Welfare Assessment: Correspondence Analysis of Welfare Score and Hematological and Biochemical Profiles of Dairy Cows in Sardinia, Italy

Simple Summary Defining animal welfare definition is still controversial, and a definition is difficult to establish. Furthermore, welfare detection is often complicated and subject to different interpretations. This work aimed to provide valid indicators to evaluate the welfare of dairy cows. The Animal Welfare and Biosecurity Evaluation form (AWB-EF) checklist developed and validated by the Italian National Centre of Reference for Animal Welfare (CReNBA) was submitted to 16 Sardinian dairy cattle farms. Blood samples from the 230 Holstein breed dairy cattle housed in these farms were analyzed for hematological parameters. Correlation analysis revealed a strong association between AWB-EF (considered as gold standard) and laboratory parameters, indicating correspondence between the health and welfare status of the animals. Our study clearly indicates that the use of a validated checklist in combination with the identification of well-known laboratory parameters can be a fundamental tool for veterinarians to detect stress conditions early. Abstract The need for animal welfare definition and assessment is increasing worldwide, and several studies have been conducted to help fill the knowledge gaps regarding the welfare of cattle. However, further studies are needed to provide valid synthetized measures for welfare evaluation. The aim of this study was to assess the welfare status of 16 Sardinian dairy cattle farms, based on the developed Animal Welfare and Biosecurity Evaluation checklist (AWB-EF) and the corresponding hematological, biochemical, and electrophoretic profiles of these animals. Considering the AWB-EF as gold standard, blood samples were collected from 230 Holstein breed dairy cattle, aged between 3 and 8 years, out of the periparturient period, and with no clinical signs of specific pathologies. Principal Component (PC) and correlation analyses were performed to simplify phenomena interpretation and assess positive/negative associations. Four PCs were able to explain 76% of the total variability, and six laboratory parameters were strongly associated with the AWB-EF score (Spearman’s correlation coefficient ≥ 0.40, p-Value < 0.05), reflecting the real health status of the animals. Given the complexity of animal welfare as a multidimensional concept and the need to include both animal-based and non-based measures in welfare evaluation, the present work represents a sound basis for future evaluation and veterinary health planning.


Introduction
The original concept of animal welfare, developed over 50 years ago by Brambell (1965) [1], was updated by Lawrence and Stott [2], who defined animal welfare as an ethical concern for the mental and physical health of animals over which we have a degree of control, in 2010. However, defining and measuring animal welfare remains controversial [3]. As underlined by Devitt et al. in 2018, if, on the one hand, there is a need to take the relationship between farmers and animals into account when considering farm animal welfare standards, then on the other hand, there is limited understanding of how the nature of this relationship influences welfare outcomes, and thus welfare assessment would require a multidisciplinary approach [4].
The need for animal welfare has been underlined by the World Organization for Animal Health (OIE), who recommended that "veterinarians should be the leading advocates for the welfare of all animals, recognizing the key contribution that animals make to human society through food production, companionship, biomedical research, and education" [5]. Increased attention has been given to farm animal welfare in developed countries, especially given the expansion of intensive animal production systems that improve profit and efficiency but challenge the conscience of many consumers [6].
Furthermore, the need for measurable outputs relevant to animal health and welfare, and that are able to determine whether the welfare program is effective, efficient, and transparent, is increasing worldwide [7].
Poor environmental conditions can affect several homeostatic functions and reduce the productive and reproductive performances of livestock. Stress factors and poor welfare can also compromise the host immune system and lead to increased susceptibility to diseases among animals [8]. Farm animal welfare should be viewed as a global condition, where the effects of infectious and non-infectious stressors cannot be easily discriminated and can overlap, challenging the host's immune system [9]. In fact, the innate immune system can rapidly respond to both infectious and non-infectious stressors, such as metabolic stress conditions, psychological stress, high/low temperatures, oxidative stress, and hypoxia [8].
