Laminitis results in lamellar (laminar) dysadhesion within the suspensory apparatus of the DP that leads to anatomic change within the affected foot.1,2 Anatomic changes include alterations in the internal relationships among the osseous structures of the foot and their relationships with the hoof capsule or changes to the morphometric characteristics of the DP. Objective methods of radiographic assessment are required to ensure accurate diagnosis of anatomic change within a laminitic foot, especially when mild changes are present. Previous studies3–6 have attempted to provide a quantitative basis for the diagnosis of anatomic change associated with laminitis in horses and ponies. However, progress has been limited because of the use of various assessment methods and limitations in data analysis. Hence, the objective diagnosis of laminitis-associated anatomic change remains imprecise.
A lateromedial radiographic view represents the gold-standard image for the detection of anatomic change within a laminitic foot. However, an increasing number of distinct pathological changes are known to be associated with laminitis, including so-called capsular rotation (generally defined as angular deviation between the dorsal aspect of the DP and the dorsum of the hoof wall), phalangeal rotation, distal displacement of the DP, and combined rotation and displacement events.7 Various radiographic measurements have been used to document these pathological changes. This has resulted in the emergence of a complex data matrix that makes diagnostic interpretation difficult. Consequently, there has been a tendency among researchers to assess single or several measurements in isolation,3–5 rather than to analyze the combined information as a definitive measure of anatomic change.
Among horses7 and donkeys,8,9 there is extensive variation in the clinical signs of laminitis; therefore, it is highly unlikely that any single radiographic variable would effectively capture the entire extent of anatomic change within an affected foot. This is of importance because prognosis is believed to be dependent on the nature and extent of anatomic change within a laminitic foot.3–5,10 Some progress has been made to address this problem by use of stepwise regression methods to incorporate information from several radiographic measurements.6 However, this statistical approach, which, in the authors' opinion, is not readily transferable to clinical settings, only highlights the perceived critical importance of a single radiographic variable. Hence, there is a need to find an alternative means of evaluating the combined information derived from several radiographic measurements. Multivariate statistical techniques may provide the means to achieve this important objective. These techniques have been used successfully in veterinary medicine to assess other complex data matrices and have aided in the diagnosis of navicular syndrome11,12 and hind limb lameness in horses.13 However, these techniques have not been used previously to assess radiometric data, to our knowledge.
Multivariate exploration and classification techniques enable variable- and object (individual)-directed analyses of multivariable data.14 These statistical techniques have been described in detail.15 Overall, exploration techniques provide information regarding individual and group variability and promote understanding of those measurements, which are responsible for this variability. Conversely, classification techniques are intended to mathematically model this variability and perform individual discrimination and group allocation tasks.
The effectiveness of these analyses is dependent largely on selection of appropriate measurements. Hence, the critical step is the application of a priori knowledge to define an appropriate set of measurements. For laminitis, these measurements must characterize the key elements of radiographic images of the anatomic characteristics of the foot and enable detection of specific pathological events that are thought to be pathognomonic of the disease or of prognostic value. In our opinion, it appeared likely that multivariate statistical techniques could be used to develop a novel and robust method for the diagnosis and quantification of laminitis-associated anatomic change within donkey feet on the basis of lateromedial radiographic data.
Thus, the purpose of the study reported here was to establish and validate an objective method of radiographic diagnosis of anatomic changes in laminitic feet of donkeys on the basis of data from a comprehensive series of radiographic measurements. The intent was to investigate various multivariate methods to model the radiographic data, to determine the methods best suited for the identification of anatomic change within a laminitic donkey foot, and to assess whether the multivariate approach could be used successfully to develop a measurement-based diagnostic technique for the detection and appraisal of anatomic change within a laminitic donkey foot.
Materials and Methods
Sample size—Pre hoc sample size calculation for a 2-sample t testa was based on a preliminary multivariate analysis of radiometric data, assuming SD of 2.1, α of 0.05, power (1 − β) ≥ 0.8, and a desired detection difference in means of 1. This calculation gave a minimum required sample size of 71 donkeys/study group. However, to account for potential experimental errors in data collection, radiographs were obtained from an initial group of 85 nonlaminitic and 85 laminitic donkeys. In this way, it was expected that a minimum statistical power of 0.8 would be achieved.
Donkeys—Two hundred thirty-two donkeys were randomly selected from the donkey herd managed by The Donkey Sanctuary, Sidmouth, Devon, England. One hundred seventy donkeys were selected for baseline data determinations; 62 donkeys were selected to provide data for validation analyses. For baseline data determinations, 2 groups were established: a group of donkeys with clinically normal forefeet (n = 85) and a group of donkeys with unilateral or bilateral forefeet laminitis (85). The first recorded incidence of acute-phase laminitis within the laminitic group ranged from 6 months to 11 years previously. For the validation analyses, a randomly selected cohort of 62 donkeys was established; this cohort included donkeys with clinically normal forefeet (n = 18) and donkeys with unilateral or bilateral forefeet laminitis (44). These donkeys were not used in the initial baseline data determination.
All donkeys were maintained under a standardized farriery regimen by a group of farriers experienced in trimming donkey feet. Case selection was designed to provide a random sample of nonlaminitic and laminitic feet with assorted anatomic characteristics and was established by reviewing medical records. Nonlaminitic donkeys were determined to be unaffected on the basis of an absence of a history of foot-related problems and visual inspection to confirm the absence of gross pathological changes of the hoof capsule. Visual evidence of a foot-related problem resulted in exclusion from the clinically normal group. Laminitic status was determined on the basis of clinical history of acute-phase laminitis. Specifically, the clinical diagnosis of acute-phase laminitis was determined on the basis of a combination of clinical signs, including stance and gait irregularities, elevated digital pulse amplitude, increased sensitivity to hoof testers in the dorsal aspect of the foot, elevated hoof temperature, signs indicating supracoronary depression, and general behavioral changes indicative of pain. Each hoof underwent visual inspection for signs consistent with laminitis, including the presence of divergent growth rings (expanding circumferentially from the dorsum to the heels), expansion (widening) of the white line, extensive flattening of the sole, solar hemorrhage in the region adjacent to the dorsodistal margin of the DP, distortion of the hoof capsule, dorsal concavity of the hoof wall, and perioplic hyperproliferation.
