Relationship Between Backfat and Sow Longevity in Canadian Yorkshire and Landrace Pigs

J.R Brisbane and J.P. Chesnais
Canadian Centre for Swine Improvement,
Ottawa, Ontario, Canada

1. Introduction

Increases in longevity, defined as lifetime number of litters, lead to increased profit through reduced sow replacement cost. There is some concern among swine breeders that genetic selection for leanness may be reducing sow longevity. If longevity declines, then replacement costs increase at an accelerating rate. For example if the farrowing rate is 2.1 litters/year, a decrease in longevity from 6 litters to 5.8 increases the replacement rate by only 1.2%, but if longevity is allowed to drop to 3 litters then a subsequent decrease to 2.8 increases the annual replacement rate by 5%. Also, as pigs in Canada become closer to an optimum level of leanness, the economic value of leanness decreases, which means other traits, including sow longevity in dam lines, receive a higher relative weighting.

In a purebred selection herd, longevity is a function of voluntary culling for genetic improvement as well as involuntary culling due to health and fertility problems. Kennedy et al (1996) analysed backfat and age at 100kg data on Canadian purebred swine and identified long generation intervals as a major factor limiting rates of genetic improvement. Given that this is true, it is possible that selection herds with greater longevity may be fatter because they have a lower rate of genetic improvement due to a long generation interval with a low amount of voluntary culling. This is completely due to herd management, and any genetic correlation between backfat and longevity must operate through involuntary culling.

In order to estimate the relationship between backfat and involuntary culling, this study looks at differences in longevity within herd-management groups. This is a preliminary analysis before estimating genetic correlations. The approach used follows that of Richard (1995, pers. comm.) who carried out similar calculations for multiplier herds in Québec. Richard (1995, pers. comm.) showed that gilts with leaner backfat probe measures in performance test, went on to have fewer litters as multiplier sows.

2. Methods

All farrowing records and backfat probe records on Yorkshire and Landrace females were extracted from the national database. These records had passed various edits for inclusion in the database, including checks for valid animal identification, valid herd number, sex conflicts, valid ranges on some numeric fields, and valid ranges and intervals between dates of birth, probing, farrowings and weanings. The probe records included backfat adjusted to 100kg probe weight and estimated age at 100kg. The farrowing records included the farrowing date and herd number.

All herd numbers which did not appear in any farrowing records from 1996 were identified, and all farrowing records on sows with these herdnumbers were deleted. This removed data which would cause underestimation of longevity because the herds ceased recording sow productivity. For each remaining herd with farrowing records, the number of consecutive years prior to 1996 in which the herd had records was counted. When a year was encountered where there were no farrowing records from the herd, then any farrowing records from that herd prior to that year were deleted. After this stage, all probe records from herds which did not have any farrowing records were deleted, and all probe records where the year of probing was prior to the year of the earliest farrowing records in the same herd were deleted. All probe records with the backfat figure missing were deleted. Then augmented probe records were produced by adding a field to each remaining probe record to indicate whether or not the gilt had at least 1 farrowing record in the edited data. Augmented probe records with birthdates in 1996 were deleted because if selected, these animals would not have enough time to appear in the farrowing records. The remaining augmented probe records formed what will be referred to as the gilt selection dataset. The herd in this dataset is the herd of probing.

To produce the sow longevity dataset, the farrowing records were taken after having passed through the edits mentioned above, and all records on sows which had some farrowing records missing (non-consecutive parity numbers) or which had the parity number missing in some of their records, were deleted. All records on sows where the days of age at farrowing (calculated from the farrowing date and the birthdate) in any parity was not between 200+85(parity-1) and 560+300(parity-1) were deleted. Then a smaller dataset was extracted with one record per sow, consisting of the sow identification, the last parity number, the year of last farrowing, the herd of last farrowing and whether or not the herd was the same for all farrowings. All records on sows where the year of last farrowing was 1996 were deleted. These sows were likely still breeding at the data cutoff. If it is assumed that farrowing records in 1996 contain representative proportions of each parity, then this is a valid procedure to avoid the underestimation of longevity. All records on sows which farrowed in more than one herd were deleted. For the remaining sows with records, the year, parity number and herd of last farrowing were added to the probe record on the sow. When searching for the probe record on the sow, probe records where the year of probing was prior to the year of the earliest farrowing records in the same herd were included. Sows which did not have a probe record were then removed from the data. The augmented probe records on the sows formed what will be referred to as the sow longevity dataset. In this dataset herd refers to the herd in which all farrowings of the sow took place.

