DETERMINING THE RELATIONSHIP BETWEEN BAROMETRIC PRESSURE CHANGES AND THE INCIDENCE OF EQUINE COLIC. Justine Cianci

Similar documents
THAL EQUINE LLC Regional Equine Hospital Horse Owner Education & Resources Santa Fe, New Mexico

Rice Yield And Dangue Haemorrhagic Fever(DHF) Condition depend upon Climate Data

South Shore Equine Clinic and Diagnostic Center

Bird strikes Swedish Airspace

SEASONAL PRICES for TENNESSEE FEEDER CATTLE and COWS

Wind Resource Assessment for NOME (ANVIL MOUNTAIN), ALASKA Date last modified: 5/22/06 Compiled by: Cliff Dolchok

Equine colic: What to Expect

Legendre et al Appendices and Supplements, p. 1

Journal of Human Sport and Exercise E-ISSN: Universidad de Alicante España

Wind Resource Assessment for CHEFORNAK, ALASKA

Special Acknowledgement: J. Richard Trout, Ph.D. (biostatistics)

Standardized catch rates of U.S. blueline tilefish (Caulolatilus microps) from commercial logbook longline data

REACT REDUCING THE RISK OF COLIC. For more information

COMPARISON OF FIXED & VARIABLE RATES (25 YEARS) CHARTERED BANK ADMINISTERED INTEREST RATES - PRIME BUSINESS*

Running head: DATA ANALYSIS AND INTERPRETATION 1

Chapter 12 Practice Test

Competitive Performance of Elite Olympic-Distance Triathletes: Reliability and Smallest Worthwhile Enhancement

CHAP Summary 8 TER 155

y ) s x x )(y i (x i r = 1 n 1 s y Statistics Lecture 7 Exploring Data , y 2 ,y n (x 1 ),,(x n ),(x 2 ,y 1 How two variables vary together

Assessment of Guide Reporting & Preliminary Results of Lion Monitoring

THAL EQUINE LLC Regional Equine Hospital Horse Owner Education & Resources Santa Fe, New Mexico

THEORY OF TRAINING, THEORETICAL CONSIDERATIONS WOMEN S RACE WALKING

Colic is considered one of the most important equine

Colic Fact Sheet One hell of a belly ache

Statistical Analysis of PGA Tour Skill Rankings USGA Research and Test Center June 1, 2007

INFLUENCE OF ENVIRONMENTAL PARAMETERS ON FISHERY

Adaptation to climate variation in a diversified fishery:

NBA TEAM SYNERGY RESEARCH REPORT 1

Risk Factors Associated with Impaction Colic in Horses at North Western Area of Libya

Neighborhood Influences on Use of Urban Trails

Properties. terc.ucdavis.edu 8

MALL CROSSING STUDY. Testing the Effectiveness Of the 4th Street East Crossing. For: City of Charlottesville Neighborhood Development Services

Prevalence, Demographics, and Risk Factors for Colic

EXPLORING MOTIVATION AND TOURIST TYPOLOGY: THE CASE OF KOREAN GOLF TOURISTS TRAVELLING IN THE ASIA PACIFIC. Jae Hak Kim

IMPROVING POPULATION MANAGEMENT AND HARVEST QUOTAS OF MOOSE IN RUSSIA

Hair Shedding Scores Relating to Maternal Traits and Productivity in Beef Cattle. An Undergraduate Honors Thesis in the. Animal Science Department

Lab Report Outline the Bones of the Story

The probability of winning a high school football game.

A SURVEY OF 1997 COLORADO ANGLERS AND THEIR WILLINGNESS TO PAY INCREASED LICENSE FEES

Stats 2002: Probabilities for Wins and Losses of Online Gambling

Foal and Mare Behavior Changes during Repeated Human-Animal Interactions in the First Two Weeks after Foaling

Analysis of Highland Lakes Inflows Using Process Behavior Charts Dr. William McNeese, Ph.D. Revised: Sept. 4,

Draft Kivalina Wind Resource Report

GAZIFÈRE INC. Prime Rate Forecasting Process 2015 Rate Case

Signs are difficult to spot but they can include poor appetite, impaired performance, poor body condition, change in temperament and colic.

Osceola County 4-H Record Book Horse Senior (15-19)

Influence of Feeding Practices on Behavior and Activity Levels of Quarter Horse Mares

Site Description: Tower Site

Influence of Forecasting Factors and Methods or Bullwhip Effect and Order Rate Variance Ratio in the Two Stage Supply Chain-A Case Study

Critical Gust Pressures on Tall Building Frames-Review of Codal Provisions

GALLUP NEWS SERVICE GALLUP POLL SOCIAL SERIES: WORLD AFFAIRS

ESTIMATION OF THE DESIGN WIND SPEED BASED ON

Navigate to the golf data folder and make it your working directory. Load the data by typing

TOP 10 Gifts Your Horse Wants for Christmas

Forecasting and Visualisation. Time series in R

The Effect of a Seven Week Exercise Program on Golf Swing Performance and Musculoskeletal Screening Scores

GALLUP NEWS SERVICE GALLUP POLL SOCIAL SERIES: WORLD AFFAIRS

Palythoa Abundance and Coverage in Relation to Depth

Analysis of Variance. Copyright 2014 Pearson Education, Inc.