At The Donkey Sanctuary, donkeys are under the care of a team of veterinarians, and there are specific management guidelines (including protocols for feeding and grazing access) for donkeys with laminitis. Palliative care and treatment of secondary pathological alterations, including recrudescent changes in the feet (chronic abscess formation), were under the direction of the attending veterinarian at all times. Use of orthopedic shoes and foot supports was determined on an individual animal basis.
Radiography—Lateromedial radiographic views of 1 forefoot of each donkey were obtained. For donkeys with unilateral forelimb laminitis, the affected foot was evaluated; for donkeys with bilateral forelimb laminitis, 1 forefoot was randomly selected for evaluation. For all donkeys, lateromedial radiographic views were obtained by a previously described procedure.9 In brief, radiographic assessment was conducted at a random stage during the normal 6-week farriery cycle. Each forefoot undergoing evaluation was cleaned prior to radiography. A soft wire marker of known length was attached to the dorsal aspect of the hoof wall. The marker was positioned at the palpable proximal limit of the hoof wall. This enabled radiographic calibration (to account for any magnification effects) and for discrimination of both the dorsal aspect and proximal limit of the hoof wall. The foot was placed on a wooden block (8 cm in height) containing a wire ground line marker. Care was taken to ensure that the limb was fully weight bearing, with the long axis of the metacarpal region positioned perpendicular to the ground. The contralateral limb was also supported on a wooden block to allow even weight distribution between the forelimbs.
The forefoot undergoing evaluation was positioned so that the bulbs of the heel were perpendicular to the x-ray film cassette. In this manner, a true lateromedial radiographic view of the foot was obtained, with beam obliquity lower than the critical value of 10°.16 Exposure factors of 53 kV at 1.2 mA were used for a 70-mm-wide hoof. Kilovoltage was adjusted by 1 kV (plus or minus) for a corresponding 5-mm (plus or minus) change in hoof width. All radiographs were obtained at a film-focal distance of 80 cm, with the beam focused midway between the dorsal and palmar aspect of the foot and midway between the coronary band and the weight-bearing border. All radiographic procedures were performed by a single experienced operator (SNC) to optimize reproducibility and control for repeatability effects. Radiographs were scanned at a resolution of 400 dots/inch on a flatbed scanner, and the resultant digitized images were analyzed with a software program.b
Radiographic measurements were selected on the basis of previous study findings9 (Figures 1 and 2; Appendix). The measurements were selected to enable radiographic characterization of key anatomic features of a donkey forefoot and, by use of an a priori selection, enable detection of specific pathological events that are thought to be pathognomonic of laminitis or of prognostic value. The software program provided 2 measurement protocols that could be used to evaluate radiographic variables: a guided mark-up protocol and a free user-defined mark-up protocol. By use of the guided mark-up protocol, radiographic variables that are based on predefined anatomic reference points (selected by the user) can be evaluated, and relationships between those structures and the ground and anatomic interrelationships are computed. The free user-defined mark-up protocol allows features of interest to be measured by the user with measurement tools to derive linear or angular measurements.
Multivariate statistical techniques—Data analyses were conducted with scan chemometric analysis software.a Data were assessed with the exploration method of PCA, and 7 supervised classification methods were evaluated, as described elsewhere.17–19 These were nearest mean classification, linear discriminant analysis, quadratic discriminant analysis, RDA, soft independent modeling of class analogues, K-nearest neighbors classification, and classification and regression trees.
PCA—The principal component method is a vector-based unsupervised analysis technique that provides an optimized method of multivariable data inspection, in which interrelationships between measurements are anticipated.20 Although multivariable data cannot be visualized simultaneously via conventional statistical methods, the principal component method provides a visual basis by which this can be readily achieved. This allows group, individual, and variable evaluations to be easily made. Principal component analysis overcomes the inherent problem of handling multivariable data by calculating a series of new orthogonal theoretical axes (ie, component axes), which maximize the variability present within the original data (optimized). This is achieved by a rotational transformation of the original data axes, with the new component axes representing linear weighted combinations of all measured variables. In this way, each variable contributes in part to each of the new component axes. The plane of maximum variation within the original data is thus defined by component 1 and 2 axes. This can be displayed graphically and represents the optimal inspection window for data assessment. Component scores are calculated automatically for each individual on the basis of their weighted value for each measured variable, thereby enabling projection of the individual onto the inspection window. Likewise, the vectorial contribution of each variable (loading) can also be projected onto the inspection window. In this way, the spatial distribution and interrelationships of groups and individuals can be observed and the influence of the respective measured variables can be assessed.
Supervised multivariate classification—Supervised multivariate classification methods serve to allocate individuals to one of several groups by means of an optimized classification rule. The classification rule is calculated from the mathematical model that best defines the relationship between the respective groups. The mathematical model and its classification rule are established from individuals of known group membership (the training set [ie, nonlaminitic and laminitic donkey groups used for baseline data determination]). Once established, verified, and cross-validated, the rule can then be used to allocate individuals of unknown membership (the validation set [ie, nonlaminitic and laminitic donkey groups used for validation analysis]) as being either nonlaminitic or laminitic. The classification rule can be further optimized with regard to misclassification errors. In this manner, the rule can be weighted so that either type 1 (false-positive) or 2 (false-negative) errors are further reduced. Seven multivariate classification methods were initially assessed, and the best performing method (ie, the method that had the highest classification success and the lowest misclassification error) was selected for evaluation of the validation data.
