Reducing antimicrobial usage in small-scale chicken farms in Vietnam: A three-year intervention study

Indiscriminate antimicrobial use (AMU) in animal production is a driver of antimicrobial resistance globally, with a need to define sustainable AMU-reducing interventions in small-scale farms typical of low- and middle-income countries. We conducted a before-and-after intervention study on a random sample of small-scale chicken farms in the Vietnamese Mekong Delta from 2016 to 2019. A baseline was established before providing farms (n=102) with veterinary advice on chicken health and husbandry, and antimicrobial replacement products. Thirty-five (34.2%) farms entered the intervention phase; the remainder no longer continued raising chickens. The intervention reduced AMU (−66%) (hazard ratio [HR]=0.34; p=0.002) (baseline 343.4 Animal Daily Doses per 1,000 chicken-days) and mortality (−40%) (HR=0.60; p=0.005) (weekly baseline 1.60 per 100). Chicken bodyweight increased by 100g (p=0.002) in intervention flocks. Our findings demonstrate that in the Vietnamese context, AMU can be substantially reduced in small-scale chicken farms without compromising flock health by providing veterinary advice.


Introduction 32
In many low-and-middle income countries (LMICs) small-scale poultry farming plays a crucial 33 role in supporting the livelihoods of rural communities (Wong et al., 2017). Compared with other 34 species, poultry production has relatively low investment and production costs (Hilmi et al., 35 2011). Globally, poultry (mainly chicken) is the second most consumed type of meat (117 36 million tonnes in 2017), and by 2026 it is expected to surpass pork (OECD-FAO, 2017). 37 Antimicrobial use (AMU) in animal production has been recognized as a driver of antimicrobial 38 resistance (AMR) globally (Marshall & Levy, 2011;O'neill, 2015). In terms of frequency, 39 chickens are the target of the highest AMU levels of all animal food species (Cuong et al., 2018). 40 In addition, many antimicrobial active ingredients (AAI) regarded as critically important for 41 human medicine by the World Health Organization (WHO, 2019) are often used in chicken 42 production (Cuong et al., 2019). 43 In Vietnam, it has been estimated that three quarters (72%) of all AMU (3,842 tonnes in 2015) 44 are aimed at animal production (Carrique-Mas et al., 2020). Studies in the Mekong Delta region 45 of Vietnam have described very high amounts of antimicrobial to small-scale chicken flocks 46 (Carrique-Mas et al., 2015;Trung et al., 2015;Cuong et al., 2019;Nhung et al., 2016). The high 47 (SE±0.2) to 1.64 (SE±0.2) (+2.4%), although the difference was not significant (one-sided 131 Wilcoxon test, p=0.999). 132 The unadjusted overall mortality increased slightly during the intervention. However, the 133 number of farms that experienced a reduction in mortality exceeded (19/31) than those that 134 increasing it (12/31). The changes in (flock average) values of ADDkg per 1,000 kg chicken-135 days, mortality and bodyweight between the baseline and intervention phases are displayed in 136 Figure 4. Among intervention flocks, there were 3/77 (3.9%) with an average weekly mortality 137 greater than 12% (12.8%, 24.8% and to 26.0%) and a cumulative mortality of >98%; two of 138 these flocks were detected with Highly Pathogenic Avian Influenza (HPAI) and one with 139 Avibacterium paragallinarum, compared with 2/87 flocks experiencing >10% weekly mortality 140 among baseline flocks (one 12.4% and one 12.8%) and cumulative mortality of 100% in these 141 two baseline flocks. 142 In the four farms that were allocated to the control arm, a total of 13 flocks were investigated 143 during the baseline phase, and 12 during the intervention phase. AMU in these decreased from 144 216.8 (SE±71.8) to 182.5 (SE±56.3) (Wilcoxon Test, p=0.857); weekly mortality changed from 145 1.17 to 1.29 per 100 birds (Wilcoxon test, p=0.493) and bodyweight changed from 1,680 g to 146 1,600 g (Wilcoxon test, p=0.511). 147