For each breed, Table 1 shows the number of breeding sows and unselected gilts in the data by period of birth. There are no unselected gilts in the sow longevity dataset.

The records of the gilt selection dataset were grouped by backfat level into 6 classes, and the proportion of gilts selected for breeding in each class was calculated (Table 2). The records of the sow longevity dataset were grouped in the same way, and in each class the proportion of sows surviving beyond each parity was calculated (Table 3).

The records of the gilt selection dataset were sorted into herd-birthyear groups. Where there was only 1 record in a group, that record was deleted. The proportion of gilts selected for breeding in each remaining herd-birthyear group was calculated, and the proportion selected from each backfat class within each herd-birthyear was also calculated. Then the deviation of the proportion selected in each backfat class within the herd-birthyear from the proportion selected in the herd-birthyear overall was calculated. These deviations were averaged across herd-birthyears for each backfat class, weighting each one by the number of animals involved. i.e.

where dij is the deviation of the proportion selected in backfat class i in herd-birthyear j from the overall proportion selected from herd-birthyear j, nij is the number of candidate gilts in backfat class i and herd-birthyear j, and nhyr is the number of herd-birthyear groups. The weighted average deviations (di) for each backfat class are given in Table 4. A similar process was repeated for the sow longevity dataset. This involved calculating the proportion of sows in each herd-birthyear group which survived beyond each parity, and the proportion of sows within each backfat class within each herd-birthyear which survived beyond each parity. Then the deviation of the proportion surviving in each backfat class within the herd-birthyear from the proportion surviving in the herd-birthyear overall was calculated, and the weighted average deviation for each backfat class across herd-birthyears was calculated as in equation (1). The weighted average deviations for each backfat class are given in Table 5.

It was necessary to use separate datasets for the study of gilt selection and the study of longevity because the gilts selected from those probed, included animals with farrowing records in 1996. Those animals therefore had to be included in the gilt selection data, but they could not be included in the longevity data because they were still breeding at the data cutoff. Similarly, the gilts selected from those probed, included sows which changed herds, but these sows were not included in the longevity analysis because they appeared in more than one herd-birthyear class.

Table 2 includes data on 3 Yorkshire gilts and 2 Landrace gilts which were deleted from the analysis for Table 4 because they were each the only animal in their herd-birthyear group. Table 3 includes data on 60 Yorkhire sows and 58 Landrace sows which were deleted from the analysis for Table 5 because they were each the only animal in their herd-birthyear group.

3. Results

3.1 Production of the datasets

Initially, for Yorkshire there were 135,393 farrowing records and 677,770 probe records. After removing farrowing records from herds which ceased sow productivity recording there were 95,344 farrowing records from 25,752 sows. After removing farrowing records prior to the period of consecutive years of records for the herd, there were 95,300 records. After removing probe records from herds which did not have any farrowing records there were 206,027 probe records, after removing probe records where the year of probing was prior to the first year of farrowing records in the herd, there were 171,561, after removing probe records where the backfat was missing there were 171,403, and after removing probe records with birthdates in 1996 there were 164,476. A field indicating if the gilt had at least 1 farrowing record, was added to each record, to form the Yorkshire gilt selection dataset. There were 23,468 gilts in this data which appeared in the edited farrowing records. The proportion of probed gilts which became breeding sows was therefore 23,468/164,476 = 14.3%

After removing farrowing records on sows which had parity numbers missing from their records or had non-consecutive parity numbers, there were 90,627 farrowing records. After removing all records on sows with an invalid age at farrowing there were 90,539 records from 24,847 sows. The smaller dataset then produced consisted of a record containing the identification, last parity number, year of last farrowing, herd of last farrowing and whether or not the herd was the same for all farrowings, for each of these sows. After removing all records on sows with 1996 as the year of last farrowing there were 18,361 records, and after deleting sows which changed herds there were 18,266 records. 17,414 of these sows had probe records with non-missing backfat, and records on these sows formed the Yorkshire longevity dataset.

