 

m LIBRARY
‘26 Micnigan State
1 0° ‘ University

 

 

 

This is to certify that the
dissertation entitled

FACTORS AFFECTING MEASURES OF LONGEVITY AND
STAYABILITY IN YORKSHIRE SOWS.

presented by

Mark D. Hoge

has been accepted towards fulﬁllment
of the requirements for the

Doctoral degree in Animal Science

 

 

ﬂw M M ﬁfe

Major Professor’s Signature

A/ﬂpld/ 3/, 2.0&of

Date

MSU is an Afﬁrmative Action/Equal Opportunity Employer

.._.-.-.-.-.-.-.-.-.-._.-.-.-.—.-.-.-.-.-.-.-.-.-.-

 

 

PLACE IN RETURN Box to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

DATE DUE DATE DUE DATE DUE

 

FACTORS AFFECTING MEASURES OF LONGEVITY AND STAYABILITY IN
YORKSHIRE sows.
By

Mark D. Hoge

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY

Animal Science

2009

ABSTRACT

FACTORS AFFECTING MEASURES OF LONGEVITY AND STAYABILITY IN
YORKSHIRE SOWS.

By
Mark D. Hoge

Adult sow lifetime is recognized as both an economic and welfare concern.
However, there are not consistent deﬁnitions of sow longevity. The purpose of this study
was to assess the relationship of developmental performance factors with longevity and to
determine the genetic variation within six different descriptions of longevity. Longevity
deﬁnitions included stayability (probability of producing 40 pigs or reaching four
parities), lifespan (number of parities accumulated before culling), lifetime proliﬁcacy
(number of pigs born alive during a female’s lifetime), herdlife (time from ﬁrst farrowing
to culling), and productive life (pigs produced per day of life).

The edited data consisted of 14,262 records from Yorkshire females with at least
one farrowing record, from both nucleus and multiplication herds across 21 farms.
Within a subset of the data, information was available on the female’s litter birth record
and corresponding growth and composition data. Furthermore, complete pedigrees were
available on 7,487 females. The data were subdivided into two data sets; Data A:
female’s farrowing records, and Data B: Data A and the information from a female’s
litter birth record and her growth and backfat data. Due to a reduced number of records
with complete pedigree information, two separate analyses were completed.

Initially, a Cox proportional hazards model was used to determine the relationship
of developmental factors and ﬁrst farrowing record with longevity. Those factors that

were signiﬁcantly (P <0.0001) associated with longevity, regardless of deﬁnition, were

A

age at ﬁrst farrowing, litter size at ﬁrst farrowing and last farrowing, number of stillborn
in her ﬁrst litter, adjusted 21 day litter weight of ﬁrst litter, herd type (nucleus or
multiplier), and standardized variables for backfat and growth. Within a contemporary
group, fatter, slower growing gilts had a decreased risk of being culled. Additionally,
sows that had more pigs born alive, fewer stillborn, and heavier litters at weaning in their
ﬁrst litter had a decreased risk of being culled. Furthermore, sows from nucleus herds
experienced a greater risk of being culled.

Heritability was estimated for four different descriptions of longevity (lifespan,
lifetime proliﬁcacy, herdlife, and pigs produced per day of life) with survival analysis by
using a proportional hazards model with an underlying Weibull distribution. A threshold
model was used to estimate heritability for two descriptions of stayability on a subset of
the original data (n=5,889) that did not include censored observations. Fixed effects
within the two model types were similar to those reported previously. Random effects
were sire and the interaction of herd by year. Fixed effects were consistent in their
signiﬁcance, regardless of statistical model or deﬁnition. Heritability estimates ranged
from 0.039 to 0.050 and from 0.076 to 0.231 for true and functional deﬁnitions of
survival analyses, respectively and from 0.117 to 0.200 for threshold analyses.

Results indicated that there are early indicators that can be monitored to provide
insight to the future productivity or length of productive life of Yorkshire females.
Additionally, sufficient genetic variation exists, regardless of deﬁnition, to improve sow
longevity. Pork producers can implement management protocols which can extend the
productive life of breeding females resulting in improved proﬁtability and animal

welfare.

This dissertation is dedicated to my parents for their continued support and to my family,
Katie, Carter, and Nolan who all have made great sacriﬁces to make this possible.

 

ACKNOWLEDGEMENTS

The opportunities and experiences that have been granted to me at Michigan State
University have been greatly appreciated. Dr. Ron Bates has been an ideal major
professor. He provided a positive learning atmosphere, guidance when needed, and has
never given up on me completing my degree. He is truly a great person and a valuable
asset to Michigan State University. Dr. Ken Stalder, Dr. Rob Tempelman, Dr. Dennis
Banks, Dr. John Kaneene, Dr. Cathy Ernst, and Dr. Harlan Ritchie have served on my
Ph.D. guidance committee and have answered questions, provided insight, and took time
to meet with me, when needed. Additionally, Dr. Vincent Ducrocq and Dr. Tirno
Serrenius have provided great insight regarding the Survival Kit.

This project would not have been possible without access to production data from
Waldo Farms, DeWitt, NE, Premier Swine Systems, Michigantown, IN,
Whiteshire/Hamroc, Albion, IN, and the Zierke Co., Morris, MN. The National Swine
Registry granted access to Yorkshire pedigree information and The American Yorkshire

Club and The National Pork Board provided partial funding for this project.

TABLE OF CONTENTS

LIST OF TABLES ................................................................................... vii
LIST OF FIGURES ................................................................................. ix
LIST OF ACRONYMS .............................................................................. x
INTRODUCTION .................................................................................... 1
CHAPTER 1. LITERATURE REVIEW .......................................................... 5
Deﬁnitions of Longevity ...................................................................... 5
Survival Analysis .............................................................................. 9
Economic Impact on Longevity ............................................................. 12
Reasons for Removal ........................................................................ 16
Factors Affecting Longevity ................................................................ 20
Genetic Control of Sow Longevity ......................................................... 29
True versus Functional Herd Life .......................................................... 35
Summary ....................................................................................... 36
CHAPTER H. CHARACTERIZATION OF SOW LONGEVITY AND THE
DEVELOPMENTAL FATORS THAT INFLUENCE IT ....................................... 37
Abstract ........................................................................................ 37
Introduction .................................................................................... 38
Materials and Methods ....................................................................... 40
Description of Data ........................................................................ 40
Statistical Analysis ......................................................................... 42
Results and Discussion ...................................................................... 43
Data A ....................................................................................... 44
Data B ....................................................................................... 47
CHAPTER III. HERITABILITY OF LONGEVITY IN YORKSHIRE F EMALES......6S
Abstract ........................................................................................ 65
Introduction .................................................................................... 66
Materials and Methods ....................................................................... 68
Description of Data ........................................................................ 68
Statistical Analysis ......................................................................... 70
Results and Discussion .................................................................... 74
SUMMARY AND CONCLUSIONS ............................................................ 89
LITERATURE CITED ............................................................................. 93

vi

LIST OF TABLES
Table 11.1. Deﬁnitions of variable acronyms used in model development ................ 51

Table 11.2. Number of observations, means, standard deviations, minimums, maximum

values of the traits studied in the Yorkshire Population ....................................... 52
Table 11.3. Estimated hazard ratios for Data A for LPL ...................................... 53
Table 11.4. Estimated hazard ratios for Data A for SNBA ................................... 54
Table 11.5. Estimated hazard ratios for Data A for PLIFE ................................... 55
Table 11.6. Estimated hazard ratios for Data A for MAXPAR .............................. 56
Table 11.7. Estimated hazard ratios for Data A for PRNBA40 .............................. 57
Table 11.8. Estimated hazard ratios for Data A for PR4PAR ................................ 58
Table 11.9. Estimated hazard ratios for Data B for LPL ...................................... 59
Table 11.10. Estimated hazard ratios for Data B for SNBA .................................. 60
Table 11.11. Estimated hazard ratios for Data B for PLIFE .................................. 61
Table 11.12. Estimated hazard ratios for Data B for MAXPAR ............................. 62
Table 11.13. Estimated hazard ratios for Data B for PRNBA40 ............................. 63
Table [1.14. Estimated hazard ratios for Data B for PR4PAR ............................... 64
Table [11.1. Deﬁnitions of variable acronyms used in model development ................ 79

Table 111.2. Number of observations, means, standard deviations, minimums, maximum
values of the traits studied in the Yorkshire Population ........................................ 80

Table 111.3. Likelihood ratio test, including P values and reliabilities (R2) for Data A
when the covariates were tested by excluding one at a time (last) from the full model....81

Table 111.4. Likelihood ratio test, including P values and reliabilities (R2) for Data B
when the covariates were tested by excluding one at a time (last) from the full model. ..82

Table 111.5. Level of statistical signiﬁcance of the covariates included in the analyses of
PRNBA40“ and PR4PARa for Data A and Data B .............................................. 83

vii

Table IH.6. Additive sire variance (of) and heritability (hz) estimates obtained from
survival analysis and threshold analysis ......................................................... 84

viii

LIST OF FIGURES

Figure 111.1. Graphical test for validity of Weibull distribution for LPL. . . . . . . . . . ..........85
Figure 111.2. Graphical test for validity of Weibull distribution for MAXPAR ........... 86
Figure 111.3. Graphical test for validity of Weibull distribution for SNBA ................ 87
Figure 111.4. Graphical test for validity of Weibull distribution for PLIFE .............. 88

LIST OF ACRONYMS
LPL = herd life deﬁned as the number of days between date of ﬁrst farrowing and culling
MAXPAR = the highest parity completed prior to removal
PLIFE = pigs born alive/day/herd life
PRNBA40 = discrete variable indicating a female produced 40 live pigs prior to removal
PR4PAR = discrete variable indicating a female completed 4 parities prior to removal
SNBA = lifetime pigs born alive
AGEF F = days of age at ﬁrst farrowing
BF = standardized deviation for back fat within a contemporary group
DAY = standardized deviation for days to 113.6 kg within a contemporary group
D21 W = adjusted 21 day litter weight of female’s ﬁrst litter
HT = herd type
OFFTEST = number of days between off test and ﬁrst farrowing
PGR = birth litter gilt ratio
PHY = herd by birth year
PLBW = birth litter birth weight
PNBA = birth litter number born alive
LNBA = last litter farrowed number born alive
NBA = ﬁrst litter number born alive
NSTL = ﬁrst litter number of stillboms
SHY = herd by ﬁrst farrowing year

WNAGE = days of lactation allowed for ﬁrst litter

Introduction

Longevity of adult females is a trait with a signiﬁcant impact on swine farm
proﬁtability. An improvement in longevity can result in decreased replacement costs and
a higher proportion of mature sows in a herd that have reached their maximum
productivity. The swine industry has a hierarchical structure, with many pork producers
purchasing out-of-herd replacement gilts. This investment constitutes a major budget
decision and ideally, replacement gilts should be maintained until this investment is
recuperated. In practice, 40 to 50% of the sows are removed before they complete three
to four parities (D’Allaire et al., 1987; Boyle et al., 1998), an age when most
replacements barely recover their initial cost (Stalder et al., 2002). Similarly, reports
from commercial swine records have shown that sow-culling rates within commercial
herds are around 50% and the reported average parity of cull sows in the US is 3.8
(PigCHAMP, 2000-2005). This is alarming since decreased sow longevity is a welfare
concern as well as a proﬁtability dilemma. In a simulation of farrow to ﬁnish swine
farms, improved sow lifetime by one parity increased proﬁt potential by $34 to $150 per
sow lifetime, depending on the assumptions made regarding pigs per litter sold, market
price and replacement gilt costs (Stalder et al., 2000). Incentives to increase sow
longevity include larger litters with heavier pigs in older parities, fewer unproductive
days, acquired immunity to herd diseases, higher sow salvage value, and lower
replacements costs (D’Allaire etal., 1987; Lucia et al., 2000).

Increased sow mortality, combined with reproductive problems such as failing to
cycle in a timely manner, not conceiving, not farrowing, poor performance or physical

problems are the major reasons for the increase in replacement rates in commercial sow

units (Dial and Koketsu, 1996; Friendship et al., 1996). These high replacement rates
result in the need for larger gilt pools and therefore the purchase or production of more
breeding gilts. In addition, breeding herd inventories usually have high proportions of
low-parity females. Thus, the majority of the female removals occur at low parities and
are mostly attributed to reproductive failure (D’Allaire et al., 1987; Dijkhuizen et al.,
1989; Patterson et al., 1996).

Improving economic efﬁciency and animal well being are good reasons to
examine what factors might contribute to opportunities to improve longevity. In addition,
sows that survive fewer parities are more likely to produce fewer pigs, compared to a sow
that remains in the breeding herd for a longer period of time. Fewer parities result in
reduced opportunity for sows to be sufﬁciently productive to achieve a return on their gilt
investment cost (Stalder, 2004).

Longevity can be deﬁned in a number of ways and measured using different
systems of analysis. Aside from the challenges of deﬁning and quantifying longevity,
there are additional inﬂuencing factors to account for when measuring longevity or herd
life. The relative performance of a sow compared to that of her herd mates can have an
impact on culling decisions. Number born, number born alive, number weaned, and
wean to estrus intervals relative to herd averages can all play a role in the longevity of a
sow. Selection for type traits has an impact on longevity due to their indirect effects.
Traits such as underline quality, feet and legs, and lean growth can have an effect on a
producer’s decision to cull an animal. There is some indication (Pond and Mersmann,

2001) that natural longevity of pigs is 12-15 years when performance is not a factor in

removal from the breeding herd. Poor longevity, regardless of the causes, does result in
increased breeding herd female replacement rates.

Regardless of the traits being measured or the deﬁnition being used, the need to
utilize records from sows that have not completed their lifetime performance at the time
of data acquisition poses a challenge concerning traditional statistical approaches. There
are certain aspects of survival analysis data, such as censoring and non-normality that
generate great difficulty when trying to analyze the data using traditional statistical
methods. The non-normality aspect of the data violates the normality assumption of most
commonly used statistical models. To date, methods used to conduct survival analysis
have included linear models, proportional hazards models, and random regression
models. In the case of linear model analysis, most studies have applied multiple-trait
Restricted Maximum Likelihood (REML) to a sire or animal model to estimate genetic
parameters. Cox's semiparametric model is widely used in the analysis of survival data to
explain the effect of explanatory variables on hazard rates or relative risk of a sow being
culled ﬁ'om the breeding herd in this case. This has been extended to include random
effects, as in regular mixed linear models (Ducrocq and Solkner,1998a). Survival data
can be ﬁtted with either a Cox proportional hazards model, which allows an unspeciﬁed
form for the underlying survivor function, or a parametric model in which the distribution
of the random disturbance can be taken from a class of distributions that includes the
Weibull distribution.

There is clearly an economic incentive to quantify factors that delay a sow’s
culling. There is suitable statistical methodology that will support this type of research.

However, there are not consistent deﬁnitions of longevity used across the literature. The

purpose of this research was to simultaneously evaluate the relationship of developmental
factors and parity one performance with six different deﬁnitions of longevity. A unique
data set was assembled to complete this project. The data set is an accumulation of both
nucleus and multiplication records of Yorkshire sows from four swine seedstock systems
(Waldo Farms, DeWitt, NE; Premier Swine Systems, Michigantown, IN;
Whiteshire/Hamroc, Albion, IN; and The Zierke Co., Morris, MN). The objectives of
this study were as follows:

1. To assess six different measures of longevity and determine what relationship
developmental performance characteristics may have with these differing deﬁnitions.

2. To estimate the genetic variation for the six different measures of longevity

and determine if non-genetic factors inﬂuence the heritability estimate.

CHAPTER 1
LITERATURE REVIEW
Deﬁnitions of Longevity

Before sow longevity can be improved or the reasons that sows vary in their
longevity can be examined, the trait itself needs to be deﬁned (Stalder, 2004). Depending
on the type of study being conducted, many different terms have been used to describe a
female’s survival including: herd life, survivability, lifespan, length of productive life,
stayability, and lifetime performance. Longevity of sows, deﬁned as the ability to delay
involuntary culling, summarizes the effects of functional traits, and of reproductive
performance through the removal of sows with inferior fertility or a low capacity to
produce pigs (Ducrocq and Stilkner, 1998a). Sow longevity is an example of survival or
time-to-event data (Allison, 1997), with time typically being the number of days that a
sow remains in the herd and the event being the removal from the herd (e. g., culling or
death).

Dekkers and .Iairath (1994) outlined three main categories to characterize
deﬁnitions of longevity. The ﬁrst category involves traits related to length of total or
productive life. This includes measures such as days from birth to culling, days from
entering the breeding herd to culling or days from parturition at parity one to removal
ﬁ'om the herd. The second category is the ability to reach a predetermined endpoint, for
example a speciﬁc age or lactation. This is a measure of the sow’s stayability. The ﬁnal
category used to deﬁne longevity is survival within each consecutive lactation or parity.

There are a variety of ways in which length of productive life has been measured

in domestic livestock. In dairy cattle studies, length of productive life has been deﬁned as

the number of days from ﬁrst calving to last test day or last known lactation. Due to the
lack of exact culling dates, these endpoints were considered a close approximation
(Dekkers et al., 1994; Jairath and Dekkers, 1994; Vukasinovic et al., 1997; Vollema and
Groen, 1998). Alternate measures of productive life have been the number of calvings or
completed lactations, and days or months in milk over a lifetime (Jairath and Dekkers,
1994; Vollema and Groen, 1998). A similar approach to measuring length of productive
life is the evaluation of length of total life. Total productive life is a measure of the

number of days or months from birth to date of culling (V ollema and Groen, 1998).

Stayability, the second category outlined by Dekkers and J airath (1994), is a
measure of an animal’s ability to remain in the herd to a set endpoint. Among studies on
dairy populations, endpoints used in the analysis of stayability are survival to 36, 48, 54,
72, and 84 months (Short and Lawlor, 1992; Vollema and Groen, 1998). Survival
through the ﬁrst lactation can also act as an indicator of stayability. If a cow calved for a
second time within 2 years of her ﬁrst calving date, she was considered to have survived

her ﬁrst lactation (Boettcher et al. 1999).

The ﬁnal classiﬁcation by Dekkers and J airath (1994), survival within
consecutive lactations, is not as common in longevity studies. Survival can either be
measured on the basis of survival to the next lactation (Jairath et al., 1998), or the time
spanned by the data can be divided into successive intervals (V eerkarnp et al., 2001).
With the second case, survival is coded as a series of binary data. Ifan animal survives
to the next interval, it can be coded as 1, while a failure is coded as 0.

From the initial work conducted in dairy cattle, swine longevity research has

evolved. D’Allaire et a1. (1992) suggests that when evaluating sow longevity, swine

operations could assess removal rate, culling rate, replacement rate, percent gilts in the
herd, mean parity of females in inventory, and mean parity at removal. Pork producers
have tended to focus on culling and replacement rates as a measure of how herds
compare regarding the retention of females within their operations (Deen et al., 2002).
The challenge of working with ﬁeld data, are inconsistent and incomplete recording
schemes over time (eg. birth date of purchased gilts) and the variation of management
and culling policy across operations. Also, Deen (2003) argued that parity, age at
removal, and removal rate are not appropriate measures of longevity because older, less
productive sows can be retained just to improve parity or age structure of the herd.
Holder et al. (1995), Culbertson and Mabry (1995), and Knauer (2006) took an
alternative approach to assessing longevity by assessing the number of pigs weaned per
day of life.

More recently, productive life span is a trait which has received increasing
attention in animal breeding. Ducrocq (1997) proposed a general strategy for the analysis
of a productive life span based on survival analysis as an adequate method for the genetic
evaluation of the length of productive life measurements. A number of studies have
deﬁned longevity by assessing the length of productive life, lifetime proliﬁcacy, or
stayability of a sow. Yazdi et al. (2000b), Serenius and Stalder (2004), and Tarrés et al.
(2006) have deﬁned length of productive life as the number of days from the date of ﬁrst
farrowing to the date of culling or censoring. Guo et a1. (2001) deﬁned the beginning of a
sow’s productive life in the nucleus herd as the age in days at ﬁrst conception, while
Rodriguez-Zas et al., (2003) deﬁned herd life as the total number of days from ﬁrst

service (regardless of success) until removal from the herd. Additionally, productive

days were deﬁned as the total number of days the sow gestated and lactated until removal
from the herd. Lifetime proliﬁcacy was deﬁned by Guo et al. (2001) and Serenius and
Stalder (2004) as the number of piglets (born alive) produced during the productive
lifespan of the sow and Tholen et al. (1996a) deﬁned stayability as the ability of a sow to
produce an additional litter, after producing the previous litter.

Lucia et al. (2000) took an alternative approach to deﬁning longevity. Instead of
looking at the lifetime productivity or proliﬁcacy of a female, this study determined the
percentage of a sow’s lifetime that she was non-productive. The number of days spent in
the herd was estimated by the difference between the date of removal and the date of
entry, excluding the entry-to-ﬁrst service interval. For each female, the number of days
spent in gestation and lactation during herd life was summed. As a nonproductive day is
any day spent by a female in the breeding herd in periods other than lactation or
gestation, the proportion of lifetime nonproductive days was calculated by subtracting
lactation days and gestation days from total herd days and dividing that quantity by herd
days. That value was then multiplied by 100% to express the value for lifetime
nonproductive days as a percentage.

