U.S. patent application number 10/521087 was filed with the patent office on 2005-10-13 for echocardiographic measurements as predictors of racing success.
Invention is credited to Seder, Jeffrey A..
Application Number | 20050224009 10/521087 |
Document ID | / |
Family ID | 30770924 |
Filed Date | 2005-10-13 |
United States Patent
Application |
20050224009 |
Kind Code |
A1 |
Seder, Jeffrey A. |
October 13, 2005 |
Echocardiographic measurements as predictors of racing success
Abstract
The present invention relates to methods for selecting racehorse
candidates. Provided herein are methods for increasing the
likelihood of selecting candidates that will be high earners, while
reducing the likelihood of selecting candidates that will be low
carners, on the basis of certain cardiac measurements.
Inventors: |
Seder, Jeffrey A.;
(Coatesville, PA) |
Correspondence
Address: |
Steven J Rocci
Woodcock Washburn
46th Floor
One Liberty Place
Philadelphia
PA
19103
US
|
Family ID: |
30770924 |
Appl. No.: |
10/521087 |
Filed: |
January 13, 2005 |
PCT Filed: |
June 20, 2003 |
PCT NO: |
PCT/US03/19537 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
60396592 |
Jul 17, 2002 |
|
|
|
Current U.S.
Class: |
119/174 |
Current CPC
Class: |
A61B 8/463 20130101;
A61B 8/0858 20130101; G06Q 10/10 20130101; G06Q 99/00 20130101;
A63K 3/00 20130101 |
Class at
Publication: |
119/174 |
International
Class: |
A01K 029/00 |
Claims
1. A method of screening a racehorse candidate, said method
comprising: (i) obtaining a measurement of the width of the
ventricular septal wall of said racehorse candidate; and (ii)
comparing said measurement to a collection of measurements from a
group of horses, wherein said collection of measurements comprises
ventricular septal wall width measurements for horses of about the
same age, sex, and weight as said racehorse candidate.
2. The method of claim 1, further comprising the step of selecting
said racehorse candidate if it has a ventricular septal wall width
that is greater than the mean ventricular septal wall width from
said collection of measurements.
3. The method of claim 1, the method further comprising the step of
selecting said racehorse candidate if it has a ventricular septal
wall width that is in the 75.sup.th percentile or higher when
compared to the ventricular septal wall width measurements from
said collection of measurements.
4. The method of claim 1, wherein said screening takes place at an
auction.
5. The method of claim 1, the method further comprising the step
of: (iii) obtaining a measurement of the cross-sectional area of
the left ventricle in diastole of said racehorse candidate; wherein
said collection of measurements further comprises left ventricle in
diastole cross-sectional area measurements for horses of about the
same age, sex, and weight as said racehorse candidate.
6. The method of claim 5, further comprising the step of selecting
said racehorse candidate if it has a ventricular septal wall width
and left ventricle in diastole cross sectional area that is greater
than the mean ventricular septal wall width and left ventricle in
diastole cross-sectional area measurement from said collection of
measurements.
7. The method of claim 5, further comprising the step of selecting
said racehorse candidate if it has a ventricular septal wall width
and a left ventricle in diastole cross sectional area measurement
that is in the 75.sup.th percentile or higher when compared to the
ventricular septal wall width and left ventricle in diastole cross
sectional area measurements from said collection of
measurements.
8. The method of claim 1, wherein said ventricular septal wall
width measurement is obtained by measuring, in a left parasternal
short-axis view obtained at end diastole, the distance from the
endocardial edge of the right ventricular free wall, at the point
where the wall meets the interventricular septum, through the
interventricular septum, to the point of attachment of the
moderator band in the left ventricle.
9. The method of claim 1, wherein said measurement is obtained from
a left parasternal short axis echocardiogram of the left ventricle
of said racehorse.
10. The method of claim 5, wherein said left ventricle in diastole
cross sectional area measurement is obtained by measuring the
circumference of the left ventricular chamber.
11. The method of claim 5, wherein said left ventricle in diastole
cross sectional area measurement is obtained from a left
parasternal short-axis echocardiogram of the left ventricle of said
racehorse.
12. The method of claim 1, the method further comprising the steps
of: (iii) obtaining a measurement of the cross-sectional area of
the left ventricle in systole of said racehorse candidate; wherein
said collection of measurements further comprises left ventricle in
systole cross-sectional area measurements of horses of about the
same age, sex, and weight as said racehorse candidate.
13. The method of claim 12, further comprising the step of
selecting said racehorse candidate if it has a ventricular septal
wall width and left ventricle in systole cross sectional area that
is greater than the mean ventricular septal wall width and left
ventricle in systole cross-sectional area measurement from said
collection of measurements.
14. The method of claim 12, further comprising the step of
selecting said racehorse candidate if it has a ventricular septal
wall width and a left ventricle in systole cross sectional area
measurement that is in the 75.sup.th percentile or higher when
compared to the ventricular septal wall width and left ventricle in
systole cross sectional area measurements from said collection of
measurements.
15. The method of claim 1, the method further comprising the steps
of: (iii) obtaining a measurement of the cross-sectional area of
the spleen of said racehorse candidate; wherein said collection of
measurements further comprises splenic cross-sectional area
measurements of horses of about the same age, sex, and weight as
said racehorse candidate.
16. The method of claim 12, further comprising the step of
selecting said racehorse candidate if it has a ventricular septal
wall width and splenic cross sectional area that is greater than
the mean ventricular septal wall width and splenic cross-sectional
area measurement from said collection of measurements.
17. The method of claim 12, further comprising the step of
selecting said racehorse candidate if it has a ventricular septal
wall width and a splenic cross sectional area measurement that is
in the 75.sup.th percentile or higher when compared to the
ventricular septal wall width and splenic cross sectional area
measurements from said collection of measurements.
18. The method of claim 1, the method further comprising the steps
of: (iii) obtaining a measurement of the height.times.weight of
said racehorse candidate; wherein said collection of measurements
further comprises height.times.weight measurements from horses of
about the same age and sex as said racehorse candidate.
19. The method of claim 18, further comprising the step of
selecting said racehorse candidate if both the ventricular septal
wall width and the height.times.weight measurement are greater than
the mean ventricular septal wall width and height.times.weight
measurements from said collection of measurements.
20. The method of claim 18, further comprising the step of
selecting said racehorse candidate if it has both a ventricular
septal wall width and a height.times.weight measurement that is in
the 75.sup.th percentile or higher when compared to the ventricular
septal wall width and height and weight measurements from said
collection of measurements.
21. The method of claim 18, the method further comprising the steps
of: (iii) obtaining measurements of the cross-sectional area of the
left ventricle in systole of said racehorse candidate; wherein said
collection of measurements further comprises left ventricle in
systole cross-sectional area measurements from horses of about the
same age, sex, and weight as said racehorse candidate.
22. The method of claim 21, the method further comprising the step
of selecting said racehorse candidate if it has a ventricular
septal wall width, a left ventricle in systole cross sectional
area, and a height.times.weight measurement that is greater than
the mean ventricular septal wall width, left ventricle in systole
cross sectional area, and height.times.weight measurements from
said collection of measurements.
23. The method of claim 21, the method further comprising the step
of selecting said racehorse candidate if it has a ventricular
septal wall width, a left ventricle in systole cross sectional
area, and a height.times.weight measurement that is in the
75.sup.th percentile or higher when compared to corresponding
measurements from said collection of measurements.
24. A method of screening a racehorse candidate, said method
comprising: (i) obtaining a measurement of the cross-sectional area
of the spleen of said racehorse candidate; (ii) comparing said
measurement to a collection of measurements from a group of horses,
wherein said collection of measurements comprises splenic
cross-sectional area measurements from horses of about the same
age, sex, and weight as said racehorse candidate.
25. The method of claim 24, further comprising the step of
selecting said racehorse candidate if it has a splenic cross
sectional area that is greater than the mean splenic cross
sectional area from said collection of measurements.
26. The method of claim 24, the method further comprising the step
of selecting said racehorse candidate if it has a splenic cross
sectional area that is in the 75.sup.th percentile or higher when
compared to the splenic cross sectional areas from said collection
of measurements.
27. The method of claim 24, the method further comprising the step
of: (iii) obtaining a measurement of the cross-sectional area of
the left ventricle in systole of said racehorse candidate; wherein
said collection of measurements further comprises left ventricle in
diastole cross-sectional area measurements from horses of about the
same age, sex, and weight as said racehorse candidate.
28. The method of claim 1, wherein said racehorse candidate is a
yearling or two year old.
29. A method of screening a racehorse candidate to select a high
earner router, said method comprising: (i) obtaining a measurement
of the cross-sectional area of the left ventricle in systole, the
cross-sectional area of the left ventricle in diastole, or the
percent change in ventricular area per stroke of said racehorse
candidate; (ii) obtaining a measurement of the height.times.weight
of said racehorse candidate; (iii) comparing said measurements from
said racehorse candidate to a collection of measurements from a
group of horses, wherein said collection of measurements comprises
height.times.weight measurements and cross-sectional area of the
left ventricle in systole measurements, cross-sectional area of the
left ventricle in diastole measurements, or percent change in
ventricular area per stroke measurements from horses of the same
age, weight and sex as said racehorse candidate.
30. A method for maintaining a horse registry system for
identifying the potential racing ability of a candidate racehorse,
the method comprising: (i) obtaining measurements from a group of
horses; (ii) standardizing said measurements for age, sex, and
weight; (iii) classifying said horses according to their racing
abilities; and (iv) determining the relationship between the
measurements of said horses and their racing ability.
31. The method of claim 30, wherein said measurements are selected
from the group consisting of ventricular septal wall width, height,
weight, cross-sectional area of the left ventricle in diastole,
cross-sectional area of the left ventricle in systole,
cross-sectional area of the spleen, and percent stroke volume.
32. A method of screening a racehorse candidate, said method
comprising: (i) obtaining an echocardiographic image of the heart
of said racehorse; (ii) rating the image according to at least one
cardiac parameter selected from the group consisting of the general
shape of the heart at diastole and systole, the clarity and
sharpness of contrast of left ventricle during diastole and
systole, the smoothness of the left ventricle during diastole and
systole, blood backflow from the left ventricle during diastole and
systole, valve closure, and clarity of the image in diastole; and
(iii) comparing the rating to a collection of ratings of the same
cardiac parameter from a group of horses of about the same age,
sex, and weight of said racehorse candidate.
Description
CROSS REFERENCE TO RELATED PATENT APPLICATIONS
[0001] The present application is related to and claims priority of
U.S. Patent Application Serial Number (USSN) 60/396,592, filed Jul.
17, 2002, which is explicitly incorporated herein by reference in
its entirety and for all purposes.
FIELD OF THE INVENTION
[0002] The present invention is directed to methods for screening
candidate racehorses, and improving the likelihood of selecting a
candidate that will become a successful racehorse.
BACKGROUND
[0003] For centuries, perhaps ever since the horse was first
domesticated, people have raced their horses against each other, to
see which man owned the faster animal. Countless generations of
breeding the fastest stallion to the best mares has led to the
modern racehorse, a magnificent running machine, genetically
selected to excel in the modern version of "the sport of kings," a
sport that still thrives across the globe today.
[0004] Since the sport first began, people have bought, sold and
traded young horses, with dreams of acquiring a horse that night
one day mature into a stakes winner. A huge business has developed
around the breeding, preparation, and sale of potential racehorses.
Potential buyers pore over sales catalogs, searching the horse's
parentage and pedigree in hope of selecting a horse that contains
just the right mix of speed, stamina, durability and temperament to
grow into a successful racehorse. The racing candidate is carefully
examined to make sure its legs are straight, the airway is clear,
and that there are no physical blemishes or infirmities that might
diminish its chance of future success. "Experts," sometimes
boasting of years of experience at spotting future champions long
before the horse ever sets foot on a racetrack, are regularly
consulted to scrutinize the field of candidates, and help the
would-be owner select a horse that possess the right conformation,
the right carriage, the right glint in the eye--the stuff of
champions.
[0005] But despite all the statistical analysis of pedigree, the
pre-sale poking and prodding, and the intense scrutiny of a
countless number of trained eyes, separating the future winners
from the future losers has remained an inexact science at best. In
a study of all named (i.e., registered) Thoroughbred foals born
between 1985 and 1994, only about 30% ever started a single race,
and the average earnings per start of all foals was only $1,378.
Selecting the offspring of the top 1% of stallions reduced the
percentage of non-starters to about 15%, and increased the average
earnings per start to over $3,000, but still, a large percentage of
all young horses sold at auction fail to recoup their original
purchase price. One has only to look to the results of this year's
Kentucky Derby to see how inaccurate the selection of racehorse
candidates can be: the race was won by Funny Cide, once a $75,000
purchase, while numerous horses from the same crop that sold for
much more, even ten or more times as much, failed to even win a
single maiden race.
[0006] Accordingly, better methods for screening potential
racehorse candidates are needed, particularly methods that will
increase the chances of selecting a horse that is more likely to
become a high earner, while decreasing the likelihood of selecting
a horses that will turn out to be a low earner. The present
invention is directed to these, as well as other ends.
