The assessments were conducted on commercial herds in Brazil. All the procedures were reviewed and approved by the Ethics Commission of the Animal Ethics Committee (CEUA) from the Federal University of Paraná (Setor Palotina) under the Protocol number 16/2013-CEUA/Palotina. The assessments followed the appropriated guidelines of the Comissão Nacional de Bem-Estar Animal (COBEA). Furthermore, the members involved in animal assessment hold degrees in Veterinary Medicine and are experienced in evaluating claw lesion on live animals.

A total of 129 swine herds were assessed once by two independent experts for claw lesion severity from 2013 to 2023. After parturition, sows (n = 12,364; ≥10% of the inventory size within each heard, up to a maximum of 100 animals) were randomly selected and scored utilizing a scale from 0 (no lesions) to 3 (severe lesions) for claw lesions in the hind legs (2) by one evaluator. Lesions included HOE, HSC, WL, CHW, CVW, T, and DC. A lesion index (L-Index) was calculated as the sum of the higher scores from each lesion per sow. Figure 1 shows the lesion scoring system used according to lesion type and severity. The following criteria were used to include data from a given herd in the database: herds employing similar management protocols in terms of sow movement, management, and feeding; herds with similar housing conditions during the lactation phase; sows were only included in the assessment if they (1) were not used as foster sows in previous parities, (2) had good health status (e.g., no prolapse, reproductive disorder, or abortion in previous parities), (3) had acceptable body condition score between 2.5 and 3.0 [1.0–5.0 scale; Young et al. (24)].

Figure 2 represents an illustrative diagram of the workflow of eligible herd selection for this study. Commercial herds located in nine Brazilian states [Distrito Federal (2), Goiás (23), Minas Gerais (3), Mato Grosso (1), Mato Grosso do Sul (12), Paraná (77), Rio Grande do Sul (10), Santa Catarina (32), São Paulo (8)], Chile (1), and Paraguay (1)] were assessed for claw lesions according to the methodology described above. The initial selection included a total of 26,030 sows evaluated, with an average of 102 ± 0.04 sows evaluated per herd. The average inventory size of herds was 2,275 sows (ranging from 250 to 10,000 sows; mode: 500). After the assessment, relevant herds were critically evaluated in terms of fitness to the objectives of the present multivariate analysis approach. Since one of the main purposes of the present study was to identify patterns of claw lesion responses in Brazilian sow herds, the herds in Chile and the herd in Paraguay were excluded. Furthermore, it was determined that 10 herds had incomplete records on the number of sows assessed. Therefore, incomplete records was used as a second selection criterion, and resulted in the exclusion of an additional 10 herds. Subsequently, a checklist was performed in the selected herds to define their inclusion in the meta-analysis. The main criteria for herd exclusion were: (a) herds rearing undefined genetic lines, which consisted basically of rudimentary lines (n = 10), (b) herds where mixed genetic lines were raised (e.g., PIC and DanBred within the same herd; n = 15), (c) an unbalanced number of parities was assessed (n = 7), (d) breeding herds (n = 2), (e) herds evaluated more than once (n = 5). The final database ended up with 129 herds (n = 12,364 sows) for this study. After performing the screening procedure, the information relative to the proposed model, which is discussed in detail below, and outcome variables (i.e., HOE, HSC, WL, CHW, CVW, T, DC, and L-Index) were tabulated using a database from an electronic data spreadsheet.

Table 1 shows each specific lesion's average, median, mode, and SD within a year. Table 2 shows the number of sows and herds assessed per year according to the factors extracted for analysis, which are explained in detail below.