Length of productive life provides a complete measure of survival. Using this
deﬁnition of survival, a value for a female’s life is not determined until a sow has been
removed from the herd. It is necessary to either wait until a sow has been culled or use a
method of analysis that accounts for incomplete or censored records. If culling dates are
not recorded, an estimation such as the last farrow date or last service date must be used.
Stayability is a measure of herd life that removes the problem of accounting for

incomplete records. However, it does not offer insight into survival before or after a

predetermined endpoint. Subsequently, the appropriate measure of longevity, for a given
study, is dependent on the data available and the objectives of the study.
Survival Analysis

Survival time can be broadly deﬁned as the time to the occurrence of a given
event. Statistical techniques used in survival analysis are aimed at modeling and analysis
of response times. The goal of survival analysis is to analyze positive measures
describing in some sense the width of the interval between an origin point and an
endpoint. Often, the endpoint corresponds to death or culling and the length from the
origin to the end is measured in number of days or probability of reaching a
predetermined point in production (Ducrocq, 1997). Famula (1981) was the ﬁrst author
who proposed survival analysis as a statistical method to analyze length of productive life
in dairy cattle. Smith (1983) and Smith and Quaas (1984) used survival analysis
techniques to estimate breeding values of sires based on the length of productive life of
their daughters. The techniques were further developed and adjusted for large scale
applications by Ducrocq (1994) and Ducrocq and Sdlkner (1998b).

Genetic evaluation for the length of productive life in livestock species requires
speciﬁc statistical methods due to: 1) some animals are still alive at the time of evaluation
and their complete length of productive life is not known, which implies that records of
such individuals must be treated as censored; 2) effects inﬂuencing length of productive
life do not act linearly and vary with time; and 3) the distribution of length of productive
life is often unknown and extremely skewed (V ukasinovic, 1997).

A speciﬁc feature of survival analysis is that it can accommodate censoring and

uses information available on animals that are still alive. That is, the survival time may

either be known to be greater than a certain amount (right censored), less than a certain
amount (left censored), or be within a certain range (double censored) (Kachman, 1999).
The most frequent type of censoring is right censoring. There are two special cases of
right censoring. Type I censoring occurs when the experiment is stopped after a ﬁxed
time and Type II censoring occurs when the experiment is stopped after a pre-determined
number of failures. Usual causes of right censoring are the absence of failure before the
end of the study. An observation is considered censored when a female is still alive at the
end of data collection, when she disappears from the herd due to sale to another herd not
under a testing program, or when the whole herd is withdrawn ﬁom the testing program
(Vukasinovic, 1997).

Survival analysis can also accommodate left truncated records, for which the
origin point lays outside the data collection period. An example of a left truncated record
is when a female is in production prior to the beginning of the study. Such an individual
is considered to be at risk only from the beginning of data collection onward
(Vukasinovic, 1997).

Multiple authors (Lee 1992; Vukasinovic 1997; Ducrocq 1987, 1994, 1997;
Solkner and Ducrocq 1999; and Kachman 1999) provide a detailed background on the
methodology of survival analysis. The distribution of survival times is usually described
or characterized by three functions: (1) the survivorship function, (2) the probability
density function, and (3) the hazard function. These three functions are mathematically
equivalent. If one of them is given, the other two can be derived. For example, let T
denote the survival time. The distribution of T can be characterized by the following

three functions.

1. Survivorship Function (or cumulative survival rate). This function, denoted by
S(t) is deﬁned as the probability that an individual survives longer than t:
S(t) = P (T>t)
From the deﬁnition of the cumulative distribution ﬁmction F (t) of T,
S(t) = l — P (an individual fails before time t)
= 1 — F (t)
The graph of S(t) is called the survival curve.

2. Probability Density Function (or Density Function). The survival time T, a
continuous random variable, has a probability density function deﬁned as the limit of the
probability that an individual fails in the short interval t to t+ At per unit width At, or
simply the probability of failure in a small interval per unit time. It can be expressed as

P{an individual dying between (t,t + At)}
At

 

f0) = A1330
The density function is also known as the unconditional failure rate.

3. Hazard Function (instantaneous failure rate or conditional failure rate).
Models for survival analysis can be built from a hazard function. The hazard ﬁmction
h(t) of survival time T gives the conditional failure rate. This is deﬁned as the probability
of failure during a very small time interval, assuming that the individual has survived to
the beginning of the interval, t to t + At per unit time, given that the individual has
survived to time t:

P{an individual of age t fails in the time (t,t + At)
At

 

h(t) = lim
At—>0
The hazard function can also be deﬁned in terms of the cumulative distribution function

F (t) and the probability density function ﬁt):

h(t) =f(t)/ {1 - F (0}
From its deﬁnition, the hazard function must be nonnegative. Additionally, it must be
positive at time t unless there is no risk of failure at time t. The hazard rate is an un-
observed variable yet it controls both the occurrence and the timing of failure or in this
case culling or removal from the herd. It is the fundamental dependent variable in
survival analysis. The shape of the hazard function will inﬂuence the survivorship
function.
Economic impact on improving longevity
As with many aspects of animal breeding, genetic change and selection are often
driven by ways to improve economic efﬁciency within a production system. Sow
longevity or the need to identify factors impacting sow longevity is no different. Sow
longevity or the length of time that a sow remains in the herd has a major effect on
proﬁtability (Pl'a et al., 2003). Replacement decision making is a continual and complex
process based on biological and economic considerations (Rajala-Schultz and Grohn,
2001). This continual decision making process is a dynamic one that adapts to changes in
replacement rates, salvage value, replacement costs, and revenues per piglet produced
(Rodriguez-Zas et al., 2006). Regardless of economic conditions, sows that remain in the
breeding herd longer have an increased opportunity to recuperate their initial cost and
could result in producing larger and heavier litters, have fewer nonproductive days,
improve acquired immunity to diseases, have greater salvage value, and lower
replacements costs (D’Allaire and Drolet, 1999; Lucia et al., 2000; Stalder et al., 2002).
Stalder ct al. (2004) effectively summarized literature that addressed the issue of

economic impact on improving sow longevity. Stevermer (1991) demonstrated that

productivity and breakeven price increased through parity seven when a sow’s maximum
parity was allowed to increase, in a model that assumed 15% replacement rate per parity.
Sehested (1996) found that improving longevity by one parity had the same impact as
improving lean meat content by 0.5%. He reported that longevity tended to decrease with
increasing parity of culling and likely to have little economic impact once average parity
at culling was above ﬁve. Conversely, a study conducted by Parsons et al. (1990)
suggested that productivity of a stable pig herd, and hence its economic status, was
insensitive to parity distribution and its underlying culling strategies when ﬁrst parity
females represented either 26% or 30% of the female breeding herd.

Faust et al. (1992, 1993a, 1993b) implemented simulation models to examine the
effect of culling rate, replacement gilt costs and other factors on the proﬁtability of
commercial pig operations. These models demonstrated that systems with the lowest
commercial replacement rates were most proﬁtable and when high replacement rates
occurred, replacement gilts were worth no more than 175% of market value. Conversely,
when replacement rates were low, replacement gilts were worth as much as 450% of
market value. Results showed that greater beneﬁt occurred when maximum parity
increased from one to ﬁve than from ﬁve to ten.

Patterson (1997) reported economic beneﬁts of increasing lifetime production on
a piglet basis. Piglet breeding costs are reduced by US$1.61 per piglet when sows with
high lifetime production values, a trait closely related to longevity, were compared to
sows with low lifetime production values. Similarly, Lucia et al. (2000) indicated that
lifetime reproductive performance was better in herds having a higher proportion of high

parity females. Dhuyvetter (2000) suggested that the optimal economic time to cull a

sow is after her 8‘11 or 9th parity, even though the additional economic beneﬁts of keeping
a sow past the sixth parity became marginally less.

Stalder et al. (2004) summarized the “ideal” parity distribution percentage
recommendations previously published by Straw (1984), Parson et al. (1990), Muirhead
and Alexander (1997), and Morrison et al. (2002). Within an operation, 17 to 30% of the
sows should be parity zero females, while 15 to 23% should be parity one sows, 14 to
19% should be parity two sows, 13 to 16% should be parity three sows, 10 to 15% should
be parity 4 sows, 5 to 14% should be parity ﬁve sows, and the remainder should be
comprised of parity six and higher females.

Net present value is deﬁned as the difference between the present value of income
and costs including depreciation. Stalder et al., (2000) conducted a net present value
analysis, on the basis of 1996 to 2000 average production levels and costs, to determine
the number of parities a sow must remain in the breeding herd of a breed-to-wean
operation before the initial investment in her is proﬁtable. A replacement gilt must
remain in the breeding herd for 3 parities before reaching a positive net present value.
Increased length of productive life, and favorable segregated early weaning piglet price
produced a higher net present value.

Rodruguez-Zas et al., (2003) utilized a dynamic programming model to ﬁnd the
optimal parity and net present value in breed-to-wean swine herds. The model included
income and costs per parity weighted by the discount rate and sow removal rate. Three
scenarios that reﬂect a wide range of cases were considered: low removal rates per parity
with no salvage value, high removal rates per parity with no salvage value, and high

removal rates per parity with a percentage of the sows having a salvage value. The ﬁnal

14

scenario represented the optimal average replacement parity and proﬁtability for many
US breed-to-wean operations. Sow replacement cost and salvage value had the greatest
impact on optimal parity followed by revenues per piglet weaned. Optimal parity varied
ﬁ'om 1 to 2 parities in situations with low sow costs, high salvage values, and high
revenues per piglet, to 6 to 10 parities for the converse economic situation.

Many computer tools and decision support programs have been developed to
examine the impact of sow longevity or an associated trait like culling rate (Stalder et al.,
2004). Some of these tools examined the number of parities required before a sow
reaches a positive net present value (Stalder et al., 2000; Stalder et al., 2002), while
others determine economically optimal culling strategies (Dijkhuizen et al., 1986; Huirne
et al., 1988). Still other models estimated the economic value of sow longevity (De
Vries, 1989). Additional software tools have been developed to assist producers with
sow replacement decisions based on ecomonics and proﬁtability (Jalvingh et al., 1992;
Jorgensen et al., 1996; Morrison et al., 2002).

Lacy et al., (2005) created the Swine Replacement Evaluator to economically
evaluate an operation’s replacement breeding herd female decisions. The spreadsheet is
designed for a producer to enter production and ﬁnancial data speciﬁc to their operation.
Resulting net present value and internal rate of return (the discount rate that will cause an
investment to have a net present value equal to zero) are calculated based on the data
speciﬁc for individual producers' operation. The gilt replacement worksheets can assist
swine producers in determining if purchasing replacement gilts at some price is a

proﬁtable decision. Additionally, a producer can use these spreadsheets to determine if

breeding females remain in the herd for a sufﬁcient number of parities to recover the
initial investment cost of replacement gilts.

Sow longevity has been proven to have an economic impact on swine production.
It is important to remember that the exact economic impact of keeping a sow productive,
if just for one more parity, is greatly dependent on current market values and the parity
distribution within a given producer’s operation. In times when market and salvage
prices are high, there is a tendency to be stricter with culling criteria. However, when
market and salvage prices are low, the economic indicators suggest that producers should
keep sows longer (Brandt et al., 1999).

Reasons for removal

The reasons sows are removed from the breeding herd of commercial pig
operations are numerous and have been documented in the scientiﬁc literature for some
time (Stalder et al., 2004). Stalder et al. (2004) provided an informative summary
regarding the percentage of sows culled and reasons culled by study. The reasons for
culling are categorized by reproductive failure, poor performance, old age, feet, leg, and
locomotion disorders, death, farrowing problems, injury, health, disease, and milking
problems. It is clear that reproductive failure is the single biggest reason for a sow’s
removal from the breeding herd accounting for approximately 30% of removals.
Reproductive failure can encompass a variety of problems, including failure to cycle,
inability to conceive, etc.

If reasons for removal are categorized by parity, a slightly different trend can be
seen (Stalder et al., 2004). Reproductive failure and feet and leg problems are the

predominant reasons young sows (sows having three parities or less) are culled. Once

sows reach a more mature age (above 3 parities), culling for reproductive failure becomes
less of an issue and problems like poor performance, age, and even death can be
responsible for just as many or more sow removals (Boyle et al., 1998; Lucia et al.,
1996). Stalder et al. (2004) concluded that producers are more apt to cull mature sows
based on individual performance, whereas the challenge with young sows is maintaining
reproductive function.

Mortality can make up a signiﬁcant portion of the total sows removed from the
breeding herd on an annual basis (Stalder et al., 2004). Additionally, annual sow
mortality rates have been increasing in recent years (Irwin et al., 2000; Duran, 2001).
Stalder et al. (2004) provided a summary of studies that reported mortality rates in
breeding sows by causes of death. The most common causes for sow death included
torsion and other abdominal organ injuries, heart failure, and cystitis (D’Allaire and
Drolet, 1999). Mortalities due to heart failure, torsion, cystitis, and uterine prolapse
tended to be problematic in older sows, while arthritis, endometritis, pneumonia,
locomotion problems, and ulcers appeared to be primary causes for death among gilts and
young sows (Chagnon et al., 1991; Maderbacher et al., 1993). Schultz et al. (2001)
reported that 38-40% of sow deaths occurred 100 to 125 days post breeding, a time at
which a substantial economic investment had already been made. Other factors such as
management, inexperienced labor, housing, rough handling, altering feed quantity or
regimen, environmental temperature, and subjection to stressful events can affect
mortality rates in swine operations (Stalder et al., 2004). Lucia et al. (2000) analyzed
lifetime reproductive performance records from 7,973 females, over a ﬁve year period,

having distinct reasons for removal. The reasons for removal were categorized across

parity categories into reproduction, litter performance, locomotion, and disease problems.
The most common removal reason was culling attributed to reproductive disorders
(33.6%), followed by culling for sub-optimal litter performance (20.6%).

Rodriguez-Zas et al., (2003) and Engblom et al., (2007) reported similar values
for percentage of sows culled for the previously deﬁned reasons for removal. Rodriguez-
Zas et al., (2003) reported locomotion problems and old age accounted, on average, for
5.4 and 7.9% per parity or 11.9 and 17.4% per year, respectively. Engblom et al., (2007)
summarized the most common removal reasons were reproductive disorders (26.9%), old
age (18.7%) and udder problems (18.6%) followed by low productivity (9.5%), lameness
and/or foot lesions (8.6%) and traumatic injuries (7.1%). The average parity number at
removal for all sows in that study was 4.4. In lower parities, most sows were removed
due to reproductive disorders, while udder problems were the most frequent reasons for
removal of medium parity sows.

Takai and Koketsu (2007) identiﬁed at-risk females having characteristics of at
least one of the four subgroups: females with re-services, lactation length < 14 days,
weaning-to-ﬁrst-mating interval 28 days, and abortion records. From 117 farms, 102,494
parity records were categorized into at-risk females and non-at-risk females. At-risk
females had at least 11.1% lower farrowing rates than non-at-risk females among all
parities and seasons of matings (P < 0.05). As parity increased, farrowing rate decreased
by 11.2% and 5.3% for at-risk and non-at-risk females, respectively. Parity 1 females
and those mated during summer had the highest proportion of becoming at-risk females.
Furthermore, gilts and sows with abortion records had at least 39.3% lower farrowing

rates. There was no difference in the number of pigs born alive between at-risk females

and non-at-risk females proving that if at-risk females are closely managed they can be
productive. Producers could improve herd productivity by monitoring and taking added
management steps to reduce the percentage of at-risk females.

To improve longevity through a reduction in mortality within the breeding herd, it
is necessary to determine the speciﬁc cause of death by performing necropsy on sows that
have died (D’Allaire et al., 1992). The period immediately following parturition is the
period when sows are at the greatest risk of death and young sows experience the highest
rate of unplanned removal (Deen and Xue, 1999; Engblom et al., 2007). This period of
the sow’s life cycle should be of particular interest to employees and management within
the farrowing segment of the operation (Stalder et al., 2004). To make improvements on
sow mortality and culling rates, managers should document the cause for removal for the
sows sent to slaughter and also those euthanized and found dead.

Heinonen et al. (1998) collected reproductive organs on 1,708 Finnish female pigs
to determine the cause of fertility disturbances in sows and gilts by examining the
reproductive organs. A large percentage (52.3%) of the organs had no abnormalities,
while 25.1% had inactive ovaries and 6.2% of the reproductive organs possessed cystic
ovaries. This is in agreement with Knauer et al. (2006), who reported that of the sows
culled for reproductive failure, 86.2% were classiﬁed as having normal ovaries. From
these results, one could hypothesize that factors other than the health and ﬁtness of a
female’s reproductive tract is causing reproductive failure among a large percentage of

the sows removed for reproductive reasons.

Factors Affecting Longevity

There are a multitude of factors within swine production systems that can either
directly or indirectly impact a sow’s productive life. These factors include: feeding
programs, body composition at selection and body condition throughout a sow’s
productive life, gilt pool management, age at puberty and ﬁrst farrowing, lactation length,
fertility, season, housing, herd size, sow behavior, feet and leg soundness, employee
management, and disease. This section of the literature review will highlight some of the
studies summarized by Stalder et al. (2004), along with more recent publications.

Nutrition, deﬁned either as the quantity of feed at a particular stage of life or
composition of energy, protein, ﬁber, vitamins or minerals within the diet, does play a
critical role in a female’s life cycle. Nutritional management can impact productivity and
indirectly affect longevity of a female. Rozeboom (1999) provided a review of various
feeding regimens and their impact on sow longevity. The review showed that gilt
development nutritional trials have had a mixture of positive and negative impacts on
sow longevity. Conversely, Kirkwood (1990) and Edge et al. (2003) independently
concluded that there are minimal long-term nutritional treatment effects on lifetime sow
performance. Kirkwood (1990) justiﬁed that any relationship between body composition
at mating and longevity is a reﬂection of selection for improved females that are
conducive to modern conventional management. This improved management minimizes
weight and condition loss during lactation resulting in no association between live weight
or backfat depth at ﬁrst breeding.

Long et al. (1998) reported that sows fed a high energy, high protein diet

throughout their growmg phase had signiﬁcantly poorer stayability through four parities

20

than gilts fed a high energy, low protein diet. Other studies concluded that limiting
energy intake during the growing phase had favorable effects on locomotion, mammary
gland growth, and in one study, resulted in a higher proportion of sows producing four
litters (te Brake, 1986; den Hartog and Noordewier, 1984; Han et al., 2000; Stalder et al.,
1998). Tarres et al. (2003) reported that Duroc females with increased average daily gain
from the end of the growth test until mating had a greater culling rate.

Tokach et al. (2000) outlined practical recommendations for gilt development
feeding programs. These recommendations included increasing calcium and available
phosphorus levels after the gilts have been selected for breeding at approximately 68 to
82 kg. Additionally, it was recommended that gilts be limit fed from 82 kg until 190
days of age or have ad-libitum feed intake of a diet with moderate protein level from 68
to 113.5 kg body weight. Others have also reported on calcium and phosphorus levels in
the diet continue to have an impact on reproductive failure and ultimately sow longevity
(Arthur et al., 1983; Komegay et al., 1984; Koketsu et al., 1996). Boyd et al. (2002) also
provided numerous nutritional and management recommendations to maximize lifetime
sow productivity. Recommendations were provided based on sow body condition scores,
predicted energy and lysine needs for lactating ﬁrst litter sows, and intake targets based
on litter size and stage of lactation.

Patience (2003) reported that gilt body composition, the relative proportion of fat
and muscle in relation to body weight, can be altered to meet some target. He continued
that the body composition targets to optimize production and longevity have not been
established. These optimum composition levels would be dependent on the genetic

background of the sow herd and type of housing. Stalder et a1. (2004) stated that some

21

minimum level of backfat is needed for replacement gilts so that they maximize lifetime
number of piglets born alive. Challinor et al. (1996) reported that gilts that had 18 to 22
mm of backfat at a average weight of 150 kg averaged 7.2 more piglets over ﬁve parities
than did gilts with 14 tol6 mm backfat. Brisbane and Chenais (1996) reported that
increased backfat levels of gilts adjusted to 100 kg inﬂuenced survival rates of Canadian
Landrace females. Survival through the fourth parity was 10% higher in gilts from the
highest backfat category (>18 mm) when compared to gilts from the leanest backfat
category (<10 m). Tummaruk et al. (2001) reported that gilts with higher adjusted
backfat had more live born piglets in their second parity when compared to gilts with low
backfat. Lopez-Serrano et al. (2000) and Tholen et al. (1996b) concluded that there is an
antagonistic relationship between backfat and stayability. Geiger et a1. (1999) reported
that gilts with 18 mm or less backfat at selection experienced higher mortality rates
compared to those with more than 18 mm of backfat. Tarres et al. (2006) created
different hypothetical situations in order to determine the optimal animal body type at
ﬁrst farrowing to maximize longevity. They reported that optimal backfat thickness
should be more than 16 mm at the end of the growth test and maintain this level until the
ﬁrst parturition without exceeding 19 mm.

In related research, Stalder et al. (2005) reported similar results on the effects of
growth and compositional traits on reproductive performance in US. Landrace sows.
Females from the lowest backfat group (S 9.0 mm) had fewer lifetime number of piglets
born alive and females from the fattest group (2 25.0 mm) produced more pigs born alive

during their lifetime than sows from intermediate backfat classes. Additionally, the

22

fattest group averaged a greater maximum number of parities when compared to the other
ﬁve backfat groups.

On the contrary, Young et al. (1991), Kerr and Cameron (1995), Rozeboom et al.
(1996), and Yazdi et al. (2000b) reported different relationships among backfat and
longevity or lifetime productivity than the previously mentioned studies. These reports
concluded that backfat thickness of gilts did not account for a substantial amount of
variation in the number of pigs born alive in parity one, two, or three or overall, and did
not adversely affect reproduction performance or longevity.