Validation—Once the initial analyses had been performed and an optimal classification method selected, a final independent validation exercise was conducted in a blinded manner by a single experienced operator (SNC) to assess the performance characteristics of the diagnostic technique. The validation analysis was performed on the basis of data derived from a randomly selected cohort of 62 nonlaminitic and laminitic donkeys. These donkeys had not been used in the initial baseline data determination. Group membership was determined on the basis of similar group inclusion criteria. Group membership for the donkeys used in the validation analysis was confirmed only to the assessor after all analyses were completed.
ROC curve analyses—To evaluate and compare the discriminatory potential of the PCA component score of components 1 and 2 for identification of nonlaminitic individuals, data were analyzed by ROC curve analyses to assess sensitivity and specificity estimates for all cutoff points within the data on the basis of PCA component scores applied against the gold-standard health status derived from the medical history. Receiver-operating characteristic curves of sensitivity versus 1 − specificity for both PCA component scores for components 1 and 2 were generated with a formal statistical testing applied to test the null hypothesis that the areas under both ROC curves were equal. Examination of individual data values was conducted to identify the PCA component score cutoff values for optimal sensitivity and specificity. Receiver-operating characteristic analyses were conducted with software,c and values of P < 0.05 were considered significant.
Results
Donkeys—For baseline data determination, lateromedial radiographic views of 1 forefoot of 85 nonlaminitic and 85 laminitic donkeys were obtained. However, images for 2 nonlaminitic (1 gelding and 1 mare) and 11 laminitic (7 geldings and 4 mares) donkeys were excluded from the analysis due to motion blur or marked obliquity in the radiographic views. Hence, radiometric data from 83 nonlaminitic and 74 laminitic donkeys were used in the baseline data determination. There were 34 geldings and 49 mares in the nonlaminitic group (mean age, 15 years; mean body weight, 163 kg) and 40 geldings and 34 mares in the laminitic group (mean age, 21 years; mean body weight, 179 kg). The data matrix contained information for 20 angular and linear anatomic variables measured on the lateromedial radiographic view of each foot (1 foot/donkey). However, due to multicolinearity effects with dorsal hoof wall angle and the angle of the proximal phalanx, the hoof pastern axis was excluded from the analyses.
PCA analysis—The results of the PCA revealed that components 1 and 2 cumulatively explained approximately 60% of the total variation present within the entire multivariable data set. Hence, the plane of maximum variation, defined by components 1 and 2, represented a highly informative inspection window for further data interpretation, given that a single variable axis in an independent data set of 19 variables would typically account for only one-nineteenth of the total variation. The spatial distribution of the nonlaminitic and laminitic group members within this inspection window, on the basis of the derived component scores for each individual, was plotted (Figure 3).
In the baseline data determination, the nonlaminitic group appeared as a cohesive cluster with few outliers, unlike the laminitic group. This indicated that the feet of the nonlaminitic group had a degree of anatomic similarity. Conversely, the diverse distribution evident within the laminitic group was indicative of greater individual variation in radiographic findings.
Component scores for the 2 study groups were summarized (Table 1). The 2 groups were separate with respect to component 1, with little between-group separation in the direction of component 2. The separation with respect to component 1 indicated the presence of important differences in certain aspects of the radiographic characteristics between the 2 groups. Between-group comparisons via a 2-sample t test, following confirmation of normal data distribution by use of the Anderson-Darling normality test, revealed a significant (P < 0.05) difference between mean component 1 scores. Evaluation of the scatterplot suggested that there was an effective between-group boundary at a component 1 score of approximately −1.0. However, group separation at this boundary was not complete, and a degree of overlap existed between the nonlaminitic and laminitic groups.
Component 1 and 2 scores derived via PCA of data from a series of 19 radiographic measurements in a study to establish and validate an objective method of radiographic diagnosis of anatomic changes in laminitic forefeet of donkeys.
Variable | Nonlaminitic group (n = 83) | Laminitic group (n = 74) | ||||
---|---|---|---|---|---|---|
Mean ± SD | Range | 95% CI | Mean ± SD | Range | 95% CI | |
Component 1 score | 1.597a ± 1.299 | −1.662 to 4.010 | 1.312 to 1.883 | −1.747b ± 2.747 | −9.228 to 2.787 | −2.379 to −1.114 |
Component 2 score | −0.134 ± 1.947 | −4.794 to 5.719 | −0.562 to 0.294 | 0.146 ± 1.886 | −4.814 to 4.264 | −0.287 to 0.580 |
Donkeys were classified as nonlaminitic or as having unilateral or bilateral laminitis on the basis of medical record review and assessment of current clinical signs. One forefoot of each donkey was used in the analyses. Initially, 85 nonlaminitic and 85 laminitic donkeys were evaluated; data from 2 unaffected donkeys and 11 donkeys with laminitis were excluded from the analyses because of motion artifact in the radiographic views.
Different superscripts denote a significant (P < 0.05) between-group difference determined by use of a 2-sample t test.
The corresponding vector plot for the radiographic variables (Figure 4) provided an understanding of the underlying causal factors that govern both individual and group distributions within the inspection window. The vector diagram indicated that component 1 was strongly influenced by angular deviation between the dorsal aspect of the DP and the dorsum of the hoof wall, apex angle of the DP, integument depth of the dorsal aspect of the foot at the mid-dorsal site, reflex angle of the DP, and phalangeal axis. Conversely, angle of the middle phalanx, middle phalanx length, angle of the proximal phalanx, and dorsal hoof wall angle strongly influenced component 2. The remaining variables, which were oriented diagonally to the axes, exerted an influence with respect to components 1 and 2.