Herds were in the Midwest (n = 34), Southeast (n = 7), or South (n = 88) regions of Brazil. Tropical Savanna, Monsoon-influenced Humid Subtropical, and Humid Subtropical climates, respectively, characterize these regions. Inventory sizes ranged from 250 to 10,000 sows (2,907.55 ± 2,352.06). Inventory sizes were further stratified into 250–810, 811–1,300, 1,301–3,000, and 3,001–10,000 sows. Herds belonged to three different production structures, namely Cooperative (Coop; 41), Integrator (50), and Independent Producers (38). The Coop provides the pigs and feed to Coop members at the cost of production. After slaughtering and marketing the pork, profits from the Coop are apportioned back to producers based on the total number of pigs marketed. In the Integrator structure, the largest company (integrator) offers the farmer pigs, feed, and technical assistance and is responsible for slaughtering and commercializing the pigs. The farmer provides facilities, equipment, heating, water, and labor. At slaughter age, pigs are retrieved from farmers by the Integrator. In the Independent structure, the farmer makes their own decisions on pig and feed sourcing, quality assurance, sanitary management, and commercialization. These are usually smaller herds. Management of the herd was conducted either as a closed herd genetic approach, where breeds were maintained from stock on-site (internal), or through purchase of commercially available genetics (external). The latter including purchased sows, boars, or semen through breeders, live auctions, or boar studs. Sows were housed in individual crates (n = 88 herds) or group housing (n = 41 herds) during gestation. The individual housing system was fairly consistent across herds, providing ~2.40 × 0.65 m (length × width, 1.56 m²/head) of space per sow with an individual feeder and drinker in the crate. The individual housing was slightly larger than the size of the sow's body. In this sense, there was only enough room for the sow to stand or lie down in place, with no room for the sow to turn around or move freely. The gestating sows in the group housing systems were housed in pens measuring 10.5 × 14.4 m (5.04 m²/head). Sows in the group house could move freely. All the herds had similar farrowing units equipped with individual farrowing crates. Each crate measured ~1.10 × 2.41 m, with a piglet area of ~1.53 × 0.53 m, which had an accessible creep area, rubber mats, and a heat lamp. Individual bowl feeders and nipple drinkers were located at the front of each sow space.

The methodology described by Sauvant et al. (25) and Lovatto et al. (26) was used in the present study for the definition of dependent and independent variables and data encoding. The encodings were used as qualitative and quantitative variables in the analysis with the purpose of considering the herd, inter and intra variability of the compiled data. Specific sequential numbers were used for each herd inserted in the database to encode the herd effect. Within each herd, each sow received sequential numbers for labeling, which characterizes the inter-encoding. This allowed to assign the sow encoding with the specific herd encoding. The intra-encoding, following a similar pattern as described above, was attributed to each factor within sows and herds. The intra encoding was composed by the following: (1) Production structure (Coop, Integrator, Independent); (2) Region (South, Southeast, Midwest); (3) Inventory size (250–810, 811–1,300, 1,301–3,000, and 3,001–10,000 sows); (4) Breed (internal, external); (5) Gestation housing (individual, group); and (6) Parity (1–7). Design criteria included Production structure, Region, Inventory size, Breed, Gestation housing, and parity. The dependent variables extracted for analysis were HOE, HSC, CVW, CHW, WL, T, DC, and L-Index.

Data were analyzed using a multivariate approach whereby relationships between claw lesion variables (HOE, HSC, WL, CHW, CVW, T, and DC) were studied by PCA and HAC. The data were analyzed using SAS, version 9.4 (SAS Institute Inc., Cary, NC, USA) and XLSTAT2017.19.6 (AddinSoft, Paris, France). Data normality was checked based on visual inspection of the raw data histogram and QQ-plots. When variables were not considered normally distributed, categorical variables were encoded as dummy variables, to allow for them to be used as predictors in the PLS approach. For the Cluster analysis, the dissimilarity and similarity measures were tailored to categorial data, which does not require for any type of data transformation.

