Statistical approach to measure the efficacy of anthelmintic treatment on horse farms A. N. VIDYASHANKAR 1 *, R. M. KAPLAN 2 and S. CHAN 1 1 Department of Statistical Science, Cornell University Ithaca, NY 14853-4201, USA 2 Department of Infectious Diseases College of Veterinary Medicine, University of Georgia Athens, GA 30602, USA (Received 30 April 2007; revised 29 June 2007; accepted 29 June 2007; first published online 23 August 2007) SUMMARY Resistance to anthelmintics in gastrointestinal nematodes of livestock is a serious problem and appropriate methods are required to identify and quantify resistance. However, quantification and assessment of resistance depend on an accurate measure of treatment efficacy, and current methodologies fail to properly address the issue. The fecal egg count reduction test (FECRT) is the practical gold standard for measuring anthelmintic efficacy on farms, but these types of data are fraught with high variability that greatly impacts the accuracy of inference on efficacy. This paper develops a statistical model to measure, assess, and evaluate the efficacy of the anthelmintic treatment on horse farms as determined by FECRT. Novel robust bootstrap methods are developed to analyse the data and are compared to other suggested methods in the literature in terms of Type I error and power. The results demonstrate that the bootstrap methods have an optimal Type I error rate and high power to detect differences between the presumed and true efficacy without the need to know the true distribution of pre-treatment egg counts. Finally, data from multiple farms are studied and statistical models developed that take into account between-farm variability. Our analysis establishes that if inter-farm variability is not taken into account, misleading conclusions about resistance can be made. Key words: efficacy, anthelmintic resistance, horse, beta-binomial model, logit-normal model, bootstrap methods, error rates, power. INTRODUCTION In recent years, anthelmintic resistance in gastroin- testinal nematode parasites of livestock has emerged as an important problem worldwide. Multiple-drug- resistant parasites threaten small ruminant industries in many areas of the world, and resistance in parasites of horses and cattle is reaching alarming levels (Kaplan, 2004). The theoretical gold standard for diagnosing resistance to anthelmintics is achieved by counting the total number of killed worms and live worms following treatment ; however, these data can be obtained only by sacrificing the animals, which is unrealistic on a farm. The practical gold standard is to measure changes in the number of eggs being produced by the parasites ; these data can be obtained by measuring the number of eggs in a sample of feces. This procedure is called the fecal egg count reduction test (FECRT) and is the most common means of determining whether resistance is present on a farm (Kaplan, 2002). However, the fecal egg count (FEC) data are a surrogate measurement, which are subject to many sources of variability. Furthermore, the correlation between this surrogate measurement and the number of worms that are actually present in a horse is known to be weak (Lyons et al. 1983; Klei, 1986). In a FECRT, fecal egg counts are compared in the same animals both before and after treatment, or between control and treated groups at some es- tablished time-point after treatment. However, there are many sources of animal-related and farm- related variability in FEC data that can impact the interpretation of results, especially when many different farms are being studied. Some of the most important sources of variability are : non-Gaussian overdispersed distribution of parasites in host ani- mals, causing large differences in pre-treatment values between animals on the same farm ; differences in parasite infection intensities between farms, caus- ing large differences in pre-treatment values between farms ; inherent variability in parasite egg numbers within the fecal output of an animal, which results in the collection of non-uniform samples (Warnick, 1992) ; variability in fecal egg counts resulting from the non-uniform distribution of eggs in solutions used for fecal egg count analysis; overall health and body condition of animals that can impact drug pharmacokinetics and pharmacodynamics; differ- ences in age, breed, and sex of animals both on and between farms ; differences in nutritional programs * Corresponding author : Department of Statistical Science, Cornell University Ithaca, NY 14853-4201, USA. Tel: +607 255 3759. Fax: +607 255 9801. E-mail: anv4@cornell.edu 2027 Parasitology (2007), 134, 2027–2039. f 2007 Cambridge University Press doi:10.1017/S003118200700340X Printed in the United Kingdom http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/S003118200700340X Downloaded from http:/www.cambridge.org/core. George Mason University, Fairfax, on 25 Oct 2016 at 01:55:03, subject to the Cambridge Core terms of use, available at