Supervised multivariate classification—Analyses conducted via 7 multivariate classification methods revealed that the RDA method best modeled the radiographic data (Table 2). The optimal RDA model provided the lowest group discrimination error (16%) and group prediction error (17%). This RDA model was able to correctly classify 79 of 83 nonlaminitic donkeys (negative diagnosis) but could only classify correctly 52 of 74 laminitic donkeys (positive diagnosis). A second analysis, with the classification rule weighted against false-negative diagnostic error, resulted in correct classification of 66 of 74 laminitic donkeys. However, this weighting resulted in the misclassification of 14 nonlaminitic donkeys.
Group recognition and prediction performances for the 7 multivariate classification methods, compared with the critical no-model statistic, used to analyze radiographic data obtained from 83 donkeys without laminitis and 74 donkeys with laminitis in a study to establish and validate an objective method of radiographic diagnosis of anatomic changes in laminitic forefeet of donkeys.
Method | No-model error (%) | Group recognition error (%) | Group prediction error (%) |
---|---|---|---|
Nearest mean classification | 47.8 | 21.7 | — |
Linear discriminant analysis | 47.8 | 16.6 | 22.3 |
Quadratic discriminant analysis | 47.8 | 21.6 | 21.6 |
RDA* | 47.8 | 16.1 | 17.2 |
Soft independent modeling of class analogues | 47.8 | 30.6 | 35.7 |
K-nearest neighbors classification | 47.8 | — | 26.7 |
Classification and regression trees | 47.8 | 17.2 | 19.1 |
All classifications assumed equal class prior and unit loss function.
Optimal method for group recognition and group prediction.
— = Not modeled successfully.
The group allocation derived from the weighted RDA analysis projected onto the optimal inspection window was plotted (Figure 5). The weighted classification resulted in a shift to the right in the effective boundary between the nonlaminitic and laminitic groups within the inspection window, yielding a component 1 score of approximately 0.5.
Validation analysis—Validation analysis of the new objective method of radiographic diagnosis of anatomic changes in laminitic forefeet of donkeys was performed involving data from an additional 18 nonlaminitic and 44 laminitic donkeys. The performance of the method resulted in correct classification of 34 of the 44 laminitic donkeys (10 laminitic donkeys were incorrectly classified); among the 18 nonlaminitic donkeys, 5 were correctly classified (13 nonlaminitic donkeys were incorrectly classified). The weighted RDA model yielded a sensitivity of 77.27% (95% CI, 64.89% to 89.66%), specificity of 72.22% (95% CI, 51.53% to 92.91%), and positive and negative predictive values of 87.18% (95% CI, 76.69% to 97.67%) and 50.00% (95% CI, 36.26% to 76.78%), respectively.
ROC curve analyses—Based on ROC curve assessment, PCA component 1 score was considered to have performed adequately against the gold-standard health status, with an area under the ROC curve of 87.3% (95% CI, 81.6% to 92.9%; Figure 6). This was significant (P < 0.001) and superior to PCA component score 2, which was not considered adequately discriminatory, with an area under the curve of only 43% (95% CI, 33.8% to 52.1%). Sensitivity and specificity estimates for PCA component score 1 indicated an optimal cutoff value of 0.5, which gave rise to overall estimates of sensitivity of 82%, specificity of 82.5%, positive predictive value of 84%, and negative predictive value of 80.5% for this discriminatory test.
Discussion
Although multivariate statistical methods are becoming widespread in the field of human medicine21 and have recently been used in the veterinary field,11–13 the present study represents the first occasion in which these techniques have been applied to the diagnostic evaluation of radiographic data in equidae, to our knowledge. Specifically, results of the present study have indicated that these techniques can be used to successfully develop a robust means of diagnosing anatomic change in a laminitic donkey foot from lateromedial radiographic views. This represents a logical progression to former methods of objective radiographic assessment and offers a means for the quantitative appraisal of donkey feet, thereby providing an important adjunct to traditional methods of subjective clinical assessments. Confidence in these analyses can be inferred because the 2 multivariate techniques (PCA and RDA), each performed with different methods of calculation, provided results that were in broad agreement in the present study.
The principal component method provided a visual technique by which clinical appraisal of the radiographic characteristics of donkey feet can be made. It allowed group, individual, and radiographic variable evaluations to be easily made. The inspection window provided a means of visual inspection of data related to the radiographic appearance of the anatomic features of donkey feet and data related to individual anatomic variations. This inspection window accounted for 60% of the total anatomic variation recorded across all 19 variables assessed in the present study. This represents an improvement over former methods of objective assessment, which rely on evaluating either information gained from a single variable of choice or information obtained by evaluating several variables in isolation. This is because single variable assessment alone provides limited information regarding the complete radiographic appearance of the foot. Conversely, traditional methods for the interpretation of multivariable assessment are neither intuitive nor easily achieved. Typically, the ideal outcome of a PCA, as commonly used when analyzing chemical spectral data, is that approximately 90% of the variability is captured within the first few principal components. The variability explained in the present study decreased slightly below this ideal position (78% by 4 components). This shortfall may result from measurement or experimental error associated with the study. Regarding the former, errors may have arisen from inconsistencies in the selection of the anatomic reference points used by the software program.a In addition to this systematic measurement error, the performance of the PCA analysis may be affected by the presence of concurrent pathological changes affecting the anatomy of the foot. In this regard, it has been argued that laminitis in donkeys forms part of a broader continuum of a degenerate foot condition.22 Likewise, individual differences in trimming methods among farriers may add to the overall variability in foot conformation and thus the radiographic characteristics of the foot. Similarly, differences in hoof horn growth and wear rates among donkeys, irrespective of disease status, would further contribute to total variation in the radiographic data.