Modelling 153
In the univariable models for the Intervention Arm, 'Status=intervention' was associated with 154 an overall decreased AMU (HR=0.33; 95% CI=0.17-0.65; p=0.001) (-67%), decreased mortality 155 (HR=0.57; 95% CI=0.40-0.82; p=0.002) (-43%) and increased bodyweight (+100g; 95% CI 37-156 164 p=0.002). The size of the flock was negatively associated with AMU (HR=0.55, 95% CI cycle (HR=0.46; p=0.066), although the difference between both was not statistically significant 164 (p=0.298). Similarly, chicken bodyweight further increased during subsequent intervention 165 cycles (+119g per chicken sold, p=0.003) compared with the first intervention one (+77g, 166 p=0.068) (p=0.378). Levels of mortality did not change between first and subsequent cycles 167 (p=0.967). There were no significant interactions between either 'flock size' and 'district' and 168 'Status=intervention'. There was no statistical difference between in AMU and mortality 169 between flocks using Product A, Product B or those given no additional product. However, 170 flocks that were administered with either Product A and B had increased bodyweight compared 171 with flocks not given any supplementary product (data not shown). 172 In the control arm, there were no significantly associations between 'Status=intervention 173 calendar time' and any of the three outcome variables in either univariable or multivariable 174 models (all p>0.419). After adjustment of flock size and study district in multivariable models, 175 estimates of AMU and bodyweight increased (+42% and +14g, respectively) and mortality was 176 reduced (-19%). 177

Discussion 178
Through a locally delivered veterinary intervention, we achieved a 66% reduction in 179 antimicrobials (quantified as daily doses) administered to small-scale commercial chicken flocks, 180 alongside a reduction in mortality (-40%). In our crude (unadjusted) analyses AMU reductions 181 were, however, modest (-35%), since our analysis implicitly adjust for week of use and most AMU 182 took place during the early weeks (i.e. the brooding period). Similarly, the crude data indicated a 183 slightly higher mortality during the intervention (+2.4%). However, the adjusted analysis indicated 184 a ~40% reduction in mortality, and mortality was reduced in a majority (19/31) of farms. This 185 discrepancy was explained unusually high mortality in three intervention flocks. 186 Unlike other studies involving the delivery of a uniform treatment (i.e. vaccination) (Bessell et 187 al., 2017), our intervention consisted of providing farmers with veterinary advice. The nature of 188 this advice was variable across farms, and was based on specific observations and information 189 collected by project veterinarians from their flocks. This advice included measures to improve 190 flock health and productivity, whilst emphasizing the message that 'antimicrobials should not 191 be admnistered to healthy chickens '. 192 In addition to providing antimicrobial replacement products, the main advice given to farmers 193 focused on biosecurity, cleaning and disinfection, vaccination, litter management and The advice provided was based on a persuasive, rather than a restrictive advice. We believe this 197 approach is likely to be more sustainable in the mid-to-long term (Davey et al., 2013). A similar 198 holistic approach was adopted on a study on pig farms in Belgium, resulting in 52% AMU 199 reduction in pigs raised from birth to slaughter, and by 32% among breeding animals; furthermore, 200 the study resulted in additional productivity gains (Postma et al., 2017). Similarly, a study 201 conducted in four European Union (EU) countries reported AMU reductions of 3% and 54% in 202 fattening in weaned pigs, respectively following improvements of herd management practices 203 (Raasch et al., 2020). However, reductions in AMU were not seen in breeding pigs, and the authors 204 Small-scale commercial chicken production using native breeds is widespread in the Mekong 220 Delta of Vietnam, and often represents an upgrade from backyard production. The popularity of 221 this system resides in the preference of the Vietnamese consumer for meat of long-cycle native 222 birds. Native chicken meat reaches a considerably higher price compared with broiler meat (PW, 223 2018). However native chickens (and their crosses) are slow growing (>4 months), and preventing 224 disease over such a prolonged period requires sustained efforts (Carrique-Mas et al., 2019). 225 In our study, the identification and enrollment of study farms was challenging due to the fluidity 226 of this type of production system, with many households setting up chicken farms as well as 227 stopping raising chickens altogether. Because of this, a large number of farms did not remain in 228 business over the extended duration of the study. Indeed, flock mortality was an important predictor for farmers giving up raising chickens (data not shown) and a large fraction of our study 230 farms (61.8%) had gone out of business even before the start of the planned intervention phase. 231 In addition to their previous experience with disease, farmers may start or stop raising chickens 232 depending on circumstances, such as market price of day-olds, commercial feed and poultry 233 meat, income from the sale of the previous flocks or other rural activities. Furthermore, many 234 farmers raised one cycle per year, but not necessarily every year. This was reflected in the lack 235 of experience in chicken husbandry of many farmers (and farm workers). with the intervention phase in this study. This may have exerted additional pressures over our 242 study farmers. During this time, many farms in ASF-affected provinces switched to chicken 243 production, resulting in increased market availability of low-cost chicken meat, therefore 244 reducing the value of chicken production in our area. 245 The changes to the initial study design are a testament to the challenges of conducting 246 intervention studies in small-scale farming systems. Initially, we planned to allocate one third of 247 all recruited farms to the control arm in order to measure any environmental influences on AMU, 248 for example, due to public engagement initiatives (television campaigns, work in schools, etc.) 249 that took place in the province under the umbrella of this project. Exposure to these may have 250 inadvertently had an influence on the farmers' decision on AMU beyond the intervention. Given 251 the high number of farms that stopped chicken production, we opted for reducing the size of the 252 control arm to a minimum of four, thus reducing the statistical power of any analysis in that 253 group. However, the descriptive data from this small control group suggests no change between 254 baseline and intervention, and gives additional validity of the observed findings. 255 The study demonstrates that reducing current high levels of AMU through the provision of 256 veterinary advice is achievable in the Vietnamese small-scale commercial farming context. 257 There was an indication that farmers responded to the advice given, and supplementation with 258 health-enhancing products may be beneficial. Many farmers, especially the larger ones may even 259 be willing to pay for such a service, since labour costs in Vietnam are relatively low 260 (approximately 25 USD for a two-hour visit). We propose to develop a business case for an 261 advisory service targeting the main livestock-producing regions in the country (Mekong River Delta, Southeast, Central region, Red River Delta), with the value proposition that healthy 263 livestock means profitable businesses. 264