Initially, for Landrace there were 155,464 farrowing records and 530,450 probe records. After removing farrowing records from herds which ceased sow productivity recording there were 107,150 farrowing records from 25,504 sows. After removing farrowing records prior to the period of consecutive years of records for the herd, there were 106,910 records. After removing probe records from herds which did not have any farrowing records there were 194,652 probe records, after removing probe records where the year of probing was prior to the first year of farrowing records in the herd, there were 169,744, after removing probe records where the backfat was missing there were 169,673, and after removing probe records with birthdates in 1996 there were 163,005. A field indicating if the gilt had at least 1 farrowing record, was added to each record, to form the Yorkshire gilt selection dataset. There were 24,553 gilts in this data which appeared in the edited farrowing records. The proportion of probed gilts which became breeding sows was therefore 24,553/163,005 = 15.1%

After removing farrowing records on sows which had parity numbers missing from their records or had non-consecutive parity numbers, there were 102,989 farrowing records. After removing all records on sows with an invalid age at farrowing there were 102,808 records from 26,683 sows. The smaller dataset was then produced in the same way as for Yorkshire, with records containing the identification, last parity number, year of last farrowing, herd of last farrowing and whether or not the herd was the same for all farrowings, for each of these sows. After removing all records on sows with 1996 as the year of last farrowing there were 20,195 records, and after deleting sows which changed herds there were 20,028 records. 19,031 of these sows had probe records with non-missing backfat, and records on these sows formed the Yorkshire longevity dataset.

3.2 Distribution of Animals by Birthyear

Table 1 shows that the Yorkshire and Landrace datasets are similar in size, and in both cases, more animals come from the more recent birthyears, where there are more herds on performance recording. This is true for both the gilt selection and the longevity data. In the gilt selection datasets for both breeds, the ratio of the number of breeding sows to the number of unselected gilts increases with birthyear period up to 1990-1992 and then decreases. This could be due to expanding herd sizes, or a reduction in the proportion of the sow herd which is selected from gilts which are not probed, or increases in replacement rates. Further data analysis is needed to determine the reason.

3.3 Gilt Selection

From Table 2, in Yorkshire the average proportion of probed gilts selected for breeding is 14.3%, and the proportion is highest (15.8%) in the 12-14mm backfat range, and lowest (7.6%) in the fattest backfat range ( > 18mm). In Landrace the average proportion selected is 15.1%, and it is also highest (17.7%) in the 12-14mm range and lowest (9.4%) in the fattest backfat range. In both cases, lean gilts are less likely to be selected than those with an intermediate level of backfat, but discrimination against lean gilts is higher in the Landrace data.

When looking at differences in rates of gilt selection within each herd and year in Table 4, the discrimination against very lean gilts is still substantial (-3.2% in Yorkshire and -4.2% in Landrace), and so is the discrimination against fatter gilts (-4.8% and -2.9%).

3.4 Longevity

From Table 3, in both Yorkshire and Landrace the proportion of sows surviving beyond any given parity increases dramatically as the backfat level increases.

The proportion of Yorkshire surviving beyond parity 2 is about 59% on average (Table 3), and increases by 26% (from 45% to 71%) as the backfat level increases. The proportion of Yorkshire surviving beyond parity 4 is about 35% on average, and increases by 31% (from 20% to 51%) as the backfat level increases. For survival beyond parity 6 the average is about 19%, and increases by 23% (from 9% to 32%) as the backfat level increases.

In Landrace, average survival beyond parity 2 is about 62% (Table 3), and increases by 30% (from 43% to 73%) as the backfat level increases. For survival beyond parity 4 the average is about 39% and increases by 36% (from 18% to 54%) as the backfat level increases. For survival beyond parity 6 the average is about 22%, increases by 29% (from 7% to 36%) as the backfat level increases.

For Yorkshire, when examining survival within each herd and year, the survival rate increases by about 8% for parity 2, by about 10% for parity 4, and by about 7% for parity 6, as the backfat level increases (Table 5). For Landrace, the increase in survival rate as backfat increases is only 3% for parity 2, 4% for parity 4, and 3% for parity 6.

It can be concluded that there is a small but significant relationship between backfat thickness and longevity, particularly in the Yorkshire breed.

4. Discussion

4.1 Implications of the Results

Breeders tend to have a lower rate of selection of lean gilts than would normally be expected. On average they tend to select gilts in the 12 to 16mm backfat class and discriminate against very lean (<10mm) or very fat (>18mm) gilts. It is possible that breeders are not selecting the leanest gilts because of the perception that there is a negative relationship between backfat thickness and longevity.

When looking at the overall data, there is in fact a strong association between leanness and longevity. It would appear that the lean sows do not remain in the herd for as many parities. After the 4th parity, the survival rate of sows in the leanest category (<10mm) was about 30% lower than those in the fattest category (>18mm) in the Yorkshire breed and 33% lower in the Landrace breed. However, this association is due mainly to the fact that breeding herds that have a higher replacement rate also have leaner sows, probably because the genetic change for backfat is higher in those herds.