To fill the knowledge gaps regarding cattle welfare, several studies have attempted to develop scientifically valid methods for assessing welfare [10][11][12][13][14][15][16][17][18][19][20][21][22]. Some of these studies primarily focused on improving external factors, management-based measures, or non-animal-based measures (N-ABMs), which affect welfare without considering the reactions and consequences for the animals [17,18]. Indeed, most of the methods used have not been refined or validated, possibly because they aimed to detect illness rather than welfare [19,20]. Thus, resource-based assessment cannot answer questions about animal welfare. For all these reasons, attempts have been made worldwide to develop animalbased measures (ABMs) to estimate the actual welfare of animals [21,22]. Conversely, the European Food Safety Association (EFSA) has provided evidence that in some cases N-ABMs may be more efficient than ABMs; thus, it follows that both ABMs and N-ABMs are necessary to obtain a holistic approach and achieve an effective overall classification of animal welfare at the farm level [23][24][25].
Currently, about two million dairy cows are reared in the Italian national territory, with more than half of these in the north of the country [26]. Despite this large number of animals, there is no official protocol for evaluating the welfare of dairy cows that uses ABMs, N-ABMs, and risk assessment. In 2017, the Italian National Centre of Reference for Animal Welfare (CReNBA) developed a simple and easy-to-use on-farm protocol for assessing the welfare of dairy cows in loose housing systems. They then carried out expert opinion elicitation to characterize a list of management and housing factors potentially associated with negative or positive welfare outcomes in dairy cows [25].
Welfare assessment requires a multidisciplinary approach, and the CReNBA Animal Welfare and Biosecurity Evaluation form (AWB-EF) monitor both ABMs and N-ABMs in dairy cattle farms. Health is a key component of welfare and health status (e.g., presence/absence of disease, organ function, metabolic processes, and internal body condition) is primarily monitored using hematological and biochemical tests [27][28][29].
In this study, we aimed to assess whether the level of welfare estimated by the AWB-EF positively correlated with laboratory parameters (metabolic profile, electrophoresis, and blood count), which are regarded as indicators of health status. To achieve this goal, an in-depth investigation on livestock welfare and biosecurity status was conducted in a broad sample of Sardinian dairy cattle herds using the CReNBA AWB-EF as gold standard, and its correlation with laboratory parameters was assessed. Considering that there is no specific official indication about how laboratory parameters relate specifically to animal welfare (except for hemoglobin levels in calves, Italian Health Ministry D.L. 331, 1st September 1998) [30], this study could represent an important starting point for identifying simple and easily detectable parameters that can help monitor animal welfare.

Study Context
Sardinia is an Italian island in the center of the Mediterranean Sea (40 • 03 N 9 • 05 E) with a total surface area of 24,100 km 2 and a population density of 69 inh/km 2 [31]. Administratively, the Sardinian region comprises five provinces, as established by the Regional Law of 4 February 2016: Nuoro, Sassari, Oristano, South of Sardinia, and Cagliari (metropolitan city), with 377 municipalities in total [32]. Given the low population density and the unpolluted environment, the Sardinian economy is mostly based on agropastoral activities [33]. Across the whole island, around 9,200 dairy cow farms are regularly recorded in the Italian Veterinary National database, with a dairy cow population of about 260,000 animals. Sardinian cattle breeding is mainly intended for meat production (85%; 7700 farms). The main area dedicated to beef production is located in the north east of the region, (Sassari and Nuoro provinces), while most of the bovine livestock for milk production is located in the center of the island (Oristano province), where the territory mainly comprises large lowland areas [26].

The Animal Welfare and Biosecurity Evaluation Form
For the purpose of this work, the AWB-EF checklist was implemented by two trained veterinarians, during the farms' official controls, in 16 dairy cow farms in Sardinia (Italy) between 2013-2015. The checklist was provided by the CReNBA, aimed to monitor the welfare conditions of dairy cows, and consisted of 90 items divided into 5 sections (A1, A2, A3, B, C) (Table S1). In detail, section (A) is divided into (A1) farm management (22 items, i.e., number of stockpeople and their training, animal grouping, inspection, type of handling, feeding strategy, water provision, cleanliness, bedding material and calving pen management), (A2) structures and equipment (29 items, i.e., features about space availability, calving pen, bedding material, floor, feeding space, water points, facilities for sick animals, milking machine, temperature, humidity, ventilation, gas concentration, and artificial lighting), (A3) animal-based measures (18 items, i.e., details about lameness, mortality rate, body condition score, treatments for mastitis, integument alterations, and mutilations; cleanliness of flank, leg, and udder).The other sections are (B) biosecurity (15 items, i.e., general measures against rodents, insect, precautions for entry of strangers, disinfection, quarantine, carcasses management, animal movement management, and disease prevention) and (C) risks and alarm system (5 items, i.e., noise, fire and ventilation alarm, and electricity generator). Each item contributed equally to generating a percentage score for each section, constituting five different outcomes. Furthermore, general information on the number of animal breeds, as well as the animals' average age, breed, the number of lactating cows and milk production (kg of milk by animal per day), was collected by the AWB-EF checklist for each farm. Animals in the periparturient period were excluded to avoid confounding bias in laboratory parameters detection [34]. The AWB-EF was provided in paper format, and all data were subsequently collected and stored in an ad hoc password-protected electronic database using a closed response data collection instrument (Microsoft Excel, Microsoft Corporation, Redmond, WA, USA).

