Decision-tree analysis of clinical data to aid diagnostic reasoning for equine laminitis

The objective of this cross-sectional study was to compare the prevalence of selected clinical signs in 25 laminitis cases and non-laminitic but lame controls to evaluate their capability to discriminate laminitis from 26 other causes of lameness. Participating veterinary practitioners completed a checklist of laminitis-associated 27 clinical signs identified by literature review. Cases were defined as horses/ponies with veterinary-diagnosed, 28 clinically apparent laminitis; controls were horses/ponies with any lameness other than laminitis. 29 Associations were tested by logistic regression with adjusted odds ratios (OR) and 95% confidence intervals, 30 with veterinary practice as an a priori fixed effect. Multivariable analysis using graphical classification tree- 31 based statistical models linked laminitis prevalence with specific combinations of clinical signs. Data were 32 collected for 588 cases and 201 controls. Five clinical signs had a difference in prevalence of greater than 33 +50%: ‘reluctance to walk’ (OR 4.4, ‘short, stilted gait at walk’ (OR 9.4), ‘difficulty turning’ (OR 16.9), 34 ‘shifting weight’ (OR 17.7) and ‘increased digital pulse’ (OR 13.2) (all P <0.001). ‘Bilateral forelimb 35 lameness’ was the best discriminator; 92% of animals with this clinical sign had laminitis (OR 40.5, 36 P <0.001). If, in addition, horses/ponies had an ‘increased digital pulse’, 99% were identified as laminitis. 37 ‘Presence of a flat/convex sole’ also significantly enhanced clinical diagnosis discrimination (OR 15.5, 38 P <0.001). This is the first epidemiological laminitis study to use decision-tree analysis, providing the first 39 evidence-base for evaluating clinical signs to differentially diagnose laminitis from other causes of lameness. 40 Improved evaluation of the clinical signs displayed by laminitic animals examined by first-opinion 41 practitioners will lead to equine welfare improvements.


Introduction 46
Equine laminitis is a painful disease of the foot that affects equidae worldwide (Mellor and others 2001; 47 Wylie and others 2011). The insidious nature of the disease and potential for unrelenting pain often 48 necessitates euthanasia of the affected animal on welfare grounds (Hunt 1993; Menzies-Gow and others 49 2010b). Effective diagnosis is necessary to allow prompt instigation of palliative and therapeutic treatments, 50 to maximise recovery prospects. 51 In equine medicine, 'laminitis' is used to describe animals presenting with pain localised to the lamellar 52 debilitating consequences of laminitis do, however, require prompt veterinary intervention and accurate 77 diagnosis is therefore essential. 78 All the factors outlined above complicate the overall challenge of diagnostic reasoning based on clinical 79 signs, presenting the veterinary clinician with a challenge to diagnose laminitis differentially from other 80 forms of orthopaedic disorder. Therefore, the aim of this study was to compare the prevalence of selected 81 clinical signs in laminitis and non-laminitis lameness cases in order to evaluate the capabilities of clinical 82 signs to differentially diagnose laminitis from other causes of lameness. The study is presented considering

Materials and Methods 88
Data were collected from two groups: 89

Group A 90
A convenience sample of five veterinary institutions (two referral centres, two large first-opinion and referral 91 equine hospitals and a first-opinion mixed practice) were visited and invited to provide data for this study. In 92 addition, veterinary practices (n=93) that were interested in participating in a parallel epidemiological 93 investigation of equine laminitis, were contacted by telephone or email and invited to provide data on clinical 94 signs of lameness (of any origin) for the study reported here. 95 A literature review was conducted to identify previously suggested clinical signs of laminitis and differential 96 diagnoses. The resultant list was reviewed by expert equine clinicians in selected referral hospitals and 97 laminitis researchers, and a 'lameness reporting form' (LM) (Supplementary Information Item 1) was 98 designed to gather information on laminitis-relevant clinical signs from both laminitic (cases) and non-99 laminitic lame (controls) horses. 100 Part one of the LM gathered case identifying information with five subsequent sections recording whether 101 clinical signs pertaining to the foot, stance and lameness irregularities (clinical signs) were present, absent or 102 had not been assessed. Part two of the LM allowed practitioners to record their diagnosis as free text and to 103 select specific diagnostic techniques used to confirm the diagnosis from six tick-box options. A free-text 104 comments section was also included for any additional information pertinent to confirmation of the diagnosis. 105 Participating practitioners were asked to complete a LM for equine lameness of any origin seen between 106 February-April 2009, and January 2010-May 2011, with the second phase of data collection initiated to 107 increase numbers for analysis. Completed forms were returned by post using supplied reply-paid envelopes. 108 Upon arrival LMs were divided into two groups for analysis: one group containing reported laminitis cases 109 and another containing all animals for which the primary cause of lameness was not laminitis (controls). 110