SUMMARY OF THE INVENTION
[0007] It has now been discovered that certain objective
measurements that may be readily obtained from young horse racing
candidates, e.g., heart and/or spleen size, may be used as
predictors of the future racing success. Accordingly, the present
invention provides methods of screening a racehorse candidate and
selecting a racehorse candidate likely to become a high-earner
racehorse.
[0008] The method for screening racehorse candidates includes the
step of obtaining one or more measurements, including
echocardiographic measurements, from a racehorse candidate and
comparing the measurements to a collection of corresponding
measurements from a group of horses. In particular, in one
embodiment, the methods include the step of obtaining a measurement
of the width of the ventricular septal wall of a racehorse
candidate and comparing it to ventricular septal wall width
measurements from a group of horses of similar age, sex, and
weight. In another embodiment, the methods of the present invention
further comprise the steps of obtaining one or more measurements
selected from the cross-sectional area of the left ventricle in
diastole, the cross-sectional area of the left ventricle in
systole, the body size (more specifically, the
height.times.weight), or the splenic cross-sectional area of the
candidate racehorse and comparing these additional measurements to
corresponding measurements from the group of horses.
[0009] In another embodiment of the present invention, the methods
of the present invention include the step of obtaining a
measurement of the splenic cross-sectional area of a racehorse
candidate and comparing it to splenic cross-sectional area
measurements from the group of horses. In another embodiment, the
methods of the present invention further comprise the steps of
obtaining one or more measurements selected from the
cross-sectional area of the left ventricle in diastole, the
cross-sectional area of the left ventricle in systole, the body
size, or the interventricular septal wall width of the candidate
racehorse and comparing these additional measurements to
corresponding measurements from the group of horses.
[0010] After the measurements have been obtained from the candidate
racehorse, the methods of the present invention may further include
the step of selecting a racehorse candidate if its ventricular
septal wall width measurement is greater than the mean ventricular
septal wall width of the group of horses. In another embodiment,
the methods of the present invention may further include the step
of selecting a racehorse candidate if one or more of its
measurements, e.g., ventricular septal wall width, left ventricle
cross sectional area in diastole or systole, body size, splenic
cross sectional area, is greater than the mean corresponding
measurement from the group of horses of similar age, weight and
sex.
[0011] In yet another embodiment, the methods of the present
invention may further include the step of selecting a racehorse
candidate if one or more of its measurements, e.g., ventricular
septal wall width, left ventricle cross sectional area in diastole
or systole, body size, splenic cross sectional area, is in the
75.sup.th percentile or higher when compared to corresponding
measurements from the group of horses. The methods of the present
invention may also include the step of rejecting a racehorse
candidate if one or more of its measurements, e.g., ventricular
septal wall width, left ventricle cross sectional area in diastole
or systole, body size, splenic cross sectional area, is not in the
75.sup.th percentile or higher when compared to corresponding
measurements from the group of horses.
[0012] In one aspect of the present invention, the measurements
obtained from the horses are acquired using standard methods,
commonly known and accepted in the art. In another aspect, the
ventricular septal wall width measurement is obtained by measuring
a particular structure in an echocardiographic image of the heart,
the measurement running from the attachment of the moderator band
through the interventricular septum into the right ventricle to the
endocardial edge of the right ventricular free wall where the wall
attaches to the interventricular septum. In yet another embodiment,
the left ventricle in diastole and/or systole cross sectional area
is obtained by measuring the circumference of the left ventricular
chamber.
[0013] The methods of the present invention thus improve the odds
of selecting high and excluding low earner racehorses.
Additionally, from within the group of high earners, certain
parameters may be applied that enable the selection of horses more
likely to be high earners in races that extend beyond a mile. These
and other applications for the methods disclosed herein will be
made apparent in the detailed description and examples that
follow.
BRIEF DESCRIPTION OF THE DRAWINGS
[0014] FIG. 1: A left parasternal short-axis echocardiogram of the
left ventricle at peak systole from a 2-year-old Thoroughbred filly
with a resting heart rate below 40 bpm obtained from the left
cardiac window with a 3.5 MHz probe. The dotted line traces the
endocardial border of the left ventricle at peak systole.
[0015] FIG. 2: A left parasternal short-axis echocardiogram of the
left ventricle at end diastole from a 2-year-old Thoroughbred filly
with a resting heart rate below 40 bpm obtained from the left
cardiac window with a 3.5 MHz probe. The dotted line traces the
endocardial border of the left ventricle at the end of
diastole.
[0016] FIG. 3: A left parasternal short-axis echocardiogram of the
left ventricle at end diastole from a 2-year-old Thoroughbred filly
with a resting heart rate below 40 bpm obtained from the left
cardiac window with a 3.5 MHz probe. The dotted line measures the
attachment of the moderator band through the interventricular
septum into the right ventricular to the endocardial edge of the
right ventricular free wall where it attaches to the
interventricular septum.
[0017] FIG. 4: LVD (mm.sup.2) measurements for colts and fillies at
ages 12 to 28 months.
[0018] FIG. 5: LVD (mm.sup.2) measurements for colts and fillies at
weights of 850 to 1150 pounds.
[0019] FIG. 6: LVS (mm.sup.2) measurements for colts and fillies at
weights of 850 to 1150 pounds.
[0020] FIG. 7: SW (mm) measurements for colts and fillies at
weights of 850 to 1150 pounds.
[0021] FIG. 8: PS (pct.) measurements for colts and fillies at
weights of 850 to 1150 pounds.
[0022] FIG. 9: Mean weight percentiles for high earner and low
earner horses at ages 12 to 28 months.
[0023] FIG. 10: Mean LVD percentiles for high earner and low earner
horses at ages 12 to 28 months.
DETAILED DESCRIPTION OF THE INVENTION
[0024] The present invention provides new methods for screening a
racehorse candidate. The present invention relates, in part, to the
discovery that certain physical characteristics of the racehorse
candidate that can be readily measured may be used as a predictor
of the horse's future racing ability.
[0025] In particular, it has been found that certain
echocardiographic features are correlated with racing success, and
may be used to screen racehorse candidates. Accordingly, the
present invention provides, inter alia, methods of obtaining
certain physical measurements of a candidate racehorse's heart.
[0026] The present invention also relates, in part, to the
discovery that the size of a horse's spleen can also be used as a
predictor of the horse's future racing ability. Accordingly, the
present invention provides methods of screening a racehorse
candidate on the basis of its splenic cross-sectional area.
[0027] It has also been found that a horse's physical size for it's
chronological age, particularly when viewed in combination with
certain cardiac parameters or splenic cross sectional area, can be
used to predict the racing ability of a candidate racehorse.
Accordingly, the present invention provides screening methods that
further include the step of determining the physical size of a
horse. For the purposes of the present invention, the physical size
or body size of a horse or "HTWT" is determined by multiplying the
height and the weight of the horse. Methods of determining the
height and weight of a horse are known in the art, e.g., using a
scale, weight tape, height stick, or well-educated estimates based
on visual inspection by those skilled in the art.
[0028] In the methods of the present invention, the heart of a
horse is measured in terms of one or more of the following
variables: cross sectional area of the left ventricle in diastole
(LVD), cross sectional area of the left ventricle in systole (LVS),
ventricular septal wall width, and percent change in ventricular
area per stroke (PS).
[0029] The term "ventricular septal wall width" refers to the width
of the septum dividing the right and left ventricles. A
particularly preferred ventricular septal wall measurement involves
a particular cardiac structure that runs from the endocardial edge
of the right ventricular free wall, at the point where the wall
meets the interventricular septum, through the interventricular
septum, to the point of attachment of the moderator band in the
left ventricle, as shown for example in FIG. 3. This structure may
be readily identified in a left parasternal short-axis view,
preferably obtained at end diastole, although other views may also
be used to obtain measurements of this structure. Measurements of
this particular structure are referred to herein as the
"interventricular septal wall structural thickness" or "SW".
[0030] The "cross-sectional area of the left ventricle in systole"
or "LVS" is measured when the left ventricle contracts to its
smallest size in systole. The "cross-sectional area of the left
ventricle in diastole" or "LVD." is measured when the left
ventricle expands to its largest size in diastole. These
measurements can be obtained by any means known to those of
ordinary skill in the art, for example, by using the "inner edge"
method. In the "inner edge" method, linear parameters are measured
from the inner edge of endocardial surfaces and areas are traced
along the inner borders of the endocardial echoes. Thus, LVS and
LVD may be determined by freezing, for example, a left parasternal
short-axis two dimensional echocardiographic ultrasound image at
the peak of systole, and the end of diastole, respectively, and
tracing the internal perimeter of the left ventricular chamber
using calipers on the ultrasound machine. The area inside the
tracing is then calculated based on a pixel count (512.times.512
for total screen). Many commercially available diagnostic
ultrasound machines include software capable of measuring a
circumscribed area in this fashion.
[0031] Alternatively, the cardiac measurements cited herein may be
measured by any method known to those of skill in the art, as may
be described, for example, in one or more of the following: Voros
et al., (1990) Equine Vet. J. p. 392-397; Weyman, A. E. (1982)
Cross-sectional echocardiography, Lea & Febiger, Philadelphia,
p. 497-504; Wyatt, et al. (1979) Circulation 60, p. 1104-1113;
O'Grady et al. (1986) Vet. Radiol. 27, p. 34-49; Henry, W. L., et
al. (1980) Circulation 62, p. 212-217; Feigenbaum, H. (1986)
Echocardiography, 4th edn. Lea & Febiger, Philadelphia; Voros,
et al, Equine vet. J. p. 398-402; Young, L. E., and Scott, G. R.
(1998) Equine vet. J. 30 (2) p. 117-122; Slater, J. D. and
Herrtage, M. E. (1995), Equine vet. J., Suppl. 19, p. 28-32; Marr,
et al., Equine vet. J., Suppl. 30, p. 131-136; Young et al.,
(1998), Equine vet. J. 30 (2) p. 117-122; Young, L. E. (1999)
Equine vet. J., Suppl. 30, p. 195-198; Pascoe, J. R., et al.,
(1990) Equine vet. J, Suppl. 30, p. 148-152.
[0032] The "percent change in ventricular area per stroke" or "PS"
is computed by subtracting LVS from LVD, dividing the resultant
number by LVD and multiplying by 100, e.g., 1 PS = LVD - LVS LVD
.times. 100.
[0033] Thus, PS may be correlated with the volume of blood that is
ejected from the heart per stroke, at rest.
[0034] In the methods of the present invention, the splenic cross
sectional area or "SPLN" is obtained by producing a cross sectional
image of the horse's spleen, and determining the cross sectional
area of same, as discussed above with regard to the LVD and LVS
measurements.
[0035] In one embodiment of the present invention, LVS, LVD, and SW
are measured from a left parasternal short axis echocardiogram of
the left ventricle of the horse at end systole and end diastole.
The echocardiogram can be obtained from the left cardiac window
using a 3.5 MHz probe. During the electrocardiographic exam, the
ultrasound transducer can be held in the right hand with the cursor
facing caudally. The left forelimb can be advanced slightly and the
transducer can be placed in the 4th or 5th left intercostal space,
at a level just dorsal to the point of the olecranon. The
transducer beam can be directed perpendicular (horizontal) to the
longitudinal cardiac axis. The resulting image provides a nearly
circular appearance to the left ventricular lumen. Moving (angling)
the transducer beam from the apex to the base of the heart, the
moderator band(s), papillary muscle, chordae tendinae and septal
leaf of the mitral valve can be identified and used as intracardiac
reference points to obtain reproducible cardiac images in the same
tomographic plane. In other embodiments of the present invention,
alternate echocardiogram views may be obtained and the cardiac and
splenic size measured from the alternate views, e.g., right
parasternal short axis view, left or right parasternal long axis
view, apical views. Typically three to five cardiac cycles are
measured for each echocardiographic measurement. This helps
minimize error, for example, in the timing of peak systole, and end
diastole. Short axis images can be projected according to
international terminology based on the recommendations of the
American Society of Echocardiography (Henry 1980, supra; Feigenbaum
1986, supra). Short axis images recorded from the left side of the
chest can be projected as though the tomographic planes are viewed
from the base to the apex of the heart.
[0036] The accuracy of the measurements may be compromised when a
horse's heart is beating very quickly. For example, in a very
rapidly beating heart, it may be difficult to accurately freeze the
image at peak systole, or at end diastole. Accordingly, it is
preferred that the measurements be taken when the horse's resting
heart rate is less than about 50 beats per minute, with a resting
heart rate at the time of examination of less than about 40 beats
per minute being even more preferred.
[0037] The present invention also provides a collection of
measurements from a group of horses for comparison with those
obtained from the candidate racehorse. In order to create a
collection or database of horse measurements, selected measurements
are obtained from a group of horses, as discussed above. The
database preferably includes measurements of each of the variables
LVD, LVS, PS, HTWT, SPLN, and ventricular septal wall width,
particularly the variable SW, as defined above. Of course,
obtaining measurements from a large number of individuals will
minimize statistical aberrations, and therefore improve the
predictive accuracy of the methods disclosed herein. Typically, the
group of horses includes at least about 1000 individuals, with a
group of greater than 5000 horses being preferred. Even more
preferably, the database will include measurements of at least
about 7500 individuals.