GALLUP NEWS SERVICE GALLUP POLL SOCIAL SERIES: WORK AND EDUCATION

IDENTIFYING SUBJECTIVE VALUE IN WOMEN S COLLEGE GOLF RECRUITING REGARDLESS OF SOCIO-ECONOMIC CLASS. Victoria Allred

REPORT ON RED-LIGHT MONITORING SYSTEMS

A Case Study of Leadership in Women s Intercollegiate Softball. By: DIANE L. GILL and JEAN L. PERRY

Warmer temperatures, molt timing and lobster seasons in the Canadian Maritimes

GLMM standardisation of the commercial abalone CPUE for Zones A-D over the period

ISSN (online) ISBN (online) July New Zealand Fisheries Assessment Report 2017/41. P.L. Horn C.P.

SCDNR Charterboat Logbook Program Data, Mike Errigo, Eric Hiltz, and Amy Dukes SEDAR32-DW-08

Lower Coquitlam River Project Water Use Plan. Temperature Monitoring Lower Coquitlam River Project Year 2 Report

Evaluating the Influence of R3 Treatments on Fishing License Sales in Pennsylvania


Status Report on the Yellowstone Bison Population, August 2016 Chris Geremia 1, Rick Wallen, and P.J. White August 17, 2016

The Australian and New Zealand College of Veterinary Scientists. Membership Examination. Medicine of Horses Paper 1

8th Grade. Data.

Section I: Multiple Choice Select the best answer for each problem.

A Retrospective Case Study Implicating Foster Calves in a Calf Diarrhea Epidemic

Name May 3, 2007 Math Probability and Statistics

Equine Injury Database Update and Call for More Data

Prognosis and Strategies to Prevent Colic

5.1 Introduction. Learning Objectives

Commodity Market Outlook: Corn, Forage, Wheat & Cattle

Revisiting the Hot Hand Theory with Free Throw Data in a Multivariate Framework

Site Description: LOCATION DETAILS Report Prepared By: Tower Site Report Date

Math 121 Test Questions Spring 2010 Chapters 13 and 14

This file is part of the following reference:

Student Population Projections By Residence. School Year 2016/2017 Report Projections 2017/ /27. Prepared by:

Chinook salmon (photo by Roger Tabor)

ROMANIAN SPORT HORSES: EFFECTS OF COMPETITION LEVEL, SEX AND BREEDER ON THE NATIONAL DRESSAGE RANKING

NMSU Red Light Camera Study Update

Fishery Resource Grant Program Final Report 2010

Announcements. % College graduate vs. % Hispanic in LA. % College educated vs. % Hispanic in LA. Problem Set 10 Due Wednesday.

Clinical Study Synopsis

Kodiak, Alaska Site 1 Wind Resource Report for Kodiak Electric Association

Veterinary Medical Diagnostic Program

100-Meter Dash Olympic Winning Times: Will Women Be As Fast As Men?

2015 Winnebago System Walleye Report

EVects of seasonal change in rugby league on the incidence of injury

Evaluating the Design Safety of Highway Structural Supports

POWER Quantifying Correction Curve Uncertainty Through Empirical Methods

Hypothermia, the Diving Reflex, and Survival. Briana Martin. Biology 281 Professor McMillan April 17, XXXX

SCIENTIFIC COMMITTEE SEVENTH REGULAR SESSION August 2011 Pohnpei, Federated States of Micronesia

Transcription:

DETERMINING THE RELATIONSHIP BETWEEN BAROMETRIC PRESSURE CHANGES AND THE INCIDENCE OF EQUINE COLIC by Justine Cianci A thesis submitted to the Faculty of the University of Delaware in partial fulfillment of the requirements for the Degree of Bachelor of Science in Pre- Veterinary Medicine and Animal Biosciences with Distinction Spring 2018 2018 Justine Cianci All Rights Reserved

DETERMINING THE RELATIONSHIP BETWEEN BAROMETRIC PRESSURE CHANGES AND THE INCIDENCE OF EQUINE COLIC by Justine Cianci Approved: Amy Biddle, Ph.D. Professor in charge of thesis on behalf of the Advisory Committee Approved: Annie Renzetti, VMD Committee member from the Department of Animal and Food Sciences Approved: Mark Parcells, Ph.D. Committee member from the Board of Senior Thesis Readers Approved: Hemant Kher, Ph.D. Chair of the University Committee on Student and Faculty Honors

ACKNOWLEDGMENTS This work is supported by the University of Pennsylvania s New Bolton Center for Large Animals, The University of Delaware Department of Animal and Food Sciences, Undergraduate Program in Pre-Veterinary Science, Equine Science and Animal Bio-sciences, Undergraduate Research and Experimental Learning and the Biddle lab. iii

TABLE OF CONTENTS LIST OF TABLES... v LIST OF FIGURES... vi ABSTRACT... viii 1 INTRODUCTION... 1 1.1 Equine Colic... 1 1.2 Anecdotal Relationship between Equine Colic and Barometric Pressure Change... 2 1.3 Responses of Other Species to Barometric Pressure Changes... 3 1.4 Statistical Methodology... 4 2 MATERIALS AND METHODS... 5 2.1 Collection and Organization of Medical Records... 5 2.2 Collection of Weather Data... 5 2.3 Statistical Analysis... 6 2.3.1 Normality... 6 2.3.2 Spearman Correlation... 6 2.3.3 Seasonality... 6 2.3.4 Survival Analysis... 7 2.3.5 Logistic Regression... 7 3 RESULTS AND DISCUSSION... 8 4 CONCLUSION... 21 REFERENCES... 24 A RAW STATISTICAL OUTPUT... 26 iv

