Causal Analytics for Applied Risk Analysis Louis Anthony Cox, Jr

Quantifying Pig-Associated MRSA Colonization Potential

Quantitative Estimation of Colonization Potential for Professional Food Handlers in the U.S. from Pork Meat

To quantify uncertainty about the true proportion of ST398 MRSA-colonized food handlers in the U.S., we apply Bayesian conjugate prior analysis. We first assume a standard uniform prior distribution for the true but unknown proportion of colonized food handlers, and then adjust this starting assumption by conditioning on the Dutch data above (showing 0 colonizations among 95 workers). The uniform prior distribution is a deliberately conservative assumption, in that it implies that, in the absence of data, half of all food handlers are expected to be colonized, which is an order of magnitude greater than the empirically observed fractions just reviewed. Thus, we begin with a deliberately exaggerated (risk-over-estimating) assumption, and then use data to revise it downward via Bayes’ Rule. As the most recent and relevant available data source, we use the Netherlands study (Jonge et al., 2010) showing s = 0 positive observations among n = 95 workers. Bayesian conditioning of the uniform prior on these data yields a posterior distribution (a Beta(s + 1, n - s + 1) = Beta(1, 96) posterior distribution) with a mean of (s + 1)/(n + 2) = 0.0103.

To determine a fraction attributable to pork handling alone, we note that pork comprises approximately 55% of total meats consumed in the EU (USCB, 2012). MRSA prevalence on pork in the Netherlands is actually slightly lower than the average for all retail meats. A large-scale Dutch study found MRSA on 264 of 2217 (11.9%) total retail meat samples, 10.7% of retail pork samples, but even higher rates on several other retail meats: turkey (35.3%), chicken (16.0%), veal (15.2%) and also beef (10.6%), lamb and mutton (6.2%), fowl (3.4%) and game (2.2%) (de Boer et al., 2009). Based on these figures, it is reasonable to assume that somewhat less than 55%, perhaps roughly 50%, of MRSA prevalence among Netherlands food handlers could be due to pork. To account for a high degree of uncertainty, we will further assume that the true fraction can be described by a uniform probability distribution with 0.50 as a mean, and [0.30, 0.70] as a plausible range. These endpoints are somewhat arbitrary, but suffice to indicate that the true fraction might differ from the point estimate of 50%.

For the U.S., Table 6.1 summarizes data from recent retail pork sampling studies. ST398 MRSA prevalence in retail pork in the U.S. is about 37.38% (0.04/0.107) of that in the Netherlands. In addition, pork as a fraction of total meat processed in the U.S. is about half (49.11%) (0.2720/0.5538) of that in the EU (USCB, 2012). A base case estimate of ST398 MRSA prevalence from pork among U.S. meat handlers is thus about (0.0103 posterior mean for the fraction of food handlers colonized in Netherlands) x (0.50 fraction from pork) x (0.3738 ratio of prevalence in U.S. pork compared to prevalence in Netherlands pork) x (0.4911 pork-processing/consumption ratio) ≈ 0.0009.

According to the U.S. Bureau of Labor Statistics, in 2010 there were approximately 382,000 persons in the U.S. employed as “Butchers and Other Meat, Poultry, and Fish Processing Workers” (includes Slaughterers and Meat Packers) (USBLS, 2012). Based on the estimates above, we would expect that about 382,000 x 0.0009 = 357, would be colonized with MRSA attributable to handling pork.

Quantitative Estimation of Colonization Potential from Consumer Food Handling

This section applies a series of multiplicative factors to convert the MRSA colonization risk among professional meat handlers to a risk to those consumers who also handle pork, based on relative exposure durations per year. Pork is eaten approximately 98.5 times annually per capita in the United States. Approximately 21% of that pork is from fresh cooked products, with another 31% coming from ham (NPB, 2010), implying an average of approximately 51.22 servings of fresh pork or ham per person per year. An average household size of 2.58 (per 2011 U.S. census data) gives an average of about 51.22 fresh servings per person-year/2.58 person-years per household-year ≈ 19.85 raw pork preparation events per household per year. If each preparation event requires the food handler to be in contact with pork, pork juices, or related working surfaces for up to 15 minutes (a generous estimate), then the average annual contact time per preparer is approximately 4.96 hours. Comparing this to a professional meat handler, who is likely to be in contact with meat, juices, and related working surfaces for approximately 2000 hours/year, approximately 0.2720 of which is related to pork (as previously discussed), the average U.S. consumer preparer has approximately 4.96/(2000 x 0.2720) = 0.0091 as much exposure time to pork as a professional meat handler. Assuming that colonization events (e.g., from a concentration of MRSA on meat) are relatively rare, and independent from exposure to exposure, a Poisson model for colonization is appropriate, and colonization risk increases approximately linearly with exposure time. The implied plausible upper bound colonization risk to the average consumer preparer is then (0.0028 MRSA colonizations from pork per worker year for professional meat handlers) x 0.0091 = 2.55 colonizations per 100,000 consumer preparer-years (2.55 x 10^-5). This does not attempt to correct for any differences in MRSA exposure concentrations or immunity between professional and consumer meat handlers, but assumes that either can be colonized if suitable rare conditions (e.g., high concentrations of MRSA) are encountered on the meat being handled. Applied to the approximate current number of U.S. households, approximately 315M/2.58 gives an estimate of (315,000,000/2.58)*(2.55 x 10^-5) = 3,111 additional annual colonizations (but not infections) of ST398 MRSA due to consumers handling pork. To place this in context, the best estimate of total MRSA colonizations in the U.S. is approximately 2 million (Graham et al., 2006).

Table 6.2 assembles the above distributions and factors into a probability model whose output is a distribution of probabilities for the average annual per person risk of being colonized with MRSA from food handling.
Table 6.2. Probability Model for Colonization Risk from Handling Pork

Component	Distribution	Mean
Dutch food handler prevalence	Beta(1,96)	0.0103
Pork attributable fraction in Netherlands	Uniform(0.30, 0.70)	0.50
U.S. to Netherlands pork prevalence ratio	Constant (ratio of survey study averages)	0.3738
U.S. to Netherlands pork processing ratio	Constant (ratio of historical processing fractions)	0.4911
Average U.S. Meat Handler Risk/yr/person	Product of Above (apply to 382,000 workers)	9.45 E-4
Consumer to food handler time of exposure ratio	Constant (U.S. pork preparation rates)	0.0091
Average U.S. Consumer risk/yr/person	Meat Handler Risk*.0091 (apply to 315M/2.58 households)	8.6 E-6

The corresponding model equations are:

Food Handling Colonizations = (Dutch food handler prevalence x pork attributable fraction x U.S. prevalence ratio x U.S. processing ratio) x (U.S. food workers + Consumer exposure time ratio x U.S. population) (6.1)