[0038] Since most racehorse candidates are sold as yearlings or
two-year olds, it is preferred that the individuals making up the
group of horses range in age from about 12 months to about 28
months of age chronologically. It has been found, however, that
cardiac measurements vary, depending on age, sex, and weight,
making it difficult to compare horses on the basis of cardiac
measurements alone, without adjusting for the effects of these
parameters. Accordingly, an adequate comparative sample of horses
of about the same age, sex and weight as the racehorse candidate is
preferred. Preferably, the group of horses used for the comparison
are of the same breed as the racehorse candidate, and that breed is
preferably Thoroughbred. As used herein, the term "about the same
age, sex and weight" means that the individuals making up the
collection of horses used for comparative purposes have a date of
birth within about 30 days of the racing candidate, are of the same
genetic gender, and have a weight of within about 25 pounds of the
racing candidate. Preferably, the database will include cardiac
measurements of at least about 35 horses of the same age, sex and
weight as the racehorse candidate. More preferably, the cardiac
measurements of the candidate racehorse are compared to a database
that includes cardiac measurements of at least about 75, and even
more preferably at least about 150, and still more preferably, at
least about 300 horses of the same age, sex and weight as the
racehorse candidate. As a result of such a large statistical
sample, greater accuracy and predictive ability may be achieved by
the methods described herein.
[0039] Once a database of sufficient size has been generated to
assure a statistically significant number of horses of about the
same age, sex and weight as the racehorse candidate have been
obtained, a comparison can be readily made. For example, the
candidate and each horse in the group may be ranked according to
each measurement on a scale from 0 to 100. Percentiles for each
measurement may then be calculated using the following equation: 2
R - 0.5 N ,
[0040] wherein R equals rank and N equals the number of horses. For
example, if there are 100 colts of similar age and weight, and one
colt has the 93.sup.rd largest LVD, he would have an LVD percentile
of 92.5%. 3 93.0 - 0.5 100 = 92.5 % .
[0041] By reporting the measurements in terms of percentiles,
determined relative to other horses of the same sex, age (within 30
days) and weight (within 25 pounds), the cardiac measurements
become independent of a horse's age, sex and weight. Thus, the
racehorse candidate may be assigned a percentile rank for each
measurement variable, e.g., LVD, LVS, SW, SPLN, HTWT, and/or PS, as
compared to a statistically significant sample of horses of about
the same age, sex and weight.
[0042] Other methods for assigning a relative rank to the racing
candidate are known to those of skill in the art, and may be used
as an alternative to the percentile system described above. For
example, standardized scores may be obtained, wherein the
standardized score for each horse in the group is determined by
subtracting the mean measurement for the group from the observed
measurement of an individual horse, and dividing the result by the
standard deviation for that variable for the group. Such
alternative methods should be considered to merely be variants of
the percentile method described above, and do not represent
alternative embodiments of the instant invention.
[0043] This comparison can be used to predict the racing ability of
the candidate racehorse, e.g., whether the candidate racehorse will
be more likely to become a high earner or lower earner. For
example, as described more fully in the examples to follow, by
selecting a racehorse candidate having a ventricular septal wall
width greater than the mean ventricular septal wall width of a
group of horses of about the same age, sex and weight, and/or
rejecting a racehorse candidate that has a septal wall width less
than the mean, the likelihood of selecting a high earner racehorse
is significantly improved. The odds of selecting a high earner
racehorse are further improved by selecting a racehorse candidate
that has a ventricular septal wall width that is in the 75.sup.th
percentile or higher, and/or rejecting a candidate that has a
septal wall width that is lower than the 75.sup.th percentile.
Conversely, the odds of selecting a low earner are decreased by
selecting a horse that has a ventricular septal wall width greater
than the mean ventricular septal wall width of a group of horses of
about the same age, sex and weight, with the odds of selecting a
low earner even further reduced by selecting a racehorse candidate
that has a ventricular septal wall width that is in the 75.sup.th
percentile or higher.
[0044] Similarly, by selecting racehorse candidates on the basis of
measurements for other variables and/or combinations of variables
discussed herein, including LVD, LVS, SPLN, HTWT, SW and PS, the
likelihood of selecting a candidate that will be a high earner may
be increased, and the likelihood of selecting a candidate that will
be a low earner will be reduced. Thus, certain embodiments of the
present invention are directed to methods that comprise selecting
horses that exhibit one or more of the aforementioned measurements
greater than the mean measurement, and preferably fall in the
75.sup.th percentile or higher, than is seen in a group of horses
of about the same age, sex and weight.
[0045] Additionally, as discussed in more detail in the examples to
follow, using the methods of the present invention, in addition to
increasing the likelihood of identifying a racehorse candidate that
will be a future high earner racehorse, the methods of the present
invention can also be used to increase the likelihood of selecting
a horse that will be a high earner router, as opposed to a high
earner sprinter. As used herein, a "sprint" is a race of 1 mile (8
furlongs) or less, while a "route" race is one of at least about
8.5 furlongs. Since the majority of graded stakes races for
Thoroughbred horses in both the United States and Europe (i.e., the
races with the highest purses) are contested at distances of
greater than 1 mile, it may be advantageous to select a racehorse
candidate with an increased likelihood of being a high earner
router.
[0046] Moreover, after conducting measurements on over 7000
Thoroughbred yearling and two-year old racing candidates, and
subsequently following their racing careers (as discussed more
fully in the examples to follow) it has been found that extremely
high earners, i.e. horses that earned at least $250,000 by the end
of their three-year old year, were about three times more likely to
have above average HTWT and LVD measurements, when compared to a
group of horses of about the same age, sex and weight, than to have
below average HTWT and LVD measurements. Accordingly, these
variables may be used in the methods of the present invention to
increase the likelihood of selecting a racehorse candidate that
will be an extremely high earner.
[0047] The present invention also provides methods for maintaining
a horse registry system or database. Such a system can be managed
using bioinformatics. Bioinformatics is the study and application
of computer and statistical techniques to the management of
biological information. Thus, in one embodiment, the present
invention provides a method for populating a database with the
biological information obtained using the methods of the present
invention. For example, a database can be populated with LVD, LVS,
PS, HTWT, SPLN and ventricular septal wall width measurements from
a group of horses whose racing abilities are known. Once a database
of sufficient size has been generated, the racing ability of
racehorse candidates can be predicted as described above, e.g., by
comparing measurements from racehorse candidates to corresponding
measurements from a group of horses of about the same age, sex, and
weight and ranking the horses according to each measurement.
Measurements from the racehorse candidates can be optionally
entered into the database as well.
[0048] In another embodiment, the present invention also provides
an apparatus for automating the methods of the present invention,
the apparatus comprising a computer and a software system capable
of comparing and standardizing echocardiographic and other
measurements from horses. The data is inputted in computer-readable
form and stored in computer-retrievable format. The present
invention also provides computer-readable medium encoded with a
data set comprising profiles, e.g., LVD, LVS, PS, HTWT, SPLN, and
ventricular septal wall width measurements, of horses known to be
high earners, low earners, high earner routers, or high earner
sprinters. The information in the data set can be used for
comparison purposes in order to improve one's odds of selecting a
higher earner racehorse. It can also be used by handicappers or
others in order to evaluate horses for betting purposes.
[0049] The methods described herein for obtaining certain
measurements from horses provides information which can be used to
determine the racing ability of candidate racehorses. Although the
data generated from the methods of this invention is suited for
manual review and analysis, in a preferred embodiment, prior data
processing using high-speed computers is utilized.
[0050] The invention also provides for the storage and retrieval of
a collection of profiles and comparisons in a computer data storage
apparatus, which can include magnetic disks, optical disks,
magneto-optical disks, DRAM, SRAM, SGRAM, SDRAM, RDRAM, DDR RAM,
magnetic bubble memory devices, and other data storage devices,
including CPU registers and on-CPU data storage arrays.
[0051] This invention also preferably provides a magnetic disk,
such as an IBM-compatible (DOS, Windows, Windows 95/98/2000,
Windows NT, OS/2, etc.) or other format, e.g., Linux, SunOS,
Solaris, AIX, SCO, Unix, VMS, MV, Mactinosh etc., floppy diskette
or hard (fixed, Winchester) disk drive, comprising a bit pattern
encoding data collected from the methods of the present invention
in a file format suitable for retrievable and processing in a
computerized comparison or relative quantification method.
[0052] The invention also provides a network, comprising a
plurality of computing devices linked via a data link, such as an
Ethernet cable (coax or 10BaseT), telephone line, ISDN line,
wireless network, optical fiber, or other suitable signal
transmission medium, whereby at least one network device comprises
a pattern of magnetic domains and/or charge domains comprising a
bit pattern encoding data acquired from the methods of the
invention.
[0053] The invention also provides a method for transmitting data
that includes generating an electronic signal on an electronic
communications device, such as a modem, ISDN terminal adapter, DSL,
cable modem, ATM switch, or the like, wherein the signal includes
(in native or encrypted format) a bit pattern encoding data
collected using the methods of the present invention.
[0054] In a preferred embodiment, the invention provides a computer
system for performing the methods of the present invention. A
central processor is preferably initialized to load and execute the
computer program for alignment and/or comparison of results. Data
is entered into the central processor via an I/O device. Execution
of the computer program results in the central processor retrieving
the data from the data file.
[0055] The target data or record and the computer program can be
transferred to secondary memory, which is typically random access
memory. For example, a central processor can be a conventional
computer; a program can be a commercial or public domain molecular
biology software package; a data file can be an optical or magnetic
disk, a data server, or a memory device; an I/O device can be a
terminal comprising a video display and a keyboard, a modem, an
ISDN terminal adapter, an Ethernet port, a punched card reader, a
magnetic strip reader, or other suitable I/O device.
[0056] The invention also provides the use of a computer system,
such as that described above, which comprises: (1) a computer; (2)
a stored bit pattern encoding a collection of measurements obtained
by the methods of the present invention, which may be stored in the
computer; (3) a comparison control; and (4) a program for
comparison.
[0057] All publications and patent documents cited above are hereby
incorporated by reference in their entirety for all purposes to the
same extent as if each were so individually denoted.
[0058] The below examples are non-limiting and for illustrating the
present invention. Alternatives and variations of the below
examples within the scope of the present invention as per the below
claims may be carried out by a person skilled in the art.
EXAMPLES
Example 1
Selecting the Group of Horses
[0059] Selected two dimensional echocardiographic (2DE)
measurements were recorded for 5,431 yearling and 2,003
two-year-old Thoroughbred racehorses between the ages of 12 through
28 months. These were unique, unraced horses. Cardiac measurements
were recorded primarily at select public yearling and two-year-old
auctions between 1995 and 2000.
[0060] All descriptive statistics used only the most current 2DE
measurements from each horse, in order to prevent multiple
measurements of the same horse from overly influencing statistics
within small groups of horses. Using the most recent measurement of
the same horse also maximized the number of two-year-olds available
for the study. Among the 7,434 unique horses, there were 2,940
fillies (40%), 4,494 colts (60%), 5,431 yearlings (73%), and 2,003
two-year-olds (27%).
[0061] Additionally, 5,909 horses (79%) were at least three years
of age by 1 Jan. 2000. Among these horses, by the end of their
three-year-old year, 1,156 (20%) raced outside of North America
(foreign) and 4,753 (80%) stayed in North American. Among the North
American horses, 1,073 (23%) never raced and 3,680 (77%) started at
least once (see Table 1).
1TABLE 1 Number of Races through Three-Year-Old Year Among Horses
Categorized as North American Number of Races Number Percent
through of of Cumulative 3-Year-Old Year Horses Total Percentage
Unraced 1,073 22.58 22.58 1-5 1,274 26.80 49.38 6-10 1,215 25.56
74.94 11-15 784 16.50 91.44 16-20 297 6.25 97.69 21-25 95 2.00
99.69 26-30 13 0.27 99.96 31-35 1 0.02 99.98 36-40 1 0.02 100.00
Total 4,753 100.00 100.00
[0062] Data from horses with resting heart rates above 40 beats per
minute was excluded from this study. The same technician,
ultrasound equipment and measurement protocol, as described in the
materials and methods section of this paper, was used for all
horses studied. Comments regarding physical appearance, body
condition, and conformation were recorded during each examination.
Table 2, below, provides some perspective to the figures in Table
1.
2TABLE 2 Averages for the Breed Worldwide Performances of Named
Thoroughbred Foals Born in North America between 1985-1994 (Source:
Thoroughbred Times, Jun. 8, 2002, p. 31) Foals by Foals of Top 1%
Subset of Population 1985-1994 of Sires % Starters/foals 68.9%
84.8% % Stakes winners/foals 3.2% 9.1% % Graded stakes
winners/foals 0.7% 3.6% % Grade 1 stakes winners/foals 0.2% 1.2% %
2-year-old starters/foals 33.5% 46.2% % 3-year-old starters/foals
59.0% 76.6% % 4-year-old starters/foals 44.0% 57.1% % 5-year-old
and up starters/foals 26.5% 36.9% Average career starts/foal 14.51
18.7 Average career starts/starter 21.1 22.0 Average win distance
in furlongs 6.82 7.24 Average earnings/starter $29,102 $71,349
Average earnings/start $1,378 $3,242 Note: Top 1% of sires
determined by total progeny earnings for 1985-1994.
