and the implication for molecular identification. Parasitology 139, 1–11.
Drudge, J.H., Lyons, E.T., 1977. Methods in the evaluation of antiparasitic
drugs in the horse. Am. J. Vet. Res. 38, 1581–1586.
Drudge, J.H., Lyons, E.T., Tolliver, S.C., Kubis, J.E., 1982. Pyrantel in horses
– clinical trials with emphasis on a paste formulation and activity on
benzimidazole resistant small strongyles. Vet. Med. Small Anim. Clin.
77, 956–967.
Fritzen, B., Rohn, K., Schnieder, T., von Samson-Himmelstjerna, G., 2010.
Endoparasite control management on horse farms – lessons from
worm prevalence and questionnaire data. Equine Vet. J. 42, 79–83.

Herd, R.P., Gabel, A.A., 1990. Reduced efficacy of anthelmintics in young
compared with adult horses. Equine Vet. J. 22, 164–169.
Hodgkinson, J.E., Lichtenfels, J.R., Mair, T.S., Cripps, P., Freeman, K.L.,
Ramsey, Y.H., Love, S., Matthews, J.B., 2003. A PCR-ELISA for the identification of cyathostomin fourth-stage larvae from clinical cases of
larval cyathostominosis. Int. J. Parasitol. 33, 1427–1435.
Ihler, C.F., 1995. A field survey on anthelmintic resistance in equine small
strongyles in Norway. Acta Vet. Scand. 36, 135–143.
Kaplan, R.M., 2004. Drug resistance in nematodes of veterinary importance: a status report. Trends Parasitol. 20, 477–481.
Kaplan, R.M., Klei, T.R., Lyons, E.T., Lester, G., Courtney, C.H., French,
D.D., Tolliver, S.C., Vidyashankar, A.N., Zhao, Y., 2004. Prevalence of
anthelmintic resistant cyathostomes on horse farms. J. Am. Vet. Med.
Assoc. 225, 903–910.
Levecke, B., Rinaldi, L., Charlier, J., Maurelli, M.P., Morgoglione, M.E.,
Vercruysse, J., Cringoli, G., 2012a. Monitoring drug efficacy against
gastrointestinal nematodes when faecal egg counts are low: do the
analytic sensitivity and the formula matter? Parasitol. Res. 109,
Levecke, B., Rinaldi, L., Charlier, J., Maurelli, M.P., Morgoglione, M.E., Vercruysse, J., 2012b. The impact of baseline faecal egg counts on the
efficacy of single-dose albendazole against Trichuris trichiura. Trans.
R. Soc. Trop. Med. Hyg. 106, 128–130.
Lichtenfels, J.R., Kharchenko, V.A., Dvojnos, G.M., 2008. Illustrated identification keys to strongylid parasites (strongylidae: Nematoda) of
horses, zebras and asses (equidae). Vet. Parasitol. 156, 4–161.
Lind, E.O., Kuzmina, T., Uggla, A., Waller, P.J., Hoglund, J., 2007. A field
study on the effect of some anthelmintics on cyathostomins of horses
in Sweden. Vet. Res. Commun. 31, 53–65.
Lyons, E.T., Drudge, J.H., Tolliver, S.C., 1975. Field tests of three salts of
pyrantel [hydrochloride, tartrate, pamoate] against internal parasites
of the horse [Strongylus vulgaris, Parascaris equorum, Strongylus edentatus]. Am. J. Vet. Res. 36, 161–166.
Lyons, E.T., Drudge, J.H., Tolliver, S.C., 1980. Antiparasitic activity of ivermectin in critical tests in equids. Am. J. Vet. Res. 41, 2069–2072.
Matthews, J.B., McArthur, C., Robinson, A., Jackson, F., 2012. The in vitro
diagnosis of anthelmintic resistance in cyathostomins. Vet. Parasitol.
186, 25–31.
Morgan, E.R., Cavill, L., Curry, G.E., Wood, R.M., Mitchell, E.S., 2005. Effects
of aggregation and sample size on composite faecal egg counts in
sheep. Vet. Parasitol. 131, 79–87.
Morris, C., Green, R., Hickey, S., Auldist, M., Thomson, N., Cullen, N., 2004.
Relationships among faecal egg counts, anti-parasite antibodies and
milk yields in an experimental Friesian herd. New Zeal. J. Agric. Res.
47, 267–274.
Nielsen, M.K., Vidyashankar, A.N., Andersen, U.V., DeLisi, K., Pilegaard, K.,
Kaplan, R.M., 2010. Effects of fecal collection and storage factors on
strongylid egg counts in horses. Vet. Parasitol. 167, 55–61.
Roepstorff, A., Nansen, P., 1998. Epidemiology, Diagnosis and Control
Of Helminth Parasites of Swine. Food and Agriculture Organization
(FAO), Rome.
Russell, A.F., 1948. The development of helminthiasis in thoroughbred
foals. J. Comp. Pathol. Therap. 58, 107–127.
Traversa, D., Iorio, R., Klei, T.R., Kharchenko, V.A., Gawor, J., Otranto,
D., Sparagano, O.A.E., 2007. New method for simultaneous speciesspecific identification of equine strongyles (nematoda, strongylida) by
reverse line blot hybridization. J. Clin. Microbiol. 45, 2937–2942.
Traversa, D., von Samson-Himmelstjerna, G., Demeler, J., Milillo, P., Schurmann, S., Barnes, H., Otranto, D., Perrucci, S., di Regalbono, A.F.,
Beraldo, P., Boeckh, A., Cobb, R., 2009. Anthelmintic resistance in
cyathostomin populations from horse yards in Italy, United Kingdom
and Germany. Parasit. Vector 2, S2.
Vermunt, J.J., West, D.M., Pomroy, W.E., 1996. Inefficacy of moxidectin
and doramectin against ivermectin-resistant Cooperia spp. of cattle in
New Zealand. N Z Vet. J. 44, 188–193.
Vidyashankar, A.N., Hanlon, B., Kaplan, R.M., 2012. Statistical and biological considerations in evaluating drug efficacy in equine strongyle
parasites using fecal egg count data. Vet. Parasitol. 185, 45–57.
von Samson-Himmelstjerna, G., Fritzen, B., Demeler, J., Schurmann, S.,
Rohn, K., Schnieder, T., Epe, C., 2007. Cases of reduced cyathostomin
egg reappearance period and failure of Parascaris equorum egg count
reduction following ivermectin treatment as well as survey on pyrantel efficacy on German horse farms. Vet. Parasitol. 144, 74–80.
Wolstenholme, A., Fairweather, I., Prichard, R., von SamsonHimmelstjerna, G., Sangster, N., 2004. Drug resistance in veterinary
helminths. Trends Parasitol. 20, 469–476.