Many authors have reported that once a female enters the breeding herd and
becomes a productive sow, composition or body condition can play a critical role in
maximizing longevity. Depending on stage of production (early gestation, late gestation,
or lactation), there are optimum levels of body reserves (both energy and protein)
required to avoid decreased milk production, rebreeding problems, and other reproductive
malfunctions. MacLean (1968) ﬁrst described the “thin sow” problem as starting with
unusually high weight loss in lactation, followed by failure to gain enough weight prior to
farrowing the next litter. This becomes a cumulative process with the sow continuing to
lose weight in successive lactations and leads to reproductive failure, emaciation, and
possibly death. To assess body condition within a production operation Charette et al.
(1996) proposed a combination of linear scores and semi-quantitative scores. This
scoring system took into consideration both changes in backfat and body weight.
Whittrnore et al. (1980) found that sows with 6.3 mm of “mid-back” fat depth or less at
weaning was indicative of an eventual thin sow that would experience reproductive

failure, and result in an early removal from the breeding herd. Reese et al. (1982) and

23

Tantasuparuk et al. (2001) found that thin sows at weaning experienced prolonged
intervals from weaning to estrus. Esbenshade et al. (1986) concluded that fat and weight
loss was inevitable during lactation, but greater losses were experienced in early parities
and by higher producing sows. Ritter et al. (1999) concluded that sows that were a body
condition score less than four (1-5 scale) had a higher incidence of shoulder lesions.
Stalder et al. (2004) summarized the beneﬁts of improved body condition at weaning.
The list of beneﬁts included improved animal welfare, decreased sow mortality,
improved replacement rates, lower wean to estrus intervals, and better reproductive
performance in the next litter. Additionally, sows with improved body condition would
have greater economic value at culling. Knauer et al. (2006) characterized the physical
condition of cull sows from US. Midwestern sow harvest plants and concluded that
variation in body condition was associated with multiple abnormal conditions of sows.
Serenius et al. (2006b) summarized results from the National Pork Producers Council
Maternal Line National Genetic Evaluation and identiﬁed that within genetic lines, sows
with lower feed intake and greater backfat loss during lactation had a shorter productive
lifetime. It was recommended that sow feed intake and backfat loss during lactation in
nucleus and multiplication breeding herds should be considered.

Growth or average daily gain measured during the grow-ﬁnish period has been
associated with longevity. Tholen et al. (1996b), Lopez-Seranno et al. (2000), and
Tummaruk et al. (2001a) have all reported antagonistic relationships between growth and
lifetime productivity and/or longevity. However, Yazdi et al. (2000a) Young et al.
(1991), and Newton and Mahan (1993) found no relationship between pre-adult growth

and the ability of a sow to produce pigs over their productive lifespan or their lifetime

24

proliﬁcacy. Similar to backfat, varying results were found by these studies pertaining to
growth. Stalder et al. (2005) concluded that days to 113 kg provided a signiﬁcant source
of variation for lifetime number weaned, lifetime weight of weaned piglets and ﬁrst
farrowing age on Landrace females. Females were grouped into six categories based on
the standard deviation of each adjusted grth and compositional trait. Females from the
two highest days to 113 kg groups (2196 days to 113.4kg) had a greater age at ﬁrst
farrowing when compared to gilts in the faster growing groups. Females in the highest
(> 210 days) days to 113 kg group had more lifetime number of piglets weaned and
higher lifetime weight of piglets weaned.

Stalder et al. (2004) justiﬁed that these inconclusive results could be due to
differences in timing of measurement, methodology, or the measurement of the trait
itself. They continue to hypothesize that for compositional traits (e.g. growth, backfat,
and loin muscle size) measured at or near selection, it appeared that several of these traits
may have some minimum and/or maximum value that is necessary to maximize lifetime
productivity.

Several authors have reported gilts that reach puberty at an earlier age, have
improved reproductive performance or longevity (Staler et al., 2004). Chapman et al.
(1978) and Holder et al. (1995) reported similar results that reduced age at puberty
resulted in favorable productivity increases in females. Culbertson and Mabry (1995)
reported that sows had reached higher parities at removal when they were ﬁrst mated at a
younger age. LeCozler et al. (1998) concluded that reduced age at ﬁrst farrowing
resulted in more piglets produced over a sow’s lifetime and an increase in parity at

removal. Koketsu et al. (1999) conducted an observational cohort study of data from 33

25

farms and results indicated that increasing the age at ﬁrst mating was associated with low
farrowing rates of maiden females and older age at ﬁrst conception was associated with
lower parity at removal, Shorter reproductive herd life, and fewer lifetime pigs born alive.
Additionally, gilts that recycled before ﬁrst farrowing produced fewer pigs throughout
their lifetime and had a lower parity at removal. Not all studies have reported favorable
association between reduced age at ﬁrst farrowing and longevity. Brooks and Smith
(1980) and MacLean et al. (2001) reported no signiﬁcant difference in productivity given
varying ages of onset at puberty, either induced or naturally occurring.

Within modern pork production, there has been a trend to wean pigs at a younger
age. It has been thought that shorter lactation length allows pork producers to maximize
throughput through the farrowing facilities, improving economic efﬁciency within the
operation, and improve piglet health, due to depleted maternal antibody levels after 21
days of lactation (Stalder et al., 2004). Stalder et al. (2004) summarized a number of
studies and the results indicated that shortened lactation length clearly has a negative
impact on sow longevity. Shortened lactation length results in re-breeding challenges
resulting in sow removal due to reproductive failure, the primary reason for sow removal.
Many of the challenges of early weaning are associated with factors that ultimately
impact sow longevity.

Seasonal variation has been proven to impact reproductive performance within
various types of swine operations. The degree of variation is dependent on the individual
operation. More modern facilities attempt to control temperature extremes and duration
of light to alleviate some reproductive performance issues brought on by seasonal

changes (Love et al., 1993). Thacker (2002) described typical attributes of seasonal

26

infertility as l) delayed onset of puberty, 2) prolonged wean to estrus intervals, 3)
reduced farrowing rate, and 4) increased abortions. Prunier et al. (1997) reported that
high ambient temperature resulted in reduced appetite, milk production, and an increase
in body reserve mobilization. Koketsu and Dial (1997), Black et al. (1993), Xue et al.
(1994), Koketsu et al. (1995b), Koketsu (2000) and Tummaruk et al. (2000) all concluded
that sows which farrowed in the summer had lighter litter weights, longer wean to ﬁrst
service intervals, and an increased mortality risk when compared to other times of year.
As a result, sow culling has been reported to be higher during the summer months (Irwin
and Deen, 2000).

Herd size has been found to have a direct association with high breeding female
mortality risks in the US. As herd size increases by 500 females, mortality risk increased
by 0.44% (Koketsu, 2000). Babot et al., (2003) found that herd size has a signiﬁcant
effect on reproductive performance. The best performance was obtained with herds with
401 to 600 sows compared to herd sizes of < 200 or > 800 sows. The effect of herd size
could be directly related to type of management. Even though large farms can hire
specialized workers, these workers are responsible for a greater number of sows per
person and have less time to pay attention to sows showing clinical symptoms. Another
concern related to herd size is the difﬁculty in eliminating or controlling a disease,
particularly if purchased replacement gilts from another source are frequently introduced.

Type of housing seems to have some interaction with the seasonality of
reproductive ﬁtness. Love et al. (1995) and Hurtgen and Leman (1980) reported that
sows in gestation crates experience little or no depression in farrowing rate during the

summer months when compared to sows gestated in groups. On the contrary, Peltoniemi

27

et al. (1999) reported a seasonal affect on reproductive measures on sows housed either in
groups or individually in stalls.

Along with type of housing, type of ﬂooring has been reported to impact sow
longevity. Smith and Robertson (1971) and D’Allaire et al. (1989) reported an increase
in culling among sows housed on totally or partially slatted ﬂoors. Additionally,
Ehlorsson et al. (2002) reported that sows on partially slatted ﬂooring had a higher
incidence of hoof cracking and heel injuries compared to sows on solid ﬂooring. Feet
and leg soundness, locomotion problems and claw disorders can be major contributors to
sow culling (Stalder et al., 2004). Various systems for scoring feet and leg soundness
have been developed (Van Steenbergen, 1989; Andersen and Hansen, 1996; Grindﬂek
and Sehested, 1996; Lundeheim, 1996; Velander, 1996). Multiple studies have reported
that feet and leg conformation scores are moderately heritable with estimates ranging
from 0.01 to 0.47 (Bereskin, 1979; Lundeheim, 1996; Rothschild and Christian, 1988;
Huang et al., 1995; De Koning, 1996; Serenius et al., 2001; Jorgensen, 1996). Lopez-
Serrano et al. (2000) reported favorable genetic correlation between stayability and leg
score in both Landrace and Large White sows. Brandt et al. (1999) and Jorgensen (2000)
determined the feet and leg quality had a Signiﬁcant inﬂuence on the productive life of
sows. Grindﬂek and Sehested (1996) and Jorgensen (1996) reported that sows with some
ﬂex to their front pastcms have a positive inﬂuence on longevity, while sows that are
straight off their front end and rigid on their pastem have more locomotion problems
resulting in a shorter herd life. On the contrary, Karina et al. (2006) assigned
conformation scores for the forelimbs and hind limbs on 961 gilts to determine the

association between limb conformation scores in gilts and the retention through the

28

second parity. Survival times for females on the basis of conformation scores for the
forelimbs and hind limbs differed signiﬁcantly for total sow removals and removals as a
result of lameness. Risk of removal, speciﬁcally as a result of lameness, increased as
conformation score for the hind limbs became poorer. Tarres et al. (2006) reported that
exterior traits (number of teats, and feet and leg scores, on a scale from 1 to 7) had a
moderate effect on the risk of culling compared with other factors. Sows with an
undesirable leg score of 2 had a 1.4 times greater risk of being culled than sows with an
intermediate and more desirable score of 4. Knauer et al. (2006) collected hoof quality
data on 3,156 sows at two Midwestern sow processing plants and reported the most
common foot lesions observed were on the rear (67.5%) and front (32.8%) pads of
heavier conditioned, higher parity sows. Additionally, cracked toes were found on the
ﬁ'ont and rear feet of approximately 20% of sows, which were usually poorer
conditioned, lower parity sows.
Genetic control of sow longevity

Longevity, along with many other economically important traits in livestock
production, is inﬂuenced by genetics and/or breed composition. Some of the genetic
effects can impact longevity through their correlation with other important traits.
Serenius et al. (2004) estimated the genetic associations of proliﬁcacy with performance,
carcass, meat quality, and leg conformation traits on the Finnish Landrace and Large
White populations. In general, proliﬁcacy traits tended to be favorably correlated with
performance traits and unfavorably correlated with carcass traits. No clear association
was found for proliﬁcacy traits with meat quality or leg scores. It was suggested that

possible improvement in the accuracy of estimated breeding values may be achieved by

29

accounting for genetic associations between proliﬁcacy, carcass, and performance traits.
Imboonta et al. (2007) reported heritabilities for reproduction traits for ﬁrst parity
females of 0.03 for total number of piglets born, 0.04 for number of stillborn piglets, 0.06
for number of piglets born alive but dead within 24 hours, versus later parities of 0.07,
0.03, and 0.02, respectively. Heritability estimates for average daily gain and backfat
were 0.38 and 0.61, respectively. Interestingly, there were favorable genetic correlations
found between average daily gain and total born and unfavorable genetic correlations
between backfat and total born. Additionally, backfat was genetically unfavorably
correlated with number of stillborn piglets in later parities. As a result, selection for high
growth rate could increase total born, and selection for low backfat will decrease total
born and increase number of stillborn. Arango et al. (2005) studied the genetic
relationship between grouped reasons for sow removal in consecutive parities. Estimates
of heritability were 0.18 for reproductive reasons, 0.13 for non-reproductive reasons, and
0.15 for other reasons. Results suggest that additive genetic variance exists for parity at
removal and different reasons of removal.

Stalder et al. (2004) summarized a number of studies reporting genetic factors
affecting longevity. Ivkove et al. (1986) reported that crossbred sows averaged 5.3 litters
while purebred sows averaged 4.4 litters at culling. They also noted that 55.2% of culling
in purebred sows occurred in the ﬁrst three parities. During the ﬁrst three parities, only
40.4% of the overall culling occurred in the crossbred sows. Jorgensen (2000) reported
that mean age and number of litters at removal were lower in purebred Yorkshire sows
when compared to crossbred sows. Speciﬁcally, purebred sows had higher culling for

locomotion problems and reproductive failure. Sehested and Schjerve (1996) reported

30

similar results in which parity at culling averaged 3.01 for purebred sows and 3.61 for
crossbred sows. Hall et a1. (2002) conducted a study involving Meishan and Duroc
crossbred sows and reported that breed composition of crossbred females can inﬂuence
longevity. Rodriguez-Zas et a1. (2003) noted differences that approached a full parity
difference between some genetic lines. They also reported signiﬁcant differences in
instantaneous sow removal rates among the genetic lines investigated. Ehlers et a1.
(2005) estimated variance components and heritabilities for sow productivity traits on
purebred versus crossbred sows. Results suggested that pooling of purebred and
crossbred data may be considered, which may potentially increase the accuracy of
breeding value estimates.

Johnson (2000) initially reported results from the National Pork Board Maternal
Line Project which demonstrated that traits contributing to longevity and attrition were
heritable. The same report noted line differences for percentage of sows producing four
litters, live pigs per sow life, and average sow life. Goodwin (2004) extended the
analysis of the same maternal line study and found similar differences through the sixth
parity. Serenius et al. (2006b) continued work with the National Pork Board Maternal
Line Project data by applying non-parametric proportional hazards models to compare
longevity of sows from the 6 commercial genetic lines and developmental factors that
may impact longevity. Results indicated that one line of sows had a lower risk of being
culled than sows from the other lines.

Most quantitative traits in animal genetics are approximated using normal
distributions, and classical procedures can be used for the estimation of genetic

parameters such as heritability (Yazdi et al., 2002). For the Gaussian linear mixed model,

31

the best predictor of individual breeding values is linear, where the heritability of the trait
is given by the ratio of the additive genetic variance to the total phenotypic variance
(Korsgaard et al., 2000). On the contrary, the theory of the statistical analysis of survival
data is based on the use of special modeling distributions (e.g. exponential, Weibull,
lognonnal, gamma). The parametric distributions used for longevity traits often come
from the family of generalizations of exponential distributions. Survival models that
include random effects, which are required in genetic evaluation, are referred to as frailty
models. The frailty term is deﬁned as an unobserved random quantity which affects
multiplicatively the hazard of individuals or groups of individuals (Ducrocq, 1997).
When a term is deﬁned separately for each individual, the frailty component extracts part
of the unobserved variation between individuals and therefore allows for a correction of
the possible discrepancy between the true variance of the observations and the one
speciﬁed by the model. Such an extra variation is referred to as “overdispersion” (Louis,
1991; Tempelman and Gianola, 1994). Survival analysis is considered theoretically
superior in longevity analysis (Serenius, 2004; Serenius and Stalder, 2006). This
approach accounts for censored observations, non-normal distributions, and models time-
dependent effects. However currently, only single-trait analysis is possible when using
the Survival Kit (Serenius and Stalder, 2004). On the contrary, linear model analysis is
capable of implementing multiple-trait models. Serenius and Stalder (2006) summarized
the current literature regarding the genetics of sow longevity and discussed the available
breeding value estimation methods for sow longevity traits. Additionally, heritability
estimates from numerous studies are summarized by the deﬁnition of longevity used and

the genetic relationship of proliﬁcacy traits, leg conformation, and developmental traits

32

on longevity. There is a need for continued research to determine the best methodology
for estimating breeding values for sow longevity. Also, additional focus needs to be
directed towards determining the magnitude of non-additive genetic variation that affects
ﬁtness traits such as sow longevity (Falconer and Mackay, 1996). As a result, Serenius et
al. (2006a) estimated variance components, including dominance genetic variation, for
overall leg action, length of productive life and sow stayability until third and ﬁfth parity
in Finnish pig populations. These investigators recommended that the effect of
dominance should be accounted for in the breeding value estimation of sow longevity,
especially when data from crossbred animals are included in the analysis.

Reported estimates of the heritability for stayability, analyzed using linear models
ranged from 0.02 to 0.11 (Tholen et al., 1996a, 1996b; Lopez-Serrano et al., 2000;
Serenius and Stalder, 2004), while the survival analyses estimates for length of
productive life have ranged from 0.11 to 0.31 (Yazdi et al., 2000a, 2000b; Serenius and
Stalder, 2004). Serenius (2004), who justiﬁed the difference in the estimates, indicates
that environmental effects are modeled more accurately in survival analysis than in linear
model analysis, due to the ability to model the farm-year effect as a time dependent
variable and the inclusion of censored records. The Cox and Weibull frailty models have
been used for analyzing survival data. In these models, it is assumed that log frailty is the
sum of log gamma and/or normally distributed random variables. When including an
error term in log frailty, heritability can be deﬁned on the log frailty scale, as well as on
the linear scale (Korsgaard et al., 2000). Ducrocq (1987) derived the heritability on the
logarithmic scale (hzlog) for Weibull sire model as follows:

Nm=4&noﬂ+FM)

33

where 13/6 in this equation is the variance of an extreme value distribution and 625 is the
sire variance.

Stalder et al. (2004) and Serenius and Stalder (2006) summarized the following
studies with reported heritabilities. Crump (2001) reported heritability estimates ranging
from 0.11 to 0.21, depending on whether survival analysis, linear model, or generalized
linear model methods were used. Additionally, he reported unfavorable genetic
correlations between average daily gain and survival and backfat and survival. Lopez-
Serrano et al. (2000) found heritability estimates for stayability ranged from 0.07 to 0.11
in Landrace sows. Tholen et al. (1996b) reported that heritability for stayability from
parity one to two, one to three, and one to four was 0.05, 0.06 and 0.09, respectively.
Additionally, stayability was unfavorably correlated with average daily gain and backfat
and favorably correlated with weaning to conception interval. Fortin and Cue (2002)
reported heritabilities for length of productive life in Yorkshire and Landrace populations
of 0.16 and 0.13, respectively. Yazdi et al. (2000a) reported heritability estimates for
longevity ranging from 0.11 to 0.27 on the original scale. The estimates were similar
within each group of models, averaging 0.13 for the time-independent and 0.25 for the
time-dependent herd by year effect in the model. Correlations between the sires’
breeding value estimates ranged from 0.96 to 0.98 among the differing models. Gou et
al. (2001) reported that heritability for lifetime proliﬁcacy was 0.25 for length of
productive life. Serenius and Stalder (2004) estimated heritabilities using both a survival
analysis procedure and linear models. The heritability estimates for length of productive
life were lower for the linear model (0.05 to 0.10) than for survival analysis (0.16 to

0.19). The authors mentioned that in their survival analysis the residual variance was not

34

estimated but rather assumed that the residual effects followed the extreme value
distribution and they cautioned that the different heritability estimates were not
comparable. Yazdi et al. (2000b) reported heritabilities in the original scale for longevity
ranged from 0.21 to 0.31 on a study of survival of Swedish Landrace and Yorkshire sows.

Heritability estimates for longevity indicate that selection for improved longevity
is possible, but improvement would be likely to be slow (Stalder et al., 2004).
Additionally, both favorable and unfavorable genetic correlations of production traits and
exterior traits have been reported to inﬂuence the productive life of sows (N apel and
Johnson, 1997; Tholen et al., 1996a, 1996b; Serenius et al., 2004; Serenius et al., 2006a;
Tarres et al., 2006; Imboonta et al., 2007). Consideration needs to be given to how
longevity will be deﬁned (e.g. stayability, length of productive life, or lifetime
proliﬁcacy) and the effect that performance characteristics may have on each defrnition.
It appears that how longevity is deﬁned within a population can impact resulting
heritabilities.

True vs. Functional Herd life

Ducrocq et al. (1988a) made an important distinction between true and functional
measures of herd life. These distinctions were applied to the concept of stayability. This
ability is referred to as true stayability when the reason for leaving the herd is not taken
into account. The removal of an animal from the herd can be either voluntary or
involuntary. Voluntary culling is based on management decisions made by the producer.
Involuntary culling is the removal of an animal due to factors beyond the control of the
producer such as failure to breed or locomotion disorders. The capacity of an animal to

avoid involuntary culling is termed functional stayability. In measuring functional

35

 

stayability, true stayability is adjusted for factors such as backfat, growth, or number of
piglets born. For example, Serenius and Stalder (2004) conducted two separate longevity
analyses. For the ﬁrst model, the only effects included in the model were farm-year held
as a ﬁxed time-dependent variable and a random genetic sire effect to approximate true
length of productive life. For the second model, the ﬁxed effects of leg score, number of
weaned piglets in ﬁrst litter and a regression of age at ﬁrst farrowing also were included
in the model to approximate functional length of productive life. Heritabilities from the
two models were slightly higher for functional length of productive life, indicating that
the ﬁxed effects included in the model can inﬂuence heritability estimates.
Summary

It has been established that decreased sow longevity can lead to economic
inefficiency and animal well being concerns. Improved sow longevity could have an
impact on proﬁtability by reducing annual replacement gilt costs and the added
management required of purchased gilts. Sow longevity is clearly a complex trait with
many developmental factors that can inﬂuence a female’s length of productive life.
Additionally, the positive estimates of heritability indicate that there is genetic variation
among the animals that can be utilized for improving longevity. However, given the low
heritability of survival traits, it may be necessary to utilize correlated traits measured
early in life to substantially augment improvement in longevity. Furthermore, multiple
deﬁnitions have been utilized to deﬁne a female’s productive life. Length of productive
life is the most common but has not been concluded that it is the most effective
deﬁnition. The following research addresses some of the current issues facing survival

analysis in swine.