Laboratory Analysis
Laboratory parameters that were able to verify whether the farm conformed to Italian (D.Lgs.n.146/2001; D.Lgs.n.126/2011) and European legislation (Directive 98/58/EC; Directive 2008/119/EC) were chosen based on EFSA reports [23,24,[35][36][37][38][39][40][41], to evaluate whether the AWB score corresponded with the cows' welfare conditions on the basis of the five outcomes. During the veterinarian's official visits, blood sampling was performed on Holstein cows between 3 and 8 years old at different stages of lactation. Individual blood samples were collected from the animals' coccygeal vein before feeding. They were transferred into vacuum tubes containing EDTA anticoagulant for hematological profiling and then into serum gel separator tubes without anticoagulant for both biochemical profiling and analysis of the electrophoretic pattern of serum proteins. The samples were placed in a container with ice and forwarded to the laboratory within 2 h. The serum was separated by laboratory centrifugation at 3500× g for 10 min at 4 • C, placed in 1.5-mL tubes, and stored at −20 • C until analysis, while the EDTA tubes were analyzed within 3 h. Hematological determinations were made using an ADVIA 2120 automatic hematology analyzer (Bayer Healthcare, Siemens, Monaco, Germany) with software that allows blood determination in cattle. Biochemical parameters were analyzed using a Dimension RXL chemistry analyzer (Siemens), and serum protein electrophoresis was carried out using an INTERLAB G26 Automated Agarose Gel Electrophoresis Analyzer (Interlab, Rome, Italy). The set of parameters used for biochemical profiling are provided in Table S2.

Sample Size and Inclusion/Exclusion Criteria
Dairy cow livestock breeding (not mixed species farming system) business's start and end dates were available, and business activities for the entire study period (i.e., business start date before on or before 1 January 2013 and business end date after on or after 31 December 2015) and the available animal census data were essential farm characteristics to be included in this study. If the farmer had more than one farm with a unique fiscal code, or the farms were inactive for the entire study period, or the farm had missing business start data and/or end data or animal census data for one or more of the study years, the farm was excluded.
The minimum number of animals to be observed is defined by the Welfare and Biosecurity Manual for dairy cows, performed by CReNBA and specific for the AWB-EF checklist (Table S2). Otherwise, given that programmed statistical analysis (factorial analysis) is strongly influenced by the ratio of sample size (N) to the number of variables being analyzed (p) [42], a specific sample size calculation was carried out before proceedings, following MacCallum et al., 1999 [43]. Given the large number of variables collected (31) and the number of hypothesized factors (5), a total sample size of 200 animals (plus 15% of these animals to account for drop out) was considered appropriate. A total of 16 randomly selected dairy cow farms were included in the study. Animals raised on the 16 farms were included in the study, except for animals in the periparturient period and those outside of the age range (3-8 years). 230 samples from 230 randomly selected cows were collected and analyzed.
For each farm, the following baseline characteristics were collected: location (latitude and longitude), province, municipality, opening date, number of animals on farm, age, race, number of lactating cows, milk production, details of previous disease (i.e., skin lesions, lameness, mastitis), mortality rate (overall mortality year rate), and number of animals with laboratory parameters within the normal range.