Group B 111
Following this development phase, a 'laminitis reporting form' (LRF) was finalised (Supplementary 112 Information Item 2) as previously described (Wylie and others 2013a). As for the LM, the LRF consisted of 113 five distinct sections on lameness, stance characteristics, feet affected and observed laminitis-related acute 114 and chronic clinical signs. Based on the data collected from animals in Group A, some modifications to the 115 form were made, hence for the purposes of this study only those clinical signs which were reported for both 116 groups were compared. No further clinical data were recorded for the purposes of this study. 117 118 A LRF was completed for any case of laminitis, defined as a horse or pony with veterinary-diagnosed, 119 clinically apparent laminitis (i.e. an active episode of laminitis), attended by one of the participating 120 practitioners (Wylie and others 2013a). In animals with recurring laminitis, an episode of veterinary-diagnosed active laminitis was defined as new if the animal had returned to its previous/normal level of 122 soundness and had not received analgesic medication for 14 days or more between episodes (Wylie and 123 others 2013a). However, for the purposes of this study only the first episode of laminitis was included. 124 Practices were asked to complete the LRF for all eligible cases occurring from May 2009 to April 2011. 125

Statistical analysis 126
To increase the numbers for data analysis, Groups A and B were combined. Multiple different clinical signs 127 were categorised (present, not present or not assessed) under the following five sections: 128 (1) Lameness: recumbency, refusal to move unless forced, reluctance to walk, lame at walk, lame at trot, 129 short stilted gait at walk, short stilted gait at trot, difficulty turning 130 (2) Stance: shifting weight, front feet placed in front of body, reluctance to lift foot 131 (3) Feet affected: bilateral front feet, bilateral hind feet or all four feet 132 (4) Acute clinical signs: increased digital pulse, increased hoof temperature, pain on sole pressure 133 clinical sign and case or control status were tested using logistic regression models reporting adjusted odds 141 ratios (OR) taking into account veterinary practice as a fixed effect, with 95% confidence intervals (CI), and 142 Wald test P-values. All analyses were conducted in R Statistical Package (version 3.1.2 © 2014 The R 143 Foundation for Statistical Computing) using the 'epicalc' and 'tree' packages. Statistical significance was set 144 at a value of P<0.05. 145 Multivariable analysis was carried out using a multi-factorial classification -tree-based statistical models 146 (hereafter 'tree models') (Clark and Pregibon 1997). This analytical technique was chosen due to the 147 unbalanced dataset with potentially different combinations of factors present in different horses. The analysis 148 consisted of determining a binary division of the clinical signs prevalence data (laminitis vs. non-laminitis 149 lameness), such that there is the largest difference in terms of prevalence of laminitis vs. non-laminitis 150 lameness for those two subsets of data. One subset of animals with a specific clinical sign is first considered 151 (e.g. those with 'bilateral forelimb lameness') and the binary division in terms of any of the other clinical 152 signs resulting in the largest difference in prevalence of laminitis is determined. The other subset is then 153 considered (e.g. those with no 'bilateral forelimb lameness') and again the clinical signs for which binary 154 division gives the largest difference in prevalence of laminitis vs. non-laminitis lameness is determined. The clinical signs and case/controls status described above. 162 Five separate preliminary tree models were produced for the following characteristics to represent the 163 features of clinically active laminitis recorded: i) lameness, ii) stance, iii) feet affected, iv) acute signs only 164 and iv) acute and chronic signs. 'Lame at trot' and 'short stilted gait at trot' were excluded from the lameness 165 tree model due to large numbers of missing data where these signs had not been assessed (missing for 55.0% 166 and 49.4% of observations, respectively). 167 After consideration of the five preliminary trees, those variables identified in each preliminary tree as being 168 the greatest differentiators in terms of laminitis were analysed together to form two combined tree models: (i) 169 a combined model of lameness, stance characteristics, feet affected and observed laminitis-related acute 170 clinical signs to reflect active episodes of laminitis in horses with no evidence of chronic laminitis, and (ii) a chronic clinical signs to reflect active episodes of laminitis in horses with evidence of previous SADP failure 173 regional anaesthesia (nerve blocks) (13.5%: CI 2.5, 24.5), surgical/post-mortem findings (13.5%: CI 2.5, 190 24.5) and blood testing for concurrent predisposing metabolic conditions (8.1%: CI 0.01, 16.9). 191

