Understanding and attempting to improve livability of all stages of pork production has generated significant interest and investment of resources in recent years. The National Pork Board and Foundation for Food and Agriculture Research have dedicated resources for a five-year, interdisciplinary, multi-university project aimed at reducing mortality in the U.S. swine industry. As part of this project, a recent literature review was performed which summarizes what is known regarding infectious and non-infectious factors contributing to mortality.
Mortality is extremely complicated. It is the final outcome that can have a near infinite combination of predisposing and contributing factors. In order to reduce mortality in pork production, a baseline understanding of the key contributing factors is necessary, i.e., what is leading to the mortality?
Mortality can be attributed to certain causes such as infectious disease or other cause, but in nearly all cases the "true" cause of death is more complicated than we will ever understand. It is most likely a situation where a series of events came together to result in the final outcome. Human nature often tries to find the most efficient, simplest answer to the problems we face.
Just because we don't (and may never) fully understand the true complexity behind why pigs die doesn't mean we can't generate approaches to reduce the impact on animal welfare and production. Simply recognizing the limitations of our knowledge is a critical step.
Once an area is identified that could be changed to potentially reduce mortality, appropriate investigation is necessary to determine the potential benefit. Such interventions could include changes in management, nutrition, environment and animal health status. Calculations used to determine how many animals are needed for a study to reasonably expect to see statistical differences between groups can be relatively straightforward where the response is assumed to follow a standard bell-shaped curve.
Determining sample size for more complex outcomes such as mortality, which often does not have a bell-shaped curve, is much more complicated. Finding a numerical difference in tests with insufficient sample size can lead you to believe the intervention was successful and result in wasteful spending on false cures. Thus, a fresh look at sample size calculations is warranted to appropriately answer the question — how many pigs do I need to use to be reasonably sure that an intervention is successful or not?
Consider the possibility of using multiple barns within a production system to answer a specific research question. Depending on the type of research question, a number of study designs could be possible. If a production system is trying to evaluate the impact of a nutritional strategy on mortality and that each 1,200-head barn available for use is fed from two separate bins, so each bin could feed half of the barn which is the experimental unit.
As an alternative, if the strategy being evaluated can be applied to an individual pen such as stocking density or feeder adjustment, the pen serves as the experimental unit. Additionally, another potential strategy could involve an individual pig serving as an experimental unit if evaluating interventions such as antimicrobial injection assuming all pigs within a pen are independent of each other.
Results of sample size calculations determining the number of animals and barns necessary to detect significant differences can be seen in Table 1 comparing a base wean-to-finish mortality of 6% to lower levels of mortality within the second group. If we expect mortality in the treatment program to be 5% compared to 6% for our control program, it would require 16 barns if the intervention is applied to half of each barn or 14 barns if the intervention is applied to individual pens within the barn to detect a significant difference with 80% statistical power (a common threshold many fields of science use as a standard).
For larger differences in mortality between the two groups, the number of animals (and experimental units) necessary decreases as expected. Also, for any given difference in mortality between the two groups, the number of barns necessary is lower if treatment is assigned to pen compared to half-barn.
The models discussed present an opportunity for researchers in academia, production and allied industry to have a fresh look at the number of animals necessary to detect statistical significance while incorporating all aspects of study design including treatment and design structure. Conducting research in modern commercial production systems is often necessary to capture the large amount of information necessary to make meaningful decisions, especially for mortality outcomes.
Incorporating recent advances in study design and statistical analysis during the planning stages of an experiment is critical and requires a revisit to an area which can seem intimidating. When conducting studies where mortality is an important outcome, let's revisit how we go about answering the question "how many."
More information can be found in the 2020 Kansas State University Swine Day report, "Practical Application of Sample Size Determination Models for Assessment of Mortality Outcomes in Swine Field Trials" that will be released in November.