36

CHAPTER II
CHARACTERIZATION OF SOW LONGEVITY AND THE DEVELOPMENTAL
FACTORS THAT INFLUENCE IT
Abstract
The length of adult sow life is now recognized as both an economical and welfare

concern. However, there are no consistent deﬁnitions to measure sow longevity. This
study assesSed six different descriptions of longevity and determined the relationship of
developmental performance factors with them. Longevity deﬁnitions included stayability
(probability of producing 40 pigs or probability of reaching four parities), lifespan
(number of parities a female has accumulated before culling), lifetime proliﬁcacy
(number of pigs born alive during a female’s productive lifetime), herdlife (time from
ﬁrst farrowing to culling) and productive life (pigs produced per day of life). Data
consisted of 14,262 records of Yorkshire females from both nucleus and multiplication
herds across 21 farms from four seedstock systems. Within a subset of the data,
information was available on the female’s litter birth record and her growth and
composition data. Therefore, the data were subdivided into two data sets, consisting of;
A: Data from a female’s farrowing records, and B: Data A and information from a
female’s litter birth record and a female’s growth and backfat data. A Cox proportional
hazards model was used to determine the relationship of developmental factors and ﬁrst
farrowing record with longevity. Those factors that were signiﬁcantly (P <0.0001)
associated with longevity, regardless of deﬁnition, were age at ﬁrst farrowing, litter size
at ﬁrst farrowing and last farrowing, number of stillborn in ﬁrst litter, adjusted 21 day

litter weight of ﬁrst litter, herdtype, backfat, and growth. Within a contemporary group,

37

fatter, Slower growing gilts had a decreased risk of being culled. Additionally, sows that
had more pigs born alive, fewer stillborn, and heavier litters at weaning in their ﬁrst litter
had a decreased risk of being culled. Furthermore, sows from nucleus herds experienced
a greater risk of being culled. Many factors impacted longevity, regardless of deﬁnition.
Pork producers can implement management protocols which can extend the productive
life of breeding females resulting in improved proﬁtability and animal welfare.
Introduction

For the past two decades, pig genetic improvement has focused on production
(growth, carcass, and meat quality) and reproduction characteristics (litter size) (Tarres,
2005). Selection for increased lean grth along with the acceleration of the
reproductive rhythm of sows have resulted in culling rates within commercial herds in
excess of 50% (Rodriguez-Zas et al., 2003; Tarres, 2005). Sows are replaced, on
average, by the third or fourth parity (D’Allaire et al., 1987; Boyle et al., 1998) through
involuntary and voluntary culling practices. Longevity of sows summarizes the effects of
functional traits on the ability to delay involuntary culling (Ducrocq and Solkner, 1998a).
Longevity of adult females is a trait with a signiﬁcant impact on swine farm proﬁtability.
An improvement in longevity can result in decreased replacement costs and a higher
proportion of mature sows in a herd that have reached their maximum productivity. The
length of the adult productive life of a sow is recognized as both an economical and a
welfare concern.

Stalder et al., (2004) reported that there are a multitude of factors within swine
production systems that can either directly or indirectly impact a sow’s productive life.

Developmental factors such as growth rate, body composition at selection, age at puberty

38

and ﬁrst farrowing, and feet and leg soundness have been proven to inﬂuence longevity.
Additionally, parity one lactation length and wean to estrus interval are early indicators of
a female’s fertility and potential length of productive life. The most consistent source of
variation across studies is the herd by year interaction. Sow housing and behavior, herd
size, seasonal fertility, disease, gilt pool management, and human interaction have been
associated with explaining the variation in longevity. As a result, it is important to assess
each production unit independently when quantifying sow longevity. Management
practices of producers, farm employees and also breeding programs should focus on
improving sow longevity.

There has not been a clear consensus in the scientiﬁc literature regarding the
deﬁnition of longevity. Yazdi et al. (2000b), Serenius and Stalder (2004) and Tarres et
al. (2003, 2006) deﬁned longevity as length of productive life (number of days from ﬁrst
farrowing to removal from the herd). Sow longevity has been deﬁned by Serenius and
Stalder (2004) as lifetime proliﬁcacy (the total number of pigs produced in a sow’s
lifetime) and length of productive life (the number of days from the date of ﬁrst
farrowing to the date of culling or censoring). Holder (1995) and Knauer et al., (2006)
combined a measure of productivity into a continuous measure of time as pigs born alive
per day of herd life.

The objective of this study was to assess six different measures of longevity and
determine what relationship developmental performance characteristics may have with

these differing deﬁnitions.

39

Materials and Methods
Data

A unique data set has been assembled to complete this project. The data was an
accumulation of both nucleus and multiplication records of Yorkshire females from four
swine seedstock systems (Waldo Farms, DeWitt, NE; Premier Swine Systems,
Michigantown, IN; Whiteshire/Hamroc, Albion, IN; and the Zierke Co., Morris, MN).
Herd production data from computer software (HERDSMANTM) was extracted from each
of these farms and merged into a common data set for analysis purposes. All crossbred
and non-Yorkshire sows were removed. Individual pig performance and farrowing
records from 21 different farms, which were associated within these four seedstock
systems were merged into one data set for analysis.

From these 21 farms, records on Yorkshire females born from 1988 through 2002,
with at least one farrowing record, were available. Information from the female’s birth
litter along with her performance for age at 113 kg and backfat thickness adjusted to 113
kg as suggested by National Swine improvement Federation (1996) were merged with
subsequent sow performance records. Individual parity records were then summed across
parities to accumulate total sow performance and determine the last record in the
database for an individual female. Additionally, a code indicating if an animal had been
culled or was still in production was available and was used to indicate censoring. The
culling reason was not available. Variables with extreme values, more than four standard
deviations away from the mean, were deleted. Additionally, if two variables were highly
correlated (eg. number born and number born alive), the variable that had a higher

frequency of recording was retained for analysis. Due to incomplete data recording

40

across farms and even within a farm over time, not all sows had complete records for
birth litter traits, growth rate and backfat thickness. To make optimal use of the data, the
data set was split into two. After editing the largest data set (Data A), 14,262 records
from Yorkshire sows with at least one farrrowing were considered for survival analysis,
including 10,651 (74.68%) and 3,611 (25.32%) uncensored and censored records,
respectively. Information included a sow’s ﬁrst parity information along with her
lifetime productivity data. The individual record for each female included herd type
(either multiplier (1) or nucleus (2)) (HT), sow herd (the herd that the female was a
productive sow), date of ﬁrst farrowing, date of culling, age at ﬁrst farrowing (AGEF F),
number born alive in the ﬁrst litter (NBA), number of stillborn in the ﬁrst litter (N STL),
length of lactation in the ﬁrst parity (WNAGE), adjusted 21 day litter weight from the
ﬁrst litter (D21W), and litter size (born alive) at the last farrowing (LNBA).

The second data set (Data B) contained 4,272 records from Yorkshire sows with
at least one farrowing record, with 3,395 (79.47%) and 877 (20.53%) uncensored and
censored records, respectively. Data B contained all variables previously listed in Data
A, along with information related to the sow’s birth litter record and adjusted days to 113
kg and adjusted backfat thickness at 113 kg. Standardized variables were created for
growth and backfat thickness. Within a pig herd, contemporary group was deﬁned by
birth month and year. Means and standard deviations were calculated within each of
these subclasses and then a female’s individual backfat measurement and growth record
were deviated from the contemporary group mean and then divided by the standard

deviation for that particular contemporary group. This approach allowed for comparisons

41

across contemporary groups accounting for differences in methodology of data collection
and environment.

The individual record for each female included all covariates present in Data A,
along with pig herd (the herd that the female was born in), date of birth, birth litter
number born alive (PNBA), birth litter birth weight (PLBW), percentage of gilt pigs in
birth litter (PGR), standardized deviation for backfat thickness (BF), and standardized
deviation for adjusted days to 113 kg (DAY).

Six different deﬁnitions of longevity were developed. Length of productive life
was determined as the number of days from the date of ﬁrst farrowing to the date of
culling or censoring (LPL). Similarly, the lifespan of a sow was determined as the
maximum parity she completed prior to removal (MAXPAR). Lifetime proliﬁcacy was
determined as the number of piglets (born alive) produced during the LPL of the sow
(SNBA). Stayablility, has been deﬁned as a binary trait measuring whether a sow has
survived in a herd until some deﬁned ﬁxed parity or time (Serenius and Stalder, 2004).
From this deﬁnition, two different discrete measures were determined using production
based thresholds. They were the ability of a sow to achieve 4 parities prior to removal
(PR4PAR) and the ability of a sow to produce 40 pigs prior to removal (PRNBA40). A
measure of length of life and a measure of productivity were combined to determine the
ﬁnal response variable by each female’s SNBA divided by the difference between cull
date and birth date (PLIFE).

Statistical Analysis
A Cox proportional hazards model was used to determine the relationship of

developmental factors with longevity (Cox, 1972). All analyses were conducted using

the PHREG procedure within SAS 9.1 (SAS Inst. INC., Cary, NC). Initially, all possible
ﬁxed effects were tested for their level of signiﬁcance in a univariate model with each of
the six response variables. Predictors with Chi-square test signiﬁcance levels at P-value
of <0.25 were included in the ﬁnal model. A primary assumption of the Cox proportional
hazard model is proportionality, that is that the estimated hazards over time will be
constant for two females with particular values for the covariates. Proportionality was
tested by including time-dependent covariates in the model. Time-dependent covariates
are interactions of the predictors with time. It was concluded through likelihood ratio
tests, that the initial model violated the proportionality assumption. As a result, models
were stratiﬁed by the interaction of sow herd by farrow year for Data A and by the
interaction of pig herd by birth year for Data B to satisfy the proportionality assumption.
The ﬁnal stratiﬁed regression model for Data A included the ﬁxed effects of AGEF F,
NBA, NSTL, LNBA, D21W, WNAGE, and HT and for Data B the models included all
of the covariates listed for Data A with the addition BF, DAY, PNBA, PLBW, and PGR.
All covariates were included in each model for each response variable. Additionally, all
possible ﬁrst level interactions of covariates, that were not highly correlated (r 2 0.60),
were tested and were found to be not signiﬁcant.
Results and Discussion

Descriptive statistics for the data are presented in Table 11.2. The average LPL for
Yorkshire females was 488.8 days, with a mean parity at removal or censoring of 3.5.
Throughout their productive lifecycle, females produced on average 34.9 pigs with an
average age of ﬁrst farrowing of 366.2 days. Additionally, 62% and 59% of sows were

removed or considered censored before producing 40 live piglets or completing 4

43

parities, respectively. The descriptive statistics of the data are consistent with other
survival analysis studies (Yazdi, 2000b; Stalder et al., 2004; Serenius and Stalder, 2004).

Results from the proportional hazards models and the discrete logistic models are
presented in Tables 11.3 through 11.8 for Data A and Tables 11.9 through 11.14 for Data B.
The Chi-Square value and a probability of signiﬁcance value is provided to assess the
signiﬁcance of each variable included in the models. To compare the inﬂuence of each
covariate on the respective response variable, one should consider the sign (+,-) of the
parameter estimate for each covariate. The sign of the parameter estimate indicates either
an increasing (+) or a decreasing (-) hazard given a one unit increase in the covariate with
the assumption that all other covariates are held constant. Additionally, the hazard ratio
is provided for each covariate. Hazard ratios greater than 1 indicate an increase in risk of
removal while a hazard ratio less than 1 indicates a decrease in risk of removal. For
example in Data A within the LPL deﬁnition of longevity, AGEF F had a positive
parameter estimate and a hazard ratio of 1.002. This indicates that as AGEFF increased
by 1 day and all other covariates were held constant, the risk of a sow being culled
increased by 0.2%. Initially, this may not seem important, but if the average AGEFF was
increased by 10 days this would increase the risk of removal by 1.0%.
Data A

The effect of age at ﬁrst farrowing on the risk of being culled in the present study
was highly signiﬁcant (P<0.0001), regardless of the deﬁnition. Additionally, the hazard
ratios were very similar across deﬁnition ranging from 1.002 to 1.004. The positive Sign
of the parameter estimate indicates that the greater the age at ﬁrst farrowing, the greater

the risk of a sow being culled. These ﬁndings were consistent with several authors who

44

have reported gilts that reach puberty at an earlier age, or were ﬁrst mated at a younger
age, have improved reproductive performance or longevity (Culbertson and Mabry, 1995;
Tholen et al., 1996a and 1996b; LeCozler et al., 1998; Koketus et al., 1999; Yazdi et al.,
2000b; Stalder et al., 2004; Serenius and Stalder, 2007). An increased AGEF F may be an
indicator of the delayed onset of puberty or could be attributed to maiden gilts that did
not conceive on the ﬁrst service. Delayed AGEFF could be considered an early indicator
of reproductive ﬁtness or lack thereof. Sterning et al. (1998) reported a tendency for
sows that were older at puberty to show delayed return to estrus after weaning. However,
Serenius and Stalder (2007) did caution against breeding immature gilts. They
summarized previous reports which indicated that litter size at ﬁrst farrowing tended to
increase with the increase in age at ﬁrst farrowing or at ﬁrst mating (Schukken et al.,
1994; Tummaruk et al., 2001a). They suggested that an optimal age at ﬁrst conception
may be close to 200 to 210 days.

The current results indicate that the performance of a female during her ﬁrst litter
does provide insight for the rest of her productive life. The proliﬁcacy traits evaluated in
the current study were total number of piglets born alive and number of stillborn piglets.
Additionally, adjusted 21 day litter weight was included in the model. Sows that had
more pigs born alive, fewer stillborn fetuses, and weaned heavier litters, tended to have a
decreased risk of being culled across all deﬁnitions. Sows that had one more pig born
alive in their ﬁrst litter had a decreased relative culling rate that ranged from 1.4% for
LPL to 10% for PLIFE. The three deﬁnitions that included a productivity component
(PLIF E, SNBA, PRNBA40) reported higher hazard ratios than those deﬁnitions that were

more time dependent (LPL, MAXPAR, PR4PAR). The number of stillborn piglets was

45

highly signiﬁcant (P<0.0001) across all deﬁnitions. As the number of stillborn piglets
increased by one piglet, the relative culling rate increased from 4.6% to 6.2% depending
on deﬁnition. This indicator of neonatal piglet mortality rate had a strong negative
association with longevity.

Sows that weaned heavier litters, in their ﬁrst parity, experienced a decreased risk
of being culled. A one kilogram increase in adjusted 21 day litter weight tended to
decrease the risk of removal by 0.4% on average across the six deﬁnitions. The number
of pigs born in a female’s last litter was also strongly signiﬁcant (P <0.001) across all
deﬁnitions. Similarly to NBA, if a sow had one more pig in her last litter her relative
culling risk was reduced by as little as 1% for MAXPAR and as much as 7% for
PRNBA40. Once again, LNBA reported higher risk ratios for deﬁnitions that include a
productivity component when compared to deﬁnitions that were time dependent only.
These results are in agreement with Yazdi et al. (2000b) who reported a decreased risk
ratio with increasing litter size for both ﬁrst and last litter size across all models when
longevity was deﬁned as length of productive life. Additionally, Serenius and Stalder
(2007) concluded that a small number of pigs born at ﬁrst farrowing had a detrimental
impact on length of productive life. These results were not in agreement with Serenius et
al. (2006b) who concluded, on a different ﬁeld data set comparing six different genetic
lines, that there was no clear association between litter size and length of productive life
when culling due to poor reproductive performance was not practiced. According to the
present study, small litter sizes in the ﬁrst and last parity resulted in earlier removal from
the herd. Cause of culling was not available for this study, thus it was impossible to

directly distinguish between voluntary and involuntary culling of these sows.

46

Length of lactation in the ﬁrst litter (WNAGE) was not signiﬁcant for SNBA,
PLIFE, PRNBA40, and PR4PAR, but was signiﬁcant (P<0.05) for LPL and MAXPAR.
It should be noted that with the exception of PRNBA40, the Sign of the parameter
estimates were positive across all deﬁnitions. These results are not in agreement with a
number of studies summarized by Stalder et al. (2004), which indicated that shortened
lactation length had a negative impact on sow longevity. It was concluded that weaning
at early ages could adversely impact sow longevity through increased culling due to
reproductive failure. Studies that found a clear negative association of lactation length
with longevity reported on lactation lengths of 16 days or less (Koketsu, 1995a; Xue et
al., 1997). The mean lactation length for this study was 20.1 days. This longer lactation
length could explain the non signiﬁcance of WNAGE. Females that are allowed to
lactate longer in their ﬁrst parity experience shorter wean to estrus intervals and higher
conception rates.

With the exception of LPL, HT affected all other deﬁnitions. Sows from nucleus
herds had an increased relative risk of being culled by 42% for SNBA and 68% for
PLIF E and MAXPAR. Additionally, sows from nucleus herds were 3 times more likely
to be removed prior to producing 40 pigs or completing 4 parities. Without reasons for
culling, the effects of herd type were implicitly the result of the differences in herd
management and variation in culling criteria between multiplier herds and nucleus herds.
Data B

The standardized variables for backfat and days to 113 kg signiﬁcantly inﬂuenced
all deﬁnitions of longevity. The estimated hazard coefﬁcients for BF were negative

indicating that fatter gilts within a contemporary group tended to have a decreased risk of

47

being culled. Additionally, the estimated hazard coefﬁcients for DAY were negative
indicating that slower growing gilts within a contemporary group tended to have a
decreased risk of being culled. The hazard ratios for both covariates revealed that with a
one unit change the risk of being culled decreased anywhere from 6.7% to 15.8%,
depending on the deﬁnition.

The results in the present study are in agreement with a number of studies that
concluded that gilts which had higher backfat were more productive or experienced a
longer productive life. Stalder et al. (2004) concluded that there may be a minimum level
of backfat needed for replacement gilts for them to maximize lifetime number of piglets
born alive. Challinor et al. (1996), Brisbane and Chenais (1996), and Geiger et al. (1999)
all indicated that gilts that had greater than 18 mm of backfat were more productive,
experienced less mortality and had a decreased risk of being culled. Similarly, Terres et
al. (2003) reported that optimum backfat thickness was 16 mm at 96.2 kg in Duroc
females. Stalder et al. (2005) reported that females from the fattest group (_>_ 25.0mm)
produced more pigs born alive during their lifetime than sows from intermediate backfat
classes. These studies collectively support the results from the current study that
increased backfat levels in replacement gilts favor a longer, more productive herd life.
Not all literature reports are in complete agreement with the current ﬁndings. Young et
al. (1991), Rozeboom et al. (1996), Kerr and Cameron (1995), and Yazdi et al. (2000b)
reported that backfat thickness of gilts did not account for a substantial amount of
variation in the number of pigs born alive and did not adversely affect reproduction

performance or longevity. Serenius and Stalder (2007) reported that backfat thickness

48

had a very small effect on risk of removal in a Finnish crossbred population, though
females that had more backfat thickness tended to have a lower risk of being culled.

The results in the present study are in agreement with others that reported an
antagonistic relationship between growth rate and measures of longevity, either direct or
indirect (Tholen et al., 1996b; Lopez-Seranno et al., 2000; Tummaruk et al., 2001a).
However, this antagonistic relationship seems to be population dependent. Yazdi et al.
(2000b), Serenius and Stalder (2004), and Serenius and Stalder (2007) found no
signiﬁcant effect of daily gain on longevity.

Current results indicate that covariates for traits measured within a female’s birth
litter (PNBA PLBW, PGR) did not signiﬁcantly impact longevity, regardless of
deﬁnition. Serenius and Stalder (2007) also concluded that the litter size in which the
sow was born into did not signiﬁcantly impact her length of productive life.

With the exception of WNAGE, covariates for ﬁrst litter performance traits
(NBA, NSTL, D21W) included in the analysis for both Data A and Data B were highly
signiﬁcant with similar parameter estimates and hazard ratios. Sows that had more
piglets born alive, fewer stillborn piglets, and weaned heavier litters in their ﬁrst parity
had a decreased risk of removal. Additionally, an increase in the number of pigs born
alive in a female’s last litter decreased her risk of removal. Furthermore, herd type was
highly signiﬁcant for all deﬁnitions with the exception of PLIFE. Regarding the other
ﬁve deﬁnitions, sows from nucleus herds had an increased relative risk of being culled by
an average of 43%.

The current results indicate that regardless of deﬁnition, developmental and

performance of ﬁrst parity females does provide insight into the longevity of the

49

Yorkshire females in this population. Within a contemporary group of potential
replacement females, gilts that grew slower and had more backfat had a decreased risk of
being culled. Additionally, parity 1 sows that farrowed at a younger age, were more
proliﬁc, had fewer stillborn fetuses, and weaned heavier litters tended to have a decreased
risk of being removed. These results indicate that there are early indicators that can be
monitored to provide insight to the future productivity or length of productive life of
Yorkshire females. The six different deﬁnitions yielded similar results indicating that the

“best” deﬁnition for assessing longevity could be considered producer dependent.

50

Table 11.1. Deﬁnitions of variable acronyms used in model development

 

Variable

Acronym

 

Number of days between date of ﬁrst farrowing and removal, d
Number of parities a sow had completed

Pigs born alive per day of sow herd life

Binary indicator variable if a sow had produced 40 pigs (0,1)
Binary indicator variable if a sow had completed 4 parities

(0,1)

Total number born alive over lifetime

Herd type, 1 multiplier, 2, nucleus

Age at ﬁrst farrowing, d

aStandardized deviation for adjusted backfat, mm
aStandardized deviation for adjusted days to 113 kg
Adjusted 21 day litter weight in parity 1 litter, kg

Birth litter gilt ratio

Birth litter, litter birth weight, kg
Birth litter number born alive
Last litter number born alive
First litter number born alive
First litter number stillborn

First litter lactation length, d

LPL
MAXPAR
PLIFE
PRNBA40
PR4PAR

SNBA
HT
AGEFF
BF
DAY
D2 1 W
PGR
PLBW
PNBA
LNBA
NBA
NSTL
WNAGE

 

aAdjusted female record was deviated from the birth month and year mean for gilts tested

within herd and divided by the standard deviation.

51

Table 11.2. Number of observations, means, standard deviations, minimums, maximum
values of the traits studied in US. purebred Yorkshire sows.