[0063]
3TABLE 3 Sale to Racetrack Performance of 1990-1999 Graduates of
Major Yearling Sales (Source: Thoroughbred Times, Jul. 6, 2002, p.
20) Pct. SELECT YEARLING No. Median Starts Starts Pct Graded Avg.
AUCTION NAME & Horses Sale Avg. Per Per Pct. Stakes Stakes Win
LOCATION Sold Price Earnings Starters (%) Starter Foal Wnrs. Wnrs.
Wnrs. Dist. Fasig-Tipton Kentucky - July 1,792 $35,000 $61,132
1,577 (88.0%) 19.2 16.9 68.5% 6.9% 2.7% 6.99 Keeneland Kentucky -
July 1,945 235,000 112,752 1,672 (86.0%) 14.7 12.7 62.1% 11.3% 6.3%
7.91 Keeneland Kentucky - September 28,176 22,000 48,768 24,130
(85.6%) 20.0 17.1 64.9% 6.4% 2.0% 7.01 Fasig-Tipton Saratoga -
August 1,535 105,000 78,696 1,338 (87.2%) 16.8 14.7 65.8% 10.0%
4.5% 7.53
[0064]
4TABLE 4 Averages Among Horses in this Study Statistics through the
Three-Year-Old Year of Study Horses Horses not Known to have Raced
Outside of North America All Horses Subset of Study Population in
this Study % Stakes winners 6.25% % Graded stakes winners 2.90% %
Grade 1 stakes winners 1.09% % At Least Stakes Placed (including
winners) 12.57% % At Least Graded Stakes Placed (including winners)
5.14% % At Least Grade I Stales Placed (including winners)
1.48%
[0065] Note: Includes unraced horses. Race dollar amounts earned
can be compared between horses without currency or country
distortions. Compare percentages in this table to those of the top
1% of sires' progeny, shown in Table 2.
[0066] Performance records. All horses used to predict performance
had race records through their three-year-old year. Race records
included race date, racetrack, race number, distance raced, level
of race, claiming price, finish position and earnings. Horses that
raced outside of North America were identified as "foreign," and
their race records were not used, since they were often incomplete
or difficult to compare with North American records on the basis of
dollar value or race level.
[0067] Sample Bias. There were pedigree and conformation biases,
since the horses examined at "select" public auctions were
pre-selected by auction companies based on above-average commercial
assessment of pedigree and conformation. Not all horses at each
auction were measured, nor were subjects randomly selected. Horses
were further pre-selected horses for cardiac measurement based on
additional criteria.
[0068] Pre-selection biases were reflected in the percentage of
stakes winners among horses measured. For example, midway through
the 1990 foal crop's ten-year-old year, 2.3 percent had won a
stakes race (Thoroughbred Times, Jul. 22, 2000, p. 51). In
contrast, 6.7 percent of horses measured for this study, and which
were not known to have raced outside of North America, won a stakes
race before they were four years old.
Example 2
Measurement Equipment and Techniques
[0069] A Pie Medical, digital cineloop scanner 200 from Classic
Medical, (Tequesta, Fla.), with a 3.5 MHz annular array, multiring
crystal transducer with a 30-cm field of view at 22 frames per
second was used for all measurements. The depth of display varied
from 15 to 25 centimeters depending on the size of the horse. The
ultrasound recorder was equipped with electronic calipers that were
used to measure the stored images at the time of the
examination.
[0070] SAS release 6.12 (SAS Institute, Cary, N.C.), for Windows NT
(Microsoft) was used for statistical analysis. Universe (IBM) for
Windows 2000 (Microsoft) was used to manage the data. The server
was a Dell 2300 Poweredge (Dell, Atlanta, Ga.) with dual 450 MHz
Intel Pentium processors, running Windows 2000.
[0071] The 2DE imaging protocol was carried out on all horses, by
the same, experienced (>5 years) technician to reduce
measurement variability. Acoustical coupling gel (Aquasonic 100
ultrasound transmission gel--Parker, Fairfield, N.J.) was applied
liberally over the girth area in the 4th and 5th intercostal
spaces, starting just below the level of the point of the shoulder
down to the level of the olecranon. Three to five cardiac cycles
were measured for each variable. Measurements were not made if the
heart rate exceeded 40 beats/min., if the heart rhythm was
irregular, or if the images were unclear.
[0072] During 2DE examination, the ultrasound transducer was held
in the right hand with the cursor facing caudally. The left
forelimb was advanced slightly and the transducer was placed in the
4th or 5th left intercostal space, at a level just dorsal to the
point of the olecranon. From this position, a left parasternal
short axis view could be obtained by directing the transducer beam
perpendicular (horizontal) to the longitudinal cardiac axis. The
image provided a nearly circular appearance to the left ventricular
lumen. Moving (angling) the transducer beam from the apex to the
base of the heart, the moderator band(s), papillary muscle, chordae
tendinae and septal leaf of the mitral valve were identified and
then used as intracardiac reference points to obtain reproducible
cardiac images in the same tomographic plane.
[0073] Except where noted, the short axis images were projected
according to international terminology based on the recommendations
of the American Society of Echocardiography (Henry 1980, Feigenbaum
1986). Short axis images recorded from the left side of the chest
were projected as though the tomographic planes were viewed from
the base to the apex of the heart.
[0074] The 2DE measurements recorded for all 7,434 horses were
measured using electronic calipers. For all dimensions, the "inner
edge" method was used (Wyatt et al 1979, Weyman 1982, O'Grady,
Bonagura, Powers and Herring 1986), i.e., linear parameters were
measured from the inner edge of endocardial surfaces, and areas
were traced along the inner borders of the endocardial echoes.
[0075] The following variables, as shown and described in FIGS.
1-3, were measured from the stored images: Left ventricular cross
sectional area in diastole (LVD); left ventricular cross sectional
area in systole (LVS); interventricular septal wall structural
thickness in diastole (SW). Percent stroke volume (PS) was computed
using the formula: 4 PS = LVD - LVS LVD .times. 100
[0076] The ultrasound technician estimated HEIGHT and WEIGHT based
solely on visual inspection and prior experience. The variable
HTWT, which was the product of height times weight, was used in
this research as an estimate of overall body size. The ultrasound
technician, a life-long horseperson, trained horses prior to this
research. While a trainer, she had an on-site horse scale in a
40-stall training facility and took daily weight measurements of
horses, and compared scale results to weight tape measurements.
Alternatively, a five rating category system was used to describe
height and weight. For example, the horses were divided on the
basis of weight or height into the following five categories: well
below average (at least 1.0 standard deviation below the mean),
below average (from 0.5 to 1.0 standard deviations below the mean),
average (within 0.5 standard deviations of the mean), above average
(from 0.5 to 1.0 standard deviations above the mean), and well
above average (at least 1.0 standard deviation above the mean).
[0077] Each weight and height measurement was assigned a whole
number from 1 to 5, with 1 equal to "well below average" and 5
equal to "well above average." HTPLUSWT was created as the sum of
these weight and height ratings, providing an overall physical size
estimate. Each horse's cardiac measurements (i.e., LVD, LVS, SW,
and PS) were ranked (expressed as a percentile ranging from 0 to
100) relative to those of other horses of the same sex,
chronological age, and of the same 1-5 weight group. Stepwise and
discriminant results based on the 1-5 weight categories were
similar to those results based on estimation of pounds.
[0078] Most cardiac measurements varied depending upon age, sex and
weight, making it extremely difficult to compare horses on the
basis of cardiac measurements without simultaneously adjusting for
the effects of these parameters. Two statistical techniques,
percentiles and standardized scores, eliminated the effects of age,
sex and weight. These statistical techniques were only possible due
to the large number of horses studied. Percentiles and standardized
scores for LVD, LVS, SW, and PS were calculated by comparing the
subject horse to others that were:
[0079] The same sex as the subject horse
[0080] Measured within 30 days of chronological age of the subject
horse
[0081] Measured within 1 year of when the subject horse was
measured
[0082] Within 25 pounds of weight of the subject horse
[0083] Percentiles and standardized scores for WEIGHT, HEIGHT and
HTWT (HTWT is the product of height times weight) were calculated
as above, except without weight restrictions on the comparison
group.
[0084] Subject comparisons were limited to within .+-.1 year of the
measurement date in order to minimize the possible effects of
gradual small changes in calibration, methodology and external
variables acting on the subjects. Examples of external variables
that may have changed over time and affected measurements include
sales preparation techniques of horses at auctions, steroid use,
growth hormones, wear and tear on equipment, etc.
[0085] Technically, percentiles fail to maintain initial distances
between variables. Since most data in natural, biological phenomena
is located near the middle of the Gaussian-shaped distribution,
measurements in the 50.sup.th and 52.sup.nd percentiles are closer
in absolute value than those in the 95.sup.th and 97.sup.th
percentiles. Standardized scores described below maintain the
natural spacing between variables, producing a scale-free statistic
with a mean of 0, and a standard deviation of 1. 5 Standardized
Score = Observation - Mean Standard Deviation
[0086] Standardized scores could be difficult to interpret because,
while they generally ranged from -3 to +3, they tended to
congregate around zero. It seems easier to understand that a horse
is in the 70.sup.th percentile compared to his peers than to know
that his standardized score is 0.55.
[0087] Statistical analyses and tables in this text are based on
percentiles. The same analyses and tables in terms of standardized
scores produced virtually the same results (data not shown).
Example 3
Reproducibility and Sources of Measurement Variability
[0088] Variation (or differences) between cardiac measurements is
caused by a combination of within- and between-subject variation.
Within-subject variation, sometimes called measurement error,
indicates how accurately or reproducibly the technician and
equipment measures a given variable (hearts and horses are moving
targets). Between-subject variation is the range of expected
differences among a particular variable in the general population
that isn't due to error. Between-subject variation accounted for
84-92% of variation in cardiac measurements in this study, while
within-subject variation accounted for 8-16% of variation.
[0089] Measurement variability was calculated for LVD, LVS, and SW
among 1,464 horses measured in 1999. These cardiac measurements
were repeated at least three times within a period of a few
minutes. [1,571 horses were measured in 1999. Those excluded from
this variability study lacked at least three measurements for LVD,
LVS, or SW because of auction conditions, during which the
technician may have lacked time to repeat measurements, could not
sustain a resting heart rate (or behavioral cooperation), or
reported only the average.]
[0090] Table 5 summarizes between-subject variation (s.sub.B) and
within-subject variation (s.sub.W) and shows some basic statistical
equations used. Column 1 lists the variables studied. Column 2
lists the mean value of each variable among all 1,464 horses in
this part of the study. Column 3 lists between-subject variation,
which is the standard deviation associated with the mean reported
in Column 2. Column 4 lists within-subject variation. Column 5
lists total variation. Column 6 lists the percentage of total
variation due to within-subject variation (or measurement error).
S.sub.B.sup.2 and S.sub.W.sup.2, as used in the equations, are mean
squared error terms from the between- and within-subject groups
studied.
5TABLE 5 Summary of Measurement Variability for Combined Sexes (LVD
and LVS units = mm.sup.2, SW units = mm) {circle over (6)} {circle
over (1)}Variable {circle over (2)}Population Mean (n = 1,464)
{circle over (3)}6 Between-Subject Variation s B 2 {circle over
(4)}7 Within-Subject Variation s W 2 {circle over (5)}8 Total
Variation S W 2 + S B 2 9 Pct . of Variation due toWithin-Subject
Variation s W 2 s W 2 + s B 2 .times. 1 00 LVD 13,282 1,490 424
1,549 7.50% LVS 4,329 496 215 540 15.81% SW 55.5 4.54 1.98 4.96
15.97%
[0091] The within-subject variations listed in Table 5 were used to
compute confidence intervals as reported in Table 6, and to answer
the following questions:
[0092] (1) How accurately did a single cardiac measurement (i.e.,
not an average of measurements repeated over a period of a few
minutes) describe the true value? A statistical solution is to use
the "95% confidence interval for a single measurement," as shown in
Column 2 of Table 6. In this example, the value was 831 mm.sup.2
for LVD. This means that there is a 95% probability that the true
LVD lies within 831 mm.sup.2 of a single LVD measurement.
[0093] (2) How accurately did the mean of three repeated cardiac
measurements over the course of a few minutes describe the true
value? A statistical solution is to utilize the "95% confidence
interval for repeated measurements," as shown in Column 3 of Table
6. For example, this value was 480 mm.sup.2 for LVD. This means
that there is a 95% probability that the true LVD is within 480
mm.sup.2 of the mean of three repeated LVD measurements.
[0094] (3) How much of a difference between cardiac measurements
over some period of time would rule out measurement error as the
sole source of the difference? A statistical solution is to use the
"95% confidence interval for repeated measurements from two
separate dates," as shown in Column 4 of Table 6. For example, this
value was 679 mm.sup.2 for LVD. This means that if the difference
between LVD measurements on different dates exceeded 679, then
there is a 95% probability that measurement error was not the sole
source of that difference.
