Sow mortality has become a growing concern in the pig production industry over the past decade. Therefore, we aimed to describe sow mortality and associated factors in a production system in the midwestern USA.
Mortality records from 2009 to 2018 for four farrow-to-wean farms were described. Environmental, farm- and individual-level factors associated with weekly mortality and individual risk of dying throughout a sow’s lifetime were assessed.
Deaths occurred at a median of 116 days from last service, or 26 days postpartum. The median parity upon death was two. Overall, the main reasons for death were locomotion (27%) and reproduction (24%). A higher weekly number of deaths was associated with spring (incidence rate ratio [IRR] 1.27, compared to winter). Sows had a higher mortality when they were exposed to at least one porcine reproductive and respiratory syndrome (PRRS) outbreak during their lifetime (IRR 1.55) and when housed in groups (pens) during gestation (IRR 1.32). Conversely, they had a lower mortality when housed in filtered farms (IRR 0.76), accounting for an interaction term between parity at removal and PRRS outbreak exposure.
Issues with data completion and information accuracy were present, and prospective data collection throughout sows’ lifetimes is still needed.
Efforts to reduce infectious diseases within the herd and manage environmental stressors should help reduce mortality.
Sow mortality has been a growing concern in the pig production industry over the past decade, from both an animal wellbeing and economic standpoint. From an economic standpoint, costs associated with sow mortality can be direct, such as the cost of replacing the dead animals, or indirect, such as opportunity costs (i.e., costs associated with the production deficit resulting from those deaths and the lack of salvage value associated with dead sows).1 It is estimated that about three to five parities are necessary to achieve an investment return for the producer.2, 3 Still, the annual culling rate ranges from 15% to 93%, with most removals occurring for parity 0 or one females,4–6 that is, before a positive net present value is achieved.
Regarding removals, particularly those registered as a death, different factors have been shown to contribute to sow mortality. Studies have shown that the majority of deaths occur in the week before farrowing and during the first 3 weeks after,1, 7 suggesting that the peripartum period represents higher risk for sow mortality. Mortality tends to increase during summer months,1, 8, 9 indicating a possible seasonality effect likely associated with heat stress. Parity may also play a role, although studies have shown gilts both at lower and higher risk of death, likely explained by differences in culling practices and available gilt pool.1, 7, 10, 11 Factors associated with management and herd health, such as exposure to infectious diseases and herd size, may also influence mortality.11–13
Most studies assess mortality within a limited time frame of a couple of years at the most. Thus, the main goal of this study was to describe sow mortality over the course of 9 years and investigate factors possibly associated with high counts of sow deaths and individual sows’ risk of dying.
MATERIALS AND METHODS
Historical production data recorded on PigCHAMP (Ames, IA, USA) from April 2009 to October 2018 (totalling 470 weeks) were obtained for four commercial farrow-to-wean farms from one production system that is representative of current production practices and is located in the midwestern United States. Individual sow records and service records of all sows in the herd during the study period were available. Data considering only females that were serviced at least once (comprising bred gilts and sows) were used and will henceforth be referred to generally as sows. During the study period, the average herd size was 3700 for farm 1, 2437 for farm 2, 2505 for farm 3 and 5442 for farm 4. Removal, as recorded through PigCHAMP, refers to females that were removed either due to death or a cull/sale event. Soft data, such as removal reasons for animals that died, were classified into disease/health, intestinal, locomotion, performance, reproduction and others (i.e., accident, unknown, behaviour) according to Ketchem et al.,14 but were not further examined. Briefly, disease/health reasons include discharge, downer, disease, mastitis/metritis; reproduction reasons include no heat, not in pig, abortion, repeat service, vaginal/uterine prolapse, rectal/anal prolapse, difficult farrowing; performance reasons include age, low total born, retained pig, low weaned, poor milker; intestinal reasons include ulcer, off feed; and locomotion reasons include unsound and lame/injured. Deaths and culls were compared based on their lifetime contribution time (i.e., the sum of the female life days between the first service to the removal date) and parity at removal.