LIST OF TABLES Table 1: Summary of Equine Data Collected from the New Bolton Center s Field Service Examinations from 1/1/05-1/1/17... 8 Table 2: Backward stepwise logistic regression results display the best predictors of colic being age, the atmospheric pressure low 24 hours prior to the event and the horse s latitude... 20 Table 3: Summary Table of Normality of Data from the New Bolton Center s Field Service Data and the Weather Center from 1/1/05-1/1/17... 26 Table 4: Summary Table of Spearman Rank Correlation Output... 26 Table 5: Summary Table of Spearman Rank Correlation Output (continued)... 27 Table 6: Summary Table of Calculations of the Average Number of Exams Seen per Month from 1/1/05-1/1/17... 27 Table 6: Summary of Moran s I Statistical Cluster Analysis Determining Autocorrelation of Barometric Pressure Measurements... 28 Table 7: Univariate Logistic Regression Output from R for All Variables... 28 Table 7: Univariate Logistic Regression Output from R for All Variables (continued)... 29 Table 8: Backward Stepwise Regression Output from R to Fit the Best Predictive Model... 29 v

LIST OF FIGURES Figure 1: Breed Distribution for the Study Population. Compared to the overall equine population in the area, the New Bolton Center sees a disproportionate number of Thoroughbred horses.... 9 Figure 2: Map of all unique zip codes used in the study, marking the approximate locations of the horses with black points and the locations of the National Weather Centers in Philadelphia (PA), Coatesville (PA) and Wilmington (DE) with yellow stars.... 10 Figure 3: Density of the animals used in the study. Red areas show the most dense collection of patients and green shows the least dense area, while the study population is taken from the approximate area within the blue outline.... 11 Figure 4: The number of monthly diagnoses made by the New Bolton Center for both colic and control exams from 1/1/05 through 1/1/17, including variance. There is no statistical difference for either variable throughout the year, signifying that there is no overall seasonality.... 13 Figure 5: Stratified Boxplot depicting the average daily atmospheric sea level pressure, measured in inches mercury from 1/1/05 through 1/1/17. There is statistically no difference throughout the year, corresponding to the lack of seasonality in diagnoses.... 14 Figure 6: Stratified Boxplot depicting the average of daily drop in atmospheric pressure at sea level, measured in inches mercury, from 1/1/05 through 1/1/17. There is statistically no difference throughout the year, corresponding to the lack of seasonality in diagnoses.... 14 Figure 7: Survival of colic diagnoses compared to all control diagnoses when looking at the drop in atmospheric sea level pressure, measured in inches mercury, 12 hours before the event. There is no significant difference seen.... 15 Figure 8: Survival of colic diagnoses compared to all control diagnoses when looking at age. There is no significant difference seen in this study s population.... 16 Figure 9: Original Model-probability of horses with and without colic; the identical distribution of the two groups leaves the model unable to distinguish between control and experimental diagnoses.... 17 vi

Figure 10: Best Fit Model-probability of horses with and without colic; the identical distribution of the two groups leaves the model unable to distinguish between control and experimental diagnoses.... 18 Figure 11: Receiver Operating Characteristic Curve determining the discrimination power of the logistic model. The area under the curve (0.6), represents minimal distinguishability between control and experimental horses.... 19 vii

ABSTRACT Colic is a universal term for abdominal pain and is the top equine emergency seen today with a fatality of almost 7%. The cause of colic could be anything from diet to disease or stress, but nothing in horses has been studied and documented. Other mammals, like cows and rats, have been shown to have an increased response to pain and have onset of parturition within twelve hours when a low pressure system surrounds the animal. Sets of data from the New Bolton Center s Field Service database along with data from the National Weather Service from 2005 through 2016 were combined to statistically look for connections between the two datasets to support anecdotal hypotheses found throughout the equine industry. Results of the study showed that (when age and breed type are normalized) there is a positive relationship (odds ratio of 0.39) between the barometric pressure change within 12 hours of the incidence of colic. For every unit of decrease in barometric pressure, the odds of colic increase by about 2.5 times. Pressure difference within 12 hours of the event shows significance to the incidence of colic (P value of 0.01) and indicates that there is a difference between incidence of colic and other diagnoses when considering barometric pressure differences. The horse s age and geographical location are also significantly associated with the incidence of colic. This study provides evidence that changes in barometric pressure could be a contributing factor for colic, enabling horse owners and veterinarians to intervene earlier for colic-prone, senior or stressed horses. viii

Chapter 1 INTRODUCTION 1.1 Equine Colic Colic is a universal term for abdominal pain; this can be sourced from anything in the abdomen including stress, trauma, parasites, or problems with a bodily system like the urinary system and the gastrointestinal system (Marcella 2018). In horses, colic can range from mild discomfort to a deadly sign of a disease process as the third most common cause of death in horses after old-age and injury (Egenvall et. al. 2008). Typical symptoms of colic include mild indicators like going off feed and not defecating to more severe signs like kicking or biting at the abdomen and excessive rolling or thrashing. Lesions associated with colic are placed in categories that include obstruction, strangulation, enteritis, peritonitis and ulceration (Tinker et. al. 1997). However, most specific colic lesions are unknown because diagnosis without necropsy or surgery is very difficult. As many as 30% of colic cases are never seen by the veterinarian because of transience, meanwhile more than 16% of colic incidences require surgery (Marcella 2018). The extreme variability in the severity of colic is a concern for veterinarians and owners. Surgery for any horse carries physical risk, but the substantial financial expense is guaranteed. The surgery and aftercare could cost over ten-thousand dollars and does not guarantee a permanent solution 11% of surgical cases are euthanized or pass away anyway (Marcella 2018). Colic is the top equine emergency seen today with a fatality of almost 7% (Marcella 2018). In order to keep horses as healthy as possible, knowing that nothing can be prevented completely, 1