[0095] (4) How much of a difference between repeated cardiac
measurements of two different horses would rule out measurement
error as the sole source of the difference? A conservative
statistical solution is to use the 95% confidence interval just
mentioned, as listed in Column 4 of Table 6. For example, this
value was 679 mm.sup.2 for LVD. This means that if the difference
between horses' LVDs exceeded 679 mm.sup.2, then there is a 95%
probability that measurement error was not the sole source of that
difference.
6TABLE 6 95% Confidence Intervals (C.I.) Associated with
Within-Subject Variations Reported in Table 5 (LVD and LVS units
mm.sup.2, SW units = mm) {circle over (3)} {circle over (4)}
{circle over (1)}Variable {circle over (2)}10 95 % C . I . for a
Single Measurement 1.96 S W 2 11 95 % C . I . for a Measurement
Repeated 3 Times Over the Course of a Few Minutes ( n = 3 ) 1.96 s
w 2 n 12 95 % C . I . for a Horse Measured on 2 Separate Dates or
for Measurements of 2 Different Horses ( n = 3 ) 1.96 2 s w 2 n LYD
831 480 679 LYS 421 243 344 SW 3.9 2.3 3.2
[0096] The average percent change in cardiac measurements for
horses measured twice within the same month of age was calculated
for horses 14-17 months of age. These were the only individual
months of age with at least five different horses represented.
[0097] The change in the cardiac measurements of these horses fell
within the range of expected measurement error described in Table
5. Most change was positive, indicating that growth may have
occurred in addition to measurement variation. Measurement
variation among horses measured twice within the same month was
also influenced by other factors, e.g., some horses were
re-measured because the ultrasound technician wasn't satisfied with
the initial measurement, likely due to the horse's behavior (i.e.,
suspected illness, medications, or elevated heart rate after start
of exam).
Example 4
Statistical Overview of Cardiac Data--Descriptive Statistics
[0098] Means and standard deviations of cardiac raw data for
combined sexes are presented in Table 7-Table 8 by months of
age.
7TABLE 7 Means of Cardiac Measurements for Combined Sexes - by
Months of Age (Total n = 7,434) Number Cardiac Measurement Means
Months of Age LVD LVS SW PS WEIGHT HEIGHT of Age Horses SPLN
(months) (mm.sup.2) (mm.sup.2) (mm) (pct.) (lbs.) (hands) 12 81 171
12.5 11,534 3,823 49.4 66.82 801 14.52 13 155 174 13.5 12,025 3,982
50.7 66.86 875 14.87 14 399 175 14.6 12,362 4,038 50.9 67.32 944
15.12 15 758 176 15.6 12,395 4,024 51.1 67.52 970 15.26 16 1,279
178 16.5 12,689 4,133 51.9 67.41 986 15.37 17 1,196 182 17.5 12,843
4,182 52.4 67.41 996 15.42 18 856 179 18.5 12,948 4,203 52.4 67.52
1,001 15.47 19 551 186 19.4 13,285 4,330 53.5 67.36 1,005 15.51 20
248 190 20.5 13,504 4,431 53.9 67.16 1,013 15.59 21 337 196 21.5
13,428 4,344 54.0 67.64 1,016 15.60 22 440 201 22.5 13,633 4,411
54.7 67.64 1,026 15.64 23 485 200 23.5 13,706 4,384 54.7 68.04
1,032 15.69 24 333 198 24.5 13,646 4,413 54.6 67.65 1,036 15.65 25
184 201 25.4 13,803 4,409 55.2 68.08 1,046 15.77 26 95 199 26.5
13,657 4,420 54.6 67.66 1,045 15.77 27 37 215 27.3 13,638 4,415
55.1 67.65 1,036 15.75
[0099]
8TABLE 8 Standard Deviations of Cardiac Measurements for Combined
Sexes - by Months of Age Corresponding to Means shown in Table 7
(Total n = 7,434) Number Cardiac Measurement Standard Deviations
Months of Age LVD LVS SW PS WEIGHT HEIGHT of Age Horses SPLN
(months) (mm.sup.2) (mm.sup.2) (mm) (pct.) (lbs.) (hands) 12 81
26.8 0.2791 1,232 424 4.09 1.989 116.2 0.592 13 155 28.3 0.2924
1,392 505 4.46 2.206 104.6 0.493 14 399 33.4 0.2897 1,408 519 5.12
2.329 73.9 0.394 15 758 34.1 0.2840 1,548 553 5.16 2.223 59.6 0.357
16 1,279 36.5 0.2841 1,567 560 5.50 2.297 50.9 0.355 17 1,196 38.1
0.2902 1,541 551 5.29 2.328 49.4 0.369 18 856 37.4 0.2783 1,595 581
5.45 2.470 50.1 0.368 19 551 39.5 0.2800 1,504 526 5.24 2.278 49.4
0.367 20 248 38.9 0.2898 1,347 494 4.34 2.204 45.8 0.358 21 337
45.7 0.2908 1,459 545 4.84 2.239 46.1 0.367 22 440 46.1 0.2760
1,404 547 4.59 2.283 41.7 0.354 23 485 44.3 0.2889 1,366 554 4.56
2.202 44.1 0.359 24 333 46.9 0.2965 1,493 587 4.63 2.819 48.2 0.366
25 184 45.6 0.2886 1,519 606 4.53 2.657 37.1 0.362 26 95 46.3
0.2887 1,410 557 4.74 2.076 36.7 0.328 27 37 48.0 0.2498 1,590 606
4.74 2.081 40.5 0.375
[0100] Growth Curves--FIG. 4 compares LVD for colts vs. fillies,
and is typical of sex-related differences. Most growth curves were
described well (R.sup.2.gtoreq.0.90) by second-degree polynominal
equations, as shown on the graphs. The growth curves should be
limited to application over the period from 12 through 27 months of
age for which they were calculated (i.e., not used to estimate
average LVD at 32 months of age).
[0101] Anomalies appeared in the data patterns of cardiac
measurements versus age at 20 and 21 months of age. These horses
were primarily measured during October through December, between
the timing of select yearling and select two-year-old auctions.
Horses often enter training during those interim months. Training
regimens, and thus each heart's response to training, likely varied
greatly during this time (Young, 1999). Puberty may play a role
among fillies at this age. Most horses were measured during this
period at private farms, without any pre-selection based on
conformation or pedigree. The ratio of colts to fillies (60% colts
to 40% fillies) in this study closely matches those at auctions.
This ratio may favor colts because breeding farms keep some of the
best-bred, best-conformed fillies for their breeding programs.
Therefore, relative to auctions, the fillies seen at private farms
may be of higher quality, overall, since they may include the
best-bred, best-conformed fillies that never make it to
auctions.
[0102] Among the total population measured, generally, the bigger
the horse, the bigger its heart, all else being equal. FIGS. 5-8
provide a visual overview of the relationship between cardiac
measurements and the weight of the horse. FIGS. 9-10 compare LVD
and WEIGHT percentiles for high earners vs. low earners, and are
typical of performance-related differences (except for PS). Not
only were future high earners heavier than low earners, but, even
when normalized by sex, age and weight, high earners still had
higher cardiac measurements. Sample sizes of high earners in these
graphs were small at 19 and 20 months of age.
Example 5
Statistical Overview of Cardiac Data--T-Tests
[0103] T-tests compared high vs. low earners of combined sexes and
ages, using data normalized for sex, age and size. Significant
differences (P-values .ltoreq.0.0001) existed between high and low
earners for all of the cardiac parameters listed in Table 9, except
for PS. Stepwise analysis, as discussed in this paper, identified
SW or SPLN, LVS and HTWT as the most significant discriminant
variables when differentiating between high and low earners. High
earners were defined as horses that raced at least three times,
with earnings per start of at least $10,000.
9TABLE 9 T-tests - Percentiles (Data Adjusted for Age, Sex and
Weight) High Earners (Earnings Per Start .gtoreq. $10,000) vs. Low
Earners (Earnings Per Start .ltoreq. $2,000) LOW EARNERS HIGH
EARNERS VARS n Mean SD n Mean SD P-Value LVD 1061 45.93 28.61 418
53.12 28.32 0.0000 LVS 1061 46.45 28.61 418 52.72 28.43 0.0001 SW
1061 46.22 27.60 418 53.29 27.17 0.0000 PS 1061 50.09 29.56 418
49.89 29.03 0.9050 SPLEEN 1025 42.78 22.99 405 48.87 22.71 0.0000
WEIGHT 1091 47.25 29.40 424 60.11 27.38 0.0000 HEIGHT 1091 53.47
28.58 424 65.81 26.07 0.0000 HTWT 1091 45.32 29.02 424 58.53 27.24
0.0000
[0104] T-tests also compared high earner routers vs. high earner
sprinters of combined sexes and ages, using data standardized for
horses of the same age, sex and size. Significant differences
(P-values .ltoreq.0.05) existed between high earner routers and
sprinters for the cardiac variables of LVD, LVS, WEIGHT, HEIGHT and
HTWT, as shown in Table 10. Stepwise analysis, as discussed in this
paper, identified LVD, LVS, HTWT and PS as the most significant
discriminant variables when differentiating between high earner
routers and sprinters. High earner routers raced at least three
times at distances of at least 8.5 furlongs, with earnings per
start at those route distances of at least $10,000. High earner
sprinters raced at least three times at distances below seven
furlongs, with earnings per start of at least $10,000 at those
sprint distances.
10TABLE 10 T-Tests - Percentiles (Data Adjusted for Age, Sex and
Weight) High Earner Routers (Raced .gtoreq. 8.5 Furlongs) vs. High
Earner Sprinters (Raced < 7 Furlongs) SPRINTERS ROUTERS VARS n
Mean SD n Mean SD P-Value LVD 180 48.68 28.69 134 56.95 27.83
0.0110 LVS 180 47.68 28.31 134 58.17 28.11 0.0012 SW 180 52.08
27.51 134 56.60 27.27 0.1495 PS 180 53.04 29.31 134 47.79 27.55
0.1085 SPLEEN 176 47.38 22.84 128 50.04 23.24 0.3207 WEIGHT 180
55.10 26.99 134 64.12 24.11 0.0024 HEIGHT 180 59.99 26.83 134 69.51
24.60 0.0014 HTWT 180 52.85 27.15 134 63.06 24.50 0.0007
Example 6
Statistical Overview of Cardiac Data--Standardizing Cardiac
Measurements to Eliminate the Effects of Age Sex and Weight
[0105] The high correlation between age and size vs. most cardiac
measurements (see Table 11) was not present among percentiles (see
Table 12). Therefore, when looking at horses of different age, sex
and size, it is possible to compare their cardiac measurements by
standardizing their data (i.e., using percentiles, as described on
page 8). To measure the correlation between age and size vs. most
cardiac measurements, Pearson correlation coefficients (r) were
computed between cardiac measurements for raw data and percentiles
(standardized for sex, age and weight). Tests for significance of
correlation coefficients produced P-Values<0.0001. Correlation
coefficients were squared and multiplied times 100 to compute
coefficients of determination (R2), as shown in Table 11-Table
12.
[0106] Essentially, due to the standardization of the data, where
there was a high degree of correlation throughout the MONTHS column
and bottom three rows (WEIGHT, HEIGHT and HTWT) of Table 11, there
was little correlation shown in the same column and rows of Table
12 (see shaded areas).
11TABLE 11 Coefficients of Determination (R.sup.2) (n ranged
between 7,288-7,434) Among Raw Data (Unadjusted for Sex, Age and
Weight) Coefficients of Determination for Cardiac Measurements -
Raw Data Variables MONTHS SPLEEN LVD LVS SW PS WEIGHT HEIGHT LVD 9
LVS 6 74 SW 6 70 51 PS 1 1 19 1 SPLEEN 5 33 28 50 WEIGHT 22 3 21 14
10 1 HEIGHT 19 3 23 16 13 1 73 HTWT 22 3 24 15 12 1 97 86
[0107]
12TABLE 12 Coefficients of Determination (R.sup.2) (n ranged
between 7,288-7,434) Among Percentiles (Standardized for Sex, Age
and Weight) Coefficients of Determination for Cardiac Measurements
- Percentiles Variables MONTHS SPLEEN LVD LVS SW PS WEIGHT HEIGHT
LVD 0 LVS 0 62 SW 0 53 30 PS 0 1 20 2 SPLEEN 0 24 14 35 1 WEIGHT 0
0 0 0 0 0 HEIGHT 0 0 1 0 0 0 67 HTWT 0 0 0 0 0 0 96 79
Example 7
Statistical Overview of Cardiac Data--Covariance Analysis of Means
to Assess the Effects of Age and Sex on Cardiac Measurements
[0108] Analysis of covariance showed that age- and weight-adjusted
means for cardiac measurements were significantly different
(P-values .ltoreq.0.01) between colts and fillies 12 through 27
months old, as shown in Table 13.