We included the porcine reproductive and respiratory syndrome (PRRS) weekly status of the farms. Such status was obtained through the Morrison Swine Health Monitoring Program. Briefly, this programme was established in 2010 and is a US producer-driven initiative, coordinated by the University of Minnesota, aimed at monitoring endemic swine diseases such as PRRS, where participants voluntarily share prospective and historical weekly sow herd PRRS infection status. Status reporting follows the American Association of Swine Veterinarians PRRS status terminology,15 in which a herd is classified as positive unstable from the first viral detection to sustained weaning of negative pigs. Because the risk period for sow mortality once the virus is introduced into a herd might be shorter than the positive unstable period, we considered PRRS outbreak weeks as the weeks between first viral detection up to when all production parameters reached baseline for the first time or for up to 12 weeks, whichever was longer. We did not consider PRRS outbreaks as those periods of time in which the herd was classified as positive unstable and lasted less than 8 weeks since PRRS control and elimination typically requires longer periods of time. Similarly, those that were recorded less than 12 weeks apart from the end of the previous unstable period (i.e., which were considered to have been misreported as stable) were not counted as new outbreaks. Baseline production parameters were defined using the exponentially weighted moving average control chart method using the average of the 21 weeks prior to PRRS detection.16 Baseline was calculated individually for total weaned/litter, total liveborn/litter, total mummies/litter, total stillborn/litter and total born/litter. For the purpose of this study, all these production parameters had to have reached baseline production for a PRRS outbreak to be considered over. Since we used the first 21 weeks of data as the baseline for the production parameters in order to assign PRRS outbreak weeks, data from this period were removed from the risk factor analysis. The baseline was recalculated every time a new PRRS outbreak was recorded, considering the 21 weeks prior to its detection. Thus, for this purpose, the studied period comprised 470 weeks, that is, between 19 October 2009 and 30 December 2018.
Daily temperature data from the nearest airport were obtained from the National Oceanic and Atmospheric Administration website. Daily data were summarised to generate weekly highest daytime temperature (i.e., highest recorded temperature in each week between 6 a.m. and 6 p.m.), weekly lowest nighttime temperature (i.e., lowest recorded temperature in each week between 6 p.m. and 6 a.m.) and weekly maximum minimum temperature (i.e., the highest of the daily low temperature during each week). Seasons were defined based on vernal and autumnal equinox and summer and winter solstice dates. In-feed medication records were available from 2014 to 2018 and consisted of dates in which medication was included in either the gestation or lactation diet. Labour data were recorded as total regular and overtime hours per employee and were available for the period between 2017 and 2018. Total hours of work and total employees were assigned to the subsequent week since data were recorded biweekly. Average sow/hours/employee was calculated using (tsow/temployees)/(thours/temployees), where ‘tsow’ is the total number of sows in the herd that week, ‘temployees’ is the total number of employees and ‘thours’ is the cumulative hours worked for all employees during that week. Farm housing type (stalls vs. pens) history and if and when air filtration was adopted in each farm were also obtained. Information on herd composition in regards to percentage of gilts, percent gestating and percent in peripartum (defined by –4 to +28 days from farrow) was summarised for each week.
Deaths were described by removal reasons, by month, days from last service, days from last farrow and parity at removal. Weekly mortality was described as the percentage of animals within the herd that died within a given week. Similarly, this metric was converted to annualised mortality by considering weekly absolute mortality fixed throughout the year with the formula (tdied × 52.1429)/tsow × 100, where ‘tdied’ is the total number of sows that died that week, ‘tsow’ is the total number of sows in the herd that week and 52.1429 is the number of weeks in a year considering leap years. Risk factors associated with mortality were assessed using two different models. First, we used a Poisson model specifying week as a panel variable to model the number of deaths per week by each possible associated factor using farm as a cluster variance estimator and the total number of sows in inventory during that week as the exposure. Second, factors associated with the sow’s individual risk of dying throughout the lifetime of the 70,467 animals studied were assessed through a multilevel Poisson regression in which the contribution time of each animal in sow-years was defined as the exposure and farm was accounted as a random effect. Associations are displayed as incidence rate ratio (IRR) or the relative difference in incidence rate of deaths between groups. For both these models, variables associated with the outcome at p < 0.10 were selected for the multivariable analysis, where a cutoff of p < 0.05 was defined for the final model using a backwards selection method. However, for the individual analysis, because older animals are more likely to have been previously exposed to PRRS at some point in their lifetime, an interaction term between parity and PRRS exposure was included in the multivariable model.