The main objective of the present study was to implement PLS based on clustering methods; thus, herds were first categorized in accordance with their individual claw lesion scores. The claw lesion scores were averaged per herd prior to any clustering analysis and subsequently used for clustering using HAC. To achieve this, the PROC CLUSTER procedure of SAS was used. This analysis is a task of exploratory data mining, and our main objective was grouping (clustering) the data set in such a way that the herds in the same group (cluster) were more similar to each other than to those in other groups (clusters). First, the function builds a hierarchical tree; then, the sum of the within-cluster inertia is calculated for each partition. Thereby, the suggested partition is the one with the higher relative loss of inertia. The herds were grouped according to similarity in three clusters, i.e., C1, C2, and C3. The three clusters were projected on the graph defined by the principal components (PC1 and PC2; Figure 3). The number of clusters was supported by practical and biological interpretation.

PLS regression (27) was used to examine claw lesion data on the three clusters defined by the HAC. In this stage, the number of components is automatically determined using the Q2 statistic, assessing the significance of a new PLS component in predicting the entire Y set. PLS projects the predictive and observed variables into a new space by finding pairs of weight vectors that maximize the covariance between the two projections. By examining this new space, the underlying relationship between clusters and claw lesions can be explored (28). Subsequently, confidence intervals for PLS regression coefficients (bkj) are calculated through jackknife. The contribution of each factor to the model in terms of the variance explained was indicated by the variable importance in projection (VIP) and the standardized regression coefficients were estimated as well to confirm the selection of the variables. In this sense, a VIP ≥ 1 was assumed as a selection threshold. To achieve this, the PROC PLS procedure of SAS was used.

Table 1 shows the average, median, and mode values for each lesion per evaluation year. Heel overgrowth and erosion average score was 1.05 ± 0.38, ranging from 0 to 3, with mode 1. Heel-sole crack average score was 0.16 ± 0.15, ranging from 0 to 3, with mode 0. White line average score was 1.18 ± 0.57, ranging from 0 to 3, with mode 1. Horizontal wall cracks average score was 0.73 ± 0.45, ranging from 0 to 3, with mode 0. Vertical wall cracks average score was 0.32 ± 0.22, ranging from 0 to 3, with mode 0. Overgrown toes average score was 0.33 ± 0.37, ranging from 0 to 3, with mode 0. Dewclaws average score was 0.55 ± 0.47, ranging from 0 to 3, with mode 0. It is important to highlight that herds were assessed at different time points throughout the year since this was a long-term evaluation project. Since the month of evaluation did not have an effect on the severity of any of the claw lesions (Figure 4), it was not included in the model.

Table 2 summarizes the number of sows and herds assessed per year stratified according to the factors examined (i.e., production structure, parity, inventory size, etc.). The number of assessments per factor was reasonably consistent throughout the years, except for breed and gestation housing types. Parity was equalized as much as possible within sample time, ranging from 1 to 7 in all of the sample times. The number of sows and farms (in parenthesis) assessed for internal and external breeds was 2,430 (29) and 9,934 (100), respectively, and 8,571 (96) and 3,793 (33) for individual and group gestation housing, respectively. These are expected, as the breeding herd component may represent ~20% of the total feed produced and can be cost-prohibitive for many production systems (31). In this sense, purchasing the breeding stock from external sources is generally more advantageous. Recently, Klein et al. (32) reported that gilt replacement protocol may impact sow farms' claw lesions. Regarding gestation group housing, the Brazilian Ministry of Agriculture, Livestock and Food Supply recently released the first pig welfare legislation (Normative Instruction # 211, February 1, 2021), which establishes that group housing during gestation is mandatory and sows should not be kept in individual crates for more than 35 days of gestation. This is also consistent with findings from the present study, with an increase in the percentage of herds using group housing in 2022 (52% of herds evaluated) and 2023 (71% of herds evaluated) compared to the previous year (1% of herds evaluated; 2013–2021).