The inspection window used in the present study readily highlighted differences in anatomic features between nonlaminitic and laminitic forefeet and provided a basis for the clinical assessment of donkeys by allowing direct graphic comparison of radiographic data with known anatomic features of clinically normal and laminitic donkey feet. The inspection window indicated a division between nonlaminitic and laminitic foot characteristics in donkeys along the component 1 axis, at a value of a 0.5. The position of a donkey with respect to the component 1 axis reflects the degree of anatomic change in the forefoot associated with laminitis. For donkeys that have a component 1 score ≥ 0.5, the radiographic appearance of the anatomic features of the foot is considered normal. Conversely, for donkeys that have a value ≤ 0.5, there is evidence of anatomic change in the foot. The actual component 1 score relates directly to the severity of the anatomic change evident within the foot. Hence, a score ≤ −3.5 is indicative of extreme anatomic change and a score within a range from −2.0 to 0.0 denotes modest change. Finally, nonlaminitic donkeys with a score within a range from 0.5 to 1.0 should be considered at risk because they may have initial signs of subclinical anatomic change associated with laminitis. Such marginal cases would warrant further detailed medical appraisal. In the present study, the diverse distribution of the laminitic donkeys within the inspection window indicated that considerable between-animal variation in the anatomic features of the foot was present with respect to scores for components 1 and 2. This corresponds with the large coefficients of variations reported previously for the respective radiographic variables in clinically normal and laminitic donkey feet.9
In the present study, the vector plot (Figure 2) provided information regarding the individual and combined effect of a radiographic variable on the anatomic variation among feet within the inspection window. The precise location of an individual is determined by the weighted interaction of the variable vectors. The plot highlighted the variables that are responsible for the anatomic differences between nonlaminitic and laminitic feet. Specifically, the vectors that act in the direction of component 1 were primarily responsible for the between-group separation. Conversely, variables that aligned with the component 2 axis had a minimal effect on group separation, but instead were major factors in determining separation on the basis of differences in conformation. Variations in the variables that lie between the 2 component axes contributed to both group and stance separation. The respective length of the variable vector indicates the relative importance in separation. Hence, the vector plot furthers understanding of anatomic change in laminitic donkey feet, in which angular deviation between the dorsal aspect of the DP and the dorsum of the hoof wall, lamellar wedge formation, and osseous change to the DP are predominant.
The overlap between the 2 groups within the inspection window in the present study cannot be explained definitively. This overlap may have arisen because the plane of maximum variation accounted for only 60% of the total information contained within the original data. Hence, additional separation may occur along other component axes. Conversely, this overlap may indeed be a true representation of the anatomic features in the evaluated forefeet. In this regard, the retrospective group allocation method used in the present study, in which a history of acute-phase laminitis was used as the selection criterion for inclusion in the laminitic group, was designed to deliver a continuum of anatomic change (from mild to severe) associated with laminitis disease progression. However, this approach was anticipated to result in potential group allocation error. It is possible that several of the acute-phase cases resolved successfully without anatomic change, thereby resulting in no anatomic abnormalities of the feet following disease recovery. It is also conceivable, given the stoic nature of donkeys, that the nonlaminitic group may have contained animals with undetected or subclinical laminitis in which mild anatomic change had developed without overt clinical signs. Hence, an overlap in anatomic features between groups was to be expected.
Data from a new clinical case can be used to generate component scores so that the individual can be positioned within the inspection window. This allows direct comparison with respect to the critical diagnostic anatomic boundaries between nonlaminitic and laminitic feet and enables assessment of the severity of anatomic change present in the affected donkey. Specifically, it can be inferred that the degree of anatomic change within the foot relates directly, in an inverse manner, to the component 1 score. Thus, the lower the component 1 score, the greater the severity of anatomic change. The results of the ROC analyses were consistent with the examination of the data by use of the optimal data inspection window method and indicated that component score 1 was significantly superior to component score 2 in discriminating between nonlaminitic and laminitic donkeys and yielded high diagnostic sensitivity, specificity, and positive and negative predictive values.
Component 1 scores may also provide an objective basis for treatment. Treatment objectives for a donkey with a score ≤ −3.5 should be directed primarily at support of the degenerative foot, whereas objectives for a donkey with a score of −2.0 to 0.0 should be directed at prevention of further anatomic change. Variation in component 2 scores reflects individual anatomic differences not associated directly with laminitis. Those scores reflect differences in foot conformation, characterized by variation in dorsal hoof wall inclination and the angular alignment of the middle and proximal phalanges. It is not known whether such variation represents inherent differences in foot development or the effects of farriery. It is important to note that radiographic views used in the present study were obtained at a random stage in the normal farriery cycle and that studies23,24 in horses have revealed that normal foot trimming specifically alters the interrelationships between osseous structures of the foot. A similar farriery effect is to be expected in donkeys; thus, between-individual difference in the stage of the farriery cycle may have further contributed to the variation evident in component 2 scores.
Component 2 scores also reflect variation in foot size, as measured by the middle phalanx length. However, the middle phalanx length is not related to donkey body weight.9 Collectively, the radiographic variables that influence component 2 are not of direct diagnostic value. However, differences in foot conformation are likely to affect foot function and clinical attempts to stabilize the affected foot. Hence, they may ultimately prove to be of prognostic relevance in the management of a laminitic foot. Further studies are needed to assess the importance of conformation on recovery from laminitis.