Statistical Analysis
Data quality and completeness were tested based on an extensive data check. Descriptive analyses based on mean (SD), median (I-III quartile) and frequency (percentages) were performed to evaluate the distribution of the parameters at baseline compared with the laboratory reference values. A correlation matrix was performed to evaluate association between variables. The linear or nonlinear nature of the relationship between the dependent variable (i.e., percentage score for each section) and each of the continuous independent variables was assessed graphically. When a linear relationship was assumed, its strength was initially evaluated through bivariate analyses by means of the Spearman non-parametric correlation coefficient. When relationships were assumed to be other than linear, mathematical transformations (i.e., log-normal) were applied. The homogeneity of the samples between farms was graphically evaluated to exclude bias generated by the within group correlation or sub-group populations.
The main hypothesis to be tested by the statistical analyses was the correlation between the percentage score for each section of the AWB-EF checklist with the laboratory results (electrophoresis, metabolic profile, and blood count; Table S3). A principal component analysis (PCA) with a varimax rotation was used to make the factors orthogonal and more interpretable, and to extract the laboratory patterns and confirm the number of dimensions underlying the set of variables [44]. The number of retained components was determined according to eigenvalues (≥2.0), scree plot examination and interpretability [45].
Given the non-normal distribution of all the variables included, to evaluate the strength of the relationship (association) between the AWB-EF scores of each checklist section and the laboratory parameters, the nonparametric Spearman's correlation coefficient was estimated. A correlation matrix and correlation graphs were produced. Any outlying observation that appeared to deviate markedly from other observations in the sample was checked. Only 8 values significantly deviated from the mean, and 3 of these were related to reporting errors (and consequently corrected). The other five were included, considering that, in such cases, outliers may be due to random variation or may indicate something scientifically interesting [46]. A high statistically significant correlation (coeff. ≥ 0.7, p-Value < 0.05) means that two variables have a strong relationship, while a weak correlation means that the variables are hardly related. All statistical tests were two-sided, and a p-Value less than 0.05 was considered statistically significant. Statistical analyses were performed using Stata13 (StataCorp, Stata statistical software, Release 13; StataCorp LP, College Station, TX, USA) and R-software (Version 3.6.2; R-Foundation for Statistical Computing, Vienna, Austria).

Descriptive and Principal Component Analysis Results
The baseline characteristics of each farm and thirty-four laboratory parameters were collected from 230 dairy cows raised on 16 extensive farms. The farm baseline characteristics are reported in Table 1, while the laboratory parameters and comparisons with the reference ranges are reported in Table 2. Farms included in this study were well-distributed around the Sardinian island. Totally, 5110 Holstein cows were bred on these 16 farms, with an average of 216 animals (SD = 131) and a median age of 4 years (I-III quartile = 3-9), as established by the inclusion criteria. On each farm, an average of 109 (SD = 63) animals were in the lactating period, with an associated average milk production of 31 (SD = 2.44) kilograms of milk per cow per day. About 15% of the animals on each farm reported skin lesions, 6% lameness, and 30% has undergone treatment for mastitis. The overall mortality rate was about the 2% of the overall population. In total, 217 out of 230 cows (94%) showed laboratory parameters within the normal range. Overall mortality year rate 2 (1)(2)(3)(4)(5) No. of animals with normal range parameters 217 (94%) Estimates are reported as mean (SD); median (I-III quartile); frequency (%); percentage (minimum-max) collected by farm. The five AWB-EF sections were evaluated and summarized in Figure 1. The average value of section A1 (farm management) was 76.9% (SD = 8.8), while section A2 (structures and equipment) showed an average percentage of 64.5% (SD = 12.3) and section A3 (animal-based measures) had a lower average value of 62.9% (SD = 10.1). Finally, sections B (biosecurity) and C (risks and alarm system) of the AWB-EF checklist showed lower averages, with values of 52.3% (SD = 11.3) and 54% (SD = 11.3), respectively. Based on PCA analysis, according to the eigenvalues and the scree plot (Figure 2), four components are able to explain why 76% of the total variance was maintained. Considering the main parameters that characterized each PCA factor, the four dimensions highlighted by the PCA were named as "Overall Welfare", "Electrophoresis", "Blood count" and "Metabolic profile", respectively, rather than dimension 1, 2, 3 and 4 as conventionally used [44].