Study design 266
The intervention was designed as a randomized 'before-and-after' controlled study on farms 267 raising chickens for meat in two districts (Cao Lanh and Thap Muoi) within Dong Thap province 268 (Mekong Delta, Vietnam) (Figure 1 and Supplementary file 1). The study was designed in two 269 stages, a 'baseline' followed by an 'intervention' phase. Two intervention arms (Arm 1 and Arm 270   2)  In these meetings, the project aims and methods were outlined. Farmers willing to enroll in the 278 study were asked to contact project staff as soon as they restocked with day-old chicks. Farmers 279 that restocked with chicken flocks (defined as a group of birds raised together in the same 280 building) meeting the criterion '>100 meat chickens raised as single age' were enrolled. 281

Description of the baseline and the intervention 282
During the baseline phase of the study, routine AMU and productivity data were collected from 283 enrolled farms without the provision of any advice. Using a random number generator, we 284 allocated enrolled farms to either an intervention or a control arm. All farms allocated to the 285 intervention arm were supported with a Farmer Training Programme (FTP), where farm owners 286 were invited to participate in six workshops where a poultry veterinarian instructed them on the 287 principles of chicken husbandry, prevention, control of infectious diseases and waste 288 management and a Farm Health Plan (FHP), where each farm was assigned to a Project 289 Veterinarian (PV) who was responsible for providing specific advice to farmers. The PV visited 290 each farm on three different occasions for each flock cycle: (i) early-brooding (weeks 1-2), (ii) 291 late brooding (weeks 3-4), and (iii) grow-out (>2 months) periods. Prior to each visit, the PV 292 reviewed records of productivity and disease over previous cycles, inspected the flock and 293 house/pen, reviewed farmers' records, discussed with farm owner about current the PV proposed the farm owner the use of an antimicrobial replacement product, either an 296 essential oil-based (Product A) or a yeast fraction-based product (Product B) used 3 days/week 297 over the first 10 weeks of the production cycle. The allocation of either of these products was 298 based an assessment of the history of disease in the flock (Product A) if the farm had a history 299 of diarrhoea in previous flocks; or Product B, all other flocks. In all visits, the PVs reminded the 300 farmers that healthy birds should not be given any antimicrobials. 301

