Materials and Methods

Three commercial swine herds that used the PigCHAMP^a record-keeping program and recorded reasons for sow removal were used in the study (Table 1). All 3 herds were located in the upper Midwest and consisted of Landrace and Landrace × Yorkshire females sourced from a single genetic supply company. None of the herds had any evidence of porcine reproductive and respiratory syndrome virus infection or Mycoplasma hyopneumoniae infection during the study period.

Records for the 3 herds were searched from January 2004 forward to identify sows for which removal reason and parity at the time of removal had been reported. Individual sows were included in the study as case sows if the reason for removal was related to concerns about fertility, fecundity, or age. Fertility concerns included “no heat,” “failure to conceive,” “abortion,” “pregnancy test negative,” and “recycle or repeat breeder.” Fecundity concerns included “small litter size” and “live born.” Age concerns included “old age or parity.” Sows were excluded from the study if the recorded reason for removal included any indication of a health problem, such as vaginal discharge or lameness.

One thousand case sows from each herd were included in the study to provide sufficient sample size to detect a difference of 0.8 PBA/MF/Y, given an D value of 0.05 and power of 0.8. To allow for a sufficient number of control sows, only sows that were parity 1 through 6 at the time of removal were included as case sows.

Each case sow in each of the 3 herds was matched with a control sow and a replacement gilt from the same herd. Control sows were included in the study to generate an estimate of counterfactual reproductive performance (ie, reproductive performance that could have been expected had case sows been retained in the herd). Control sows were matched with case sows on the basis of genetic type, parity, number of inseminations required to generate the case's final litter (for case sows, the litter prior to removal; for control sows, the litter used to match case and control sows), and weaning classification (ie, weaned a second litter, weaned the current litter, or did not wean any pigs). Control sows were also matched with case sows on the basis of number of pigs born alive and number of pigs weaned for the parity of interest. In addition to these within-parity matching criteria, control sows were matched with case sows on the basis of lifetime average number of pigs born alive per litter and lifetime average farrow-to-farrow interval (for parity 1 sows, farrow-to-farrow interval was recorded as 0 days). Replacement gilts were matched with case sows on the basis of genetic type and breeding date, such that the date of first insemination for the replacement gilt was approximately the same as the date of removal for the case sow.

Subsequent reproductive performance of control sows in each herd was compared with reproductive performance of the replacement gilts. Because of the high degree of cross-fostering among litters that occurred and because of herd differences in farrowing practices, comparisons were made on the basis of number of pigs born alive per litter. Litter size was annualized to account for differences in first insemination-to-farrow intervals for the replacement gilts and standardized for multiparous sows to account for differences in lactation duration and weaning-to-first-insemination interval. Thus, for control sows, reproductive performance (ie, number of PBA/MF/Y) was calculated with the following equation: (number of pigs born alive in the next litter × 365)/(farrow-to-farrow interval – 25). In this equation, the farrow-to-farrow interval was adjusted by 25 days to account for the expected duration of the lactation period and the weaning-to-first-insemination interval. For replacement gilts, reproductive performance (ie, number of PBA/MF/Y) was calculated with the following equation: (number of pigs born alive in the parity 1 litter × 365)/(first-insemination-to-farrow interval).

Control sows were followed up through the next farrowing and litter or until removed from the herd. Replacement gilts were followed up through the first farrowing and parity 1 litter or until removed from the herd. For each case sow, removal was classified as successful or unsuccessful on the basis of whether reproductive performance was higher for the matched control sow or matched replacement gilt.

Statistical analysis—For each herd, subsequent reproductive performance of control sows (ie, PBA/MF/Y) was compared with reproductive performance of replacement gilts (ie, PBA/MF/Y) within individual categories for removal of case sows with standard software.^b Analysis of variance was used to test differences in means. Logistic regression was used to determine the odds of a higher number of PBA/MF/Y for replacement gilts, with the 95% confidence interval calculated as e^{(β₁ ± 1.96*SE β₁)}. Values of P ≤ 0.05 were considered significant.