they should be kept on strict daily routines, fed a high quality diet, provided plenty of exercise (and turnout), have changes made slowly and have stress reduced during competition, transport and other events (Marcella 2018). With this information, knowing the response of many species to barometric pressure, it is important to note that these pressure variations may change a horse s behavior and exaggerate a preexisting health concern. Horses are prey animals and will not reveal pain until it is significant, so keeping their daily routine as normal as possible and keeping an eye on higher-risk horse is vital. 1.2 Anecdotal Relationship between Equine Colic and Barometric Pressure Change The root of colic could be anything from excessive sugar in spring grass to pendulous lipomas developed over time or stress put on the animal from transportation or managerial changes, but nothing has been thoroughly studied and documented. All of the anecdotal evidence comes from field impressions of unseasonal temperatures and storm fronts passing through (Marcella 2018). This has led many veterinarians to believe that managerial changes cause stress to the horses, which may spur colic. The University of Liverpool in the United Kingdom in 2006 studied the seasonality of colic. The study concluded with data that supports seasonal trends with certain types of colic that appear to coincide with either times of managerial change or periods where horses are more likely to be intensively managed (Archer et. al. 2006) but more studies are needed to identify the influences of the seasonality. The debate on the barometric pressure link to equine colic is briefly mentioned in many publications, including veterinarian Bradford Bentz s Understanding Equine Colic, where he writes that clinical experience and some epidemiologic evidence suggest an 2

association. A study published in 2012 from the Szent Istvan University in Hungary found that pressure changes did influence the incidence of colic, but the study was limited in its size (1089 horses over two years) (Eisenreich 2012). This supported the idea that there is a connection that has yet to be studied comprehensively. 1.3 Responses of Other Species to Barometric Pressure Changes The changing behavior of other animal species, including mammals, fuels the hypothesis that horses may respond in a similar way. A study was published in 2013 looking at the modified sexual behavior of various insects in response to changing atmospheric pressure. Dropping the atmospheric pressure (to a natural level seen in the environment) found that locomotion, calling behavior and courtship behavior all decreased, but with significant interspecific differences in which behaviors were altered (Pellegrino et. al. 2013). Another study published in 2013 determined that the White-Throated Sparrow also adjusts its behavior when barometric pressure changes. When pressure dropped (to mimic winter conditions during migration), the birds significantly increased their movement and feeding behavior (Metcalfe et. al. 2013). Although outdated, a study finding that large drops in atmospheric pressure stimulate parturition in cattle due to an increase in physical activity during rising pressure systems is also important (Dvorak 1978). Dvorak poses the idea that, similar to humans, the low pressure elicits corticoid secretion from the dam s adrenal cortex, which will stimulate parturition. In 2001 a study was done testing the pain tolerance of rats. It noted that when the barometric pressure dropped, the rats have a greater response to spontaneous pain (Sato et. al. 2001). These studies suggest that both behavior and any physiological problems may be exaggerated and change when atmospheric pressure drops. 3

1.4 Statistical Methodology The statistical analysis of the data is looking for correlations and associations between horse characteristics and barometric pressure measurements. The monotonicity of the data and the binary outcome of the diagnosis (colic or control) limit the statistical testing because many tests are either inappropriate or unable to run with this type of data. The analysis was predominately logistic regression in order to model the best predictors of the data and their odds of influencing colic. A receiver operating characteristic curve and pseudo r squared value present the quality of the model and its ability to distinguish between experimental and control diagnoses. Survival analysis curves were created to visualize the differentiation of survival probability between colic cases and horses seen for other diagnoses. Due to the difference in topography across the New Bolton Center s field servicing area, spatial variations are important considerations. Generalized linear spatial models take variation in location into account and is able to make better predictions on the influence of colic. However, due to the complexity of the model and the variability and format of the data structure, generalized linear spatial models were not created for the study. 4

Chapter 2 MATERIALS AND METHODS 2.1 Collection and Organization of Medical Records The horses used in this study were provided by the New Bolton Center s Field Service database. Query searches collected the medical records needed. Each record was read for the appropriate years and all information designated by the study was recorded. This information included exam, date, patient location and the demographics of age, breed and sex. For exams to be considered separate incidences, they must be clearly defined in the medical record as distinct occurrences with no association between the two. Wounds and lacerations associated with one another were counted as the same incidence. Control exams were picked on the basis of having no relation to colic or typically showing colic-type symptoms. One limitation to the data is that all specified diagnoses were not recorded due to being outside the time frame of the study or being treated by another veterinary service outside of the New Bolton Center. For a dataset that fully represents the area s equine population, more hospitals need to be surveyed. 2.2 Collection of Weather Data Weather information was provided freely from the National Weather Service s online historical database. Information from three stations was collected to cover the New Bolton Center s field servicing area. These stations were located in Philadelphia (PA), Coatesville (PA) and Wilmington (DE). The age of the station and sporadic downing of the systems leaves some dates without data. This missing data was not available with any other service. The missing data limited the study sample marginally, but it was not extrapolated because of the addition of significant bias. 5

2.3 Statistical Analysis 2.3.1 Normality The normality of the data was found using the Shapiro-Wilk test. This test allowed the testing of the data to see if it comes from a normal distribution and is assumed to be random. Normal data will represent a random and representative sample of horses from the New Bolton Center s field service veterinary cases. 2.3.2 Spearman Correlation Direct correlations between weather and equine variables was looked at using Spearman Ranked Correlation. This test is for monotonic relationships between variables those without linearity. Coefficient values, rho, approaching 1 or -1 represent a very strong correlation and those approaching 0 signify a very weak relationship. P-values are also created with significance at 0.05 or below. The statistical test can produce a low rho value with high significance, in which the result is significant, but the relationship is very weak. 2.3.3 Seasonality Yearly seasonality of colic, control diagnoses or barometric pressure was calculated by using averages and standard deviations per month. Exam average and deviation is shown on a line graph in order to show both experimental and control diagnoses on the same plot for direct comparison. Barometric pressure average and deviation was shown using stratified boxplots. Each barometric pressure measurement type shown to be significant in other studies looking at animal response to pressure is shown in a separate plot. 6