Data collection 302
Each farmer was provided by project staff with a diary to weekly record data on farming 303 practices, including number of chickens purchased, number of chickens in and out of the flock 304 (number of dead and sold chickens), as well as the types and quantities of antimicrobial products 305 used. The average bodyweight of slaughter-age chickens was also measured by average of total 306 bodyweight of chickens divided for total number of chickens sold. Project staff visited study 307 farms four times (different from PV visits) to verify the data collected, which was subsequently 308 transferred to validated questionnaires and double-entered into a web-based database. 309

Statistical analyses 310
The initially proposed sample size was based on previous quantitative data on AMU in Mekong 311 Delta chicken farms (Carrique-Mas et al., 2015). We aimed at recruiting 120 farms and estimated 312 a total of 40 farms for each arm. A sample size of 40 farms per arm, each contributing with 2 313 cycles investigated during baseline and 2 during the intervention, and a two-sided significance 314 level of 5%, will have 82% power to detect a ~33% reduction, and a 91% power to detect a 50% 315 reduction. Since the study design exploits within-farm correlation of unknown magnitude, the 316 true power was expected to be higher. 317 The primary outcome was the weekly number doses of antimicrobial active ingredient (AAI) 318 corresponding to 1 kg of live chicken administered to a flock (Weekly ADDkg). Secondary 319 outcomes were 'Weekly mortality', calculated by dividing the number of chickens dying over 320 the week by the total chicken present at the beginning of each week (%), and 'Weight of the 321 birds (in units of 100g) at the time of sale'. The latter was calculated by dividing the total flock 322 weight by the number of chickens sold at the end of the cycle. The correlation between all three 323 outcomes at flock and at week level was investigated using the Spearman's rank correlation 324 coefficient. 325 Weekly antimicrobial consumption (ADDkg) was calculated by multiplying the amounts of 326 antimicrobial product administered by the farmer to the flock (g) through water or feed, indicated the amount of product to be diluted in water (Volume of water /Weight of product) 329 (l/g) or mixed with feed (Weight of feed /Weight of product) (kg/g) based on the product label. 330 The obtained amounts were then divided by the estimated daily consumption of water (0.225 l) 331 or feed (0.063 kg) by a 1 kg-chicken (Cuong et al., 2019). 332 No. ADDkg = Amount of antimicrobial product administered (g) * dilution factor in water (l/g) or feed (kg/g) Daily consumption of water (0.225l) or feed (0.063kg) of 1 kg chicken

No. ADDkg
Estimated weight (kg) of the flock *1,000 334 We built Poisson regression models with for 'Weekly ADDkg' and 'Weekly mortality'. For the 335 former the offset was the (weekly) total number of chicken-kg days (log); for the latter it was 336 the number of chickens at the beginning of the week (log). In addition, a linear regression model 337 was developed with bodyweight of chickens at the point of sale (kg) as outcome. In all cases, 338 'Farm', 'Flock cycle' and "Week" were modelled as random effects, where 'Week' was nested 339 within 'Flock cycle', and the latter was nested within 'Farm'. The main variable of interest was 340 the impact of the intervention delivered; therefore, we investigated 'Status' (baseline, transition, 341 and intervention) as an explanatory variable in Intervention Arm and 'Status' (baseline, 342 intervention calendar time) as an explanatory variable in the Control Arm. 'Status=transition' 343 was assigned to those flocks that were not exposed to all three advisory visits for Intervention 344 Arm farms. This occurred to a number of flocks at the beginning of the intervention phase, given 345 that some advisory visits (typically the first and second) were missed. In order to account for the 346 potential confounding effects of 'District' and 'Flock size' these were forced into a multivariable 347 model; we tested the interactions between 'Status=intervention' with 'District' and 'Flock size' 348 to investigate whether the observed effects were dependent on the geographical location or the 349 size of the flock. Moreover, we investigated whereas subsequent cycles over the intervention 350