A fundamental statistical requirement in multivariate classification is that the classification method can successfully model the training data and hence outperform the no-model situation. The optimal RDA classification method ultimately adopted in the present study met this criterion, indicating that a robust mathematical model could be derived to define the respective groups according to anatomic features. In addition, the resultant classification rule gave adequate levels of diagnostic group recognition and group prediction capabilities.
Initial unweighted analyses revealed that some limitations existed in achieving a high level of group recognition within the laminitic group of donkeys. This is of concern because there is an inherent welfare cost associated with the failure to diagnose, and hence a failure to treat, laminitis. The weighted classification rule achieved major improvements on the former position and reduced the number of false-negative diagnostic errors considerably. In this way, detection rates for modest anatomic change in the foot were further enhanced. This is of primary importance in achieving prompt intervention and treatment for an affected animal.9 This is important for donkeys because clinical signs of pain associated with the development of chronic lameness are often absent until major degenerative anatomic change has occurred within the foot.9,25
In the present study, most misclassification errors involved individuals for which component scores were close to the boundary between the nonlaminitic and laminitic groups. This is because the classification rule was established on the basis of mathematical probability, and group assignment is least certain at the between-group boundaries. Measurement error, within-group variation, and between-group overlap increase the level of uncertainty in the boundary region and hence increase the potential for diagnostic error. These issues reinforce the need for the adoption of a weighted classification rule to minimize the likelihood of false-negative diagnoses. As a consequence of this weighting, however, 14 nonlaminitic donkeys were classified in the laminitic group. It was not possible to confirm whether this represented an inherent limitation of the diagnostic model or a true indication of subclinical anatomic change within the nonlaminitic group. In light of this uncertainty, these individuals must be viewed as potentially at risk, warranting further medical assessment and diligent management.
The results of the validation analysis highlighted the robust nature of the new diagnostic technique in terms of diagnostic sensitivity and positive predictive performance. These diagnostic capabilities are essential to ensure clinical intervention in response to anatomic change. Explanations for the 10 false-negative errors in the present study are equivocal. Component 1 scores for most of those 10 donkeys (n = 7) were close to the between-group boundary. Hence, they could represent either true diagnostic error or recovery from laminitis without anatomic change. The component 1 scores for the 3 remaining donkeys identified as false-negative results were > 2.5 (ie, values typical of feet with normal anatomic features). It is therefore highly probable that these 10 cases represented donkeys that had recovered from laminitis without anatomic change. Further studies are now required to test these assertions by comparing the diagnostic value of this new multivariate approach against traditional methods of subjective and objective radiographic assessment of nonlaminitic and laminitic (both acute and chronic phase) donkeys.
The findings of the present study have suggested that this novel multivariable approach can provide the basis for an objective radiographic diagnostic technique to identify and evaluate anatomic change within donkey feet. The RDA-weighted multivariate classification rule was used to perform an initial diagnostic evaluation of the anatomic features of the donkey feet in the present study. The weighted rule minimized the likelihood of failure to treat. The PCA inspection window enabled direct anatomic comparison between nonlaminitic and laminitic donkeys and provided an understanding of the extent of anatomic change within the affected feet. This method is likely to be of particular clinical relevance because results of previous studies3,4,6 in horses with laminitis have indicated that treatment outcome and return to former performance levels are related to the extent of anatomic change within a laminitic horse foot.
By use of all the information derived from a lateromedial radiographic view via this method, it may be possible to relate anatomic change more accurately to recovery. It is widely recognized that locomotor function and the anatomic features of a foot are closely interrelated and that anatomic change affects locomotor function. Although the precise functional effects of anatomic change associated with laminitis are poorly defined, it is likely that they are determined by both the nature and extent of anatomic change. Hence, different anatomic changes are expected to have different locomotor effects. Empirical evidence suggests that donkeys with laminitis are more tolerant of distal displacement of the DP than they are of rotational dislocation.26 Additional studies are needed to assess the biomechanical effects of anatomic changes in laminitic feet and to establish the precise prognostic value of those changes.
The diagnostic technique evaluated in the present study potentially provides a means of delivering improvements in the management of laminitic donkeys. Specifically, detection of modest change would lead to prompt medical and farriery intervention at an early stage, when prospects of recovery and return to former performance levels are most favorable. The identification of at-risk animals that have signs of subclinical anatomic change would enable appropriate veterinary inspection and diligent foot management to be initiated.
For those clinicians who currently use the software programa used in the present study, it would be possible to introduce this multivariate approach with little additional time needed to derive the radiographic data. With that data, the component scores for a given individual can be simply and immediately generated with a proprietary spreadsheet, then plotted directly onto the optimal inspection window for visual assessment and diagnostic interpretation.
Given successful validation, this new technique could also provide an ideal method for delivering additional improvements in prospective monitoring of laminitic equids and in the assessment of treatment effectiveness in preventing additional anatomic change. Better knowledge of the anatomic changes that are currently associated with a guarded prognosis will allow identification of donkeys that are most likely to remain refractory to current treatments and also help focus and direct the future development of enhanced treatments. Overall, these advances will collectively allow the desired shift in management emphasis from a reactive to a proactive position. By achieving this, it may be possible to further minimize the debilitating effects of laminitis and to optimize recovery prospects in equidae.
ABBREVIATIONS
CI | Confidence interval |
DP | Distal phalanx |
PCA | Principal component analysis |
RDA | Regularized discriminant analysis |
ROC | Receiver-operating characteristic |
Minitab, Minitab Inc, State College, Pa.
Metron, version 2.06, Epona Tech, Creston, Calif.
Stata, version 9.2, StataCorp, College Station, Tex.