When examining the survival of sows on a within-herd basis, the association between degree of leanness and longevity is much weaker. In the Landrace breed, the trend is very small. The difference in survival after the 4th parity between the lowest backfat category (<10mm) and the highest (>18mm) is about 4% in favour of the latter. In the Yorkshire breed the trend is more pronounced. The difference in survival after the 4th parity between the highest and lowest backfat categories is about 10% in favour of the latter.

4.2 Future Work

Given the above, there is merit in further investigating the relationship between backfat thickness and survival rates in sows. To investigate genetic relationships, differences in survival rates will be correlated to backfat EBV instead of phenotypic backfat. Another approach will be to develop a genetic model to estimate the genetic correlation between backfat and longevity. If, based on the genetic correlation estimate and the economic values of the two traits, it is decided that longevity should be included in the breeding goal, then the genetic model would be used to calculate EBV for longevity.

Three alternative types of genetic models exist for longevity. One is a multiple binary trait system defining survival or culling at each parity (Everett et al, 1976), as is currently done in genetic evaluations of herdlife in Canadian dairy cattle. For best results, this approach requires a threshold model. The use of a conventional linear model with binary traits gives evaluations of lower accuracy. The second is a single trait approach modelling the lifetime number of parities, with censored records being replaced by projected records (Van Raden and Klaaskate, 1993; Van Raden and Wiggans, 1995), as is currently done in genetic evaluations of herdlife in dairy cattle in the US. A third alternative is to use survival analysis methods. Smith and Quaas (1984) estimated heritability using these methods, and Ducrocq and Solkner (1994) have developed freely available software which will estimate genetic parameters and predict breeding values using this approach with an animal model.

Cull codes are recorded in some of the farrowing data, and future work could include an analysis of the reasons for culling in each backfat group. Currently there is no cull code representing culling for genetic improvement purposes in the national data formats, and to include one in future would allow for more accurate investigation of longevity.

Before sow longevity is included in the breeding goal for dam lines however, it will be important to validate this work by examining the relationship between the longevity of F1 sows and the backfat EBV of their purebred parents. This will require the use of sow productivity data from commercial herds where the parents of each F1 can be traced back to the multiplier herds.

5. References

Ducrocq, VP and J. Solkner (1994) "The Survival Kit". A Fortran Package for the Analysis of Survival Data. Proc. 5th World Cong. Gen Appl. Livest. Prod. 22:51

Everett, RW, JF Keown and EE Clapp (1976) Production and Stayability Trends in Dairy Cattle. J. Dairy Sci. 59:1532

Kennedy, BW, VM Quinton and C Smith (1996) Genetic Changes in Performance Tested Pigs for Growth Rate and Fat Depth. Canadian J. of Animal Science 76:41

Smith, SP, and RL Quaas (1984) Productive Lifespan of Bull Progeny Groups: Failure Time Analysis. J. Dairy Science 67:2999

Van Raden, PM, and EJH Klaaskate (1993) Genetic Evaluation of Length of Productive Life Including Predicted Longevity of Live Cows. J. Dairy Sci. 76:2758

Van Raden, PM, and GR Wiggans (1995) Productive Life Evaluations: Calculations, Accuracy and Economic Value. J. Dairy Sci. 76:2758

Table 1. Number of Breeding Sows and Unselected Gilts by Birthyear

a) Gilt Selection Datasets
Landrace Yorkshire
BirthyearBreeding sows (A) Unselected gilts (B)Ratio (A/B) Breeding sows (A)Unselected gilts (B) Ratio (A/B)
before 19842,07317,447 0.121,44311,405 0.13
1984-19862,52816,330 0.151,87713,624 0.14
1987-19894,73426,831 0.184,68226,848 0.17
1990-19926,99229,765 0.237,22136,856 0.20
1993-19958,22648,079 0.178,24552,275 0.16
Total24,553138,452 0.1823,468141,008 0.17

b) Sow Longevity Datasets
BirthyearLandrace
breeding sows
Yorkshire
breeding sows
before 19842,3601,469
1984-19862,5911,979
1987-19894,7484,737
1990-19926,4476,361
1993-19952,8852,868
Total19,03117,414



Table 2. Proportion of Gilts Selected for Breeding by Backfat Class (number of candidates in parenthesis)

