Causal Analytics for Applied Risk Analysis Louis Anthony Cox, Jr

Background on Toxoplasmosis

The causative agent of Toxoplasma infection, Toxoplasma gondii, is a common parasite in most warm-blooded animals, including people, worldwide. It is believed to be transmitted primarily via eating contaminated raw and undercooked meat (especially pork and poultry in the U.S.), as well as through contact with contaminated cat feces. More recently, drinking water and other potential environmental sources have received increased attention (Tenter, et al., 2000). However, since as many as 2.8 million Toxoplasma-infected pigs enter the U.S. food chain each year (Hill and Dubey, 2013), pork still must be considered a major risk factor. Approximately 9% of the US population (ages 12-49) are infected with the parasite, and much higher prevalence can be found in other parts of the world (Jones et al., 2007). Although risk of parasite-mediated diseases such as toxoplasmosis and trichinosis can be kept at or near zero in modern intensive swine production systems, open systems are less controllable, and increased risk of toxoplasmosis is part of the price of these alternative production systems (Davies, 2011).

Most infected adult humans suffer few or no detectable ill effects from toxoplasmosis. However, infection can be deadly for immunocompromised patients and devastating for the unborn children of pregnant women. In the mid 1990s, it was estimated that about 1 in 10 AIDS patients in the US died from toxoplasmosis (Hays, 1996). For a woman infected during pregnancy, there is a 20% to 50% chance of the baby being born with the infection and having increased risk of blindness, mental retardation, or other medical problems. In 1994, it was estimated that about 0.8 babies in 10,000 in the US were born with the infection (Guerina et al., 2004). Of all creatures infected with Toxoplasma gondii, only cats shed T. gondii oocysts, which can withstand the external environment and spread the disease. If infected cat feces contaminate the feed of pigs, humans can subsequently become infected by eating or handling raw or undercooked pork containing the parasite.

Enhanced bio-security during pork production greatly reduces the incidence of infectious diseases. There has been a steady increase in use of bio-security measures to protect human and animal health, including disinfection, all-in/all-out livestock rotation, visitor control, rodent control, and housing control (using confined spaces that keep out all dogs, cats, birds, etc) (USDA-APHIS, 2008). Practices used in alternative pork production systems prescribe continuous outdoor access for pigs and preclude the use of confinement facilities. Although some bio-security measures are still possible, such as herd isolation, visitor control, and quick removal of diseased animals, there are also more opportunities for swine to come in contact with sources of T. gondii infection.

Data and Methods

To compare the T. gondii-related health outcomes for consumers of pork derived from pigs in total confinement systems versus pork derived from pigs in open/free range systems, we modeled the influential input parameters as probability distributions, and then linked these distributions in a probabilistic simulation model that relates health outcomes (more infections, birth defects, and deaths -expressed as Quality Adjusted Life Years) to inputs. Figure 7.1 outlines the main structure of the resulting simulation model. The fraction of pigs potentially shifted from total confinement to open/free range production systems is an exogenously specified input to the model. Varying this fraction and simulating the resulting changes allows the probable trade-off between greater animal welfare (higher fraction of pigs in open facilities) and greater human health risk to be estimated quantitatively. To account for uncertainty in the input parameters, each individual simulation run performs a random draw from all of the parameter distributions and generates the resulting linked outcomes. Performing a large number of independent simulation runs builds up distributions for the health outcomes of interest. This approach automates the process of sensitivity analysis by simultaneously combining virtually all possible combinations of inputs. The data needed to form each parameter distribution is drawn from the available scientific literature on T. gondii prevalence in pigs and pork, human infection rates, and human health effects of infections, as described in the sections below. These parameters are also summarized in Table 7.1.

Distributions for T. gondii Prevalence in Pigs

Literature and Data Review

In a nationwide survey performed in 1983-1984, 23% of all market pigs tested positive for T. gondii antibodies, indicating past (or present) infection (Dubey et al., 1991). A decade later, a 1995 USDA study involving pigs from 17 major pork producing states (Patton et al., 1996) found that 3.2% of market pigs tested positive, while only .9% tested positive in a similar 2000 study (Patton et al., 2002). The 2006 version of the USDA study (Hill et al., 2010) found a mean within-herd prevalence of T. gondii antibodies of 2.7%. Thus, a sharp decline in prevalence of T. gondii antibody prevalence has coincided with increased use of modern farming practices that emphasize biosecurity, such as confinement rearing, rodent control, hygienic feed handling, and exclusion of cats (USDA-APHIS, 2008; Davies, 2011).

The prevalence values above reflect nationwide averages over a variety of pork production practices and housing types. According to the USDA, in 2006, 73.6% of hog production sites in the U.S. used either total confinement (53.2%) or open buildings with no outside access (20.4%) (USDA-APHIS, 2008). The rest were classified as “open building with outside access” (23.3%), lot with hut or no building (1.8%), or pasture with no hut or no building (1.3%). Not surprisingly, studies show that the prevalence of T. gondii in pigs is much lower for pigs isolated from potential infection sources. In a 1995 study, samples from pigs in total confinement production systems in North Carolina were found to be positive for T. gondii in only 1 out of 1,752 (0.06%) cases (Davies et al., 1998). A 2006 USDA swine survey found that, among pigs in total confinement facilities, 0.3% tested positive, while in other facilities (open buildings with or without outside access, hut, lot, or pasture), the rate was more than twenty-fold greater, at 6.5% (USDA-APHIS-VS-CEAH, 2011). A 2008 study of over 74,000 market pigs in the United States detected T. gondii in 0.8% of samples overall, ranging from 0.5% in the largest systems to 2.6% in the smallest (less than 1000) (McKean et al., 2009). Larger systems are more likely to be confinement production systems. At present, it appears that pigs raised in total confinement can be T. gondii free, provided that there is adequate rodent control and prevention of contamination of food and water with oocysts (Dubey, 2009).