Heel overgrowth and erosion had the highest prevalence throughout the years, followed by WL and CHW. This is consistent with multiple reports in the literature showing that the most frequently observed claw lesions in sows are HOE, WL, and CHW (6, 9). Furthermore, the lowest scores were observed for T, DC, and CVW, which is similar to the findings from Henningsen (33) in a retrospective study on U.S. sow farms. The authors observed that toe abnormalities were not as severe as previously believed and that histopathology did not indicate a direct relationship between these abnormalities and weight-bearing. This was due to the fact that similar lesions were present in both the lateral toes and lateral dew claws.

Interestingly, L-Index had a progressive decrease from 2013 to 2023 (Table 1). This finding may be due to a combination of factors. First, it is important to consider that claw size and hoof growth have gained traction as selection criteria due to their important genetic background, directly impacting sow culling (4). Secondly, despite being speculative, improvements in housing (flooring) conditions have certainly been made throughout the years, reducing the predisposition of sows to claw lesions (34). Also, refinements in feeding practices during the past 10 years may have been allowed for a better support of claw health in these herds. For example, nutrient requirements for gestating and lactating sows were solely based on growth and reproductive performance for many years, and advancements in mineral and vitamin nutrition have shown that they play a pivotal role in maintaining optimum foot health in sows (35, 36). The herds were grouped in three clusters, i.e., C1, C2, and C3. From the multivariate perspective, the spatial distribution of each observation related to the first and second principal components and variable axis are presented as a biplot in Figures 3A, B, representing a condensed summary of the database structure. These two components represented a cumulative inertia of 71.98%. Figure 3A also shows that HOE, CHW, and CVW contributed most to the first component, with coordinates of 0.86, 0.81, and 0.75, respectively, while WL, DC, and T contributed most to the second component, with coordinates of 0.86, −0.62, and −0.55, respectively. It is also possible to compare clusters regarding claw lesion variables by observing their centroid position in these 2 dimensions. Hence, it is possible to conclude that C3, compared to the other clusters, is represented by herds with lower scores for all the lesions except for DC. It is also possible to evaluate variable relationships since the angle between their axes reveals how they relate. The smaller the angle between the 2 variables, the more positively correlated they are, which may be concluded from the relation between T and DC, and HOE and CHW. Indeed, HOE is one of the most prominent lesions affecting the outer claw of the hind leg due to weight-bearing biomechanics (9, 37). The softer heel horn, comprising fewer tubules, bears most of the weight (38), unlike other cloven-hoofed animals where the sole is predominant. Since the occurrence of HOE is highly associated with flooring surface abrasiveness (39, 40), it is common to observe its cooccurrence with wall cracks (9), which is in line with our findings. Moreover, and on the contrary, variables with opposite directions are negatively correlated, which means that for this database, the higher the WL score for a given sow is, the lower the score for T and DC may be. This also makes biological sense, as T and DC are lesions particularly observed in older sows (10), while WL is generally present earlier and leads to sow culling in many situations (41–43).

In a univariate approach, C3 herds were characterized by having lower scores for all the lesions compared to C1 and 2, except for DC, which were basically the same across clusters (Table 3). Specifically, HOE, HSC, WL, CHW, CVW, and T were decreased by 17, 25, 11, 25, 21, and 17%, respectively, in C3 compared to C1 and 2 combined. In comparing C1 and C2, they had similar scores for individual lesions. The L-Index was greater in C2 compared to C1, which indicates that overall claw quality was further deteriorated in C2 herds despite C1 being affected by multiple lesions.