Group B 192
The recruitment of cases is described in detail in Wylie et al. (2013a). In brief, LRFs were received for 551 193 unique horses/ponies from 30 first-opinion veterinary practices over the two-year period. 194

Clinical signs 195
The prevalence of the presence of each clinical sign in laminitis cases and non-laminitis lame controls, 196 excluding records where the sign was not assessed, and difference in prevalence between the two groups are 197 provided in Table 1. The overall prevalence of specific clinical signs ranged from 2.7% (CI 1.5, 3.9) for 'sole 198 prolapse' (number assessed = 706) to 85.0% (CI 81.4, 88.7) for 'lame at trot' (number assessed = 367). The 199 difference in prevalence between cases and controls ranged from -14.1% for 'lame at trot' (sign more 200 common in controls) to +71.9% for 'short stilted gait at walk' (found more often in cases than controls). 201 There were five clinical signs with a difference in prevalence of greater than +50%: three lameness-related 202 signs ('reluctance to walk', 'short, stilted gait at walk' and 'difficulty turning'), one stance-related sign 203 ('shifting weight') and one acute clinical sign ('increased digital pulse'). 204 The logistic regression results are provided in Table 2 The best discriminator in the stance tree was 'shifting weight'; 98.1% (CI 96.6, 99.6) of animals with that 217 clinical sign had laminitis. In animals that were not 'shifting weight', 'front feet placed in front of the body' 218 identified 94.2% (CI 89.2, 99.1) as laminitis cases. 219 In the 'acute clinical signs' tree, 91.0% (CI 88.5, 93.5) of animals with 'increased digital pulses' had 220 laminitis, and 'pain on sole pressure' in the absence of 'increased digital pulses' identified 69.0% (CI 52.1, 221