Pigs that are raised using so called natural, open, organic, or free range production practices have, by definition, greater exposure to the open environment and potential sources of T. gondii. Gebreyes et al. (2008) compared serum samples taken from swine reared in conventional intensive indoor production systems and outdoor anti-microbial free (ABF) production systems. The study analyzed 616 samples from three different states in the US: Wisconsin, Ohio, and North Carolina. The percentage of samples testing positive for T. gondii was 1.1% (3/292) in swine from conventional systems versus 6.8% (22/324) in swine from ABF systems. Pigs raised on two organic farms in Michigan, (Dubey et al., 2012) were found to have antibodies to T. gondii in 90.1% (30/33) of cases . Dubey et al. (2002) found a T. gondii seroprevalence rate of 76% (19/25) in randomly selected free-range pigs on a farm in Massachusetts previously known to have T. gondii infections, and from 70.1% (34/48) of free range pigs on a poorly run farm in Maryland (Dubey et al., 2008). It is also worth noting that a cross-sectional study involving 3247 feral pigs in 32 states found an average T. gondii seroprevalence of 17.7% (Hill, et al., forthcoming). Feral pigs could be considered the ultimate in free-range swine, however it is plausible that they would not have as much contact with T. gondii carriers such as rodents and cats as would open production farm pigs.

Research in the Netherlands also shows wide variation in swine seroprevalence, depending on the housing type. Kijlstra et al. (2004) compared T. gondii prevalence among swine from free range, organic, and intensive/conventional. The rates were 4.7% (30/635), 1.2% (8/660), and 0% (0/621) respectively. A few years later, van der Giessen et al. (2007) found rates of 5.62% (10/178) for free range, 2.74% (11/402) for organic, and 0.38% (1/265) for intensive/conventional.

Derived Distributions for Average T. gondii Prevalence in Pigs

Using relevant data from the studies cited above, we estimated the average T. gondii prevalence in two contrasting categories of pigs: 1) a randomly sampled subset of pigs raised in total confinement production systems, versus 2) a randomly sampled subset, of identical size, of pigs raised in open systems with outside access such as those used in organic, natural, or free-range production. As noted above, various midway housing solutions exist, such as open buildings with no outside access, however, these are not directly relevant to our framework. To account for high degrees of estimation uncertainty, we developed beta probability distributions describing the average prevalence within each of these two groups of pigs. Beta distributions can be used to estimate continuous values that lie between 0 and 1, such as proportions and percentages. They are often used for developing distributions from limited empirical data since a minimum, maximum, and mean value are sufficient to characterize the distribution. This is especially relevant in a meta-analysis, where data and results from different studies cannot always be explicitly combined, and expert judgment may be used to inform some parameters. When appropriately parameterized, they also provide the useful property of a long “right hand tail”, which is suitable for situations where the main weight of evidence is for relatively low values, yet higher possible values must be considered, but at lower weight. The beta distribution parameters, alpha and beta, were computed from estimated values for the minimum, maximum, mean, and variance of average prevalence by using the equations described in Davis (2008). The variance was assumed to follow the standard (beta) PERT form, . The minimum, maximum, and mean value parameters were estimated as described below and are summarized in Table 2.

Distribution for Closed/Total Confinement Systems

A plausible upper bound on the total confinement subset average prevalence is the value of 2.7% reported in the 2006 USDA survey as the average over all facility types. We reason as follows: since this average mixes in values from pigs in a variety of open facility types, where average prevalence is known to be higher, it is very likely to be an upper bound on the average for total confinement facilities alone. For a lower bound on the average, we use zero, reflecting the fact that zero or near zero T. gondii prevalence has been reported in a number of cases (Davies, 2011; Kijlstra et al., 2004; van der Giessen et al., 2007) , together with the fact that elimination of this risk appears to be realistic for current total confinement systems (Davies, 2011; Dubey, 2009). For the mean of the average total confinement subset prevalence, we used the average value for total confinement facilities of 0.3% reported in the 2006 USDA survey. We did not combine this with the value of 0.06% (1/1752) reported in Davies (2011) since that study was based on swine in North Carolina alone. The McKean et al (2009) study also cannot be used explicitly since it differentiated by herd size rather than confinement type (although the two are correlated). However, the size, geographical diversity, and relative recentness of the USDA study support the validity of these estimates.

Distribution for Open/Free Range Systems

In the case of open/free range facility types, a plausible lower bound on the average subset prevalence is the same 2.7% average 2006 survey value used as the upper bound for total confinement facilities. Using reasoning similar to above, since the overall overage is derived from a sample of which over half were pigs in total confinement facilities, the 2.7% average would likely form a lower bound on the average prevalence of pigs drawn from open/free range situations alone. To estimate the mean of the average for open/free range systems, we combined sample data from the available U.S. studies described above that involved outdoor ABF (Gebreyes et al., 2008), free range (Dubey et al., 2002; Dubey et al., 2008), or organic production (Dubey et al., 2012). The combined data from these studies provides an estimated mean seroprevalence rate of (22+30+19+34) / (324+33+25+48) = 105/430 = .244. This estimate seems quite plausible since it is very close to the nationwide average prevalence reported for 1983-84 of 23% (Dubey et al., 1991), which is before the major shift to production methods that emphasize confinement and biosecurity. As an upper bound for the average subset prevalence for open/free range facilities, we used the highest observed fraction of 90.1%

Figure 7.1 shows the beta distributions implied by the base case parameter values in Table 7.2. These represent the frequency distributions of average T. gondii antibody prevalence rates for subsets of pigs in each category, total confinement and open/free range, respectively.