We obtained a total of 357,425 service records from 85,608 sows. Of these, 70,467 sows were removed during the study period as 11,852 and 58,615 sows died or were culled, respectively. The median number of days between the first service and removal was 388 (interquartile range [IQR]: 148–619.5) for females that died and 601 days (IQR: 293–881) for the ones that were culled. Similarly, the median parity at removal was 2 (IQR: 0–4) and 4 (IQR: 2–6) for deaths and culls, respectively. Out of the 11,852 deaths, 26.9% (n = 3190) were due to locomotion reasons, 24.0% (n = 2846) due to reproduction, 5.6% (n = 667) due to disease/health, 3.8% (n = 455) due to performance and 3.7% (n = 434) due to intestinal disorder reasons. However, 35.9% (4260) of the removals were due to other reasons, over half of which were unknown or missing. Listed reasons included, but were not limited to, accident-trauma, heat stress and inventory adjustments. Figure 1 shows the relative frequency of removal reason categories per year throughout the studied period for sows that died. Although disease/health reasons fluctuated throughout the years, reproduction reasons seemed to have had a relative decrease in recent years compared to the beginning of the study period, whereas locomotion reasons increased.
Deaths occurred at a median of 116 days (IQR: 94–125) from last service or 26 days (IQR: 5–103) postpartum (Figure S1). The distribution of deaths had two peaks, with the highest peak at 1–10 days postpartum followed by a much smaller peak at 131–140 days from the last farrow. The median parity upon death was 2 (IQR: 1–4), with 12.5% of all deaths recorded comprised of sows that never farrowed (i.e., bred gilts) and a periparturient risk peaking after the first farrow, with over 20% of the deaths comprising sows that farrowed only once. Among deaths within the bred gilt category, death occurred within a median of 102 days after their first service (IQR: 60–114).
The average annualised mortality per month represents the average percent mortality if all deaths that occurred in that month were constant throughout the course of 1 year. Mortality ranged from 1.79% to 3.29% for all farms combined (Figure 2a). A higher mortality peak was observed in July (summer), followed by a smaller peak during April (spring). The spring peak was explained by a higher mortality in 2015 that was not observed in other years (Figure 2b). Similarly, the magnitude of the summer peak in mortality was not consistently high throughout the years, with some years presenting no peak at all during the summer months. This was also evidenced when looking at mortality at the farm level (Figure S2).
The median weekly mortality was 0.18% (minimum 0.00%, maximum 4.48%) for farm 1, 0.16% (minimum 0.00%, maximum 0.98%) for farm 2, 0.16% (minimum 0.00%, maximum 1.92%) for farm 3 and 0.13% (minimum 0.00%, maximum 0.44%) for farm 4. This would be equivalent to a median of 0.27%, 0.35%, 0.33% and 0.27% annualised mortality for farms 1, 2, 3 and 4, respectively (Figure 3). We found that a higher weekly death count was associated with spring and summer weeks with incidence rates 1.27 and 1.37 times higher than those in winter weeks, respectively (Table 1). We also found that weeks in which medication was included in the feed had 1.38 times the weekly mortality of those weeks when feed medication was not present. However, although feed medication improved the model by reducing the Akaike information criterion (AIC) from 4748.329 to 4655.541 when considering the period in which these data were available, because feed medication data covered only 2014–2018, an additional multivariable model without feed medication was constructed for the entire period. In this model, only season was associated with a higher number of weekly deaths (AIC 11,674.87). These results are represented by the univariable analysis in Table 1.
|Environmental weekly variables
|Maximum minimum temperature >10°C
|Daytime high temperature >25°C
|Night time low temperature >0°C
|Gilts comprising >20% of the herd
|Having ≥80% of the inventory in pig
|≥8% of the inventory in peripartum
|Any feed medication
|Any feed medication in the previous week
|Any feed medication during lactation
|Any feed medication during gestation
- Abbreviations: CI, confidence interval; IRR, incidence rate ratio; PRRS, porcine reproductive and respiratory syndrome.
Over the 10 years of study, a total of 111,522 sow-years were followed and the overall mortality was 10.69 per 100 sow-years when adjusting for farm (Table 2). Mortality decreased with increasing parity at removal. In the multivariable analysis model, animals with parities of one or less had 44 times the annual risk of death than those with parities of seven or more. We identified that animals exposed to a PRRS outbreak had 1.5 times the risk of dying than those unexposed. This association only appeared after the inclusion of an interaction term with parity. The mortality when animals were housed in groups (i.e., pens) during gestation was 1.3 times that of those housed in individual stalls. This association did not change drastically in the univariate or multivariate models. Filtered farms had a mortality 1.5 times greater than unfiltered farms in the univariable model; however, in the multivariable model, the mortality on filtered farms was 0.76 of that observed in unfiltered ones. Because only one farm transitioned to group housing during gestation, sow mortality on this farm before and after its implementation is shown in Table S1. Briefly, the mortality in sows that were removed before the change to pen gestation was 15.34 per 100 sow-years (95% confidence interval [CI] 14.28–16.48), whereas the mortality for sows that were first serviced only after the change to pen gestation was 16.35 per 100 sow-years (95% CI 15.78–16.94). Overall, associations remained very similar to what was reported in Table 2. However, filtration was not assessed since it was implemented close to the change to pen gestation.
|Mortality rate (deaths/100 sow-years)
|Parity at removal
|Animal ever exposed to a PRRS outbreak
|Pen gestation since first service
|Farm filtered since first service
|Season at first service
- Note: Multivariate model includes interaction term between parity at removal and porcine reproductive and respiratory syndrome (PRRS) outbreak exposure.