2.3.4 Survival Analysis Survival analysis curves using Kaplan-Meier Analysis were created for all exams by age and pressure difference within 12 hours of the exam. This will allow the distinction of death rate between colic cases and horses seen for other diagnoses. 2.3.5 Logistic Regression This test allows the prediction of the odds of colic occurrence by looking at the relationship between a colic event and the barometric pressure pattern for dichotomous (both binary and non-binary) or continuous and mutually exclusive data. For determining the independent predictors used in the final model, univariate analysis was performed and those with P values of less than 0.2 were chosen for multivariable logistic regression. This regression used a backward stepwise approach to model the best predictors of colic. Significance here was found for predictors with P-values less than 0.05 and confidence intervals not containing 1. Intervals containing 1 signify that unit changes in predictors have no real influence on the incidence of colic. A receiver operating characteristic (ROC) curve and pseudo r squared value present the quality of the model and its ability to distinguish between experimental and control diagnoses. The ROC curve is looking for the discrimination between colic and other diagnoses with the logistic model created. It calculates this by finding the area under the curve values approaching 1 represent excellent distinguishability between colic and control exams, but values approaching 0.5 (similar to a 50/50 chance) signifies a very poor discrimination. 7

Chapter 3 RESULTS AND DISCUSSION The data used in the study is normally and evenly distributed. Normality of the data (including barometric pressure, day of year, age and the distribution of exams) was determined with the Shapiro-Wilk test. However, the New Bolton Center treats a disproportionate number of Thoroughbred horses compared to the general equine population of the survey area. The prominence of Thoroughbred racing and breeding in the area presents the New Bolton Center with many valuable animals which presents a bias, in the form of owner behavioral differences, in the amount of care given to those animals compared to the overall population. For example, owners of expensive racehorses are more likely to spend the extra money for expert care than typical horse owners. However, this allows the statistical testing of Thoroughbred horses, historically known as sensitive, against other breeds. Table 1: Summary of Equine Data Collected from the New Bolton Center s Field Service Examinations from 1/1/05-1/1/17 Diagnosis Number of Horses Variable Mean Age Colic 866 Experimental 14 Choke 117 Control 18 Laceration 381 Control 11 Wound Repair 403 Control 10 Ophthalmologic 502 Control 14 Lameness 653 Control 12 8

Unknown Stock 9% 5% Thoroughbred 31% Pony 23% Warmblood 12% Draft 9% Non- Thoroughbred Hotblood 11% Figure 1: Breed Distribution for the Study Population. Compared to the overall equine population in the area, the New Bolton Center sees a disproportionate number of Thoroughbred horses. 9

Figure 2: Map of all unique zip codes used in the study, marking the approximate locations of the horses with black points and the locations of the National Weather Centers in Philadelphia (PA), Coatesville (PA) and Wilmington (DE) with yellow stars. 10

Figure 3: Density of the animals used in the study. Red areas show the most dense collection of patients and green shows the least dense area, while the study population is taken from the approximate area within the blue outline. Pairwise spearman correlation (significance with Rho 0.3 or -0.3) was used within equine data to determine association of equine characteristics and between equine and weather data to determine if a correlation exists between barometric pressure parameters and equine attributes. This test yielded some significant correlation between colic and month and reoccurring exams, but rho values were miniscule. This signifies that, although significant, the correlation in very weak. 11

Pressure averages and differences are statistically similar throughout the year seen in boxplots. However, there is large variation in both and there are many outliers in the plot depicting pressure difference. Those months with lower deviation had more outliers above the plot, signifying many horses were seen by veterinarians when there was greater pressure differences. After removal of outliers, there were statistically similar number of colic examinations per month throughout the time course as well as control diagnosis. Overall, neither colic nor control diagnosis show seasonality, however individual types of colic may be more likely at different times of the year. Therefore, time of year was not considered a factor in the study. 12

Figure 4: The number of monthly diagnoses made by the New Bolton Center for both colic and control exams from 1/1/05 through 1/1/17, including variance. There is no statistical difference for either variable throughout the year, signifying that there is no overall seasonality. 13

Figure 5: Stratified Boxplot depicting the average daily atmospheric sea level pressure, measured in inches mercury from 1/1/05 through 1/1/17. There is statistically no difference throughout the year, corresponding to the lack of seasonality in diagnoses. Figure 6: Stratified Boxplot depicting the average of daily drop in atmospheric pressure at sea level, measured in inches mercury, from 1/1/05 through 1/1/17. There is statistically no difference throughout the year, corresponding to the lack of seasonality in diagnoses. 14

Survival analysis curves showed that the number of colic deaths did not differ significantly from control exams by age or change in pressure. These variables were chosen because of their historic and suggestive evidence through the responses of other species. Death was no longer looked at as a significant variable for further testing. Figure 7: Survival of colic diagnoses compared to all control diagnoses when looking at the drop in atmospheric sea level pressure, measured in inches mercury, 12 hours before the event. There is no significant difference seen. 15