Results of Correlation Analysis
Five correlation matrices were computed to evaluate the association between A1, A2, A3, B, and C scores with the laboratory parameters of Pearson's correlation coefficient and correspondent p-Values (Table 3).
The correlation matrix was interpreted by considering the positive relationships between variables as 'very strong' when the Spearman's correlation coefficient was higher than 0.70, 'strong' when it was between 0.40-0.69, 'moderate' when it was between 0.30-0.39, 'weak' when it was between 0.20-0.29, 'negligible' when it was < 0.20, and likewise for the negative relationships between negative correlation values [47]. In particular, 16 parameters were statistically correlated with the score A1-management factor, and the Spearman's correlation coefficient of 0.54 and −0.48 indicates a strong positive relationship with amplitude of hemoglobin distribution (HDW) and strong negative relationship with GPT (Table 3, Score A1), respectively. All 12 variables statistically correlated with the score A2-housing factors and showed a weak or negligible positive or negative coefficient, except for GPT and LYMPH, which showed a negative and positive moderate correlation, respectively (Table 3, Score A2), as well as neutrophil-lymphocyte ratio (NRL). A strong negative correlation was described by a 0.52 coefficient between the score A3-animal-based measures and A1GB, while the A2GB was moderately positively correlated with the score (coefficient = 0.37) ( Table 3, Score A3), and the other five variables showed moderate (BGB, MCHC, BAS) or weak correlation (GGB, MPV). Between the 10 variables statistically correlated with score B-biosecurity, a very strong negative relationship was identified with A1GB (Table 3, Score B), a very strong positive relationship with A2GB (Table 3, Score B), and a strong negative relationship with GPT, with Spearman's correlation coefficients of −0.71, 0.72 and −0.51, respectively. The score C-risk and alarm system were statistically significantly correlated with seven variables, but only blood urea nitrogen (BUN) and HDW showed strong positive relationships (coefficients = 0.66 and 0.40, respectively) ( Table 3, Score C). Table 3. Correlation matrix based on (a) A1-management factor, (b) A2-housing factors, (c) A3-animal-based measures, (d) B-biosecurity, (e) C-risk and alarm system scores as outcomes, and fact1-4 as explicative variables. Data are presented as Spearman's correlation coefficient for statistically significant variables correlated with score values (*** p-Value < 0.0001, ** p-Value range (0.0001-0.005) * p-Value range (0.005-0.05)).

Discussion
The present study evaluated the welfare status of 230 dairy cows on 16 dairy cattle farms in Sardinia, testing the association between the welfare score detected by the AWB-EF checklist and the individual hematological and biochemical parameters observed. Considering the welfare protocol score as gold standard, and by correlating it with laboratory results, we were able to estimate whether the data recorded by the protocol reflected the health status of the animals. The five different AWB-EF sections reflecting the laboratory patterns individuated with PCA analysis, and each score showed strong and statistically significant association with one or more of the laboratory parameters.
Considering the PCA results, welfare status in farm management could be well described by the electrophoresis, blood count and metabolic profile. Besides, the correlation matrix results described a strong association between AWB-EF checklist scores and laboratory parameters such as A1GB, A2GB, GPT, HDW, LYMPH and BUN.
Serum protein levels, including α1-globulin and α2-globulin, are correlated with important functions of organic synthesis (mainly in the liver), and their levels depict the animals' capacity to cope with growth and production demands [48]. Electrophoresis has been used to monitor the ability of young bulls to adapt to a different environment, and researchers observed statistically significant variations in levels of either albumin, α-globulin, β-globulin or γ-globulin several times after arrival at the new farming center [49]. Another study described abnormality in electrophoretic parameters in calves after transportation stress, and particularly α-globulin peak was the most predictive of diseases in Holstein Friesian cattle among the clinical immunological parameters evaluated [50]. In the same study, hematological parameters were monitored in animals after transportation stress, and researchers observed that WBC was the most predictive of disease among the hematological parameters evaluated. Nevertheless, the differences between leucocyte subsets were not monitored in that study [50]. Overall, WBC and the variation of different leukocyte populations can be monitored to evaluate the sanitary status of the herds [51]. In our study, a strong association of AWB-EF was also observed with two biochemical parameters: GPT and BUN. Evaluation of the serum activities of hepatic enzymes, including GPT, is routinely used to monitor liver health status, whereas BUN levels are inversely correlated with the decline of kidney function [51][52][53][54]. Correct functionality of these two vital organs is strongly correlated with animal welfare.