- Abbreviations: CI, confidence interval; IRR, incidence rate ratio.
Studies that have characterised sow mortality throughout the past couple of decades found a higher mortality during warmer months.1, 8 In our study, the majority of deaths occurred during summer months, and there was a higher probability of a high mortality during warmer weeks, particularly in spring. Discrepancies in the effect of warmer months on mortality between systems and locations have been reported1 and were also evidenced here between farms and between years. A higher mortality during warmer periods is likely related to factors aggravating thermal stress,9 in which case management practices such as higher ventilation rates and cooling can potentially translate to lower mortality. Those factors could explain why mortality is so high in some summers but not in others. Additionally, duration and intensity of exposure to extreme temperatures increased mortality risk in female cattle17 and might also play a role in sow mortality. However, to truly assess that association, further investigations on the actual barn temperatures and humidity and interventions in relation to monthly mortality are needed, information that unfortunately is not routinely recorded on most farms.
When describing deaths, we found that most sows that died were lower parity animals. Over 30% of deaths in this study occurred among females that never farrowed or farrowed only once. The authors’ experience suggests that the losses of younger females are common and are a factor that hinders cost competitiveness of sow farm operations, as female break-even cost is typically at or after their third litter. However, we found a bimodal distribution of deaths in regards to days from last farrow. This likely represents deaths that occurred during the prepartum phase of potentially unrecorded services or nonproductive sows that were unnecessarily kept in the herd. Alternatively, this could be happening due to a late entry of mortality in PigChamp (i.e., farm workers noticing the missing sow only later when it should be farrowing). However, the weekly percentage of gilts and the weekly percentage of animals in peripartum in the herd were not associated with higher mortality weeks, likely because these parameters tend to remain relatively constant over time in a herd unless an acute event happens.
When looking at the individual risk of sow death, a higher mortality amonglower parity sows was also found. This potentially captures a mixture of causes of death that are age dependent as younger animals are more susceptible to diseases, while older animals that might have survived previous exposures become immune and less susceptible to reinfections, and performance/reproduction reasons such as increased stillbirths and lactation issues become more relevant (Figure S3). We observed that while death due to disease/health comprised 8.3% of all removal reasons in sows with parities of one or less, they comprised 2.6% of removal reasons for sows with parities of seven or more (Fisher’s exact, p = 0.19). However, deaths due to either performance or reproductive reasons comprised 23.7% in sows with parities of one or less and 36.9% in sows with parities of seven or more (chi-square, p < 0.001). However, soft data, such as recorded removal reason, should be interpreted with caution since it is heavily susceptible to subjectivity and data quality may be subpar. An interaction term between parity and PRRS previous exposure was added to our models to partially capture this effect; however, not all diseases or health issues included in this removal category would be affected by a sow’s parity. It is important to note that these data represent only animals that were recorded as deaths in the field and exclude animals that were actively removed through culling. This may help to explain why we identified a lower risk of death in older (parities of seven or more) animals, since important causes of removal of animals from a herd (e.g., locomotion and performance reasons) tend to be chronic in nature and thus allow for the removal of an animal before its death on the farm.
Weeks in which in-feed medication was implemented were associated with higher weekly death count than when feed medication was absent. Here, medication likely acts as a proxy for the herd’s overall health, meaning that medication is being used as therapeutic treatment rather than prophylaxis. Information on individual treatment courses given for a disease or health issue is not systematically recorded but could potentially be useful to the broader understanding of sow mortality and the health stressors that sows undergo.