13TABLE 13 Means Adjusted for Age and Weight LS MEANS VARIABLES
COLTS FILLIES P-VALUES N LVD 13,315 12,832 0.0001 7,434 LVS 4,318
4,179 0.0001 7,434 SW 53.86 52.12 0.0001 7,434 PS 67.55 67.41
0.0100 7,434 SPLEEN 193 183 0.0001 7,220
[0109] Analysis of covariance showed that sex- and weight-adjusted
means for cardiac measurements were sometimes significantly
different between horses of different months of age. The
significance of differences varied depending on the variables
studied and the number of months apart. In most cases, significant
differences (P-values .ltoreq.0.05) were rare or weak when
comparing yearlings to yearlings, or two-year-olds to
two-year-olds, while differences were significant when comparing
yearlings to two-year-olds.
Example 8
Stepwise Discriminant Analysis of the Relationship of Cardiac
Measurements to Performance
[0110] It is impossible to know the level of ability of most horses
measured that subsequently never raced, or raced just a couple of
times. For this reason, when forming groups of high vs. low earners
or routers vs. sprinters, horses used had to have raced at least
three times. Raising the minimum number of starts (up to 6) did not
improve or weaken discriminant analyses.
[0111] Stepwise analysis was conducted for colts, fillies and
combined sexes, using percentiles for the variables: LVD, LVS, SW,
PS, SPLN and HTWT (HTWT is the product of height times weight).
[0112] Stepwise analysis was used to identify statistically
significant variables that could differentiate between groups of
horses categorized as high and low earners, defined as:
[0113] High Earners. Raced at least three times, with earnings per
start of at least $10,000.
[0114] Low Earners. Raced at least three times, with earnings per
start of $2,000 or less.
[0115] Among the horses in this study that raced at least three
times in North America, 34 percent earned $2,000 or less per start
(categorized as "low" earners) and 13 percent earned at least
$10,000 per start (categorized as "high" earners). Table 2 provides
average worldwide performance statistics for the Thoroughbred
breed.
[0116] For high vs. low earners, stepwise analysis identified the
following significant variables (listed in order of statistical
significance):
[0117] Combined Sexes. HTWT, SW or SPLN, LVS
[0118] Colts. HTWT, SW or SPLN
[0119] Fillies. HTWT, SW or SPLN
[0120] Stepwise analysis was used to identify statistically
significant variables that could differentiate between groups of
horses categorized as high earner sprinters and high earner
routers, defined as:
[0121] High Earner Sprinters. Raced at least 3 times at distances
<7.0 furlongs, earned at least $10,000 per start at distances
<7.0 furlongs, and earned less than $2,000 per start at
distances .gtoreq.8.5 furlongs.
[0122] High Earner Routers. Raced at least 3 times at distances
.gtoreq.8.5 furlongs, and earned at least $10,000 per start at
distances .gtoreq.8.5 furlongs.
[0123] For high earner sprinters vs. high earner routers, stepwise
analysis identified the following significant variables (listed in
order of statistical significance):
[0124] Combined Sexes. HTWT, LVS
[0125] Colts. LVD, HTWT
[0126] Fillies. PS, HTWT
Example 9
Multivariate Discriminant Analysis of the Relationship of
Measurements to Performance
[0127] Discriminant analysis was used to classify high earners vs.
low earners, and high earner routers vs. high earner sprinters, as
defined in the stepwise analysis section.
[0128] Discriminant results were based on a classification
threshold of 50%. A classification threshold is the minimum
acceptable probability (as defined by the model user) required to
classify a horse into a particular group. Thus, no horse was
classified into a group unless the models assigned it at least a
50% probability of belonging to that group. Generally, the higher
the threshold, the better the models performed (i.e., a horse with
a 70% high earner probability was more likely to be a high earner
than a horse with a lower probability. As the threshold increases
for a particular group, the models generally misclassify more
members of that group. At public auctions, a high "high earner"
threshold would minimize the chances of buying poor performers
(Type II errors), while increasing the chances of rejecting good
performers (Type I errors).
[0129] Z-statistics were computed to determine the reliability of
discriminant results using the formula below (shown for high
earners): 13 Z H = P H post - P H pre P H pre ( 1 - P H pre ) N CH
post
[0130] Where:
[0131] N.sub.Hpre=Number of high earners in model
[0132] N.sub.Tpre=Total number of horses in model
[0133] N.sub.HCCPost=Number of high earners correctly classified by
model
[0134] N.sub.CHpost=Number of horses classified as high earners by
model
[0135] P.sub.Hpre=Pre-model probability (N.sub.Hpre/N.sub.Tpre)
[0136] P.sub.Hpost=Post-model probability
(N.sub.HCCpost/N.sub.CHpost)
[0137] P-values associated with the Z-statistics were reported in
place of Z-statistics (Table 14-Table 25).
[0138] The model parameters were:
[0139] Horses had to be born by 1997 (so would have racing data
through three-year-old year)
[0140] Horses had to have at least 3 starts (i.e., sound enough to
race multiple times)
[0141] Two types of discriminant analyses, called blind and
non-blind tests, were conducted for each model.
[0142] Non-Blind Test. A non-blind test is one in which the horses
classified by a model were used to create the model. Thus, the
models "saw" those horses before. A non-blind test is the best-case
scenario of how well a model performs.
[0143] Blind Test. A blind-test is one in which the horses
classified by a model were not used to create the model. Thus, the
models did not "see" those horses before.
[0144] Three tables were used to summarize each discriminant
analysis in terms of blind and non-blind tests. The first table
presents non-blind test results based on all horses available for
the study. The second table presents non-blind test results based
on horses with names beginning with the letters A-M. The third
table presents blind-test results, for which the A-M model was used
to classify horses with names beginning with the letters N-Z, which
the models hadn't seen previously. Each table presents summary
statistics as described below:
[0145] Pre-Model Probability. Discriminating between two groups (A
and B), the pre-model probability is the ratio of all Group A or
Group B horses to the total number of horses in the model. This is
the probability, using a random selection technique without
statistically created models, of correctly classifying a Group A or
Group B horse. This probability is shown as a Ratio and a Percent.
For example, if there are 7 Group A horses and 93 Group B horses,
there is a 7% probability of randomly selecting a Group A horse.
For Group A horses, this would be shown as a ratio of 7/100 and as
a percent of 7.00.
[0146] Post-Model Probability. Discriminating between two groups (A
and B) the post-model probability is the ratio of Group A or Group
B horses correctly classified by the models to the total number of
horses classified by the statistically created models as Group A or
Group B horses. This is the probability with discriminant models of
correctly classifying Group A or Group B horses. Using the example
above, a discriminant model classifying the same 100 horses might
classify 25 horses into Group A, of which 5 horses actually
belonged to Group A. In this case, the ratio for Group A horses
would be 5/25, or 20 percent. Thus, in this example, the
discriminant models improved the odds of correctly identifying
Group A horses from 7% without models to 20% with models. Likewise,
they improved the odds of correctly classifying Group B horses from
93% without models to 73/75, or 97.3% with models.
[0147] P-value. The P-value was listed corresponding to the
Z-statistic computed for each model.
[0148] The variables HTWT, SW, and LVS, as identified by stepwise
analysis, were used in the following high earner vs. low earner
discriminant models, e.g., Tables 14-25. The predictive results of
the blind and non-blind tests were similar. Results showed that as
long as data was first standardized (using percentiles) for each
subject's sex, age and size, each subject's data could be compared
with data from subjects of different sex, age and size. This made
the combined sexes discriminant models just as powerful as separate
colt and filly models.
[0149] Further comparisons of earnings groups, including $10,000+
earnings per start vs. less than $7,500- earnings per start
produced similar results. Thus, the models, using the same
independent variables, successfully differentiated between stakes-
and allowance-caliber horses, as well as between stakes- and
claiming-caliber horses.
[0150] Horses generally earned more and raced less frequently as
the probability of being high earners, as assigned by the
discriminant model, increased.
[0151] Table 14-Table 16 summarize discriminant results for
non-blind and blind tests of high earners and low earners,
comprised of colts and fillies combined, that had raced at least
three times (i.e., had three "starts"). High earners earned at
least $10,000 per start and low earners earned $2,000 or less per
start. The improvement associated with discriminant modeling was
statistically significant for both high and low earners for all
groups studied (P-values .ltoreq.0.0027).
[0152] Non-Blind A-Z. Table 14 shows that among 1,479 total horses,
non-blind discriminant models improved the odds of correctly
classifying high earners from 28.26% without models to 37.32% with
models. They improved the odds of correctly classifying low earners
from 71.74% without models to 79.57% with models. The improvement
associated with discriminant modeling was statistically significant
for both high and low earners (P-values <0.0001).
14TABLE 14 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests -- Combined Sexes - Names Starting with
Letters A-Z Pre-Model Post-Model Probability Probability Category
Ratio Pct. Ratio Pct. P-Value High Earners 418/1479 28.26 256/686
37.32 0.0000 Low Earners 1061/1479 71.74 631/793 79.57 0.0000
[0153] Non-Blind A-M. Table 15 shows that among horses with names
beginning with the letters A-M, non-blind discriminant models
improved the odds of correctly classifying high earners from 27.75%
without models to 37.65% with models. They improved the odds of
correctly classifying low earners from 72.25% without models to
80.80% with models. The improvement associated with discriminant
modeling was statistically significant for both high and low
earners (P-values <0.0001).
15TABLE 15 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests -- Combined Sexes - Names Starting with
Letters A-M Pre-Model Post-Model Probability Probability Category
Ratio Pct. Ratio Pct. P-Value High Earners 245/883 27.75 154/409
37.65 0.0000 Low Earners 638/883 72.25 383/474 80.80 0.0000
[0154] Blind N-Z. Table 16 shows that among horses with names
beginning with the letters N-Z, blind discriminant models based on
the A-M horses improved the odds of correctly classifying high
earners from 29.03% without models to 37.77% with models. They
improved the odds of correctly classifying low earners from 70.97%
without models to 78.62% with models. The improvement associated
with discriminant modeling was statistically significant for both
high and low earners (P-values .ltoreq.0.0027).
16TABLE 16 Discriminant Model Results - High Earners vs. Low
Earners Blind Test - Combined Sexes - Names Starting with Letters
N-Z Pre-Model Post-Model Probability Probability Category Ratio
Pct. Ratio Pct. P-Value High Earners 173/596 29.03 105/278 37.77
0.0013 Low Earners 423/596 70.97 250/318 78.62 0.0027
[0155] Table 17-Table 19 summarize discriminant results for high vs
low earners among colts. Table 17 shows that among 880 colts,
non-blind discriminant models improved the odds of correctly
classifying high earners from 26.70% without models to 34.96% with
models. They improved the odds of correctly classifying low earners
from 73.30% without models to 80.47% with models. The improvement
associated with discriminant modeling was statistically significant
for both high and low earners (P-values .ltoreq.0.0004).
17TABLE 17 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests - Colts - Names Starting with Letters A-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 235/880 26.70 143/409 34.96 0.0002
Low Earners 645/880 73.30 379/471 80.47 0.0004
[0156] Non-Blind A-M. Table 18 shows that among colts with names
beginning with the letters A-M, non-blind discriminant models
improved the odds of correctly classifying high earners from 26.47%
without models to 33.33% with models. They improved the odds of
correctly classifying low earners from 73.53% without models to
79.51% with models. The improvement associated with discriminant
modeling was statistically significant for both high and low
earners (P-values .ltoreq.0.0226).
18TABLE 18 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests - Colts - Names Starting with Letters A-M
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 140/529 26.47 82/246 33.33 0.0147
Low Earners 389/529 73.53 225/283 79.51 0.0226
[0157] Blind N-Z. Table 19 shows that among colts with names
beginning with the letters N-Z, blind discriminant models based on
the A-M horses improved the odds of correctly classifying high
earners from 27.07% without models to 38.41% with models. They
improved the odds of correctly classifying low earners from 72.93%
without models to 82.89% with models. The improvement associated
with discriminant modeling was statistically significant for both
high and low earners (P-values .ltoreq.0.0022).
19TABLE 19 Discriminant Model Results - High Earners vs. Low
Earners Blind Test - Colts - Names Starting with Letters N-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 95/351 27.07 63/164 38.41 0.0011
Low Earners 256/351 72.93 155/187 82.89 0.0022
[0158] Table 20-Table 22 summarize discriminant results for high
vs. low earners among fillies.
[0159] Non-Blind A-Z. Table 20 shows that among 599 fillies,
non-blind discriminant models improved the odds of correctly
classifying high earners from 30.55% without models to 42.22% with
models. They improved the odds of correctly classifying low earners
from 69.45% without models to 79.03% with models. The improvement
associated with discriminant modeling was statistically significant
for both high and low earners (P-values .ltoreq.0.0002).
20TABLE 20 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests - Fillies - Names Starting with Letters A-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 183/599 30.55 114/270 42.22 0.0000
Low Earners 416/599 69.45 260/329 79.03 0.0002
[0160] Non-Blind A-M. Table 21 shows that among fillies with names
beginning with the letters A-M, non-blind discriminant models
improved the odds of correctly classifying high earners from 29.66%
without models to 44.16% with models. They improved the odds of
correctly classifying low earners from 70.34% without models to
81.50% with models. The improvement associated with discriminant
modeling was statistically significant for both high and low
earners (P-values .ltoreq.0.0005).