References
1 Collins SN, van Eps AW, Pollitt CC, et al. The lamellar wedge. Vet Clin North Am Equine Pract 2010; 26: 179–195.
2 van Eps AW, Collins SN, Pollitt CC. Supporting limb laminitis. Vet Clin North Am Equine Pract 2010; 26: 287–302.
3 Kameya T. Clinical studies on laminitis in the racehorse. Exp Rep Equine Health Lab 1973; 10: 19–40.
4 Stick JA, Lann HW, Scott EA, et al. Pedal bone rotation as a prognostic sign in laminitis of horses. J Am Vet Med Assoc 1982; 180: 251–253.
5 Linford RL, O'Brien TR, Trout DR. Qualitative and morphometric radiographic findings in the distal phalanx and digital soft tissues of sound Thoroughbred racehorses. Am J Vet Res 1993; 54: 38–51.
6 Cripps PJ, Eustace RA. Factors involved in the prognosis of equine laminitis in the UK. Equine Vet J 1999; 31: 433–442.
7 Herthel D, Hood DM. Clinical presentation, diagnosis, and prognosis of chronic laminitis. Vet Clin North Am Equine Pract 1999; 15: 375–394.
8 Walker M, Taylor TS, Slater M, et al. Radiographic appearance of the feet of mammoth donkeys and the findings of subclinical laminitis. Vet Radiol Ultrasound 1995; 35: 32–37.
9 Collins SN, Dyson SJ, Murray RC, et al. Radiological anatomy of the donkey foot: objective characterisation of the normal and laminitic donkey foot. Equine Vet J 2011; 43: 478–486.
10 Baxter GM. Equine laminitis caused by distal displacement of the distal phalanx: 12 cases (1976–1985). J Am Vet Med Assoc 1986; 189: 326–329.
11 Williams GE, Silverman BW, Wilson AM, et al. Disease-specific changes in equine ground reaction force data documented by use of principal component analysis. Am J Vet Res 1999; 60: 549–555.
12 Williams GE. Locomotor characteristics of horses with navicular disease. Am J Vet Res 2001; 62: 206–210.
13 Pfau T, Robilliard JJ, Weller R, et al. Assessment of mild hind limb lameness during over ground locomotion using linear discriminant analysis of inertial sensor data. Equine Vet J 2007; 39: 407–413.
14 Deene JS. Classification, postgraduate continuing professional training in industrial data modelling. Leicester, England: Industrial Data Modelling Unit, De Montfort University, 2000;180.
15 Scan chemometric analysis software user's guide. State College, Pa: Minitab Inc, 1998;287.
16 Koblik PD, O'Brien TR, Coyne CP. Effects of dorsopalmar obliquity on radiographic measurement of distal phalangeal rotation angle in horses with laminitis. J Am Vet Med Assoc 1988; 192: 346–349.
17 Timm NH. Applied multivariate analysis. New York: Springer-Verlag Inc, 2002;693.
18 Breiman LJ, Friedman CJ, Stone R, et al. Classification and regression trees. Boca Raton, Fla: Chapman and Hall/CRC, 1984;168.
19 Hastie T, Tibshirani R, Friedman J. The elements of statistical learning: data mining, inference, and prediction. 2nd ed. New York: Springer-Verlag Inc, 2009;739.
20 Multivariate analysis: principal components and calibration, postgraduate continuing professional training in data modelling. Leicester, England: Department of Biomedical Statistics, De Montfort University, 2000;364.
21 Webb AR. Statistical pattern recognition. 2nd ed. Oxford, England: John Wiley & Sons Ltd, 2002;534.
22 Crane M. Hoof disorders of the donkey, in Proceedings. 10th Geneva Cong Equine Med Surg 2007;153–156.
23 Kummer M, Geyer H, Imboden I, et al. The effect of foot trimming on radiographic measurements of the front feet of normal Warmblood horses. Vet J 2006; 172: 58–66.
24 Moleman M, van Heel M, van Weeren, et al. Hoof growth between 2 shoeings leads to a substantial increase of the moment about the distal, but not the proximal, interphalangeal joint. Equine Vet J 2006; 38: 170–174.
25 Reilly JD. The donkey's foot and its care. In: Svendsen ED, ed. The professional handbook of the donkey. 2nd ed. London: Whitter Books, 1997;71–93.
26 Eley JL. Keeping your donkey healthy. Edenbridge, Kent, England: Donkey Breed Society, 1998;156.
Appendix
Direct and derived radiographic variables and morphometric characteristics of the DP in the forefoot of donkeys assessed in a study to establish and validate an objective method of radiographic diagnosis of anatomic changes in laminitic feet of donkeys on the basis of data from a series of 19 radiographic measurements. (Adapted from Collins SN, Dyson SJ, Murray RC, et al. Radiological anatomy of the donkey foot: objective characterisation of the normal and laminitic donkey foot. Equine Vet J 2011;43:478–486. Reprinted with permission.)