Mortality was higher among sows that were exposed to at least one PRRS outbreak in their lifetime compared to sows that had never been exposed to a PRRS outbreak when controlling for parity. This can be rationalised as the mortality risk increasing in the face of health challenges such as this systemic infectious disease. Parity may serve as a proxy for how many infections an animal may have faced (older animals were likely exposed to more pathogens than younger ones), and the effect modification identified after the introduction of an interaction term between parity and PRRS exposure suggests that animals that were exposed to a PRRS outbreak and died were likely the younger ones. Similarly, we also found an effect modification in that sows housed in filtered farms had lower mortality than those in non-filtered farms in the multivariate model, likely because these and other infectious health challenges are reduced with filtration and that PRRS exposure is partially capturing this effect. We also found that sows housed in groups had higher mortality than those housed in individual stalls. This finding can be a consequence of behavioural changes that might occur because of the hierarchical social structure these animals need to establish when sharing an environment, such as fighting for resources such as feed and water, which would lead to injuries. These acute and chronic social stresses caused by group housing are complex and difficult to measure, but some effects on reproductive success have been reported.18 Additionally, within-herd transmission of infectious diseases might be favoured when contact between animals is increased. However, it is important to note that although welfare concerns play a role in addressing mortality, other factors, such as environmental enrichment, public perspective and local legislation, must be considered in addressing both mortality and welfare. Due to the nature of some of the most common reasons for removals due to death, such as locomotion and reproductive reasons, it is reasonable to assume that the mortality data comprised a mix of natural death and euthanasia, although no distinction was available. Regional and national welfare legislations on both type of housing and requirements for on-site euthanasia differ in other countries, in which case representability of some of our findings might be limited.
Some limitations of the study include that among the four farms, only one transitioned to group housing of animals (farm 1, in 2013), and all transitioned from non-filtered to filtered within the same year (2012). This raises the possibility that some of the differences in risk may have occurred due to long-term changes in PRRS occurrence rather than due to these variables themselves. However, the weekly mortality was treated as panel data to partially capture the temporality dependency, while mortality throughout the study period did not appear to have any overall increasing or decreasing trend (Figure 3). Additionally, the four farms from one production system that were included in this study had relatively low mortality, ranging between 1.79% and 3.29% annually, when contrasting to mortality reported elsewhere. From a US perspective, sow mortality in 2007 was estimated at 8.76%, ranging from 1.4% to 22.7% depending on the herd,19 while the median mortality in 2017 was estimated at 10.15%.20 Globally, sow mortality for combined herds from the USA, Canada, Australia and the Philippines was estimated at 13.56% in 2021, but ranged between 7.32% and 11.78% from 2012 to 2018.21 Although mortality trends might vary between different farms and systems, in our study, mortality seems to be reasonably stable at around 0.12%–0.35%, depending on the farm, with the few acute events of high spikes in mortality distributed throughout the studied period. This highlights that longer monitoring periods might be important to contextualise short-term increases in mortality. Furthermore, factors associated with large spikes in mortality are likely different from factors associated with baseline mortality, and these scenarios might need to be investigated separately. Similarly, previous studies have described an increasing trend in the percentage of deaths due to pelvic organ prolapse since 2015,22, 23 but when put into a broader context, its occurrence is similar to what was observed in 2009 and 2010 in this dataset (Figure S4).
Issues with data completion and information accuracy were certainly present in this dataset and should be considered when interpreting the results. This is a common issue with retrospective studies using production records and can affect both soft and hard data analysis. The decision regarding cause of death involves several aspects, such as the recorder’s training, knowledge and availability of supporting data for such diagnosis. Thus, these limitations must be considered when describing sow mortality. Relying solely on death records hinders the assessment of exposure to diseases as a risk factor for mortality as no reliable information on disease occurrence among the animals that survived is available. Efforts to monitor a myriad of diseases over time, as has been done for PRRS, would allow a more comprehensive understanding of the relative contribution of each disease to the sow mortality burden. Likewise, environmental and other contributing factors presumed to affect mortality risk, such as barn temperature and nutrition, need to be more accurately recorded to truly assess their association. Prospective data collection targeting these factors is needed for a more in-depth analysis of factors involved in sow mortality in order to generate interventions and thus reduce sow mortality.
Data analysis was performed by Mariana Kikuti, who wrote the first draft of the paper. Cesar A. Corzo and Juan Carlos Pinilla contributed to conception of the study. Mariana Kikuti, Guilherme Milanez Preis, Juan Carlos Pinilla, John Deen and Cesar A. Corzo reviewed and approved the final version of the manuscript.
The authors would like to acknowledge the Morrison Swine Health Monitoring Program (MSHMP) for sharing their health and production data. The MSHMP is a Swine Health Information Center funded project (SHIC, www.swinehealth.org, project #20-172 SHIC). This project was funded by the Pig Improvement Company North America.
CONFLICTS OF INTEREST
Juan Carlos Pinilla is employed by the Pig Improvement Company North America. The remaining authors declare they have no conflicts of interest.
No ethical approval was required as this article describes data collected routinely as part of the standard care farms offer to their animals.