 

 

Traita No. Mean SD MIN MAX
LPL,d 14,262 488.8 41 1.4 10 3,548
MAXPAR 14,262 3.5 2.4 1 14
SNBA 14,262 34.9 27.4 0 182
PLIFE 14,262 0.036 0.016 0
0.277

PRNBA40”

0 8,857

1 5,405
PR4PARb

0 8,409

1 5,853
AGEFF,d 14,262 366.2 46.2 251 600
NBA 14,262 9.4 3.0 0 20
NSTL 14,262 0.7 1.2 0 14
LNBA 14,262 9.5 3.3 0 20
021w, kg 14,262 51.1 14.3 4.54 115.9
WNAGE, d 14,262 20.1 5.0 0 40
BF 5,861 -021 0.912 -3.16 3.85
DAY 5,962 -0.33 0.855 -3.21 3.58
PGR 7,906 0.54 0.15 0.07 0.92
PNBA 8,465 10.8 2.8 1 20
PLBW, kg 6,879 16.12 4.42 1.36 34.1

 

aSee Table 11.1 for acronym deﬁnitions.

52

Table 11.3. Estimated hazard ratios for Data A for LPLa

 

 

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
AGEF F 0.0019 <0.0001 1.002 1.001 1.002
NBA -0.0139 0.0005 0.986 0.978 0.994
NSTL 0.0445 <0.0001 1.046 1.027 1.065
LNBA -0.016l <0.0001 0.984 0.978 0.990
D21W -0.0037 <0.0001 0.996 0.995 0.997
WNAGE 0.0081 0.0115 1.008 1.002 1.014
HT 0.0879 0.1846 1.091 0.959 1.241

 

8'Number of observations = 12,851.
bSee Table 11.1 for acronym deﬁnitions.

53

Table 11.4. Estimated hazard ratios for Data A for SNBA“

 

 

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
AGEF F 0.0017 <0.0001 1.002 1.001 1.002
NBA -0.0643 <0.0001 0.938 0.930 0.945
NSTL 0.0501 <0.0001 1.051 1.032 1.071
LNBA -0.0599 <0.0001 0.942 0.936 0.948
D21W -0.0030 <0.0001 0.997 0.996 0.998
WNAGE 0.0047 0.1378 1.005 0.998 1.011
HT 0.3513 <0.0001 1.421 1.254 1.610

 

1’Number of observations = 12,851.

bSee Table 11.1 for acronym deﬁnitions.

54

Table 11.5. Estimated hazard ratios for Data A for PLIFEa

 

 

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
AGEFF 0.0042 <0.0001 1 .004 l .004 1.005
NBA -0.1054 <0.0001 0.900 0.893 0.907
NSTL 0.0437 <0.0001 1.045 1.026 1.064
LNBA -0.1016 <0.0001 0.903 0.898 0.909
D21W -0.0016 <0.0001 0.998 0.998 0.999
WNAGE 0.0033 0.3106 1.003 0.997 1.010
HT 0.5196 <0.0001 1.681 1.488 1.900

 

1'Number of observations = 12,851.

bSee Table 11.1 for acronym deﬁnitions.

55

Table 11.6. Estimated hazard ratios for Data A for MAXPARa

 

 

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
AGEF F 0.0021 <0.0001 1.002 1 .002 1.003
NBA -0.0209 <0.0001 0.979 0.970 0.988
NSTL 0.0557 <0.0001 1.057 1.035 1.081
LNBA -0.0102 0.0066 0.990 0.982 0.997
D21 W -0.0039 <0.0001 0.996 0.995 0.997
WNAGE 0.0075 0.0447 1.008 1.000 1.015
HT 0.5206 <0.0001 1.683 1.459 1.941

 

2}Number of observations = 12,851.
bSee Table 11.1 for acronym deﬁnitions.

56

Table 11.7. Estimated hazard ratios for Data A for PRNBA40a

 

 

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
AGEF F 0.0027 <0.0001 1.002 1.001 1.003
NBA -0.0590 <0.0001 0.943 0.930 0.956
NSTL 0.0587 0.0004 1.061 1.027 1.096
LNBA -0.0724 <0.0001 0.930 0.919 0.941
D21W -0.0048 <0.0001 0.994 0.994 0.997
WNAGE -0.0021 0.7024 0.998 0.987 1.009
HT 1.3336 <0.0001 3.799 3.079 4.687

 

1*Number of observations = 12,851.
bSee Table 11.1 for acronym deﬁnitions.

57

Table 11.8. Estimated hazard ratios for Data A for PR4PARa

 

 

Parameter Prob. > Hazard 95% Hazard Ratio

Variableb Estimate Chi-Square Ratio Conﬁdence Interval

AGEFF 0.0024 <0.0001 l .002 1 .002 1.003
NBA -0.0323 <0.0001 0.968 0.995 0.981
NSTL 0.0597 0.0002 1.062 1.028 1.096
LNBA -0.0338 <0.0001 0.967 0.956 0.978
D21 W -0.0050 <0.0001 0.995 0.993 0.996
WNAGE 0.0016 0.7680 1.002 0.991 1.012
HT 1.3154 <0.0001 3.726 3.019 4.600

 

2‘Number of observations = 12,851.
bSee Table 11.1 for acronym deﬁnitions.

58

Table 11.9. Estimated hazard ratios for Data B for LPLa

 

 

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
BF -0.1001 <0.0001 0.905 0.868 0.943
DAY -0.1093 <0.0001 0.896 0.857 0.938
PNBA 0.0136 0.2194 1.014 0.992 1.036
PLBW -0.0033 0.3508 0.997 0.990 1.004
PGR 0.2104 0.0760 1.234 0.978 1.557
AGEF F 0.0031 <0.0001 1.003 1.002 1.004
NBA -0.0048 0.4640 0.995 0.983 1.008
NSTL 0.0681 <0.0001 1.071 1.038 1.105
LNBA -0.0184 0.0004 0.981 0.972 0.992
D21W -0.0047 <0.0001 0.995 0.994 0.997
WNAGE -0.0088 0.1649 0.991 0.979 1.004
HT 0.4859 <0.0001 1.626 1.470 1.798

 

8Number of observations = 4,272.

bSee Table 11.1 for acronym deﬁnitions.

59

 

 

Table 11.10. Estimated hazard ratios for Data B for SNBA“

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
BF -0.1082 <0.0001 0.897 0.861 0.935
DAY -0.0933 <0.0001 0.91 1 0.871 0.952
PNBA 0.0066 0.5503 1.007 0.985 1.029
PLBW -0.0029 0.4053 0.997 0.990 1.004
PGR 0.1475 0.2101 1.159 0.920 1.460
AGEFF 0.0025 <0.0001 1 .003 1 .002 1.003
NBA -0.0548 <0.0001 0.947 0.935 0.959
NSTL 0.0738 <0.0001 1.077 1.004 1.1 1 l
LNBA -0.0583 <0.0001 0.941 0.934 0.953
D21W -0.0044 <0.0001 0.996 0.994 0.997
WNAGE -0.0104 0.0991 0.990 0.977 1.002
HT 0.2948 <0.0001 1.343 1.217 1.482

 

*Number of observations = 4,272.
bSee Table 11.1 for acronym deﬁnitions.

60

 

 

Table 11.11. Estimated hazard ratios for Data B for PLIFEa

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
BF -0.1052 <0.0001 0.900 0.864 0.938
DAY -0.0693 0.0021 0.933 0.893 0.975
PNBA -0.0076 0.4936 0.992 0.971 1.014
PLBW -0.0004 0.8894 1.000 0.993 1.006
PGR 0.1148 0.3293 1.122 0.891 1.413
AGEF F 0.0046 <0.0001 1.005 1.004 1.005
NBA -0.1007 <0.0001 0.905 0.893 0.916
NSTL 0.0594 0.0002 1.061 1.029 1.095
LNBA -0.1004 <0.0001 0.904 0.895 0.914
D21W -0.0030 <0.0001 0.997 0.995 0.998
WNAGE -0.0151 0.0175 0.985 0.973 0.997
HT 0.0135 0.7850 1.014 0.919 1.118

 

1*Number of observations = 4,272.
bSee Table 11.1 for acronym deﬁnitions.

61

 

 

Table 11.12. Estimated hazard ratios for Data B for MAXPARa

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
BF -0.0948 <0.0001 0.909 0.873 0.948
DAY -0.0825 0.0003 0.921 0.881 0.963
PNBA 0.0084 0.4428 1.008 0.987 1.031
PLBW -0.0019 0.5744 0.998 0.991 1.005
PGR 0.1137 0.2542 1.143 0.908 1.439
AGEFF 0.0022 <0.0001 1 .002 1 .001 1 .003
NBA -0.0062 0.3297 0.994 0.981 1.006
NSTL 0.0641 <0.0001 1.066 1.034 1.100
LNBA -0.0085 0.1066 0.992 0.981 1.002
D21W -0.0039 <0.0001 0.996 0.995 0.998
WNAGE -0.0101 0.1104 0.990 0.978 1.002
HT 0.2801 <0.0001 1.323 1.199 1.460

 

2*Number of observations = 4,272.
bSee Table 11.1 for acronym deﬁnitions.

 

 

Table 11.13. Estimated hazard ratios for Data B for PRNBA40a

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
BF -0.1608 <0.0001 0.851 0.794 0.913
DAY -0.1437 0.0003 0.866 0.802 0.935
PNBA 0.0213 0.2530 1.022 0.985 1.059
PLBW -0.0071 0.2365 0.993 0.982 1.005
PGR 0.3583 0.1036 1.401 0.933 2.108
AGEF F 0.0024 0.001 1 1.002 1.001 1.004
NBA -0.0461 <0.0001 0.955 0.934 0.997
NSTL 0.1036 0.0003 1.107 1.049 1.193
LNBA -0.0705 <0.0001 0.931 0.914 0.950
D21W -0.0067 <0.0001 0.993 0.991 0.996
WNAGE -0.0274 0.0088 0.972 0.953 0.993
HT 0.3964 <0.0001 1.486 1.259 1.755

 

aNumber of observations = 4,272.
bSee Table 11.1 for acronym deﬁnitions.

63

 

 

Table 11.14. Estimated hazard ratios for Data B for PR4PARa

Parameter Prob. > Hazard 95% Hazard Ratio
Variableb Estimate Chi-Square Ratio Conﬁdence Interval
BF -0. 1722 <0.0001 0.842 0.786 0.902
DAY -0.1418 0.0002 0.868 0.804 0.936
PNBA 0.0196 0.3244 1.018 0.982 1.055
PLBW -0.0063 0.2688 0.994 0.982 1.005
PGR 0.2937 0.1395 1.356 0.898 2.003
AGEF F 0.0024 <0.0001 1.003 1.001 1.004
NBA —0.0209 0.0647 0.979 0.958 1.001
NSTL 0.0831 0.0023 1.085 1.030 1.146
LNBA -0.0289 0.0022 0.970 0.954 0.990
D21W -0.0068 <0.0001 0.993 0.991 0.996
WNAGE -0.0251 0.0152 0.975 0.956 0.995
HT 0.3037 0.0002 1.355 1.152 1.594

 

3Number of observations = 4,272.
l”See Table 11.1 for acronym deﬁnitions.

64

CHAPTER III
HERITABILITY OF LONGEVITY IN YORKSHIRE FEMALES
Abstract

The purpose of this study was to assess the genetic variation within six different
sow longevity descriptions. Heritability was estimated for four different descriptions of
longevity using a proportional hazards model with an underlying Weibull distribution.
The deﬁnitions included herdlife (time from ﬁrst farrowing to culling), lifespan (number
of parities a female has accumulated before culling), lifetime proliﬁcacy (number of
piglets born alive during the lifetime of the sow), and a combination of proliﬁcacy and
length of productive life (number of pigs produced per day of life). Data consisted of
7,487 records of Yorkshire females with at least one farrowing record, from both nucleus
and multiplication herds across 21 farms from four seedstock systems. A threshold
model was used to estimate heritability for two descriptions of stayability (probability of
producing 40 pigs or probability of reaching 4 parities) on a subset of the original data
(n=5,889) that did not include censored observations. Terms for both models included
ﬁrst litter performance effects of: age at ﬁrst farrowing, number born alive, number of
stillboms, adjusted 21 day litter weight, and length of lactation along with the number
. born alive at last recorded farrowing. Additionally, number born alive, litter birth weight
and percentage of gilts born in a female’s birth litter and her growth and backfat records
were included. These variables were treated as ﬁxed and time-independent. The random
effects of sire and the interaction of herd by year were included in all models.
Heritability estimates ranged from 0.043 to 0.059 and from 0.120 to 0.231 for true and

functional deﬁnitions of survival analyses, respectively, and from 0.117 to 0.200 for

65

threshold analyses. Sufﬁcient genetic variation exists, regardless of deﬁnition, to
improve sow longevity.
Introduction

In animal breeding, economic incentive drives selection pressure. Producers
strive to produce high quality, consumer acceptable products in an efﬁcient, humane
manner. As consumers have demanded a leaner pork product that is inexpensive, the
swine industry has responded by increasing lean content and maximizing the number of
pigs produced per sow per year over the past 20 years. Resulting removal rates in
commercial herds are commonly between 40 and 55%, with the majority of sows being
removed prior to completing 4 parities (Lucia, 1996; Rodriguez-Zas et al., 2003; Tarres,
2006). Consequently, higher replacement rates due to poor longevity increase the
number of replacement gilts needed and breeding herd inventories have a high proportion
of low parity females. According to Lucia et al. (2000) and Stalder et al. (2002) at least
three litters are required from a sow before her developmental and variable costs are
recovered and there is positive cash ﬂow for the producer.

Longevity of sows summarizes the effects of functional traits, deﬁned as the
ability to delay involuntary culling (Ducrocq and Solkner, 1998a). One method for
analyzing longevity data is survival analysis which allows inclusion of both censored and
uncensored records of animals (Cox, 1972). This approach relies on the concept of
hazard, instantaneous or age-speciﬁc failure rate (Lawless, 1982; Lee, 1992) or, the
animal’s risk of being culled at time t, conditional upon survival to time t (Ducrocq,
1987; Ducrocq and Solkner 1998b). Proportional hazard models can be semiparametric

or parametric depending on whether on not the baseline hazard distribution is assumed to

66

be arbitrary or parametric (Yazdi et al., 2002). Proportional hazards models have been
extended to incorporate time-dependent covariates, random effects, and genetic
relationships between animals. Alternatively, generalized linear mixed sire models have
been used to analyze measures of stayability in beef cattle and pigs (Kadermideen et al.,
2004; Martinez et al., 2005; Guerra et al., 2006). Reproductive performance is a major
component of culling decisions, and stayability is an indicator of reproductive
performance (Martinez et al., 2005). Threshold models have been developed for
binomial data to estimate or predict effects on a continuous underlying “liability”
variable, on which linearity is imposed (Kadarrnideen et al., 2004).

Sow longevity and/or reproductive traits in general are lowly heritable and some
production traits such as backfat thickness and daily gain have been reported to be
unfavorably correlated (Lopez-Serrano et al., 2000). Heritability is necessary to calculate
expected responses to selection and to predict breeding values. Stalder et al., (2004) and
Serenius and Stalder (2006) provided summaries of heritability estimates from the
literature. Heritability estimates for stayability ranged from 0.05 to 0.11 (Lopez-Serrano
et al., 2000; Tholen et al., 1996a, 1996b) and from 0.11 to 0.31 for survival analysis
(Yazdi et al., 2000a, 2000b; Serenius and Stalder, 2004).

There has not been a clear consensus in the scientiﬁc literature regarding the
deﬁnition of longevity. Additionally, it is not clear what inﬂuence other indicator traits
or productivity measures have on longevity. The objective of this study was to estimate
the genetic variation for six measures of longevity and determine if correction for non-

genetic factors inﬂuenced the magnitude of the heritability estimates.

67

Materials and Methods

Data

The data was an accumulation of both nucleus and multiplication records of
Yorkshire females from four swine seedstock systems (Waldo Farms, DeWitt, NE;
Premier Swine Systems, Michigantown, IN; Whiteshire/Hamroc, Albion, IN; and the
Zierke Co., Morris, MN).
From these 21 farms, records on Yorkshire sows born from 1988 through 2002, with at
least one farrowing record, were available. Information from the female’s birth litter,
along with her performance for age at 113 kg and backfat thickness adjusted to 113 kg as
suggested by National Swine improvement Federation (1996), were merged with
subsequent sow performance records. Further details of data editing and variable
deﬁnition can be found in Hoge (2009a). Due to incomplete data recording across farms
and even within a farm over time, not all sows had complete records for birth litter traits,
growth rate and backfat thickness. To make optimal use of the data, the data set was
divided into two. After editing the largest data set (Data A), 7,487 records of Yorkshire
sows with at least 1 farrrowing were considered for survival analysis, including 5,889
(78.65%) and 1,598 (21.35%) uncensored and censored records, respectively. The
average censoring rate across all four survival deﬁnitions was approximately 21%. The
data set comprised a total of 697 sires with an average of 10 daughters each (range 2-
175). Information included sows’ ﬁrst parity information along with her lifetime
productivity data. The individual record for each female included herd type (either
multiplier (l) or nucleus (2)) (HT), sow herd (the herd that the female was a productive

sow), date of ﬁrst farrowing, date of culling, age at ﬁrst farrowing (AGEF F), number

68

born alive in ﬁrst litter (NBA), number of stillborn in ﬁrst litter (N STL), length of
lactation in ﬁrst parity (WNAGE), adjusted 21 day litter weight of ﬁrst litter (D21W),
and litter size (born alive) at last farrowing (LNBA).

The second data set (Data B) contained 3,886 records of Yorkshire sows with at
least 1 farrowing record, with 2,995 (77.07%) and 891 (22.93%) uncensored and
censored records, respectively. Additionally, the data set comprised a total of 520 sires
with an average of 7 daughters each (range 2-131). Data B contained all variables
previously listed in Data A, along with information related to the sow’s birth litter record
and adjusted days to 113 kg and adjusted backfat thickness at 113 kg. Standardized
variables were created for growth and backfat thickness (see Hoge 2009a for details).
Within a pig herd, contemporary group was deﬁned by birth month and year.

The individual record for each female included all covariates present in Data A,
along with pig herd (the herd that the female was born in), date of birth, birth litter
number born alive (PNBA), birth litter birth weight (PLBW), percentage of gilt pigs in
birth litter (PGR), standardized deviation for backfat thickness (BF), and standardized
deviation for adjusted days to 113 kg (DAY).

A subset of Data A and Data B was used for generalized linear mixed model
analyses. These data sets excluded censored observations and sires with less than 2
daughter records. There were 5,889 sow records included in the analyses for Data A with
540 sires represented and 2,995 sow records included in the analyses for Data B with 346
sires represented.

Six different deﬁnitions of longevity were developed. Length of productive life

was determined as the number of days from the date of ﬁrst farrowing to the date of

69

culling or censoring (LPL). Similarly, the lifespan of a sow was determined as the
maximum parity she completed prior to removal (MAXPAR). Lifetime proliﬁcacy was
determined as the number of piglets (born alive) produced during the LPL of the sow
(SNBA). Stayablility, has been deﬁned as binary trait measuring whether a sow has
survived in a herd until some deﬁned ﬁxed parity or time (Serenius and Stalder, 2004).
From this deﬁnition, two different discrete measures were determined using production
based thresholds. They were the ability of a sow to achieve 4 parities prior to removal
(PR4PAR) and the ability of a sow to produce 40 pigs prior to removal (PRNBA40). A
measure of length of life and a measure of productivity were combined to determine the
ﬁnal response variable by each female’s SNBA divided by the difference between cull
date and birth date (PLIFE).

Statistical Analysis -

In the current study, statistical analyses were based on survival analysis to
identify and evaluate the impact of factors inﬂuencing removal of Yorkshire females for
LPL, MAXPAR, PLIFE, and SNBA. The proportional hazard model following a
Weibull distribution was applied to the data using the Survival Kit (Ducrocq and Solkner,
2001). A discrete-time model that assumes survival data are measured on a discrete scale
with few classes, described by Prentice and Gloeckler (1978), was used in the analysis for
MAXPAR. The validity of the Weibull distribution was assessed by plotting the log(-
log(S(t))) against the log(t), where S(t) is the Kaplan-Meier survivor function and t is
time (Figure III.1 through 111.4). The plots of this relationship revealed approximately
straight lines indicating the Weibull distribution was assumed to ﬁt the data

appropriately. Preliminary survival analyses were carried out using a Cox proportional

70

hazards model to examine the confounding and interaction between ﬁxed effects. No
signiﬁcant interactions were found between ﬁxed effects. The hazard function on a sow

was modeled according to Ducrocq et al., 1988:
h(t. w» = MW“ exp{w<t)'6}

Where h(t, w(t)) is the hazard function of an individual depending on time t (or a

function of time), and 210(k)" 7' is the baseline hazard function which is assumed to

follow a Weibull distribution, where l» and p are location and shape parameters of the
baseline Weibull hazard function. Vector 0' = {b ’u'} is a vector of ﬁxed (b) and random
(u) covariates with a corresponding incidence matrix w(t)' = {x(t)' z(t)'}. For Data A,
time independent ﬁxed effects included in b were AGEF F, NBA, NSTL, LNBA, D21W,
WNAGE, and HT and for Data B the models included all of the covariates listed for Data
A with the addition BF, DAY, PNBA, PLBW, and PGR. The genetic effect of sire and
the time dependent herd-year interaction were treated as random effects and included in
u. The sire effect was assumed to have zero mean and var(sire) = A 02,, where 02, is the
sire variance, and A is the additive relationship matrix among the sires. Tempelman
(1998) encouraged treating contemporary groups as random to allow useful
implementation of generalized linear animal models. However, this is in contrast to what
has historically been done by animal breeders due to concerns regarding prediction bias
when contemporary groups and genetic values are correlated. Tempelman (1998)
justiﬁed the treatment of herd-year-season effects as random and included the possibility
of modeling autoregressive or other temporal correlation structures over sequential year-
seasons to alleviate the extreme category problem, and lessening the impact of model

misspeciﬁcation to make the statistical analysis more robust. Additionally, Yazdi et al.