Results

Herd sizes ranged from 2,600 to 3,000 sows (Table 1). For each of the 3 herds in the study, 1,000 case sows were enrolled in the study. In herd A, 777 of the case sows had been removed from the herd for reasons related to fertility, 187 had been removed for reasons related to fecundity, and 36 had been removed for reasons related to age. In herd B, 873 of the case sows had been removed for reasons related to fertility, and 127 had been removed for reasons related to fecundity. In herd C, 787 of the case sows had been removed for reasons related to fertility, 197 had been removed for reasons related to fecundity, and 16 had been removed for reasons related to age.

For herd A, mean ± SE number of pigs born alive in the matched litters was 9.25 ± 0.11 for case sows and 9.27 ± 0.11 for control sows. For herd B, mean number of pigs born alive in the matched litters was 10.22 ± 0.11 for case sows and 10.21 ± 0.11 for control sows. For herd C, mean number of pigs born alive in the matched litters was 9.19 ± 0.11 for case sows and 9.27 ± 0.11 for control sows. For herd A, mean ± SE farrow-to-farrow interval at the matched parity was 107.4 ± 2.14 days for case sows and 106.0 ± 2.14 days for control sows. For herd B, mean farrow-to-farrow interval at the matched parity was 98.1 ± 2.16 days for case sows and 98.0 ± 2.16 days for control sows. For herd C, mean farrow-to-farrow interval at the matched parity was 92.5 ± 2.32 days for case sows and 91.1 ± 2.34 days for control sows. There were no differences between case and control sows with regard to mean number of pigs born alive (P > 0.6) or mean farrow-to-farrow interval (P > 0.6).

For herd A, mean ± SE number of PBA/MF/Y was 29.78 ± 0.01 for control sows and 32.11 ± 0.01 for replacement gilts. For herd B, mean number of PBA/MF/Y was 30.09 ± 0.01 for control sows and 32.55 ± 0.01 for replacement gilts. For herd C, mean number of PBA/MF/Y was 27.12 ± 0.01 for control sows and 28.42 ± 0.01 for replacement gilts.

For control sows and replacement gilts matched with case sows that had been removed for reasons related to fertility, number of PBA/MF/Y was significantly lower for control sows than for replacement gilts in herds A (mean ± SE, 30.54 ± 0.40 and 32.21 ± 0.40, respectively; P = 0.035) and B (30.75 ± 0.41 and 32.71 ± 0.41, respectively; P < 0.001), but not in herd C (28.02 ± 0.41 and 28.37 ± 0.41, respectively).

For control sows and replacement gilts matched with case sows that had been removed for reasons related to fecundity, number of PBA/MF/Y was significantly lower for control sows than for replacement gilts in herds A (27.47 ± 0.82 and 32.18 ± 0.82, respectively; P < 0.001), B (25.50 ± 1.06 and 31.52 ± 1.06, respectively; P < 0.001), and C (23.53 ± 0.83 and 28.79 ± 0.83, respectively; P < 0.001). For control sows and replacement gilts matched with case sows that had been removed for reasons related to age, number of PBA/MF/Y was not significantly different between control sows and replacement gilts in herd A (25.48 ± 1.87 and 29.55 ± 1.87, respectively; P = 0.124) or herd C (26.21 ± 4.80 and 23.99 ± 4.80, respectively; P = 0.745).

The odds that sow removal was successful (ie, that reproductive performance of the replacement gilt was higher than reproductive performance of the control sow) were 2.105 (95% CI, 1.761 to 2.516) for case sows in herd A, 2.097 (95% CI, 1.754 to 2.506) for case sows in herd B, and 1.562 (95% CI, 1.309 to 1.863) for case sows in herd C. For sows removed for reasons related to fertility, the odds that sow removal was successful were 1.806 (95% CI, 1.477 to 2.208) for herd A, 1.955 (95% CI, 1.617 to 2.365) for herd B, and 1.305 (95% CI, 1.072 to 1.589) for herd C. For sows removed for reasons related to fecundity, the odds that sow removal was successful were 3.607 (95% CI, 2.355 to 5.525) for herd A, 3.437 (95% CI, 2.052 to 5.756) for herd B, and 3.365 (95% CI, 2.226 to 5.088) for herd C. For sows removed for reasons related to age, the odds that sow removal was successful were 3.999 (95% CI, 1.501 to 10.655) for herd A and 1.000 (95% CI, 0.104 to 9.614) for herd C.