Figure 8: Survival of colic diagnoses compared to all control diagnoses when looking at age. There is no significant difference seen in this study s population. Logistic Regression was used for a predictive analysis of the binary variable of colic (colic or other diagnosis) compared to horse characteristics and barometric pressure measurements. When looking at each parameter separately, in a univariate analysis, variables displaying p values of less than 0.2, which suggests association, were included in a multivariate backward stepwise regression to find the best predictor of the incidence of colic. This test removes predictors one by one until the model AIC (akaike information criterion) an estimator of model quality is unchanging. Factorial variables (breed and sex) were compared to a standard to note differences in categories. Sexes were compared to mares and breeds were compared to hotbloods (Arabians, Thoroughbreds and the historically high-strung breeds). The results of the backward stepwise regression are given a p-value, odds ratio and confidence interval. Those predictors with significant p-values (less than 0.05) and 16

Frequency 0 100 200 300 400 500 600 700 confidence intervals not including 1.00 (this value denotes that changes in that variable do not significantly affect the colic response) were used as the best predictors for the incidence of colic. The odds ratio of these predictors determines the increase or decrease in odds of a colic event with changes in that predictor. Initial models showed significance in the pressure low 12 hours before the exam and the pressure average 24 hours prior to the exam. These variables were found to be highly correlated and the model was unable to make a distinction between experimental or control horses. The model was then adjusted to remove duplicate predictors. The pseudo r squared value was minimal, at less than 0.01, signifying a poor-quality model. Probability of Animals With or Without Colic 0.0 0.2 0.4 0.6 0.8 1.0 Probabiltiy Figure 9: Original Model-probability of horses with and without colic; the identical distribution of the two groups leaves the model unable to distinguish between control and experimental diagnoses. 17

The model that best fits the data gives a probability distribution almost identical to the original model. This signifies that, although considered a better model due to a larger pseudo r squared value (0.023), it is still not a predictive model of colic. However, the model still found associations between colic and atmospheric pressure measurements. A resistor operating characteristic curve was also created to determine the ability of the model to discriminate between control and experimental horses. The area under the curve, 0.6, signifies that the model has very poor distinguishability. Figure 10: Best Fit Model-probability of horses with and without colic; the identical distribution of the two groups leaves the model unable to distinguish between control and experimental diagnoses. 18

Figure 11: Receiver Operating Characteristic Curve determining the discrimination power of the logistic model. The area under the curve (0.6), represents minimal distinguishability between control and experimental horses. The best model created with the data found that age has an odds ratio of 0.95, meaning that for each change in age (one year) the odds of a horse colicking increases by a factor of 1.05. The pressure difference within 12 hours of the colic event has an odds ratio of 0.40 and a confidence interval that does not include the integer 1. This is the most significant pressure parameter to which colic responds. An odds ratio of 1.48 for repeated diagnoses is significant in that there is a positive relationship. The more duplicate diagnoses a horse gets, the less likely that the diagnoses are colic. The last predictor found to be significant is the horse s north/south location. With an odds ratio 19

of 2, each degree increase in latitude decreases the odds of a horse colicking by a factor of 2. Table 2: Backward stepwise logistic regression results display the best predictors of colic being age, the atmospheric pressure low 24 hours prior to the event and the horse s latitude Variable Pr (> z ) Odds Ratio 95% Confidence Interval Age 3.2e-11 0.96 0.95 0.97 Farm Latitude 0.041 2.05 1.03 4.07 Pressure Drop 0.006 0.40 0.21 0.77 Repeated Event 7.4e-06 1.48 1.24 1.75 20

Chapter 4 CONCLUSION The best predictors of colic found in this study, after model adjustments, are age, the horse s geographical latitude, the incidence of repeated colic events for one horse and the barometric pressure drop within 12 hours of the event. For each change in age (one year), the odds of a horse colicking increases by a factor of 1.05. This seems minimal but over the course and events of a horse s life, this factor may be a large scalar. For every unit decrease (inches mercury) of the atmospheric pressure, the odds of colic increase by a factor of 2.5. This is the most significant pressure parameter to which colic responds. As pressure fronts move through, horses that are senior, prone to colic or stressed will have a very large factor multiplying their probability of colic. As for repeated incidence of the same diagnosis for a particular horse, the more duplicate diagnoses a horse gets, the less likely that diagnoses are colic. Surveying the horses used in this study, the animals with the most repeated diagnoses were not colic cases. They may have been horses with multiple laceration repairs or lameness exams, for example. The last predictor found to be significant is the horse s north/south location. Each degree increase in latitude decreases the odds of a horse colicking by a factor of 2. The farther north a horse lives, the more variable the climate and the more stress the animal experiences adapting to its environment. The different plants growing in the distinct climates and the variation in terrain will also affect the prevalence of colic. Low pressure fronts, known as cold fronts, occur when cold, dense air pushes warm air up into the atmosphere. This usually leads to the release of moisture from the air and severe weather. These events are short-lived but severe and cause the drops in 21

barometric pressure that we observe. The pressure will drop when a cold front moves in, then increase and stabilize for a short time before dropping again. These depressions are associated with the incidence of equine colic because they are an environmental stressor to the horse s physiology. Significant managerial changes may also contribute, as the animals will have inconsistent exercise, turnout and a different level of feed for their temporary living situation. These managerial changes put stress on the animal s gastrointestinal system by quickly changing the body s normal routine. Horses are unable to adapt quickly to significant changes to exercise and food and will experience problems because of the inability to quickly change digestive enzymes and cecum bacterial populations, especially paired with stall confinement (which is common for horses during severe weather). Knowing that latitude is significant, looking into the frequency and intensity of cold fronts as latitude increases, as well as microclimates within different types of terrain (for example, mountainous versus open) can elaborate on this finding. After model adjustments were made to create the highest quality predictor model, the pseudo r squared value was still marginal (about 0.023) and the receiver operating characteristic curve had an area of 0.6 (minimal). The best predictive model found here cannot be used for diagnostic tests or predictability measures; the fit is not significant enough to draw those conclusions. However, the logistic model fit does show an association between colic and atmospheric pressure drops. This allows the proposition of new questions that relate horses to previously studied species when considering barometric pressure responses. These topics include parturition, pain tolerance and behavior. 22