Backfat class (mm)
< 10(10,12) (12,14)(14,16)(16,18) >18Over all pigs
Yorkshire12.0%
(16,544)
14.8%
(40,179)
15.8%
(47,513)
15.4%
(34,042)
12.2%
(16,584)
7.6%
(9,614)
14.3%
(164,476)
Landrace10.8%
(15,343)
15.4%
(34,927)
17.7%
(42,133)
16.3%
(34,097)
14.4%
(21,375)
9.4%
(15,130)
15.1%
163,005

Table 3. Proportion of Sows Surviving after each Parity by Backfat Class

a) Yorkshire
ParityBackfat class interval (mm), followed by number of pigs in parenthesis
< 10
(1061)
(10,12)
(3769)
(12,14)
(5663)
(14,16)
(4366)
(16,18)
(1837)
> 18
(718)
Over all pigs
(17414)
165.4%70.1% 76.1%77.2%80.2% 83.4%75.2%
245.0%51.7% 59.2%61.8%66.5% 71.4%58.6%
329.6%38.6% 45.0%49.1%54.1% 59.6%45.3%
420.5%28.6% 34.1%38.7%44.8% 51.1%35.1%
513.2%20.2% 25.5%29.9%35.6% 41.8%26.4%
69.0%14.4% 17.8%22.3%26.2% 31.8%19.1

b) Landrace
ParityBackfat class interval (mm), followed by number of pigs in parenthesis
< 10
(952)
(10,12)
(3395)
(12,14)
(5559)
(14,16)
(4731)
(16,18)
(2898)
> 18
(1496)
Over all pigs
(19031)
161.2%69.3% 76.1%79.6%81.8% 83.4%76.5%
243.2%50.2% 60.3%66.6%69.8% 72.6%61.6%
329.5%36.9% 47.2%54.7%58.4% 63.3%49.3%
418.2%25.6% 36.4%43.9%48.4% 53.9%38.7%
511.4%17.1% 26.7%35.1%40.0% 45.1%29.8%
67.0%10.8% 19.3%26.6%30.1% 36.0%21.9%

Table 4. Within-Herd-Birthyear Deviations in Proportion of Gilts Selected for Breeding by Backfat Class (number of candidates in parenthesis).
Backfat class interval (mm)
< 10(10,12) (12,14)(14,16)(16,18) > 18
Yorkshire-3.2%
(16544)
+0.1%
(40178)
+1.4%
(47513)
+1.4%
(34042)
-1.2%
(16582)
-4.8%
(9614)
Landrace-4.2%
(15343)
-0.4%
(34927)
+1.6%
(42133)
+1.3%
(34095)
+0.5%
(21375)
-2.9%
(15130)

Table 5. Within-Herd-Birthyear Deviations in Proportion of Sows Surviving by Backfat Class

a) Yorkshire
ParityBackfat class interval (mm), followed by number of pigs in parenthesis
< 10
(1056)
(10,12)
(3757)
(12,14)
(5649)
(14,16)
(4351)
(16,18)
(1828)
> 18
(713)
Over all pigs
(17354)
1-3.0%-1.9% +0.7%+0.6%+1.3% +1.9%75.2%
2-4.0%-2.8% +0.5%+1.0%+2.6% +3.6%58.6%
3-5.1%-2.2% -0.1%+1.3%+2.9% +4.7%45.3%
4-4.4%-2.0% -0.7%+1.2%+3.7% +5.7%35.1%
5-3.9%-2.0% -0.5%+1.2%+3.3% +4.9%26.4%
6-2.7%-1.4% -0.8%+1.2%+2.4% +4.1%19.1%

b) Landrace
ParityBackfat class interval (mm), followed by number of pigs in parenthesis
< 10
(949)
(10,12)
(3384)
(12,14)
(5546)
(14,16)
(4721)
(16,18)
(2890)
> 18
(1483)
Over all pigs
(18973)
1-4.0%-1.1% +0.9%+0.5%+0.1% 0.0%76.5%
2-3.0%-2.4% +0.8%+1.2%+0.3% +0.3%61.6%
3-2.6%-2.1% +0.4%+1.2%0.0% +1.1%49.3%
4-2.4%-2.2% +0.6%+1.0%-0.2% +1.4%38.7%
5-1.8%-2.2% 0.0%+1.4%+0.2% +1.3%29.8%
6-1.4%-2.2% +0.3%+1.5%-0.5% +1.1%21.9%