21TABLE 21 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests - Fillies - Names Starting with Letters A-M
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 105/354 29.66 68/154 44.16 0.0001
Low Earners 249/354 70.34 163/200 81.50 0.0005
[0161] Blind N-Z. Table 22 shows that among fillies with names
beginning with the letters N-Z, blind discriminant models based on
the A-M horses improved the odds of correctly classifying high
earners from 31.84% without models to 39.50% with models. They
improved the odds of correctly classifying low earners from 68.16%
without models to 75.40% with models. The improvement associated
with discriminant modeling was not statistically significant for
high or low earners (P-values .ltoreq.0.0819).
22TABLE 22 Discriminant Model Results - High Earners vs. Low
Earners Blind Test - Fillies - Names Starting with Letters N-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 78/245 31.84 47/119 39.50 0.0735
Low Earners 167/245 68.16 95/126 75.40 0.0819
[0162] The variables HTWT and LVS, as identified by stepwise
analysis, were used in the high earner routers vs. sprinters
discriminant models for combined sexes (see Exhibits PM05290204
& PM05290205).
[0163] Table 23-Table 25 summarize discriminant results for high
earner routers vs. sprinters. Table 23 shows that among 314 high
earner horses, non-blind discriminant models improved the odds of
correctly classifying routers from 42.68% without models to 55.03%
with models. They improved the odds of correctly classifying
sprinters from 57.32% without models to 68.48% with models. The
improvement associated with discriminant modeling was statistically
significant for both routers and sprinters (P-values
.ltoreq.0.0037).
23TABLE 23 Discriminant Model Results - High Earner Routers vs.
High Earner Sprinters Non-Blind Tests - Combined Sexes - Names
Starting with Letters A-Z Pre-Model Post-Model Probability
Probability Category Ratio Pct. Ratio Pct. P-Value Routers 134/314
42.68 82/149 55.03 0.0023 Sprinters 180/314 57.32 113/165 68.48
0.0037
[0164] Non-Blind A-M. Table 24 shows that among high earner horses
with names beginning with the letters A-M, non-blind discriminant
models improved the odds of correctly classifying routers from
37.78% without models to 51.85% with models. They improved the odds
of correctly classifying sprinters from 62.22% without models to
73.74% with models. The improvement associated with discriminant
modeling was statistically significant for both routers and
sprinters (P-values .ltoreq.0.0183).
24TABLE 24 Discriminant Model Results - High Earner Routers vs.
High Earner Sprinters Non-Blind Tests - Combined Sexes - Names
Starting with Letters A-M Pre-Model Post-Model Probability
Probability Category Ratio Pct. Ratio Pct. P-Value Routers 68/180
37.78 42/81 51.85 0.0091 Sprinters 112/180 62.22 73/99 73.74
0.0183
[0165] Blind N-Z. Table 25 shows that among high earner horses with
names beginning with the letters N-Z, blind discriminant models
based on the A-M horses improved the odds of correctly classifying
routers from 49.25% without models to 60.94% with models. They
improved the odds of correctly classifying sprinters from 50.75%
without models to 61.43% with models. The improvement associated
with discriminant modeling was not statistically significant for
routers or sprinters (P-values .ltoreq.0.0735).
25TABLE 25 Discriminant Model Results - High Earner Routers vs.
High Earner Sprinters Blind Test - Combined Sexes - Names Starting
with Letters N-Z Pre-Model Post-Model Probability Probability
Category Ratio Pct. Ratio Pct. P-Value Routers 66/134 49.25 39/64
60.94 0.0588 Sprinters 68/134 50.75 43/70 61.43 0.0735
[0166] The variables HTWT, SPLEEN and LVS, as identified by
stepwise analysis, were used in the following high earner vs. low
earner discriminant models, e.g., Tables 26.
[0167] Table 26-Table 28 summarize discriminant results for
non-blind and blind tests of high earners and low earners,
comprised of colts and fillies combined, that had raced at least
three times (i.e., had three "starts"). High earners earned at
least $10,000 per start and low earners earned $2,000 or less per
start. The improvement associated with discriminant modeling was
statistically significant for both high and low earners for all
groups studied (P-values .ltoreq.0.0002).
[0168] Non-Blind A-Z. Table 26 shows that among 1,430 total horses,
non-blind discriminant models improved the odds of correctly
classifying high earners from 28.32% without models to 37.78% with
models. They improved the odds of correctly classifying low earners
from 71.68% without models to 79.95% with models. The improvement
associated with discriminant modeling was statistically significant
for both high and low earners (P-values <0.0001).
26TABLE 26 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests - Combined Sexes - Names Starting with
Letters A-Z Pre-Model Post-Model Probability Probability Category
Ratio Pct. Ratio Pct. P-Value High Earners 405/1430 28.32 252/667
37.78 0.0000 Low Earners 1025/1430 71.68 610/763 79.95 0.0000
[0169] Non-Blind A-M. Table 27 shows that among horses with names
beginning with the letters A-M, non-blind discriminant models
improved the odds of correctly classifying high earners from 27.87%
without models to 37.47% with models. They improved the odds of
correctly classifying low earners from 72.13% without models to
80.39% with models. The improvement associated with discriminant
modeling was statistically significant for both high and low
earners (P-values .ltoreq.0.0001).
27TABLE 27 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Tests - Combined Sexes - Names Starting with
Letters A-M Pre-Model Post-Model Probability Probability Category
Ratio Pct. Ratio Pct. P-Value High Earners 238/854 27.87 148/395
37.47 0.0000 Low Earners 616/854 72.13 369/459 80.39 0.0001
[0170] Blind N-Z. Table 28 shows that among horses with names
beginning with the letters N-Z, blind discriminant models based on
the A-M horses improved the odds of correctly classifying high
earners from 28.99% without models to 38.21% with models. They
improved the odds of correctly classifying low earners from 71.01%
without models to 79.73% with models. The improvement associated
with discriminant modeling was statistically significant for both
high and low earners (P-values .ltoreq.0.0009).
28TABLE 28 Discriminant Model Results - High Earners vs. Low
Earners Blind Test - Combined Sexes - Names Starting with Letters
N-Z Pre-Model Post-Model Probability Probability Category Ratio
Pct. Ratio Pct. P-Value High Earners 167/576 28.99 107/280 38.21
0.0007 Low Earners 409/576 71.01 236/296 79.73 0.0009
[0171] Colts
[0172] Table 29-Table 31 summarize discriminant results for high
vs. low earners among colts.
[0173] Non-Blind A-Z. Table 29 shows that among 859 colts,
non-blind discriminant models improved the odds of correctly
classifying high earners from 26.66% without models to 34.89% with
models. They improved the odds of correctly classifying low earners
from 73.34% without models to 80.75% with models. The improvement
associated with discriminant modeling was statistically significant
for both high and low earners (P-values .ltoreq.0.0004).
29TABLE 29 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Test - Colts - Names Starting with Letters A-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 229/859 26.70 142/407 34.89 0.0002
Low Earners 630/859 73.34 365/452 80.75 0.0004
[0174] Non-Blind A-M. Table 30 shows that among colts with names
beginning with the letters A-M, non-blind discriminant models
improved the odds of correctly classifying high earners from 26.45%
without models to 34.58% with models. They improved the odds of
correctly classifying low earners from 73.55% without models to
80.58% with models. The improvement associated with discriminant
modeling was statistically significant for both high and low
earners (P-values .ltoreq.0.0078).
30TABLE 30 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Test - Colts - Names Starting with Letters A-M
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 137/518 26.45 83/240 34.58 0.0042
Low Earners 381/518 73.55 224/278 80.58 0.0078
[0175] Blind N-Z. Table 31 shows that among colts with names
beginning with the letters N-Z, blind discriminant models based on
the A-M horses improved the odds of correctly classifying high
earners from 26.98% without models to 36.65% with models. They
improved the odds of correctly classifying low earners from 73.02%
without models to 81.67% with models. The improvement associated
with discriminant modeling was statistically significant for both
high and low earners (P-values .ltoreq.0.0091).
31TABLE 31 Discriminant Model Results - High Earners vs. Low
Earners Blind Test - Colts - Names Starting with Letters N-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 92/341 26.98 59/161 36.65 0.0058
Low Earners 249/341 73.02 147/180 81.67 0.0091
[0176] Fillies
[0177] Table 32-Table 34 summarize discriminant results for high
vs. low earners among fillies.
[0178] Non-Blind A-Z. Table 32 shows that among 571 fillies,
non-blind discriminant models improved the odds of correctly
classifying high earners from 30.82% without models to 42.01% with
models. They improved the odds of correctly classifying low earners
from 69.18% without models to 79.14% with models. The improvement
associated with discriminant modeling was statistically significant
for both high and low earners (P-values .ltoreq.0.0002).
32TABLE 32 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Test - Fillies - Names Starting with Letters A-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 176/571 30.82 113/269 42.01 0.0001
Low Earners 395/571 69.18 239/302 79.14 0.0002
[0179] Non-Blind A-M. Table 33 shows that among fillies with names
beginning with the letters A-M, non-blind discriminant models
improved the odds of correctly classifying high earners from 30.06%
without models to 43.05% with models. They improved the odds of
correctly classifying low earners from 69.94% without models to
80.54% with models. The improvement associated with discriminant
modeling was statistically significant for both high and low
earners (P-values .ltoreq.0.0017).
33TABLE 33 Discriminant Model Results - High Earners vs. Low
Earners Non-Blind Test - Fillies - Names Starting with Letters A-M
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 101/336 30.06 65/151 43.05 0.0005
Low Earners 235/336 69.94 149/185 80.54 0.0017
[0180] Blind N-Z. Table 34 shows that among fillies with names
beginning with the letters N-Z, blind discriminant models based on
the A-M horses improved the odds of correctly classifying high
earners from 31.91% without models to 40.34% with models. They
improved the odds of correctly classifying low earners from 68.09%
without models to 76.72% with models. The improvement associated
with discriminant modeling was statistically significant for both
high and low earners (P-values .ltoreq.0.0488).
34TABLE 34 Discriminant Model Results - High Earners vs. Low
Earners Blind Test - Fillies - Names Starting with Letters N-Z
Pre-Model Post-Model Probability Probability Category Ratio Pct.
Ratio Pct. P-Value High Earners 75/235 31.91 48/119 40.34 0.0488
Low Earners 160/235 68.09 89/116 76.72 0.0455
Example 10
Chi-Square Analysis of Performance vs. Heart Size and Physical
Size
[0181] The statistical methods described to this point, and which
have shown the predictive nature of cardiac measurements, are
perhaps less intuitive than the following examples. Once the key
variables of HTWT (the product of height times weight--used as a
measure of physical size), LVD, LVS, PS, SPLN, and SW were
standardized for age, sex and weight, on a scale from 0 (small) to
100 (large), groups of horses based on these variables could be
created. For example, groups of horses could be created with above
or below average LVD, or horses could be grouped into quartiles
(i.e., from the bottom 25% to the top 25%) based on specific heart
measurements or physical size. Questions such as: "Was there as
high a percentage of high earners among horses with below average
LVD as among horses with above average LVD?" could then be
answered.
[0182] Table 35 shows the percentage of horses that earned at least
$10,000 per racing start among horses grouped by physical size and
heart size. Overall, 13.3 percent of the horses in this study's
sample earned at least $10,000 per start.
35TABLE 35 Percentage of Horses that Earned at least $10,000 Per
Start Based on Percentiles for Individual Variables Percentiles
0-25% 25-50% 50-75% 75-100% HTWT 7.6 12.8 14.5 17.8 LVD 11.6 11.1
13.4 17.5 LVS 11.4 11.8 13.9 16.3 SW 10.8 13.1 13.1 16.3 PS 14.3
11.4 14.0 13.3 Average* 10.4 12.2 13.7 17.0 *Average was calculated
excluding PS, which wasn't usually predictive.
[0183] Table 35 shows that as physical size and heart size
measurements increased, except for PS, so did the percentage of
high earners. This table shows that 17.8% of horses with HTWT in
the 75-100% percentile range earned at least $10,000 per start. The
percentage of horses that earned at least $10,000 per start was
below average (13.3% was average for all horses studied) for groups
with cardiac variables below the 50.sup.th percentile. Horses with
cardiac variables in the 75.sup.th and higher percentiles were more
likely to earn at least $10,000 per start.
[0184] Next, horses were first grouped by physical size, and then
by heart measurement size. Table 36 shows that all groups of horses
with HTWT percentiles of 75-100% (right-hand column) produced
higher than average percentages of horses with earnings per start
(EPS).gtoreq.$10,000. All groups of horses with HTWT percentiles of
0-25% (left-hand column) produced fewer than average percentages of
horses with EPS.gtoreq.$10,000, regardless of heart measurement
size.
[0185] Shaded areas in Table 36 show groups with higher than
average percentages of horses with EPS.gtoreq.$10,000. Horses with
HTWT percentiles in the 25-50% range generally performed as well as
average as long as cardiac variables were above average.
[0186] The highest percentages of high earners occurred when
percentiles for both HTWT and heart size were at least 75%. In
cases where HTWT and heart size percentiles were at least 75%, the
average percentage of horses with EPS.gtoreq.$10,000 was 23.0%
(excluding PS)--a 73% improvement over random odds of selecting
high earners (13.3% vs. 23.0%).