In preparation for continued research in spatial modeling, Moran s I test found that all pressures are presented as random except the high and low barometric pressure 24 hours prior to the event. Those are statistically more clustered than expected if random and have to be accounted for when creating spatial models. This will be important for generalized linear spatial models, which take spatial distribution into account, whereas logistic regression does not. 23

REFERENCES Archer, Debra C, Gina L Pinchbeck, Christopher J Proudman, and Helen E Clough. Is Equine Colic Seasonal? Novel Application of a Model Based Approach. BMC Veterinary 2, no. 27 (August 2006). Boscolo-Berto, Rafael, Fabrizio Del Moro, Alessandro Abate, Goran Arandjelovic, Franco Tostato, and PierFrancesco Bassi. Do Weather Conditions Influence the Onset of Renal Colic? A Novel Approach to Analysis. Urologia Internationalis 80, no. 1 (January 2008). Bradford G. Bentz. Understanding Equine Colic. Lexington, KY: Blood-Horse Publications, 2004. Dvorak, R.A. A Note on the Relationship Between Barometric Pressure and Calving Incidence. Animal Reproduction Science 1, no. 1 (May 1978): 3 7. Eisenreich, Josefa. The Correlation of Weather Changes with the Incidence of Colic in Horses. Hungarian Veterinary Archive, 2012. Marcella, Kenneth L. Equine Colic. PowerPoint presented at the Southern States FeedMaster Program, Richmond, VA, January 10, 2018. Metcalfe, Jessica, Kim L Schmidt, Wayne Bezner-Kerr, Christopher G. Guglielmo, and Scott A MacDougall-Shackleton. White-Throated Sparrows Adjust Behaviour in Response to Manipulations of Barometric Pressure and Temperature. Animal Behavior 86, no. 6 (December 2013): 1285 90. Pellergrino, Ana, Maria Penaflor, Cristiane Nardi, Wayne Bezner-Kerr, Christopher G. Guglielmo, Jose Bento, and Jeremy McNeil. Weather Forecasting by Insects: Modified Sexual Behaviour in Response to Atmospheric Pressure Changes. PLos One 8, no. 10 (October 2013). Sato, Jun, Keisuke Takanari, Sayaka Omura, and Kazue Mizumura. Effects of Lowering Barometric Pressure on Guarding Behavior, Heart Rate and Blood Pressure in a Rat Model of Neuropathic Pain. Neuroscience Letters 299, no. 1 2 (February 2001): 17 20. Tinker, Mary K, N.A. White, P. Lessard, C.D. Thatcher, K.D. Pelzer, Betty Davis, and D.K. Carmel. Prospective Study of Equine Colic Incidence and Mortality. Equine Veterinary Journal 29, no. 6 (1997): 448 53. USDA. Demographics of the U.S. Equine Population, 2015. Animal and Plant Health Inspection Service, February 2017. 24

Vainio, K, B Sykes, and A.T. Blikslager. Primary Gastric Impaction in Horses: A Retrospective Study of 20 Cases (2005 2008). Equine Veterinary Education 23, no. 4 (2011): 186 90. Van Der Linden, Marianne A, Celine M Laffont, and Marianne M Sloet van Oldruitenborgh-Oosterbaan. Prognosis in Equine Medical and Surgical Colic. Journal of Veterinary Internal Medicine 17 (May 2003): 343 48. 25

Appendix A RAW STATISTICAL OUTPUT Table 3: Summary Table of Normality of Data from the New Bolton Center s Field Service Data and the Weather Center from 1/1/05-1/1/17 Test Shapiro-Wilk Statistic (W) P value DayBeforeDiff 0.81625 2.20E-16 PressDiff 0.82773 2.20E-16 DayBeforeAvg 0.99306 4.64E-11 DayBeforeLow 0.97808 2.20E-16 DayBeforeHigh 0.99587 1.40E-07 LowPress 0.98804 1.58E-15 AvgPress 0.99592 1.63E-07 HighPress 0.9965 1.22E-16 Age 0.96189 2.20E-16 Day 0.96866 2.20E-16 Month 0.95605 2.20E-16 Sex 0.67794 2.20E-16 Diagnosis 0.85378 2.20E-16 Breed Category 0.79476 2.20E-16 Table 4: Summary Table of Spearman Rank Correlation Output Variable 1 Variable 2 P value Spearman s Rho Coefficient Colic DayBeforeDiff 6.84E-05-0.07136 Colic DayBeforeLow 0.009047 0.04681489 Colic DayBeforeHigh 0.6171 0.0089721 Colic DayBeforeAvg 0.1146 0.02831087 Colic PressDiff 2.47E-05-0.07556828 Colic HighPress 0.5908 0.009647 Colic LowPress 0.0148 0.04371263 Colic AvgPress 0.09726 0.02975084 Colic Age 4.41E-08-0.01044 Colic Breed Category 0.1219-0.02907089 Colic Sex 0.6293-0.008950087 Colic Day 0.0008515 0.05980028 26