71

(2000b) reported that the model that best described the data in their study was when herd-
year was treated as random and time dependent. Records of sows were more thoroughly
corrected for the effect of herd-year when it was treated as time-dependent versus time
independent. Furthermore, Yazdi et al. (2000b) concluded that the decrease in the sire
variances observed when herd-year was changed from ﬁxed to random indicated that
differences in the longevity of sows related to herd-year had partly genetic origin and
modiﬁed the variability of longevity due to sire.

Two separate longevity analyses were carried out for each response variable. The
ﬁrst was to approximate true longevity; thus, random time—dependent herd-year and
genetic sire effects were the only effects included in the model. To approximate
functional longevity, all covariates were included in each model for each response
variable. The heritability of longevity was calculated from the sire variance component
as a proportion of phenotypic variance of the Weibull distribution. The heritability (hz)
on the logarithmic scale from survival analysis was calculated as shown by Ducrocq and
Casella (1996):

h2 = 4025/(1122/6'i'0'25)
Where 7122/6 is the variance of the extreme value distribution (Lawless, 1982).

Single-trait threshold analyses were conducted to estimate heritability of two
measures of stayability (PRNBA40 and PR4PAR) using a generalized linear mixed sire
model for binary data with a logit link. All analyses were conducted using the
GLIMMIX procedure within SAS (SAS Inst. 1NC., Cary, NC). Conditional on the ﬁxed
and random effects, stayability traits were assumed to follow a binomial distribution.

The linear mixed mode] predictor used was modeled according to Martinez et al. (2005):

n = Xb + 25
where n is the vector of linear predictors, which is related to predictions on the
observational scale through the inverse link function; b is p X] vector of ﬁxed effects; 5 is
a q x 1 vector of random effects, and X, Z are known incidence matrices for b and s,
respectively, relating the observations in y, the n X 1 vector of observations, or its
conditional expectation to ﬁxed and random effects, respectively. Covariates were
treated similarly to survival analysis with Data A, ﬁxed effects included were AGEFF,
NBA, NSTL, LNBA, D21W, and WNAGE and for Data B the models included all of the
covariates listed for Data A with the addition BF, DAY, PNBA, PLBW, and PGR. For
both Data A and Data B, the ﬁxed effect HT was removed to satisfy convergence
criterion. Time dependent herd-year interaction and sire were treated as random effects.
The residual variance on the underlying scale was assumed to be 1 (Kadarmideen et al.,
2004; Martinez et al., 2005). The link between observed threshold of producing 40 pigs
or completing 4 parities and the underlying liability is a “threshold point”. This is
deﬁned by the link function which relates E(y|[3,u,)t) to the observed mean. Estimates of
heritability with threshold models were obtained as described by Kadarmideen et al.

(2004) using a logit model:

2
2_ 40's
It — 7 ,

2 2 7!“
(0's +08 *r‘3—“)

 

where 082 = l and as2 was the estimate of the sire variance.

To test the signiﬁcance of the sire variance estimates, a Z statistic was calculated

by dividing the estimate of the sire variance by its standard error.

73

Results and Discussion

Descriptive statistics for the data are presented in Table 1112. The average LPL
for Yorkshire females was 505.2 days. This relates closely to the mean parity at removal
of 3.6. Females farrowed their ﬁrst litter at approximately 1 year of age with an average
of 9.4 pigs born alive. Throughout their productive lifetime females produced just over
36 pigs born alive with an adjusted litter 21 day weight of 51.7 kg. The mean age at
weaning was just under 21 days. The descriptive statistics of the data are consistent with
other survival analysis studies (Yazdi, 2000; Stalder et al., 2004; Serenius and Stalder,
2004; Serenius and Stalder 2007; Engblom et al., 2008).

To test the signiﬁcance of different covariates, a likelihood-ratio test was carried
out for models used in survival analysis. The results of the log likelihood tests from
survival analysis are presented in tables 111.3 for Data A and 111.4 for Data B. The P-
values and R2 of Maddala (measure of proportion of variation explained by the model)
are based on log likelihood differences when one effect at a time was excluded from the
model. For Data A, all covariates included in the model were signiﬁcant for all response
variables with the exception of WNAGE for MAXPAR, SNBA, and PLIFE. For Data B,
current results indicate that a female’s birth litter covariates included in the analysis
(PNBA, PLBW, PGR) were not signiﬁcant, regardless of deﬁnition. The standardized
variables for backfat and days to 113 kg were signiﬁcant (P<0.001) across all deﬁnitions
of longevity. These results are in agreement with Hoge (2009a) where a Cox
proportional hazards model was used on a larger data set from the same population for
preliminary variable selection. Furthermore, the sign (+ or - ) of the hazard coefﬁcients

for each covariate were the same for the Cox proportional hazards models used in Hoge

74

(2009a) and the Weibull frailty models used in this study. Within a contemporary group,
fatter, slower growing gilts had a decreased risk of being culled. Additionally, sows that
had more pigs born alive, fewer stillborn, and heavier litters at 21 days of lactation for
their ﬁrst litter had a decreased risk of being culled. The primary focus of this study was
to assess genetic variation; therefore hazard rates for each covariate were not presented.
For further details related to the inﬂuence of covariates on various measures of longevity
on this population and a comparison with other studies, see Hoge (2009a).

The levels of statistical signiﬁcance of the covariates tested in the threshold are
presented in table 1115. Similar to survival analysis, using either Cox or Weibull
proportional hazards models, all ﬁrst litter performance traits were signiﬁcant, with the
exception of WNAGE for PRNBA40. Covariates related to a female’s birth litter
performance were not signiﬁcant for either response variable, while BF and DAY were
signiﬁcant (P<0.01) for PR4PAR.

Estimates of sire variance components and heritability are presented in table 1116.
Heritability estimates were low for true longevity (range 0.043 to 0.059). When ﬁxed
effects were included for functional estimates of longevity, heritability estimates were
higher for Data A with a range from 0.120 for MAXPAR to 0.190 for LPL and for Data
B with a range 0.132 for MAXPAR to 0.231 for LPL. This increase in heritability
estimates indicates that correcting for a female’s developmental factors and ﬁrst litter
performance proves beneﬁcial when estimating sire variance. Heritability estimates
yielded from threshold analysis, when censored records were excluded, were 0.156 and
0.117 for Data A and 0.176 and 0.200 for Data B for PRNBA40 and PR4PAR,

respectively. Caution should be used when making direct comparisons of heritability

75

estimated from the threshold analysis to those obtained from survival analysis, due to
differences in censoring rates. Gou et al., (2001) reported heritability estimates for length
of productive life using a linear mixed model with censoring was 0.25 for the actual level
of censoring (1 5.5%), and it tended to decrease to 0.18 and 0.16 as simulated censoring
increased to 25 and 35%, respectively. When all censored records were excluded, the
heritability estimate for length of productive life was 0.45. These results showed that as
censoring rates decreased, estimates of heritability were increased, indicating that
censoring rates can inﬂuence estimates of sire variance in the present data.

Results from this study are in agreement with previous ﬁndings where heritability
estimates of length of productive life from survival analysis has ranged between 0.10 and
0.31 (Yazdi et al., 2000b; Serenius and Stalder, 2004; Serenius and Stalder, 2007).
Additionally, Gou et al., (2001) and Serenius and Stalder (2004) reported that heritability
estimates of lifetime proliﬁcacy were slightly lower than those for length of productive
life, which agrees with results from the current study. Heritability estimates were slightly
higher for LPL (0.190 for Data A and 0.231 for Data B) than for SNBA (0.172 for Data
A and 0.181 for Data B) and PLIFE (0.168 for Data A and 0.192 for Data B), both
measures of lifetime proliﬁcacy. Serenius and Stalder (2004) also reported that length of
productive life and lifetime proliﬁcacy were genetically highly correlated (rg>0.95)
indicating that selection for one will result in genetic gain in the other. Additionally,
Tholen et al., (1996) and Lopez-Serrano et al. (2000) used linear models and reported
heritability estimates for stayability (the probability of the sow surviving in the herd from
parity 1 to parity 4 and from parity 1 to parity 3, respectively) to be 0.08 and 0.10.

Fernandez de Sevilla et al., (2008a) reported the variance component for the sire genetic

76

effect was 0.03, using a Cox semiparametric proportional hazards model. Heritability
estimates from this study ranged from 0.05 to 0.07, using three different equations for
calculating heritability. Fernandez de Sevilla et al., (2008b) used a competing risk
approach on the same population of Duroc sows. These investigators reported estimates
of heritability for longevity related to low productivity culling and these estimates ranged
ﬁ'om 0.008 to 0.024. Heritability estimates for low fertility culling ranged from 0.017 to
0.083. These low heritability estimates could be partially attributed to the high censoring
rates within the data (81.09% and 93.19%, respectively). Fernandez de Sevilla et al.,
(2008) concluded that there were substantial discrepancies in the sources of variation and
genetic background of sow longevity depending on the cause of failure.

Martinez et al., (2005) estimated genetic parameters for stayability to speciﬁed
ages for Hereford cows using a generalized linear mixed model for binary data and a
linear mixed model. Investigators reported that estimates of heritability for stayability
were higher for threshold model analysis (range of 0.09 to 0.30) as compared to estimates
obtained from linear mixed model analysis (range 0.05 to 0.09). Guerra et al., (2006)
estimated the heritability of calving rate and calf survival in a multibreed population with
linear, threshold, and logistic, generalized mixed sire models. Reported heritabilities for
calving rate were 0.063, 0.150, and 0.130 and calf survival were 0.049, 0.160, and 0.190
for linear, threshold, and logistic models, respectively. Heritability estimates for calving
rate and calf survival from the linear mixed sire model were similar to threshold and
logistic model estimates when transformed to an underlying scale.

The relative culling rate for daughters of sires ranged from 0.77 to 1.32 across

deﬁnitions of longevity in survival analysis. This corresponds to the lowest and highest

77

risk of culling for daughters of a common sire compared to daughters of an average sire
which have a relative risk of culling equal to 1. Daughters from the poorest sires had

1.71 times higher risk of culling compared with daughters of the best sires. The ranges of
relative culling rates for daughters were similar to those reported by Yazdi et al., (2000b).

The fact that herd-year environmental effects impact longevity and/or stayability
has been readily documented (Yazdi et al., 2000b; Serenius and Stalder, 2004; Serenius
and Stalder, 2007; Serenius et al., 2008). Operations have different culling policies and
management practices. Additionally, these culling policies and management practices
may change from year to year. The current results indicate relative culling rates for herd-
year ranged from 0.54 to 1.48 across deﬁnitions of longevity. Within the worst herd-year
combination, females had 2.74 times higher risk of culling compared to the best herd-
year combination, indicating that herd-year had a large inﬂuence on the risk of culling,
regardless of deﬁnition. Serenius et al., (2008) recommended that if breeding value
estimation is routinely conducted, solutions for herd-year interaction classes should be
available for producers to utilize so as to make more effective management decisions.
Investigators do caution that a signiﬁcant amount of data is required before herd-year-
season solutions can be routinely utilized to make decisions.

The current results indicate that regardless of deﬁnition, the estimated
heritabilities suggest that direct genetic improvement for improved longevity is feasible.
Additionally, correcting for developmental factors and performance of ﬁrst parity
lactation does increase estimates of sire variances and results in higher heritability
estimates. Furthermore, covariates included in the model had similar inﬂuence when

using a Cox semi-parametric model, Weibull frailty model, or a threshold model.

78

Table [11.1. Deﬁnitions of variable acronyms used in model development

 

Variable

Number of days between date of ﬁrst farrowing and removal, d
Number of parities a sow had completed

Pigs born alive per day of sow herd life

Binary indicator variable if a sow had produced 40 pigs (0,1)
Binary indicator variable if a sow had completed 4 parities (0,1)
Total number born alive over lifetime

Herd type, 1 multiplier, 2, nucleus

Age at ﬁrst farrowing, d

aStandardized deviation for adjusted backfat, mm
aStandardized deviation for adjusted days to 113 kg

Adjusted 21 day litter weight in parity 1 litter, kg

Birth litter gilt ratio

Birth litter, litter birth weight, kg
Birth litter number born alive
Last litter number born alive
First litter number born alive
First litter number stillborn

First litter lactation length, d

Acronym
LPL
MAXPAR
PLIFE
PRNBA40
PR4PAR
SNBA

HT

AGEF F
BF

DAY

D21 W
PGR
PLBW
PNBA
LNBA
NBA
NSTL
WNAGE

aAdjusted female record was deviated from the birth month and year mean for gilts tested

within herd and divided by the standard deviation.

79

Table 111.2. Number of observations, means, standard deviations, minimums, maximum

values of the traits studied in the Yorkshire Population

 

 

Traita No. Mean SD MIN MAX
LPL,d 7,487 505.2 407.7 10 3487
MAXPAR 7,487 3.6 2.4 1 14
SNBA 7,487 36.1 27.1 0 182
PLIFE 7,487 0.038 0.017 0 0.277
PRNBA40

0 3,462

1 2,427
PR4PAR

0 3,242

1 2,647
AGEFF,d 7,487 364.9 43.7 253 597
NBA 7,487 9.4 3.1 0 l9
NSTL 7,487 0.7 1.2 0 13
LNBA 7,487 9.5 3.5 0 20
D21W, kg 7,487 51.6 14.5 4.5 115.9
WNAGE, (1 7,487 20.3 4.7 0 40
BF, dev 3,886 -0.26 0.92 -3.19 3.89
DAY, dev 3,886 -0.41 0.82 -3.14 3.58
PGR 4,148 0.53 0.14 0.08 0.92
PNBA 4,148 10.8 2.9 1 20
PLBW, kg 4,148 35.9 9.8 3.0 75

 

aSee Table 111.1 for acronym deﬁnitions.

80

 

Table 111.3. Likelihood ratio test, including P values and reliabilities (R2) for Data A
when the covariates were tested bLexcluding one at a time (last) from the full modela

 

 

Variableb LPL MAXPAR SNBA PLIFE
P value R2 P value R2 P value R2 P value R2
HT < 0.001 0.23 < 0.001 0.20 < 0.0001 0.19 < 0.001 0.27

AGEFF < 0.001 0.23 < 0.001 0.21 < 0.0001 0.18 < 0.001 0.25
NBA < 0.001 0.23 < 0.001 0.22 < 0.0001 0.17 < 0.001 0.23
NSTL < 0.001 0.23 < 0.001 0.22 < 0.0001 0.19 < 0.001 0.27
LNBA < 0.001 0.23 0.001 0.21 0.0021 0.16 < 0.001 0.22
D21W < 0.001 0.24 < 0.001 0.21 < 0.0001 0.19 < 0.001 0.28
WNAGE < 0.001 0.24 0.024 0.21 0.0125 0.19 0.213 0.27

 

aNumber of observations = 7,487.
t’See Table 111.1 for acronym deﬁnitions.

8]

Table 111.4. Likelihood ratio test, including P values and reliabilities (R2) for Data B
when the covariates were tested bLexcluding one at a time (last) from the full modela

 

 

 

Variable” LPL MAXPAR SNBA PLIFE
P value R2 P value R2 P value R2 P value R2
HT < 0.001 0.17 < 0.001 0.21 < 0.001 0.19 < 0.001 0.27

AGEFF < 0.001 0.16 < 0.001 0.21 0.002 0.18 < 0.001 0.25
NBA < 0.001 0.18 < 0.001 0.22 < 0.001 0.17 < 0.001 0.23
NSTL < 0.001 0.17 < 0.001 0.22 < 0.001 0.19 < 0.001 0.27
LNBA < 0.001 0.17 < 0.001 0.22 < 0.001 0.16 < 0.001 0.22
D21W < 0.001 0.17 < 0.001 0.21 < 0.001 0.18 < 0.001 0.27
WNAGE < 0.001 0.18 0.002 0.22 0.002 0.19 0.363 0.27
PNBA 0.984 0.18 0.410 0.22 0.65] 0.19 0.012 0.27
PLBW 0.323 0.18 0.639 0.22 0.244 0.19 0.463 0.27
PGR 0.134 0.17 0.146 0.22 0.312 0.19 0.300 0.27
BF 0.006 0.17 < 0.001 0.22 < 0.001 0.19 0.002 0.27
DAY < 0.001 0.16 < 0.001 0.22 < 0.001 0.19 < 0.001 0.27

 

23Number of observations = 3,882.
bSee Table 111.1 for acronym deﬁnitions.

82

Table 111.5. Level of statistical signiﬁcance of the covariates included in the analyses of
PRNBA40a and PR4PARa for Data A and Data B

 

 

Variableal PRNBA40 PR4PAR

Data A Data B Data A Data B
AGEFF *** *** *IIHII ***
NBA #31”! *IIUII *IIUIK *Il‘
NSTL ** ** ** **
LNBA * ** 1103* * **
DZIW *IIIIII *** *** **
WNAGE NS NS * *
PNBA —b NS — NS
PLBW —— NS —— NS
PGR — NS — NS
BF _ III _ **
DAY -— NS —— **

 

aSee Table 111.1 for acronym deﬁnitions.
l”Covariate was not considered in the model.
*P<0.05; **P<0.01; ***P<0.001; NS = not signiﬁcant.

83

 

Table 111.6. Additive sire variance (0'32) and heritability (hz) estimates obtained from
survival analysis and threshold analysis

 

 

Data A Data B
(,3 b SE 122 (,3 b SE h2
Survival Analysis
LPL a
True 0.020* 0.011 0.048 0025* 0.011 0.059
Functional 0.082*** 0.013 0.190 0.101 *** 0.014 0.231
MAXPARa
True 0.020* 0.011 0.048 0.021 * 0.012 0.050
Functional 0.051 *** 0.012 0.120 0.056*** 0.012 0.132
SNBAa
True 0018* 0.011 0.043 0020* 0.012 0.048
Functional 0.074*** 0.013 0.172 0.078*** 0.012 0.181
PLIFEa
True 0.021* 0.012 0.050 0.022* 0.011 0.053
Functional 0.072*** 0.013 0.168 0.083*** 0.014 0.192
Threshold Analysis
PRNBA40all 0.147M 0.046 0.156 0.106* 0.062 0.117
PR4PAR”l 0.152" 0.045 0.176 0.197** 0.071 0.200

 

aSee Table 111.1 for acronym deﬁnitions.
bPr(z s —z)*P<0.05; **P<0.01; ***P<0.001

84

Figure 111.1. Graphical test for validity of Weibull distribution for LPL

Log of Negative Log of Estimated Survivor Function

 

2 1
FA.
A
O; 0 q
.5
«s
..D
E
o" '2 -1
70‘
.2.
:2
v -4 .
DD
0
"T‘
w—v
DO
0
A d
-6

 

 

 

Log(LPL)

85

Figure 111.2. Graphical test for validity of Weibull distribution for MAXPAR

Log of Negative Log of Estimated Survivor Function

 

 

 

2
E»
E l
..o
m
.o
8
a.
'7;
a; O
a
no
.2
“3.73
o -l
,_1
T I I 1 T
0 0.5 l 1.5 2 2.5

Log(MAXPAR)

86

 

Figure 111.3. Graphical test for validity of Weibull distribution for SNBA

Log of Negative Log of Estimated Survivor Function

 

 

 

 

2
é:
:: 0‘
..D
at
..D
e
a.
.g ‘24
3%
or)
.2
J... .4,
on
o
..J

'6,

T T T
0 1 2 3 4

Log(SNBA)

87

Figure 111.4. Graphical test for validity of Weibull distribution for PLIFE

Log of Negative Log of Estimated Survivor Function

 

 

 

 

2 t
E o .
.5
:6
..O
8
O-
3 -2 .
.2.
5.
DD
3 —4 .
‘55
O
A

-6 ‘

-6 -4 -2
Log(PLIFE)

88

SUMMARY AND CONCLUSIONS

The length of adult sow productive life is now recognized as both an economic
and welfare concern. However, there are not consistent deﬁnitions to measure sow
longevity. The initial objective of this study evaluated six different descriptions of
longevity and determined the relationship of developmental performance factors with
them. The second objective was to assess the genetic variation within the six different
descriptions of longevity. Longevity deﬁnitions included stayability (probability of
producing 40 pigs or probability of reaching four parities), lifespan (number of parities a
female has accumulated before culling), lifetime proliﬁcacy (number of pigs born alive
during a female’s productive lifetime), herdlife (time from ﬁrst farrowing to culling) and
pigs produced per day of life.

To meet these two objectives, a unique data set was assembled. The data was an
accumulation of both nucleus and multiplication records of Yorkshire females from four
swine seedstock systems (Waldo F amrs, DeWitt, NE; Premier Swine Systems,
Michingantown, IN; Whiteshire/Hamroc, Albion, IN; and the Zierke Co., Morris, MN).
Individual pig performance and farrowing records from 21 different farms, which were
associated within these four seedstock systems were merged into one data set for
analysis. Due to incomplete data recording across farms and even within a farm over
time, not all sows had complete records for birth litter traits, growth rate and backfat
thickness. To make optimal use of the data, the data set was divided into two.