85.8) as cases of laminitis. 222
The best discriminator in the 'acute and chronic clinical signs' tree was 'increased digital pulses'; 91.0% (CI 223 88.4, 93.5) of animals with that clinical sign had laminitis, and the additional presence of 'divergent growth 224 rings' identified 100% as laminitis cases. 225 The tree diagram combining categories of clinical signs for acute laminitis with lameness, stance and feet is 226 provided in Figure 1. Presence of 'lameness in both forelimbs' was the best discriminator, with 93.1% (CI 227 90.7, 95.5) of animals with this clinical sign belonging to the laminitis group. Additional presence of an 228 'increased digital pulse' improved diagnostic accuracy to 99% (CI 97.9, 100) (P<0.001). A 'bilateral 229 forelimb lameness' with no 'increase in digital pulse', yet presence of a 'short stilted gait at walk' identified 230 100% of animals as laminitis cases, however statistical analysis of this sub-group and the presence of 231 'shifting weight' was not possible due to small numbers of animals with these signs. The presence of 'pain 232 on sole pressure' was not statistically associated with improved clinical discrimination (P=0.30). 233 The overall tree diagram considering both acute and chronic laminitis clinical signs with lameness, stance and 234 feet is provided in Figure 2. Presence of 'lameness in both forelimbs' was again the best discriminator; 92% 235 of animals with this clinical sign had laminitis (P<0.001). The additional presence of 'increased digital 236 pulses' improved this to 99% of cases (P<0.001). Presence of a 'flat/convex sole' also provided improved 237 clinical discrimination (P=0.002). It was not possible to assess statistical significance for 'short stilted gait at 238 walk', or 'shifting weight', again because of the small numbers of animals with these signs. This is the first study comparing the prevalence of veterinary-recognised clinical signs in laminitis and other 242 causes of lameness to evaluate the capabilities of discrimination for differential diagnosis. 243 A wide range of clinical signs were displayed by the laminitic cases, in agreement with previous reviews 244 (Baxter 1994; Eustace 2010; Hunt and Wharton 2010; Swanson 1999). There were no individual, or 245 combinations of, clinical signs present in every case. The clinical signs that were considered to be the most 246 useful on the basis of this work were three features of lameness investigation ('reluctance to walk', 'short, 247 stilted gait at walk' and 'difficulty turning'), one feature of stance ('shifting weight') and an 'increased digital 248 pulse'. All these signs had a difference in prevalence of over 50% between active laminitis cases (signs more 249 prevalent) and non-laminitic lame horses (signs less prevalent). As the clinical details forms were designed to 250 gather information on laminitis, it may be expected there was a statistically significant difference in the 251 distribution of many of the clinical signs between laminitis cases and non-laminitis lameness controls. For  Two clinical signs -'coronary band depression' and 'prolapsed sole' -were pathognomonic for laminitis in 281 this study, . were only found in 13.6% and 3.7% of cases, respectively. Both these signs can indicate disease 282 progression to chronic phase laminitis (i.e. SADP failure and distal phalanx dislocation within the hoof); 283 therefore these signs would not be expected to be present in acute cases, unless they were also suffering from 284 concurrent pathology such as chronic seedy toe/white line disease or severe club feet (Kuwano and others 285 1999). These results may help veterinary practitioners prioritise where to begin their clinical examination of 286 an active laminitis case, as primary inspection of the sole and coronary band would prevent the animal 287 undergoing lameness evaluation which could precipitate further SADP damage/failure. 288 Two overall combined trees were generated to reflect the two clinical scenarios of active laminitis, one 289 consisting of clinical signs considered to occur in the acute phase of the disease, and one that also contained 290 data reflective of lamellar damage and displacement of the SADP. In both scenarios, the presence of a 291 bilateral lameness was the most useful discriminator, followed by the presence of increased digital pulses. 292 Whilst these clinical presentations are not specific for laminitis, this work provides an evidence-base for case 293 diagnosis and future epidemiological case definitions. 294 This work did not provide evidence for some commonly cited clinical signs of diagnostic importance. In 295 particular, 'front feet in front of the body', taken to represent the classic 'laminitis stance', was found in less 296 than half of the diagnosed active laminitis cases, and did not prove to be a useful discriminator. Therefore, despite much anecdotal publicity of this visibly apparent clinical sign (Stashak 2002; Swanson 1999), 298 veterinarians, researchers and owners should be careful to avoid relying on its presence for making a 299 diagnosis of laminitis [40]. 300 The use of clinical recording forms based on evidence-based recommendations may help veterinary 301 practitioners structure their clinical examination of an active laminitis case. However, in medical practice 302 well-validated diagnostic algorithms tools are underused (Pearson and others 1994). For example, a simple 303 predictor based on seven clinical signs for ischaemia in humans was only used in 2.8% of cases (Corey and 304 Merenstein 1987). The clinical usefulness of developing such a technique would need to be established by a 305 survey of first-opinion practitioners to decide whether such a tool would provide useful assistance in laminitis 306 diagnosis in the field. 307 The limitations of this study include diagnosis by a number of different veterinary clinicians, which may have 308 different levels of experience. To take this into account veterinary practice was included in the generation of 309 the odds ratio estimates, however, misclassification bias may still occur, although this would have tended to 310 shift the odds ratios towards non-significant. Similarly, as it is not possible to obtain a definitive diagnosis of 311 active laminitis in an observational epidemiological study there was the potential for misclassification of 312 cases and controls. For this reason, veterinary recordings of the clinical signs observed was used, as 313 described in Wylie et al., (Wylie and others 2013a, b) and misclassification would have again reduced the 314 ability to detect significant differences rather than produce anomalous significant differences. Inclusion of 315 data in the tree models required the animals to have data for each included variable, resulting in smaller 316 numbers of contributing individuals as the trees became more complex. Consequently, although the variables 317 retained high statistical significance, smaller contributing sample sizes led to larger confidence intervals 318 around prevalence point estimates and the need therefore for some caution in their interpretation. 319 It is acknowledged that there may be some bias in the data if veterinary practitioners did not accurately detail 320 the clinical signs which they observed and perhaps listed clinical signs that they anticipated to reflect their 321 diagnosis. Furthermore, it would be interesting to collect greater numbers of control animals to conduct the 322 analyses between specific control lamenesses, such as forelimb foot pain only, to highlight more subtle 323 In conclusion, separate clinical signs were compared between laminitis and non-laminitis cases of lameness,