36TABLE 36 Percentage of Horses that Earned at least $10,000 Per
Start Based on Percentiles for Individual Cardiac Variables
Combined with HTWT HTWT 0-25% 25-50% 50-75% 75-100% 0-25% LVD 6.7
10.4 13.7 16.4 LVS 6.9 10.5 12.3 16.2 SW 6.7 12.3 11.3 13.5 PS 7.8
13.1 15.1 21.4 25-50% LVD 9.4 9.6 11.9 13.4 LVS 8.0 10.9 14.5 14.0
SW 7.0 11.4 14.7 18.3 PS 5.9 10.4 13.2 15.4 50-75% LVD 4.7 15.2
14.5 18.8 LVS 7.3 14.6 15.9 16.9 SW 9.1 12.3 14.4 16.4 PS 8.3 13.4
14.6 19.2 75-100% LVD 11.0 16.8 17.9 22.2 LVS 8.4 16.0 15.1 24.0 SW
7.5 15.8 17.6 22.7 PS 8.2 14.0 15.0 15.3 Above average performance
categories are shaded.
[0187] Chi-square analysis was used to examine how Thoroughbreds'
normalized heart size (as measured by LVD, LVS, PS, and SW) and
normalized physical size (as measured by HTWT, which is the product
of height times weight) relate to subsequent earnings and racing
distances. Chi-square methods were used to show the predictive
nature of each variable individually. Chi-square methods were then
used to show the predictive nature of each cardiac variable, when
used in conjunction with HTWT.
[0188] High earners and high earner routers were more likely to be
above average in normalized physical size and normalized heart size
(as measured by LVD, LVS, and SW). Low earners were more likely to
be below average in normalized physical size and normalized heart
size. High earner sprinters tended to be above average in
normalized physical size with thick heart walls (as measured by
normalized SW).
[0189] Statistics describing these relationships were summarized in
tables 37-52.
[0190] Extremely high earners--Among 3,150 horses that raced at
least three times by the end of their three-year-old year, 101
(3.2%) earned at least $250,000 and had earnings per start of at
least $20,000. The following tables show the percentage of
extremely high earners with various combinations of above and below
average normalized HTWT and normalized cardiac measurements (LVD,
LVS, SW and PS).
[0191] The following tables, Tables 37-40, show that a
disproportionately high percentage of extremely high earners were
large physically (for their sex, and chronological age), and had
large hearts even relative to other large horses, i.e., when
cardiac measurement variables were normalized for sex,
chronological age, and physical size. Extremely high earners were
three times more likely to have above average normalized HTWT and
normalized cardiac measurements than to have below average
normalized HTWT and normalized cardiac measurements. When breaking
normalized HTWT categories down further, 4% of extremely high
earners had HTWT of 0-25%, while 38% had HTWT of 75-100%--a nearly
ten-fold difference. The general population, e.g., when not looking
at racing performance success variables, is fairly evenly
distributed among the four quartiles listed in the tables.
37 TABLE 37 HTWT Below Above Average Average LVD Below 17% 19%
Average Above 15% 50% Average
[0192]
38 TABLE 38 HTWT Below Above Average Average SW Below 15% 23%
Average Above 17% 46% Average
[0193]
39 TABLE 39 HTWT Below Above Average Average LVS Below 18% 24%
Average Above 14% 45% Average
[0194]
40 TABLE 40 HTWT Below Above Average Average PS Below 12% 31%
Average Above 20% 38% Average
[0195] The following tables, Table 41-Table 44, show the percentage
of high earner routers with various combinations of above and below
average normalized HTWT and normalized cardiac measurements (LVD,
LVS, SW and PS).
[0196] These tables show that a disproportionately high percentage
of high earner routers were large physically (compared to other
subjects of the same sex and chronological age), and had large
hearts even relative to other large horses, i.e., when normalized
for sex, chronological age, height and weight. High earner routers
were four times more likely to have above average normalized HTWT
and normalized cardiac measurements than to have below average
normalized HTWT and normalized cardiac measurements. The general
population is fairly evenly distributed among the four quartiles
listed in the tables when not considering the racing performance
variables.
41 TABLE 41 HTWT Below Above Average Average LVD Below 12% 25%
Average Above 15% 48% Average
[0197]
42 TABLE 42 HTWT Below Above Average Average SW Below 10% 31%
Average Above 16% 43% Average
[0198]
43 TABLE 43 HTWT Below Above Average Average LVS Below 13% 26%
Average Above 14% 47% Average
[0199]
44 TABLE 44 HTWT Below Above Average Average PS Below 12% 37%
Average Above 14% 37% Average
[0200] The following tables, Tables 45-48, show the percentage of
high earner sprinters with various combinations of above and below
average normalized HTWT and normalized cardiac measurements (LVD,
LVS, SW and PS).
[0201] These tables show that high earner sprinters were fairly
evenly distributed by normalized physical size and the two
normalized heart size variables of LVD and LVS, especially compared
to distributions of the same variables for high earner routers.
However, the tables show that high earner sprinters were about 50%
more likely to have been big physically (normalized HTWT) with
above average normalized SW and/or PS, than to be small physically,
with small SW and/or PS. High earner sprinters were most likely to
be relatively big horses with thick heart walls (normalized SW).
The general population, i.e., all levels of racing performance, and
not just sprinters or high earner sprinters, is fairly evenly
distributed among the four quartiles listed in the tables.
45 TABLE 45 HTWT Below Above Average Average LVD Below 22% 27%
Average Above 24% 26% Average
[0202]
46 TABLE 46 HTWT Below Above Average Average SW Below 22% 22%
Average Above 25% 32% Average
[0203]
47 TABLE 47 HTWT Below Above Average Average LVS Below 23% 28%
Average Above 24% 26% Average
[0204]
48 TABLE 48 HTWT Below Above Average Average PS Below 19% 23%
Average Above 28% 31% Average
[0205] The following tables, Table 49-52 show the percentage of low
earners with various combinations of above and below average
normalized HTWT and normalized cardiac measurements (LVD, LVS, SW
and PS).
[0206] These tables show that a disproportionately high percentage
of low earners were relatively small physically, and had small
hearts even relative to other small horses. Low earners were about
1.5 times more likely to have below average normalized HTWT and
normalized cardiac measurements than to have above average
normalized HTWT and normalized cardiac measurements. The general
population is fairly evenly distributed among the four quartiles
listed in the tables when not considering subsets of different
levels of racing performance.
49 TABLE 49 HTWT Below Above Average Average LVD Below 31% 23%
Average Above 25% 20% Average
[0207]
50 TABLE 50 HTWT Below Above Average Average SW Below 31% 23%
Average Above 25% 20% Average
[0208]
51 TABLE 51 HTWT Below Above Average Average LVS Below 32% 23%
Average Above 24% 20% Average
[0209]
52 TABLE 52 HTWT Below Above Average Average PS Below 28% 21%
Average Above 29% 22% Average
Example 11
Predicting Racing Performance
[0210] Discriminant results showed that a horse's weight and height
were important predictive indices of subsequent performance, in
terms of earnings and successful distances raced. Additionally,
interventricular septal wall structural thickness (SW) as defined
in FIG. 3, or cross-sectional spleen area (SPLN), were the most
important predictive variable when differentiating between high and
low earners. In addition to physical size, the left ventricle in
diastole and systole (LVD and LVS) were the most important
predictive variables when differentiating between successful
sprinters and routers.
[0211] Several of the variables studied were highly correlated
(i.e., similar). Discriminant models typically had very similar
results when one or two variables were replaced with other
variables with which they were highly correlated (e.g. LVS and LVD,
or WT and HTWT).
[0212] In most cases, combined-sex discriminant models correctly
identified the same horses that were correctly identified by the
same-sex models.
[0213] Blind tests showed that cardiac parameters predicted
subsequent racing performance with far greater accuracy than
possible selecting horses from these groups at random. Models
successfully differentiated not only between stakes- and
claiming-caliber horses, but also between stakes- and
allowance-caliber horses.
[0214] On average, blind test discriminant models improved random
odds of identifying high earners (or routers) by 35 percent (i.e.,
going from a 30% probability of correctly identifying high earners
without models to a 40% probability with models).
[0215] Stepwise and discriminant analyses beyond those presented
here sometimes produced exceptional results for one group in the
comparison, but unexceptional results for the other group. For
example, a high vs. low earners model may accurately predict high
earners, while just meeting random expectations among low earners.
Multiple models differentiated by level of earnings may be needed
in such instances. Model limitations have to be assessed relative
to potential applications. Z-tests were helpful in determining the
statistical strength of discriminant results for each individual
group represented in the models.
Example 12
Using Subjective Visual Cardiac Parameters to Predict Racing
Performance
[0216] Subjective ratings (ranging from 1=poor to 5=excellent) to
describe the images on the ultrasound machine--visual impressions
of ecogenicity (e.g., clarity, sharpness of contrast, type and
symmetry of shapes, smoothness of functioning of structures) of the
2D images were recorded. These ratings were recorded as:
Ecogenicity (EC and VEC); general shape of the image at diastole
and systole (CATE and SQ); clarity and sharpness of contrast of
left ventricle during diastole and systole (DCL and SCL);
smoothness of left ventricle during diastole and systole (DSM and
SSM); blood backflow from left ventricle during diastole and
systole (DBF and SBF); double-beat wave (XB); overall irregularity
of the heart image (IRRG); how well the valve closes (NVC); and
overall clearness of image in diastole (PVAR).
[0217] Stepwise analysis identified statistically significant
variables that could differentiate between groups of horses
categorized as high and low earners.
[0218] The variables considered in the analysis were LVD, LVS, SW,
PS, HTWT, EC, CATE, DCL, DSM, DBF, XB, VEC, SQ, SCL, SSM, SBF,
IRRG, NVC, and PVAR. Among these variables, LVD, LVS, SW, PS were
standardized for sex, age and weight, and HTWT was standardized for
sex and age.
[0219] For high vs. low earners, with the additional consideration
of visual ratings as described above, stepwise analysis identified
the following significant variables
[0220] Combined Sexes. HTWT, PVAR, SBF
[0221] Colts. SBF, HTWT, DSM
[0222] Fillies. HTWT, PVAR, SQ, DSM
[0223] In order to work with higher numbers of horses for
discriminant analyses using the subjective visual variables
(assessed each on a scale from 1 to 5), horses for which there were
only two-year-old race records were added to the groups of raced
horses used elsewhere to assess racing performance levels in this
study. These were horses born in 1998. Thus, unlike everywhere else
in this monograph, this analysis of raced horses had some horses
with two- and three-year-old race records and others with just
two-year-old race records.
[0224] The variables used in discriminant analysis were those
identified as significant by stepwise analysis. Only combined sex
models were analyzed due to limited number of horses.
[0225] Non-Blind A-Z. Table 53 shows that among 394 horses,
non-blind discriminant models improved the odds of correctly
classifying high earners from 33.25% without models to 43.93% with
models. They improved the odds of correctly classifying low earners
from 66.75% without models to 75.11% with models. All results were
statistically significant (P-values .ltoreq.0.0083).
53TABLE 53 Discriminant Model Results Using Subjective 1-5
Variables - High vs. Low Earners Non-Blind Tests - Combined Sexes -
Names Starting with Letters A-Z Pre-Model Post-Model Probability
Probability Category Ratio Pct. Ratio Pct. P-Value High Earners
131/394 33.25 76/173 43.93 0.0029 Low Earners 263/394 66.75 166/221
75.11 0.0083
[0226] Non-Blind A-M. Table 54 shows that among horses with names
beginning with the letters A-M, non-blind discriminant models
improved the odds of correctly classifying high earners from 34.18%
without models to 41.28% with models. They improved the odds of
correctly classifying low earners from 65.82% without models to
71.88% with models. Results were not statistically significant
(P.ltoreq.0.1499).
54TABLE 54 Discriminant Model Results Using Subjective 1-5
Variables - High vs. Low Earners Non-Blind Tests - Combined Sexes -
Names Starting with Letters A-M Pre-Model Post-Model Probability
Probability Category Ratio Pct. Ratio Pct. P-Value High Earners
81/237 34.18 45/109 41.28 0.1188 Low Earners 156/237 65.82 92/128
71.88 0.1499
[0227] Blind N-Z. Table 55 shows that among horses with names
beginning with the letters N-Z, blind discriminant models based on
the A-M horses improved the odds of correctly classifying high
earners from 31.85% without models to 43.42% with models. They
improved the odds of correctly classifying low earners from 68.15%
without models to 79.01% with models. All results were
statistically significant (P-values .ltoreq.0.0444).
55TABLE 55 Discriminant Model Results Using Subjective 1-5
Variables - High vs. Low Earners Blind Tests - Combined Sexes -
Names Starting with Letters N-Z Pre-Model Post-Model Probability
Probability Category Ratio Pct. Ratio Pct. P-Value High Earners
50/157 31.85 33/76 43.42 0.0300 Low Earners 107/157 68.15 64/81
79.01 0.0444
* * * * *