Table 5: Summary Table of Spearman Rank Correlation Output (continued) Variable 1 Variable 2 P value Spearman s Rho Coefficient Colic Month 0.0006501 0.06113046 Colic Death 2.20E-16 0.2990366 Colic Recurrence 0.006757 0.04857622 Table 6: Summary Table of Calculations of the Average Number of Exams Seen per Month from 1/1/05-1/1/17 Diagnosis Month Average Number of Exams Standard Deviation Colic Jan 6.7 3.1 Colic Feb 6.0 3.2 Colic Mar 6.8 3.2 Colic Apr 6.8 3.0 Colic May 9.6 3.9 Colic Jun 6.3 3.0 Colic Jul 6.4 3.9 Colic Aug 5.2 3.5 Colic Sep 5.3 2.8 Colic Oct 5.7 2.8 Colic Nov 4.6 2.6 Colic Dec 6.4 2.8 Control Jan 10.3 4.6 Control Feb 9.2 4.2 Control Mar 15.3 7.4 Control Apr 17.0 7.7 Control May 17.1 7.8 Control Jun 17.2 7.9 Control Jul 15.0 5.1 Control Aug 16.2 7.8 Control Sep 16.6 8.3 Control Oct 16.4 7.7 Control Nov 13.0 0.0 Control Dec 11.7 5.0 27

Table 6: Summary of Moran s I Statistical Cluster Analysis Determining Autocorrelation of Barometric Pressure Measurements Data Z-Score P value I Statistic Expectation Variance 24hrBeforeDiff 0.069914 0.4721 0.00093228-0.0003219 0.00032178 12hrPressDiff 0.47748 0.3165 0.00824536-0.0003219 0.00032193 12hrHighPress 0.83174 0.2028 0.01461143-0.0003219 0.00032236 12hrLowPress 0.90013 0.184 0.01583722-0.0003219 0.00032227 12hrAvgPress 0.94282 0.1729 0.0166051-0.0003219 0.00032233 24hrBeforeHigh 1.9544 0.02533 0.03476777-0.0003219 0.00032235 24hrBeforeLow 1.7446 0.04053 0.03099196-0.0003219 0.00032217 24hrBeforeAvg 1.1868 0.1177 0.02098497-0.0003219 0.00032231 Table 7: Univariate Logistic Regression Output from R for All Variables Variable Odds Ratio Std. Err Z value Pr (> z ) 95% Confidence Interval PressDiff 0.3243389 0.31429-3.583 0.00034 0.175175 0.6005179 HighPress 1.0679227 0.19581 0.336 0.737 0.7175556 1.567521 LowPress 1.511392 0.1796 2.3 0.0215 1.062904 2.1491172 AvgPress 1.34015854 0.189 1.549 0.121 0.9252417 1.941141 DayBeforeDiff 0.3353962 0.29602-3.69 0.000224 0.1877536 0.5991396 DayBeforeLow 1.501105 0.1743 2.331 0.0198 1.066724 2.1123691 DayBeforeHigh 1.0463075 0.19308 0.234 0.815 0.7166598 1.527586 DayBeforeAvg 1.29272803 0.1865 1.377 0.169 0.8969112 1.863223 Age 0.9676624 0.005312-6.188 6.07E-10 0.9576404 0.9777893 Farm Latitude 2.074521 0.3348 2.179 0.0293 1.076215 3.9988644 Farm Longitude 1.072198 0.18134 0.384 0.701 0.7514743 1.529804 Repeated Event 1 1.250327 0.08257 2.706 0.00672 1.063507 1.469964 Breed Category 2 1.066723 0.14802 0.436 0.66256 0.7981067 1.4257468 3 0.8582855 0.12884-1.186 0.23557 0.6667547 1.104835 4 0.9896557 0.10443-0.1 0.92069 0.8064782 1.2144388 5 0.6147616 0.17288-2.814 0.00489 0.4380758 0.8627089 Sex 2 0.997421 0.087395-0.03 0.976 0.8404028 1.183776 3 0.7026732 0.255324-1.382 0.167 0.4260111 1.159007 TB 1 1.183815 0.08941 1.887 0.0591 0.9935213 1.410555 28

Table 7: Univariate Logistic Regression Output from R for All Variables (continued) Variable Odds Ratio Std. Err Z value Pr (> z ) 95% Confidence Interval Month 2 0.9936766 0.20883-0.03 0.975767 0.6599163 1.49624 3 1.4594487 0.195602 1.933 0.053261 0.9947009 2.141338 4 1.5985895 0.193856 2.42 0.015523 1.0932671 2.337478 5 1.2345679 0.18646 1.13 0.258429 0.8566429 1.779222 6 1.7629439 0.196576 2.884 0.003923 1.1992571 2.591581 7 2.1032626 0.193993 3.833 0.000127 1.4380258 3.076241 8 2.2659323 0.203442 4.021 5.80E-05 1.5208159 3.376115 9 2.0544586 0.203785 3.533 0.000411 1.3779567 3.063086 10 1.8842659 0.201028 3.152 0.001624 1.2706538 2.794198 11 1.6565041 0.205888 2.451 0.014231 1.1064709 2.479962 12 1.1825573 0.201891 0.831 0.406231 0.7961086 1.756597 Table 8: Backward Stepwise Regression Output from R to Fit the Best Predictive Model Variable Estimate Standard Error z value Pr (> z ) Odds Ratio 95% Confidence Interval Day 0.001059 0.0004184 2.532 0.01135 1.00 1.000239 1.0018812 Age -0.03596 0.005418-6.636 3.22e-11 0.9647 0.954489 0.9749797 Farm Latitude 0.7164 0.3512 2.040 0.04135 2.047047 1.028523 4.0741903 Pressure Drop -0.9061 0.3303-2.743 0.00609 0.4040912 0.2114933 0.7720798 Repeated Event 0.3893 0.08686 4.483 7.38e-06 1.475997 1.244956 1.7499151 29