The health status of animals is one criterion of welfare assessment, and it is important to monitor management and housing factors, which are strongly related to a lower incidence of disease and mortality, mental comfort, absence of stress, good appetite, body homeostasis, and maintenance of proper animal welfare levels [55]. Assessing welfare requires detailed knowledge of the available scientific information, since the definition involves describing how well the animals experience their environment based on the best possible judgement of their situation. Such information is necessary to avoid errors in interpreting a given measure and cannot be based solely on science or on data collected from experiments or laboratory analysis [56]. Proper management conditions are essential for organisms to function normally [57], and hematological and biochemical tests may help veterinarians to understand animals' welfare status based on factors other than the presence/absence of disease [27,58,59].
Several studies used hematological and biochemical parameters to assess animal health and welfare status, both in livestock and small animals [16,22,25,48,52,[60][61][62][63][64]. Nevertheless, the use of laboratory parameters requires the collection of samples on the farm and, depending on the parameters used, can entail quite high costs in the purchase of reagents and require time for sample collection and analysis. Thus, welfare protocols are a more immediate and less expensive method for judging the living conditions of animals. Few reports have focused on welfare scores in cows, and more in-depth studies are needed to detect the causes of possible deviations from normal reference ranges. Animal-based measures alone are not enough to ensure a complete evaluation of animal welfare [41]. As a result, applying welfare protocols requires a more general farm evaluation carried out by well-trained veterinarians, which could be difficult to organize. The combination of both welfare protocol and laboratory parameters could be a valid tool to assess a robust estimation of both animal health and welfare status.
In this study, the score assigned by the welfare protocol mainly corresponded with the animals' health status, and the analysis showed that some laboratory patterns may be particularly useful indicators of welfare. Otherwise, the results of correlation analysis must be carefully considered, as their extrapolation could be dangerous. Correlation analysis only considers the linear relationship between two variables (i.e., other variables that could influence the response variable are not studied) and could be affected by outliers. Furthermore, correlation analysis does not establish if one variable is dependent and the other is independent. Therefore, correlation analysis provides information about the strength and the direction (positive or negative) of a relationship between two continuous variables, but a strong correlation does not imply a cause-and-effect relationship.
Furthermore, a limit of this study is that information (i.e., day in milk, amount of milk production, age, lactating period) is collected overall by farm and is not available for each animal. Thus, considering that the single animal was the epidemiological unit for the correlation analysis, this information was not included in the final analysis, generating a possible bias considering that these conditions could significantly influence blood and milk parameters [65].
We can conclude that the average values of the hematological and biochemical parameters fell within the range of reference values. Mainly strong or moderate associations have been highlighted between the average welfare score of the dairy farms and laboratory analysis, suggesting that the welfare protocol score mainly reflected the real health status of dairy cattle. The use of a validated checklist in combination with the identification of few well-defined parameters, able to synthetize the health and welfare status of the animals, can be a fundamental tool for veterinarians to detect stress conditions early. Although more in-depth analysis is needed to provide not only an association between measures and the checklist but to quantify and model this association, the results obtained in this study are a strong starting point for future research. Finally, it is necessary to underline that health is a key component of welfare, and welfare assessment requires a multidisciplinary approach, so a laboratory evaluation of the health status of animals cannot be considered the sole criteria for determining animal welfare.
Supplementary Materials: The following are available online at https://www.mdpi.com/2076-261 5/11/3/854/s1, Table S1. Structure of welfare protocol, Table S2. Minimum number of animals to be observed for animal-based measures (ABMs) evaluation, Table S3. List of laboratory test parameters analyzed to evaluate the correspondence with animal welfare status observed with AWB-EF, by category (metabolic profile, electrophoresis, blood count).  Institutional Review Board Statement: The present study was approved by Ethic Committee of Istituto Zooprofilattico della Sardegna "G. Pegreffi", approval number 6/2012/CE.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.