To meet the ﬁrst objective, a Cox proportional hazards model was used to
determine the relationship of developmental factors with longevity. There were 14,262

(Data A) records of Yorkshire sows that included a sow’s ﬁrst parity information along

89

with her lifetime productivity data. The second data set (Data B) contained 4,272 records
that included information from Data A, along with information related to the sow’s birth
litter record and adjusted days to 113 kg and adjusted backfat thickness at 113 kg. The
average length of productive life for Yorkshire females was 488.82 days, with a mean
parity at removal or censoring of 3.49. Females produced on average 34.93 pigs with an
average age of ﬁrst farrowing of 366.2 days. The effect of age at ﬁrst farrowing on the
risk of being culled was highly signiﬁcant (P<0.0001). The older a female was at ﬁrst
farrowing the greater the risk of being culled. Additionally, the results indicated that the
performance of a female during her ﬁrst litter does provide insight for the rest of her
productive life. Sows that had more pigs born alive, fewer stillborn fetuses, and had
heavier 21 day litter weights, tended to have a decreased risk of being culled. The length
of lactation in the ﬁrst litter was signiﬁcant (P<0.05) for length of productive life and
lifespan. With the exception of length of productive life, herd type affected the other ﬁve
deﬁnitions. Sows from nucleus herds had an increased relative risk of being culled that
ranged from 42% to 68%, depending on the deﬁnition. The standardized variables for
backfat and days to 113 kg signiﬁcantly inﬂuenced all deﬁnitions of longevity. Within a
contemporary group, fatter, slower growing gilts tended to have a decreased risk of being
culled. Current results indicate that covariates for traits measured within a female’s birth
litter did not signiﬁcantly impact longevity.

The second objective was to estimate the genetic variation for six measures of
longevity and determine if non-genetic factors inﬂuenced the magnitude of the estimates
of the genetic variances and the resulting heritability estimates. A reduced number of

sows had complete pedigree information available. As a result, Data A consisted of

90

7,487 records that were daughters of 697 sires and Data B consisted of 3,886 records that
were daughters of 520 sires. Censoring rates were 21.35% and 22.93% for Data A and
Data B, respectively. Survival analysis was used to identify and evaluate the impact of
factors inﬂuencing the removal of females and to estimate the sire variances for length of
productive life, lifespan, and lifetime proliﬁcacy. Heritability estimates derived from
survival analysis ranged from 0.039 to 0.231. Estimates of heritability were similar
across all four deﬁnitions, but were the lowest when only the random effects of sire and
herd-year were included (0.039 to 0.050). When developmental factors and ﬁrst litter
performance traits were included in the model, heritability estimates increased across all
four deﬁnitions (0.076 to 0.231).

The current results indicated that regardless of deﬁnition, developmental factors
and performance of ﬁrst parity females does provide insight into the longevity of the
Yorkshire females in this population. Additionally, the heritability estimates suggest that
direct genetic improvement for improved longevity is feasible. Furthermore, correcting
for developmental factors and performance of ﬁrst parity females did increase estimates
of sire variances, resulting in higher heritability estimates. The six different deﬁnitions
yielded similar results indicating that the “best” deﬁnition for assessing longevity could
be considered producer dependent. These results will provide guidelines on assessing
sow longevity within swine operations. Producers, wanting to improve sow longevity,
can select a set of variables and a deﬁnition of longevity that is most suitable for their
operation.

This study was not without limitations. The ﬁrst limitation was related to

inaccurate and inconsistent recording of reasons for removal. This variable was intended

91

to be utilized in the analyses, but it was concluded that the lack of completeness
eliminated the possibility of inclusion into the models. Another variable that was
intended to be evaluated was number of services prior to ﬁrst conception. Due to the
incompleteness of breeding records, this variable was eliminated from the study. The
ﬁnal limitation of the data was related to herd type. Approximately 65% of the data came
from nucleus herds versus 35% from multiplier herds. Caution should be used in
interpreting the results due to the fact that females within a nucleus herd experience more
voluntary culling in order to maximize genetic progress. Older sows have a longer
generation interval and this can potentially slow genetic change.

It has been proven that improving sow longevity holds economic incentives.
Future research efforts should include a type or structure score at selection into the
breeding herd. Selection for correctly structured females could potentially impact the
length of productive life of females. Additionally, feed consumption and body fat should
be monitored throughout a female’s ﬁrst lactation. Sows that have a greater appetite are
more likely to maintain an acceptable body composition at weaning and ultimately
decrease the number of non-productive days a female experiences between parity one and
parity two. Finally, efforts should be focused on identifying genetic markers that
inﬂuence length of productive life. If genes are identiﬁed that impact measures of
longevity, the selection for these favorable traits can be implemented in both sexes and
earlier in life. Although the amount of available literature has increased substantially,
there still is a substantial amount of research to assist the swine industry to improve

longevity within the US. sow herd.

92

LITERATURE CITED

93

LITERATURE CITED

Allison, P. D. 1997. Survival Analysis Using the SAS System. A Practical Guide. SAS
Institute Inc. Cary, NC.

Anderson, S. and U. G. Hansen. 1996. Selection for conformation and longevity in the
Danish breeding system. Proceedings of the Nordiska J ordbruksforskares
Forening Seminar 265-Longevity of Sows, Denmark: Research Centre Foulum.

Arango, J., 1. Misztal, S. Tsuruta, M. Culbertson, and W. Herring. 2005. Study of codes
of disposal at different parities of Large White sows using a linear censored
model. Journal of Animal Science 83: 2052-2057.

Arthur, S. R., E. T. Komegay, H. R. Thomas, H. P. Veit, D. R. Notter, K. E. Webb Jr.,
and J. L. Baker. 1983. Restricted energy intake and elevated calcium and
phosphorus intake for gilts during growth. IV. Characterization of metacarpal,
metatarsal, femur, humerus and turbinate bones of sows during three parities.
Journal of Animal Science 57: 1200-1214.

Babot, D., E. R. Chavez, and J. L. Noguera. 2003. The effect of age at ﬁrst mating and
herd size on lifetime productivity of sows. Animal Research 52: 49-64.

Bereskin, B. 1979. Genetic aspects of feet and legs soundness in swine. Journal of
Animal Science 48: 1322-1328.

Black, J. L., B. P. Mullan, M. L. Lorschy, and L. R. Giles. 1993. Lactation in the sow
during heat stress. Livestock Production Science 35: 153-170.

Boettcher, P. J ., L. K. Jairath, and J. C. M. Dekkers. 1999. Comparison of Methods for
Genetic Evaluation of Sires for Survival of Their Daughters in the First Three
Lactations. Journal of Dairy Science 82: 1034-1044.

Boyd, R. D., G. C. Castro, and R. A. Cabrera. 2002. Nutrition and management of the
sow to maximize lifetime productivity. Advances in Pork Production 13: 47-59.

Boyle, L., F. C. Leonard, B. Lynch, and P. Brophy. 1998. Sow culling patterns and sow
welfare. Irish Veterinary Journal 51: 354-357.

Brandt, H., N. von Brevem, and P. Glodek. 1999. Factors affecting survival rate of
crossbred sows in weaner production. Livestock Production Science 57: 127-135.

Brisbane, J. R., and J. P. Chenais. 1996. Relationship between backfat and sow longevity

in Canadian Yorkshire and Landrace pigs. Proceedings of the National Swine
Improvement Federation, Ottawa, Ontario, Canada.

94

Brooks, P. H., and D. A. Smith. 1980. The effect of mating age on the reproductive
performance, food utilisation and liveweight change of the female pig. Livestock
Production Science 7: 67-78.

Chagnon, M., S. D'Allaire, and R. Drolet. 1991. A prospective study of sow mortality in
breeding herds. Canadian Veterinary Journal 55: 180-184.

Challinor, C. M., G. Dams, B. Edwards, and W. H. Close. 1996. The effect of body
composition of gilts at ﬁrst mating on long-terrn sow productivity. Animal
Science 62: 660 Abstract.

Chapman, J. D., L. H. Thompson, C. T. Gaskins, and L. F. Tribble. 1978. Relationship of
age at ﬁrst farrowing and size of ﬁrst letter to subsequent reproductive
performance in sows. Journal of Animal Science 47: 780-787.

Charette, R., M. Bigras-Poulin, and G. -P. Martineau. 1996. Body condition evaluation in
sows. Livestock Production Science 46: 107-115.

Cox, D. R. 1972. Regression models and life table with discussion. Journal of Royal
Statistical Society Series B 34: 187-220.

Crump, R. E. 2001. A genetic analysis of sow longevity. Proceedings of the 14th
Conference of the Association for the Advancement of Animal Breeding and
Genetics, Queenstown, New Zealand.

Culbertson, M. S., and J. W. Mabry. 1995. Effect of age at ﬁrst service on ﬁrst parity and
lifetime sow performance. Journal of Animal Science 73:Supplement 1: 21.

D'Allaire, S., T. E. Stein, and A. D. Leman. 1987. Culling patterns in selected Minnesota
swine breeding herds. Canadian Journal of Veterinary Research 51: 506-512.

D'Allaire, S., R. S. Morris, F. B. Martin, R. A. Robinson, and A. D. Leman. 1989.
Management and environmental factors associated with annual sow culling rate:
A path analysis. Preventive Veterinary Medicine 7: 255-265.

D'Allaire, S., A. D. Leman, R. Drolet. 1992. Optimizing longevity in sows and boars.
Veterinary Clinics of North America 8: 545-557.

D'Allaire, S., and R. Drolet. 1999. Culling and mortality in breeding animals. Diseases of
Swine. Iowa State University Press, Ames.

De Koning, G. 1996. Selection in breeding programmes against leg problems.

Proceedings of the Nordiska Jordbruksforskares Forening Seminar 265-Longevity
of Sows, Denmark: Research Centre Foulum.

95

De Vries, A. G. 1989. A model to estimate economic values of traits in pig breeding.
Livestock Production Science 21: 49-66.

Deen, J ., and J. Xue. 1999. Sow mortality in the US: An industry-wide perspective.
Proceedings of the Allen D. Leman Swine Conference.

Deen, J ., T. Nortey, and L. Anil. 2002. How should we compare levels of sow attrition?
Proceedings of the Allen D. Leman Swine Conference.

Deen, J. 2003. Sow longevity measurement. Proceedings of the Allen D. Leman Swine
Conference.

Dekkers, J. C. M., L.K. Jairath, and B. H. Lawrence. 1994. Relationships between sire
genetic evaluations for conformation and ﬁrnctional herd life of daughters.
Journal of Dairy Science 77: 844-854.

Dekkers, J. C. M., and L. K. J airath. 1994. Requirements and uses of genetic evaluation
for conformation and herd life. Proceedings of the 5th World Congress of
Genetically Applied Livestock Production, Guelph, Ontario, Canada.

den Hartog, L. A. and G. J. Noordewier. 1984. The effect of energy intake on age at
puberty in gilts. Netherlands Journal of Agricultural Science 32: 263-280.

Dhuyvetter, K. C. 2000. What does attrition cost and what is it worth to reduce?
Proceedings of the Allen D. Leman Swine Conference.

Dial, G., and Y. Koketsu. 1996. Reproductive failure. Understanding the reasons that
sows are culled for infertility. International Pigletter 16: 15-16.

Dijkhuizen, A. A., R. S. Morris, and M. Morrow. 1986. Economic optimization of culling
strategies in swine breeding herds, using the Porkchop Computer Program.
Preventive Veterinary Medicine 4: 341-353.

Dijkhuizen, A. A., R. M. M. Krabbenborg, and R. B. M. Huirne. 1989. Sow replacement:
A comparison of farmers' actual decisions and model recommendations.
Livestock Production Science 23: 207-218.

Ducrocq, V. P. 1987. An analysis of length of productive life in dairy cattle. Ithaca, NY,
USA, Cornell University.

Ducrocq, V. P. 1994. Statistical analysis of length of productive life for dairy cows of the
Norrnande breed. Journal of Dairy Science 77:855-866.

Ducrocq, V. P. 1997. Survival analysis, a statistical tool for longevity data. Proceedings
of the 48th Annual Meeting of the European Association for Animal Production,
Vienna, Austria.

96

Ducrocq, V. P., and J. Solkner. 1998a. hnplementation of a routine breeding value
evaluation for longevity of dairy cows using survival analysis technique.
Proceedings of the 6th World Congress on Genetics Applied to Livestock
Production, University of New-England, Armidale, Australia.

Ducrocq, V. P., and J. Solkner. 1998b. The Survival Kit V3.0, a FORTRAN package for
large analyses of survival data. Proceedings of the 6th World Congress on
Genetics Applied to Livestock Production, University of New-England, Armidale,
Australia.

Ducrocq, V. P., Quaas, R. L., Pollak, E. J ., and Casella, G. 1988. Length of productive
life of dairy cows. 1. Justiﬁcation of a Weibull model. Journal of Dairy Science
71: 3061-3070.

Ducrocq, V. P., and J. Solkner. 2001. The Survival Kit V3.12. User's manual.

Duran, C. O. 2001. Sow mortality. Compendium on Continuing Education for the
Practicing Veterinarian 23: $76-$83.

Edge, H., S. Edwards, L. Taylor, and P. Gill. 2003. Feeding for sow lifetime productivity.
Proceedings of the Meat and Livestock Commission SFT Conference on Practical
Pig Nutrition.

Ehlers, M. J ., J. W. Mabry, J. K. Bertrand, and K. J. Stalder. 2005. Variance components
and heritabilities for sow productivity traits estimated from purebred versus
crossbred sows. Journal of Animal Breeding and Genetics 122: 318-324.

Ehlorsson, C. J., O. Olsson, and N. Lundeheim. 2002. Investigations of housing and
environmental factors affecting the claw health in group-housed dry sows. Svensk
Veterinartidning 54: 297-304.

Engblom, L., N. Lundeheim, A. Dalin, and K. Andersson. 2007. Sow removal in Swedish
commercial herds. Livestock Science 106: 76-86.

Esbenshade, K. L., J. H. Britt, J. D. Armstrong, V. D. Toelle, and C. M. Stanislaw. 1986.
Body condition of sows across parities and relationship to reproductive

performance. Journal of Animal Science 62: 1187-1193.

Famula T. R. 1981. Exponential stayability model with censoring and covariates.
Journal of Dairy Science 64: 538-545.

Falconer, D. S. and T. F. Mckay. 1996. Introduction to Quantitative Genetics. Essex, UK,
Longman.

97

Faust, M. A., M. W. Tess, and O. W. Robison. 1992. A bioeconomic simulation model
for a hierarchical swine breeding structure. Journal of Animal Science 70: 1760-
1774.

Faust, M. A., O. W. Robison, and M. W. Tess. 1993a. Genetic and economic analyses of
sow replacement rates in the commercial tier of a hierarchical swine breeding
structure. Journal of Animal Science 71: 1400-1406.

Faust, M. A., O. W. Robison, and M. W. Tess. 1993b. Integrated systems analysis of sow
replacement rates in a hierarchical swine breeding structure. Journal of Animal
Science 71: 2885-2890.

Fernandez de Sevilla, X., E. Fabrega, J. Tibua, and J. Casellas. 2008a. Effect of leg
conformation on survivability of Duroc, Landrace, and Large White sows.
Journal of Animal Science. 86: 2392-2400.

Fernandez de Sevilla, X., E. Fabrega, J. Tibua, and J. Casellas. 2008b. Competing risk
analyses of longevity in Duroc sows with a special emphasis on leg conformation.
Animal. 1:1-8.

Fortin, F. and R. I. Cue. 2002. A genetic study of longevity in swine. Journal of Animal
Science 85:Supplement 1: 89.

Friendship, R. M., M. R. Wilson, G. W. Almond, R. R. McMillan, R. Hacker, R. Pieper,
and S. S. Swaminathan. 1996. Sow wastage: Reasons for and effect on
productivity. Journal of Veterinary Resources 50: 205-208.

Geiger, J. O., C. Irwin, and S. Pretzer. 1999. Assessing sow mortality. Proceedings of the
Allen D. Leman Swine Conference.

Goodwin, R. N. 2004. Sow longevity differences. Proceedings of the National Swine
Improvement Federation, Nashville, Tennessee, USA.

Grindﬂek, E., and E. Sehested. 1996. Conformation and longevity in Norwegian pigs.
Proceedings of the Nordiska J ordbruksforskares Forening Seminar 265-Longevity
of Sows, Denmark: Research Centre Foulum.

Guerra, J. L. L., D. E. Franke, and D. C. Blouin. 2006. Genetic parameters of calving
rate and calf survival from linear, threshold, and logistic models in a multibreed
beef cattle operation. Journal of Animal Science 84: 3197-3203.

Guo, S.-F., D. Gianola, R. Rekaya, and T. Short. 2001. Bayesian analysis of lifetime

performance and proliﬁcacy in Landrace sows using a linear mixed model with
censoring. Livestock Production Science 72: 243-252.

98

Hall, A. D., S. Lo, and K. A. Rance. 2002. Comparative study of the lifetime productivity
and performance characteristics of Meishan and Duroc cross-bred pigs. Acta
Agriculturae Scandinavica Section A 52: 183-188.

Han, I. K., B. P. Hyun, J. D. Kim, K. S. Sohn, and S. W. Kim. 2000. Recent advances in
sow nutrition to improve reproductive performance. Asian-Australian Journal Of
Animal Science 13: 335-355.

Heinonen M., A. L., and S. Pyorala. 1998. Evaluation of reproductive failure of female
pigs based on slaughterhouse material and herd record survey. Animal
Reproduction Science 52: 23 5-244.

Holder, R. B., W. R. Lamberson, R. O. Bates, and T. J. Safranski. 1995. Lifetime
productivity in gilts previously selected for decreased age at puberty. Animal
Science 61:115-121.

Huang, S. Y., H. L. Tsou, M. T. Kan, W.K. Lin, and C. S. Chi. 1995. Genetic study on
leg weakness and its relationship with economic traits in central tested boars in
subtropical area. Livestock Production Science 44: 53-59.

Huirne, R. B. M., T. H. B. Hendricks, A. A. Dijkhuizen, and G. W. J. Giesen. 1988. The
economic optimization of sow replacement decisions by stochastic dynamic
programming. Journal of Agricultural Economics 39: 426-438.

Hurtgen, J. P., and A. D. Leman. 1980. Seasonal inﬂuence on the fertility of sows and
gilts. Journal of the American Veterinary Medical Association 177: 631-635.

Imboonta, N., L. Rydhmer, and S. Tumwasom. 2007. Genetic parameters for

reproduction and production traits of Landrace sows in Thailand. Journal of
Animal Science 85: 53-59.

Irwin, C. K., and J. Deen. 2000. Levels of culling in commercial swine herds.
Proceedings of the Allen D. Leman Swine Conference.

Ivkovi, S., M. Teodorovi, and S. Kovin. 1986. Longevity of sows according to the
management in large units. World Review of Animal Production 22: 11-25.

Jairath, L., and J. C. M. Dekkers. 1994. Genetic parameters of functional and true
lactational survival, and relationships with milk production and conformation

traits in registered Canadian Holsteins. Journal of Dairy Science 77 Supplement 1:
146 Abstract.

Jairath, L., J. C. M. Dekkers, L. R. Schaeffer, Z. Liu, E. B. Burnside, and B. Kolstad.

1998. Genetic evaluation for herd life in Canada. Journal of Dairy Science 81:
550-562.

99

Jalvingh, A. W., A. A. Dijkhuizen, J. A. M. Arendonk, and E. W. Brascamp. 1992. An
economic comparison of management strategies on reproduction and replacement

in sow herds using a dynamic probabilistic model. Livestock Production Science
32: 331-350.

Johnson, R. 2000. Role of genetics in gilt attrition. Proceedings of the Allen D. Leman
Swine Conference.

Jorgensen, E., C. Dethlefsen, A. R. Kristensen. 1996. Systems for monitoring and
decision support with respect to sow longevity. Proceedings of the Nordiska
Jordbruksforskares Forening Seminar 265-Longevity of Sows.

Jorgensen, B. 1996. The inﬂuence of leg weakness in gilts on their longevity as sows,
assessed by survival analysis. Proceedings of the Nordiska Jordbruksforskares
F orening Seminar 265-Longevity of Sows, Denmark: Research Centre F oulum.

Jorgensen, B. 2000. Longevity of breeding sows in relation to leg weakness symptoms at
six months of age. Acta Veterinaria Scandinavica 41: 105-121.

Kachman, S. D. 1999. Applications in Survival Analysis. Journal of Animal Science 77:
147-153.

Kadarmideen, H. N., D. Schworer, H. Ilahi, M. Malek, and A. Hofer. 2004. Genetics of
osteochondral disease and its relationship with meat quality and quantity, growth,
and feed conversion traits in pigs. Journal of Animal Science 82: 3118-3127.

Karina, T. I. and R. B. Morrison. 2006. Association between limb conformation and
retention of sows through the second parity. American Journal of Veterinary
Research 67: 505-509.

Kerr, J. C., and N. D. Cameron. 1995. Reproduction performance of pigs selected for
components of efﬁcient lean growth. Animal Science 48: 241-244.

Kirkwood, R. N. 1990. Early puberty, mating reasonable goal. International Pigletter 10:
9.

Knauer, M. T., K. J. Stalder, L. Karriker, C. Johnson, and L. Layman. 2006. Factors
inﬂuencing sow culling. Proceedings of the National Swine Improvement
Federation, North Carolina State University.

Koketsu, Y., G. Dial, V. King, P. Bahnson, J. L. Xue, and T. Lucia. 1995a. Reproductive

failure observed on commercial swine farms. Recent Advances in Swine
Production and Health 5: 113-123.

100

Koketsu, Y., G. D. Dial, W. E. Marsh, J. E. Pettigrew, and V. L. King. 1995b. Path
analysis of factors affecting feed intake and reproductive performance in lactating
sows. Journal of Animal Science 72: Supplement 1: 333 Abstract.

Koketsu, Y., G. D. Dial, J. E. Pettigrew, and V. L. King. 1996. The inﬂuence of nutrient
intake on biological measures of breeding herd productivity. Journal of Swine
Health and Production 4: 85-94.