Table 4 shows the characterization of the herds according to cluster group. There was an increased proportion of herds and sows belonging to C1 (56% of herds) compared to C2 (26% of herds) and C3 (18%). This remarkable finding shows that most sows may be at risk in terms of claw quality, and sows may indeed be suffering from multiple lesions. Regarding production, the three clusters followed the same trend, with a higher proportion of herds in Coop and Integrator structures than in Independent. This exemplifies a transition seen in the Brazilian swine industry, with more producers shifting to Integrator or Coop structures. This is mainly driven by price and market, which are absorbed by Integrators and diluted in Coop environments. Most herds from C1 and C2 were present in the South region, followed by the Midwest, with only a few located in the Southeast. On the other hand, C3 had almost the same number of herds located in the South and Midwest, with a few herds in the Southeast. The South and Midwest regions represent the main pork production regions in Brazil, geographically positioned closer to grains, grow/finishing facilities, and packing plants. C1 herds were homogeneously distributed across inventory sizes. C2 had the majority of herds with inventories ranging from 810 to 1,300 and no herds with more than 3,000 sows. C3 was basically characterized by a lower percentage of herds with inventory sizes of 250–810 sows. Breed type did not differ among clusters, with a higher proportion of external than internal breeds. The proportion of herds with group gestation housing increased progressively from C1 through C3.

The selection of variables in PLS regression models depends on defined thresholds, such as VIP, which was chosen in the present study (44). However, these thresholds are not strict limitations, and variables near them should also be considered (45), especially given the multifactorial nature of claw quality (8, 17, 46). Variable selection was conducted separately for each cluster to enhance generalization to other herds, favoring a more generalized approach (Figure 5). The L-Index within each cluster was used as the response variable to employ the PLS regression. This was done to investigate whether the factors leading to overall claw quality deterioration would differ between clusters.

The L-Index increased with increasing parity (VIP = 1.291, regression coefficient = 0.052), which was expected from the aging effect on claw quality previously discussed (Figure 5). However, this was observed in C1 only. This suggests that parity increase does not always lead to deterioration in claw quality and confirms the multifactorial scenario by which claw lesions are influenced. Independent production structure increased the L-Index in all three clusters (C1: VIP = 1.154, regression coefficient = 0.046; C2 = VIP = 2.056, regression coefficient = 0.116; C3: VIP = 1.140, regression coefficient = 0.060). This was concomitant with Coop and Integrator production structures, reducing the L-Index in C2 (VIP = 1.544, regression coefficient = −0.087) and 3 (VIP = 1.406, regression coefficient = −0.074), respectively. Better claw quality in Integrator/Coop structures compared to Independent may be explained by a higher standardization of processes, including genetic material, nutrition, technical expertise, and facilities. The Southeast and South regions had increased L-Index in C1 (VIP = 1.046, regression coefficient = 0.042) and 2 (VIP = 1.181, regression coefficient = 0.062), respectively, while the Midwest region had decreased L-Index in C2 (VIP = 1.189, regression coefficient = −0.048) and C3 (VIP = 1.453, regression coefficient = −0.076). It is important to highlight the higher proportion of herds in the Midwest in C3 (45%) compared to C1 and C2 (around 21%). This may also be explained by a recent trend in Brazilian herds, with the South directing more pork to exports and the Midwest raising a new, technified, integrated source of pork to meet internal demands.

Individual housing increased the L-Index regardless of cluster (C1: 1.694, regression coefficient = 0.068; C2: VIP = 1.567, regression coefficient = 0.088; C3: VIP = 1.206, regression coefficient = 0.063). This goes against findings from Anil (2), who conducted claw lesions scoring in 184 sows in farrowing stalls on day 110 of gestation, associating them with gestation housing system (individual vs. group), using logistic regression models. The authors reported that the prevalence of claw lesions among gestating sows was significantly higher in group pens with electronic sow feeding (ESF) compared to stalls, except for toe lesions. The majority of sows with various types of claw lesions, including wall, heel, white line, heel-sole junction, sole, and overgrown heels, were housed in group pens during gestation, ranging from 57.8 to 75.4%. This increased occurrence of claw lesions in sows housed in group pens with ESF systems could be attributed to fighting and aggressive behavior during mixing and feeding times. The elevated mobility afforded by the ESF system may have contributed to this discrepancy compared to sows housed in stalls.