Discussion

Results of the present study suggested that removal of sows from commercial swine herds because of fertility problems, fecundity problems, or age was not always successful, in that reproductive performance of replacement gilts was not always higher than reproductive performance that could have been expected had the removed sows been allowed to remain in the herd. Further study is warranted to determine under what circumstances performance-based removal and replacement programs have merit.

The present study was designed to evaluate the potential for performance-based removal and replacement actions to improve productivity at the level of the individual animal. Unfortunately, there was not sufficient statistical power to detect differences within parities. In addition, to obtain sufficient numbers of animals in the 2 control populations, the case population was restricted to sows that were parity 1 to 6 at the time of removal. As a result, for 2 of the 3 herds in the study, low numbers of replacement gilts and control sows matched with sows removed because of age were included in the study, and no sows removed because of age were identified in the third herd. Although we attempted to include 1,000 consecutive case sows from each herd in the study, approximately 30 potential case sows were eliminated in each herd because no suitable controls could be identified. Finally, to reduce the impact of herd differences in replacement programs, individual replacement gilts were eligible to be included in the study only if they had been inseminated at least once. From a herd perspective, this criterion favored gilt performance because not all gilts that enter a herd are bred and retained in the herd.

Study herds in the present study were selected on the basis of author familiarity and consistency of removal reasons over time. Multiplier herds were used to ensure consistent health status over the observation period. Herds A and B had both been in production for several years before the start of the study period, but herd C had been begun only 7 months before the analysis period, and the differences in herd age may account for some of the differences in removal and replacement outcome. Although fecundity-associated removal and replacement outcomes were consistent across all 3 herds, the fertility-associated removal and replacement outcome was not advantageous for the youngest herd.

An important limitation of the present study was that recorded removal reasons were not independently validated. Stalder et al^c found discrepancies between recorded removal reasons and reproductive tracts in slaughtered cull sows, and application of a decision-tree method to real-world sow removals found inconsistencies among culling programs.⁸ Similarly, Knauer et al¹⁰ have found inconsistencies between actual and recorded sow removal reasons on commercial swine herd. On the other hand, removal reasons in the present study were reviewed for consistency with the sow's recorded history, and sows with indications of health problems were not included. Although use of removal reasons recorded in PigCHAMP was a potential source of bias in the present study, the PigCHAMP classifications could be considered to reflect the reality of sow removal decisions.

For each herd in the present study, more replacement gilts than control sows generated litters, and this contributed to the higher numbers of PBA/MF/Y among replacement gilts. This was a potential source of bias in the study, in that herd personnel may have been more inclined to retain replacement gilts until a litter was generated than to retain older sows, since the gilts had not yet produced any pigs to offset their initial investment expense. Importantly, although we found significant differences in number of PBA/MF/Y in the present study, the proportion of sows removed from any individual herd for performance-based reasons may not be high enough to impact the herd's overall annual output.

Although calculated odds that removal would be successful consistently favored replacement in the present study, results for calculated reproductive performance (ie, number of PBA/MF/Y) were less consistent. Additionally, the magnitude of the differences in reproductive performance may not be biologically or economically meaningful at the level of the production system. Nevertheless, our findings do suggest that herds may benefit from improvements in assumptions about future sow performance, and further study is needed to identify conditions under which performance-based removal and replacement assumptions fail. Factors of potential interest include contemporary herd productivity, removal parity, and within-class removal reasons.

The present study was not designed to investigate associations between removal reason and parity at removal. Importantly, herd C forced in replacement gilts during the period immediately after beginning operations (ie, the herd employed a mature-herd replacement rate even during the early stages of operation), and this may have affected results for this herd. In addition, herd C used replacement gilts derived from the initial litters of the start-up population. Foxcroft¹¹ has suggested that females derived from first-parity litters may be prone to lower lifetime productivity than their later-litter counterparts.