Koketsu, Y., and G. D. Dial. 1997. Factors inﬂuencing the postweaning reproductive
performance of sows on commercial farms. Theriogenology 47: 1445-1461.

Koketsu, Y., H. Takahashi, and K. Akachi. 1999. Longevity, lifetime pig production and
productivity, and age at ﬁrst conception in a cohort of gilts observed over six

years on commercial farms. Journal of Veterinary Medical Science 61: 1001-
1005.

Koketsu, Y. 2000. Retrospective analysis of trends and production factors associated with

sow mortality on swine-breeding farms in USA. Preventive Veterinary Medicine
46: 249-256.

Koketsu, Y. 2007. Longevity and efﬁciency associated with age structures of female pigs
and herd management in commercial breeding herds. Journal of Animal Science
85: 1086-1091.

Komegay, E. T., B. G. Diggs, 0. M. Hale, D. L. Handlin, J .P. Hitchcock, and R. A.
Barczwski. 1984. Reproductive performance of sows fed elevated calcium and
phosphorus levels during growth and development. A cooperative Study. Report
S-145 of the Committee on Nutritional Systems for Swine to Increase
Reproductive Efﬁciency. Journal of Animal Science 59: Supplement 1: 253
Abstract.

Korsgaard, I. R., A. H. Andersen, and J. Jensen. 2000. On different models, on
heritability, reliability and related quantities of survival traits. 5 1 st Annual
Meeting of the European Association for Animal Production, The Hague, The
Neterlands.

Lacy, C., K. Stalder, T. Cross, and G. Conatser. 2005. Breeding herd replacement female
evaluator. www.ipic.iastate.edu.

Lawless, J. 1982. Statistical Models and methods for lifetime data. John Wiley and sons,
New York, NY.

Le Cozler, Y., J. Dargon, J. E. Lindberg, A. Aumaitre, and J. Y. Dourmad. 1998. Effect

of age at ﬁrst farrowing and herd management on long-terrn productivity of sows.
Livestock Production Science 53: 135-142.

101

Lee, E. T. 1992. Statistical Methods for Survival Data Analysis, John Wiley & Sons, Inc.
New York, NY.

Long, T. E., K. J. Stalder, R. N. Goodwin, J. Halstead, J. M. Anderson, and R. L. Wyatt.
1998. Effect of gilt development diet on stayability to fourth parity in sows.
Journal of Animal Science 76: Supplement 2: 52 Abstract.

Lopez-Serrano, M., R. Reinsch, H. Looft, and E. Kalm. 2000. Genetic correlations of
growth, backfat thickness and exterior with stayability in Large White and
Landrace sows. Livestock Production Science 64: 121-131.

Louis, T. 1991. Assessing, accomodating and interpreting the inﬂuences of heterogeneity.
Environmental Health Perspectives 80: 215-222.

Love, R. J ., G. Evans, and C. Klupiec. 1993. Seasonal effects on fertility in gilts and
sows. Journal of Reproduction and Fertility 48191-206.

Love, R. J ., C. Klupiec, E. J. Thornton, and G. Evans. 1995. An interaction between

feeding rate and season affects fertility of sows. Animal Reproduction Science 39:
275-284.

Lucia, T., G. D. Dial, and W. E. Marsh. 1996. Patterns of female removal. 1. Lifetime
productivity for reproduction adn performance-related culls. Proceedings of the
14th International Pig Veterinary Society.

Lucia, T., G. D. Dial, and W. E. Marsh. 2000. Lifetime reproductive and ﬁnancial
performance of female swine. Journal of the American Veterinary Medical
Association 216: 1802-1809.

Lundeheim, N. 1996. Conformation scoring in the Swedish pig progeny testing scheme.
Proceedings of the Nordiska Jordbruksforskares F orening Seminar 265-Longevity
of Sows, Denmark: Research Centre Foulum.

MacLean, C. W. 1968. The thin sow problem. Veterinary Record 83: 308-316.

MacLean, M., H. Willis, R. Monaghan, and G. Foxcroﬁ. 2001. The effect of gilt age at
puberty on lifetime reproductive performance. Advances in Pork Production 12.
Abstract No. 30.

Maderbacher, R., G. Schoder, P. Winter, and W. Baumgartner. 1993. Causes of pig losses
in a pig breeding unit. Deutsche Tierarztliche Wochenschriﬁ 100: 468-473.

Martinez, G. E., R. M. Koch, L. V. Cundiff, K. E. Gregory, S. D. Kachman, and L. D.
Van Vleck. 2005. Genetic parameters for stayability, stayability at calving, and

stayability at weaning to speciﬁed ages for Hereford cows. Journal of Animal
Science 83: 2033-2042.

102

Morrison, B., A. Larriestra, J. Yan, and J. Deen. 2002. Determining optimal parity
distribution with a push model of gilt supply. Proceedings of the Allen D. Leman
Swine Conference.

Muirhead, M. R. and T. J. L. Alexander. 1997. Managing pig health and the treatment of
disease.

National Swine Improvement Federation. 1996. Guidelines for uniform swine
improvement programs. Available: http://www.nsif.com.

Newton, E. A., and D. C. Mahan. 1993. Effect of initial breeding weight and management

system using high-producing sow genotype on resulting reproductive performance
over three parities. Journal of Animal Science 71: 1177-1186.

Parsons, T. D., C. Johnstone, and G. D. Dial. 1990. On the economic signiﬁcance of
parity distribution in swine herds. Proceedings of the International Pig Veterinary
Society.

Patience, J. F. 2003. Dietary approaches to altering the composition of body weight gain
in growing pigs. Proceedings of the Allen D. Leman Swine Conference.

Patterson, R., C. Cargill, and A. Pointon. 1996. Investigations into deaths and excessive
culling of sows in Australian pig herds. Proceedings of the Nordiska
Jordbruksforskares F orening Seminar 265-Longevity of Sows, Denmark:
Research Centre Foulum.

Patterson, R., B. Mullan, A. Pointon, and C. Cargill. 1997. Investigations of the effect of
sow wastage on economic efﬁciency.

Peltoniemi, O. A. T., R. J. Love, M. Heinonen, V. Tuovinen, and H. Saloniemi. 1999.
Seasonal and management effects on fertility of the sow: a descriptive study.
Animal Reproduction Science 55: 47-61.

PigCHAMP. PigCHAMP. 2000 Datashare. PigCHAMP Inc. Ames, Iowa, USA (2000)
Available at: www.pigchamp.com/2000Datashare.htm.

PigCHAMP. PigCHAMP. 2001 Datashare. PigCHAMP Inc. Ames, Iowa, USA (2001)
Available at: www.pigchamp.com/2001Datasharehtm.

PigCHAMP. PigCHAMP. 2002 Datashare. PigCHAMP Inc. Ames, Iowa, USA (2002)
Available at: wwwpigchamp.com/2002Datashare.htm.

PigCHAMP. PigCHAMP. 2003 Datashare. PigCHAMP Inc. Ames, Iowa, USA (2003)
Available at: wwwmigchamp.com/2003Datasharehtm.

103

PigCHAMP. PigCHAMP. 2004 Datashare. PigCHAMP Inc. Ames, Iowa, USA (2004)
Available at: www.pigchamp.com/2004Datashare.htm.

PigCHAMP. PigCHAMP. 2005 Datashare. PigCHAMP Inc. Ames, Iowa, USA (2005)
Available at: www.pigchamp.com/2005Datasharehtm.

Pl'a, L. M., C. Pomar, and J. Pomar. 2003. A Markov decision sow model representing
the productive lifespan of herd sows. Agricultural Systems 76: 253-272.

Pond, W. G. and H. J. Mersmann. 2001. Biology of the Domestic Pig. Cornell University
Press: 148-150.

Prentice, R. L., and Gloeckler. 1978. Regression analysis of grouped survival data with
application to breast cancer data. Biometrics 34:57-67.

Prunier, A., M. Messias de Braganca, and J. Le Dividich. 1997. Inﬂuence of high ambient
temterature on performance of reproductive sows. Livestock Production Science
52123-133.

Rajala-Schultz, P. J ., and Y. T. Grohn. 2001. Comparison of economically optimized
culling recommendations and actual culling decisions of Finnish Ayrshire cows.
Preventive Veterinary Medicine 49: 29-39.

Reese, D. E., B.D. Moser, E. R. Peo Jr., A. J. Lewis, D. R. Zimmerman, J. E. Kinder, and
W. W. Stroup. 1982. Inﬂuence of energy intake during lactation on the interval
from weaning to ﬁrst estrus in sows. Journal of Animal Science 55: 590-598.

Ritter, L. A., J. L. Xue, G. D. Dial, R. B. Morrison, and W. E. Marsh. 1999. Prevalence
of lesions and body condition scores among female swine at slaugher. Journal of
the American Veterinary Medical Association 214: 525-528.

Rodriguez-Zas, S. L., B. R. Southey, R. V. Knox, J. F. Connor, J. F. Lowe, and B. J.
Roskamp. 2003. Bioeconomic evaluation of sow longevity and proﬁtability.
Journal of Animal Science 81: 2915-2922.

Rodriguez-Zas, S. L., C. B. Davis, P. N. Ellinger, G. D. Schnitkey, N. M. Romine, J. F.
Connor, R. V. Knox, and B. R. Southey. 2006. Impact of biological and economic

variables on optimal parity for replacement in swine breed-to-wean herds. Journal
of Animal Science 84: 2555-2565.

Rothschild, M. F., and L. L. Christian. 1988. Genetic control of front-leg weakness in

Duroc swine. 1. Direct response to ﬁve generations of divergent selection.
Livestock Production Science 19: 459-471.

104

Rozeboom, D. W., J. E. Pettigrew, R. L. Moser, S. G. Cornelius, and S. M. Kandelgy.
1996. Inﬂuence of gilt age and body composition at ﬁrst breeding on sow
reproductive performance and longevity. Journal of Animal Science 74: 138-150.

Rozeboom, D. W. 1999. Feeding programs for gilt development and longevity.
Proceedings of the Minnesota Nutrition Conference 60: 151-170.

Schukken, Y.H., J. Buurman, R. B. Huirne, A. H. Willemes, J. C. Vemooy, J. van den
Broek, and J. H. Verheijden. 1994. Evaluation of optimal age at ﬁrst conception

in gilts from data collected in commercial swine herds. Journal of Animial
Science 64: 121-131.

Schultz, R., D. Dau, W. Cast, D. Hoeﬂing, O. Duran, T. Carson, L. Becton, C.
Woodward, K. Pollard, K. Busker, D. Kaster, and M. Steidinger. 2001. A sow
mortality study-the real reasons sows die; Identifying causes and implementing
action. Proceedings of the 32nd Annual Meeting of the American Association of
Swine Practitioners.

Sehested, E. 1996. Economy of sow longevity. Proceedings of the Nordiska
Jordbruksforskares Forening Seminar 265-Longevity of Sows, Denmark:
Research Centre Foulum.

Sehested, E. and A. Schjerve. 1996. Aspects of sow longevity based on analyses of
Norwegian sow recording data. Proceedings of the Nordiska J ordbruksforskares
F orening Seminar 265-Longevity of Sows, Denmark: Research Centre F oulum.

Serenius, T., A. M. Sevon-Aimonen, and E. A. Mantysarri. 2001. The genetics of leg
weakness in Finnish Large White and Landrace populations. Livestock
Production Science 69: 101-111.

Serenius, T. 2004. Genetics of sow efﬁciency in the Finnish Landrace and Large White
populations. Agrifood Research Reports 55, MTT Agrifood Research Finland. 55.

Serenius, T., and K. J. Stalder. 2004. Genetics of length of productive life and lifetime
proliﬁcacy in the Finnish Landrace and Large White pig populations. Journal of
Animal Science 82: 3111-3117.

Serenius, T., M. L. Sevon—Aimonen, A. Kause, E. A. Mantysaari, and A. Maki-Tanila.
2004. Genetic associations of proliﬁcacy with performance carcass, meat quality,
and leg conformation traits in the Finnish Landrace and Large White pig
populations. Journal of Animal Science 82: 2301-2306.

Serenius, T., and K. J. Stalder. 2006. Selection for sow longevity. Journal of Animal
Science 84: E166-E171.

105

 

Serenius, T., K. J. Stalder, and R. L. Fernando. 2006a. Genetic associations of length of
productive life with age at ﬁrst farrowing and leg soundness score in Finnish
Landrace populations. Proceedings of the 8th world congress on genetics applied
to livestock production, Belo Horizonte, Brazil.

Serenius, T., K. J. Stalder, T. J. Baas, J. W. Mabry, R. N. Goodwin, R. K. Johnson, 0. W.
Robison, M. Tokach, and R. K. Miller. 2006b. National pork producers council
maternal line national genetic evaluation program: A comparison of sow

longevity and trait associations with sow longevity. Journal of Animal Science 84:
2590-2595.

Serenius, T., and K. J. Stalder. 2007. Length of productive life of crossbred sows is
affected by farm management, leg conformation, sow's own proliﬁcacy, sow's
origin parity and genetics. Animal 1: 745-750.

Serenius T., K. J. Stalder, and R. L. Fernando. 2008. Genetic associations of sow
longevity with age at ﬁrst farrowing, number of piglets weaned, and wean to
insemination interval in the Finnish Landrace swine population. Journal of
Animal Science 86: 3324-3329.

Short, T. H. and T. J. Lawlor. 1992. Genetic parameters of conformation traits, milk
yield, and herd life in Holsteins. Journal of Dairy Science 75: 1987-1998.

Smith, W. J ., and A. M. Robertson. 1971. Observations on injuries to sows conﬁned in
part slatted stalls. Veterinary Record 89531-533.

Smith, S. P. 1983. The extension of failure time analysis to problems of animal breeding.
Ithaca, New York, USA, Cornell University.

Smith, S. P., and R. L. Quaas. 1984. Productive life span of bull progeny groups: failure
time analysis. Journal of Dairy Science 67: 2999-3007.

Solkner, J ., and V. Ducrocq. 1999. The Survival Kit: a tool for analysis of survival data.
International Workshop on Genetic Improvement of Functional Traits: Longevity,
Uppsala, Sweden.

Stalder, K. J ., T. E. Long, R. N. Goodwin, J. M. Anderson, R. L. Wyatt, and J. H.
Halstead. 1998. Effect of gilt development diet on maternal traits of primiparous
sows. Journal of Animal Science 76: Supplement 2: 51 Abstract.

Stalder, K. J ., R. C. Lacy, T. L. Cross, G. E. Conatser, and C. S. Darroch. 2000. Net
present value analysis of sow longevity and the economic sensitivity of net
present value to changes in production, market price, feed cost, and replacement

gilt costs in a farrow-to-ﬁnish operation. The Professional Animal Scientist 16:
33-40.

106

 

Stalder, K. J ., R. C. Lacy, T. L. Cross, and G. E. Conatser. 2002. Financial Impact of
Average Parity of Culled Females in a Breed-to-Wean Swine Operation Using
Replacement Gilt Net Present Value Analysis. Journal of Swine Health and
Production 11: 69-74.

Stalder, K. J., M. Knauer, T. J. Baas, M. F. Rothschild and J. W. Mabry. 2004. Sow
Longevity. Pig News Information 25: (1): lN-22N.

Stalder, K. J ., M. Saxton, G. E. Conatser, and T. V. Serenius. 2005. Effect of growth and
compositional traits on ﬁrst parity and lifetime reproductive. Livestock
Production Science 97: 151-159.

Sterning, M., L. Rydhmer, and L. Eliasson-Selling. 1998. Relationships between age at
puberty and interval from weaning to estrus and between estrus signs at puberty
and after the ﬁrst weaning in pigs. Journal of Animal Science 76: 353-359.

Stevermer, E. J. 1991. How our production cost is impacted by changes in reproductive
efﬁciency. Proceedings of the Professional Swine Managers' Conference.

Straw, B. 1984. Causes and control of sow losses. Modern Veterinary Pracitce 65: 349-
353.

Takai, Y., and Y. Koketsu. 2007. Identiﬁcation of a female-pig proﬁle associated with
lower productivity on commercial farms. Theriogenology 68: 87-92.

Tantasuparuk, W., A. -M. Dalin, N. Lundeheim, A. Kunavongkrit, and S. Einarsson.
2001. Body weight loss during lactation and its inﬂuence on weaning-to-service
interval and ovulation rate in Landrace and Yorkshire sows in the tropical
environment of Thailand. Animal Reproduction Science 65: 273-281.

Tarres, J ., J. Tibau, J. Piedraﬁta, E. F abrega, and J. Reixach. 2003. Factors affecting
longevity in maternal Duroc swine lines. Proceedings of the 54th Annual Meeting
of European Association for Animal Production, Rome, Italy.

Tarres, J ., J. P. Bidanel, A. Hofer, and V. Ducrocq. 2006. Analysis of longevity and
exterior traits on Large White sows in Switzerland. Journal of Animal Science 84:
2914-2924.

teBrake, J. H. A. 1986. Culling of sows and the proﬁtability of piglet production.
Netherlands Journal of Agricultural Science 34: 427-435.

Tempelman, R. J ., and D. Gianola. 1994. Estimation of genetic variance for
overdispersed count data with an animal breeding application. Joint Statistical
Meetings, Toronto, Canada.

107

ten Napel, J. and R. Johnson. 1997. Genetic relationships among production traits and
rebuilding performance. Journal of Animal Science 75: 51-60.

Thacker, B. J. 2002. Seasonal infertility: Management practices-causes and ﬁxes.
Proceedings of the Swine Disease Conference for Swine Practitioners, Iowa State
University, Ames, Iowa, USA.

Tholen, E., K. L. Bunter, S. Herrnesch and H. U. Graser. 1996a. The genetic foundation
of ﬁtness and reproduction traits in Austrailian pig populations. 1. Genetic
parameters for weaning to conception interval, farrowing interval, and stayability.
Australian Journal of Agricultural Research 47: 1261-1274.

Tholen, E., K. L. Bunter, S. Herrnesch and H. U. Graser. 1996b. The genetic foundation
of ﬁtness and reproduction traits in Austrailian pig populations 2. Relationships
between weaning to conception interval, farrowing interval, stayability, and other

common reproduction and production traits. Australian Journal of Agricultural
Research 47: 1275-1290.

Tokach, M., M. Webster, and B. James. 2000. The role of nutrition and the feeding
program on gilt longevity. Proceedings of the Allen D. Leman Swine Conference.

Tummaruk, P., N. Lundeheim, S. Einarsson, and A.-M. Dalin. 2000. Reproductive
performance of purebred Swedish Landrace and Swedish Yorkshire sows: 1.

Seasonal variation and parity inﬂuence. Acta Agriculturae Scandinavica Section
A, Animal Science 50: 205-216.

Tummaruk, P., N. Lundeheim, S. Einarsson, and A.-M. Dalin. 2001a. Effect of birth litter
size, birth parity number, growth rate, backfat thickness and age at ﬁrst mating of

gilts on their reproductive performance as sows. Animal Reproduction Science
66: 225-23 7.

Tummaruk, P., N. Lundeheim, S. Einarsson, and A.-M. Dalin. 2001b. Repeat breeding
and subsequent reproductive performance in Swedish Landrace and Swedish
Yorkshire sows. Animal Reproduction Science 67:267-280.

Van Steenbergen, E. J. 1989. Description and evaluation of a linear scoring system for
exterior traits in pigs. Livestock Production Science 23: 163-181.

Veerkamp, R. F., E. P. C. Koenen, and G. De Jong. 2001. Genetic correlations among
body condition score, yield, and fertility in ﬁrst-parity cows estimated by random
regression models. Journal of Dairy Science 84: 2327-2335.

Velander, I. H., and M. Holmstrom. 1996. A quantitative method of describing the

conformation of sows. Proceedings of the Nordiska Jordbruksforskares F orening
Seminar 265-Longevity of Sows, Denmark: Research Centre Foulum.

108

Vollema, A. R., and A. F. Groen. 1998. A comparison of breeding value predictors for
longevity using a linear model and survival analysis. Journal of Dairy Science 81:
33 15-3320.

Vukasinovic, N., J. M011, and N. Kunzi. 1997. Analysis of productive life in Swiss
brown cattle. Journal of Dairy Science 80: 2572-2579.

Whittemore, C. T., M. F. Franklin, and B. S. Pearce. 1980. Fat changes in breeding sows.
Animal Production 31: 183-190.

Xue, J. L., G. D. Dial, W. E. Marsh, and P. R. Davies. 1994. Multiple manifestations of
season on reproductive performance of commercial swine. Journal of the
American Veterinary Medical Association 204: 1486-1489.

Xue, J. L., G. D. Dial, W. E. Marsh, and T. Lucia. 1997. Association between lactation
length and sow reproductive performance and longevity. Journal of the American
Veterinary Medical Association 210: 935.

Yazdi, M. H., N. Lundeheim, L. Rydhmer, E. Ringrnar-Cederberg, and K. Johansson.
2000a. Survival of Swedish Landrace and Yorkshire sows in relation to
osteochondrosis: a genetic study. Animal Science 71: 1-9.

Yazdi, M. H., L. Rydhmer, E. Ringmar-Cederberg, N. Lundeheim, and K. Johansson.
2000b. Genetic study of longevity in Swedish Landrace sows. Livestock
Production Science 63: 255-264.

Yazdi, M. H., P. M. Visscher, V. Ducrocq, and R. Thompson. 2002. Heritability,
reliability of genetic evaluations and response to selection in proportional hazards
models. Journal of Dairy Science 85: 1563-1577.

Young, L. G., G. J. King, J. Shaw, M. Quinton, J. S. Walton, and I. McMillan. 1991.
Interrelationships among age, body weight, backfat, and lactation feed intake with
reproductive performance and longevity of sows. Canadian Journal of Animal
Science 71: 567-575.

109

 

1111'

3 1293 03062 8329

11111111