Variable | Symbol | Anatomic definition | Method of determination | Selection criteria |
---|---|---|---|---|
Angular variables Dorsal hoof wall angle | S | Angle subtended between the dorsal aspect of the hoof wall and the ground line | Directly determined by guided mark-up protocol | Important factor in force distribution in the distal portion of the limb; required to determine angular deviation between the dorsal aspect of the DP and dorsum of the hoof wall and hoof pastern axis (the angular difference between the dorsal hoof wall angle and long axis of the proximal phalanx); changes in dorsal hoof wall angle are associated with laminitis |
Dorsal angle of the DP | Ts | Angle subtended between the dorsal aspect of the DP and the ground line | Directly determined by free mark-up ground line procedure as defined by 2 reference points: dorsal margin of the articular surface of the DP and dorsodistal limit of the DP palmar cortex | Required to determine angular deviation between the dorsal aspect of the DP and dorsum of the hoof wall and phalangeal rotation and hoof pastern axis; changes in dorsal angle of DP are associated with laminitis |
Angle of proximal phalanx | U | Angle subtended between the long axis of the proximal phalanx and the ground line | Directly determined by guided mark-up protocol | Required to establish pastern and hoof pastern axes and to determine phalangeal rotation |
Angle of middle phalanx | C | Angle subtended between the long axis of the middle phalanx and the ground line | Directly determined by guided mark-up protocol | Required to establish pastern axis and to determine DIP rotation events |
Angle of solar aspect of the DP | SA | Angle subtended between the solar aspect of the DP and the ground line | Directly determined by guided mark-up protocol | Additional indicator of DP rotation events; control against effect of measurement errors in variable Ts due to DP remodeling |
Angle of pastern axis | PAxis | Angular difference between the long axis of the proximal phalanx and the middle phalanx | Derived as U − C | Key descriptor of the stance characteristics of the distal portion of the limb |
Angular deviation between the dorsal aspect of the DP and dorsum of the hoof wall | Ang H | Angular difference between the dorsal aspect of the DP and the dorsal hoof wall angle | Derived as Ts − S | Primary rotational event associated with DP dislocation; important diagnostic and prognostic indicator for laminitis |
Phalangeal rotation angle | Ang R | Angular difference between the dorsal aspect of the DP and the long axis of the proximal phalanx | Derived as U − Ts | Primary rotational event associated with DP dislocation; argued to be the true measure of rotation within a laminitic foot |
DIP rotation angle | Ang F | Angular difference between the dorsal aspect of the DP and the long axis of the middle axis | Derived as C − Ts | Measure of DP rotation about the DIP; control against effect of measurement error in phalangeal rotation angle due to broken pastern axis |
Linearvariables Integument depth of the dorsal aspect of the foot (proximal site) | IDA | Perpendicular linear distance between the dorsal aspect of the hoof wall and the dorsal surface of the DP, immediately distad to the distal limit of the extensor process | Directly determined by free mark-up procedure as defined by the dorsal aspect of hoof wall and a selected DP reference site distad to the junction of the extensor process | Measures combined effect of angular deviation between the dorsal aspect of the DP and dorsum of the hoof wall and distal displacement of the DP; diagnostic indicator of laminitic wedge formation in conjunction with integument depth of the dorsal aspect of the foot (distal site) |
Integument depth of the dorsal aspect of the foot (distal site) | IDB | Perpendicular linear distance between the dorsal aspect of the hoof wall and the dorsal surface of the DP proximal to the apex of the DP | Directly determined by free mark-up procedure as defined by the dorsal aspect of hoof wall and a selected reference site proximad to the apical region of the DP | Measures combined effect of angular deviation between the dorsal aspect of the DP and dorsum of the hoof wall and distal displacement of the DP; diagnostic indicator of laminitic wedge formation in conjunction with integument depth of the dorsal aspect of the foot (proximal site) |
Integument depth of the dorsal aspect of the foot (mid-dorsal site) | IDM | Perpendicular linear distance between the dorsal aspect of the hoof wall and the dorsal surface of the DP at the midpoint between the proximal and distal integument depth of the dorsal aspect of the foot sites | Directly determined by free mark-up procedure by use of measurement site established along the dorsal aspect of the hoof wall at 50% of the linear distance between the proximal and distal integument depth of the dorsal aspect of the foot sites | Measures combined effect of angular deviation between the dorsal aspect of the DP and dorsum of the hoof wall and distal displacement of the DP; important diagnostic indicator of early laminitic change in equids |
Distal displacement of the DP | D | Perpendicular linear distance between the proximal limit of the hoof wall and the extensor process of the DP | Directly determined by guided mark-up protocol | Absolute measure of distal displacement of the DP; important diagnostic and prognostic indicator for the laminitic condition in equids |
Middle phalanx length | MPL | Linear measurement of the long axis of the middle phalanx | Directly determined by guided mark-up protocol | Control indicator of potential scaling effects |
Morphometric characteristics of the DP Proximal palmar cortex angle | PPCA | Angle subtended between the proximal palmar cortex of the DP and the ground line | Directly determined by free mark-up procedure as defined by 2 reference points: articular process with the navicular bone and point of insertion of DDFT | New descriptor for the characterization of the DP; additional indicator of DP rotation events |
Reflex angle of the palmar cortex | RA | Internal angle subtended between the proximal and distal palmar cortex of the DP | Directly determined by free mark-up procedure as defined by 3 reference points: articular process with the navicular bone, point of insertion of DDFT, and dorsodistal limit of the DP palmar cortex | New descriptor for the characterization of the DP; potential indicator for remodeling and resorption of the DP |
Apex angle | AA | Internal angle subtended between the distal palmar cortex and the dorsal aspect of the DP | Directly determined by free mark-up procedure as defined by 3 reference points: point of insertion of the DDFT, dorsodistal limit of the DP palmar cortex, and dorsal margin of the articular surface of the DP | New descriptor for the characterization of the DP; potential indicator for remodeling and resorption of the DP |
Palmar cortex length | PCL | Linear distance between the apex of the DP and the articular process of the navicular joint | Directly determined by free mark-up procedure as defined by 2 reference points: dorsodistal limit of the DP palmar cortex and articular process with the navicular bone | Indicator of resorption of the DP |
Proximal palmar cortex length | PPCL | Linear distance between the point of insertion of DDFT and the articular process with the navicular bone | Directly determined by free mark-up procedure as defined by 2 reference points: dorsodistal limit of the DP palmar cortex and articular process with the navicular bone | New descriptor for the characterization of the DP; control indicator of potential scaling effects |
DDFT = Deep digital flexor tendon. DIP = Distal interphalangeal joint.