5Surveillance of the Food Supply

Publication Details


The search for the organisms that cause foodborne disease—both before and after people consume them—involves a constellation of overlapping information networks, none of which constitutes a complete surveillance system. As John Besser of the Minnesota Department of Health notes in the first paper of this chapter, the scores of existing food surveillance programs can be reduced to two basic types: food monitoring (the direct detection of microbial pathogens along the food chain) and disease surveillance (the collection of human or animal disease data, followed by analyses of case clusters and disease trends). He describes and compares various strategies and methods of both food monitoring and foodborne disease surveillance, noting their strengths and limitations as currently practiced.

Although food monitoring is theoretically capable of providing primary prevention against foodborne disease, Besser finds that such systems seldom meet this standard because of the technical complexity of the task (which can only be addressed at considerable expense). However, another workshop participant, Robert Buchanan of the Food and Drug Administration (FDA) Center for Food Safety and Applied Nutrition (CFSAN), observed that food monitoring plays an important role in verifying the effectiveness of food safety systems (see Summary and Assessment, p. 15). Although Buchanan acknowledged the many technical challenges associated with the detection of disease-causing microbes in food—from the collection of samples to the detection of tiny populations of bacteria in food that are nevertheless capable of producing severe disease if ingested—he also provided several examples of recent technological advances that have enabled researchers to identify food contaminants in “real time” (a relative term in the food industry, measured by the length of time the monitoring agency has to act upon detecting contamination), as well as targeted sampling techniques that can boost the efficiency of food safety systems.

Besser, by contrast, concludes that while microbial monitoring has its place, particularly in high-risk situations (e.g., botulinum toxin in milk; see Chapter 4), a better return on investment is likely to be achieved through increased funding of foodborne disease surveillance programs. Notable among these is PulseNet, which tracks diseases through standardized pulsed-field gel electrophoresis (PFGE) protocols (see Tauxe in Chapter 3 for descriptions of PulseNet and other foodborne disease surveillance programs). Another source of information on foodborne disease is captured by the U.S. Department of Agriculture (USDA) Consumer Complaint Monitoring System (CCMS), and is described in the second contribution to this chapter by Kimberly Elenberg and Artur Dubrawski. CCMS uses an algorithm-based tool that organizes and analyzes data concerning food-associated symptoms or foreign objects present in food in real time in order to provide early warning of a foodborne threat. The database is used to record, evaluate, and track all consumer food complaints involving meat, poultry, and egg products. Based on its proven ability to detect low-amplitude signals amid noisy data, CCMS appears poised to join the “network of networks” that constitute the U.S. foodborne disease surveillance system.


John M. Besser, M.S.1

Minnesota Department of Health

Increasing concern about threats to our food supply caused by microbial contamination, either intentional or accidental, has resulted in the establishment of many local, national, and global networks to address the problem, each with its own function and acronym (GAO, 2003, 2004), as shown in Table 5-1. Although none of these networks by itself constitutes a complete surveillance system, each serves some part of a broader food surveillance effort. Such surveillance efforts fall into two main categories: (a) those involving direct detection of microbial pathogens in food ingredients, products, or production environments (referred to as food monitoring); and (b) those involving the collection of human or animal foodborne disease data to identify problems in the food supply through analyses that detect clusters of cases and disease trends (referred to as disease surveillance). Within each of these two broad categories, various approaches are used, each with strengths and limitations.

TABLE 5-1. Networks and Resources for Food Safety.


Networks and Resources for Food Safety.

The ultimate goal of all food safety programs is to prevent contaminated products from reaching the consumer. Since the 1950s, Hazard Analysis Critical Control Point (HACCP) programs have attempted to achieve this by identifying, monitoring, and correcting hazards that may occur anywhere in the farm-to-table continuum. Although HACCP programs have been useful, no prevention program can be 100 percent effective. These efforts are challenged by the complexity and ever-changing character of food production and distribution systems, by the limitations of human behavior, and, most recently, by the potential threat of intentional attacks on the food supply. Consequently, there is an increasing demand to identify and close gaps in food safety by microbiological testing. Choosing the best strategy to accomplish this, however, is difficult. Inherent limitations of methods to detect microbial contamination make strategic choices complex and less intuitive.

Strengths and Limitations of Food Monitoring

Direct testing of food prior to consumption is currently used primarily to monitor the efficacy of HACCP programs. Examples of monitored processes includes oyster farming (beds monitored for fecal contamination), milk production (monitored for somatic cell counts as a check on herd health), and ready-to-eat meat and poultry product production. Lot-by-lot testing as a means of assuring food safety is restricted to a few high-risk foods. Effluent from sprout farms is monitored for Salmonella and E. coli O157:H7, and some beef trim is monitored for E. coli O157:H7 before being made into hamburger. The use of lot-by-lot microbial monitoring to protect food is attractive because, unlike disease surveillance, food monitoring can potentially prevent initial cases. This may be the only option for high-risk situations where initial cases are intolerable from a societal point of view. For example, Wein et al. described a model to predict what might happen if 10 grams of botulism toxin was introduced into a milk tanker or silo (Wein and Liu, 2005). Even with optimum disease surveillance, hundreds of thousands of people would be expected to get sick or die by the time the problem was identified and corrected. Food monitoring may add a layer of protection as part of a comprehensive biosecurity plan assuming the basic limitations of food monitoring can be overcome.

Intrinsic limitations of food monitoring are related to sampling and testing (see Table 5-2). Of these, sampling issues represent the greatest challenge. Because the denominator of potential food vehicles could be the total amount of food available for consumption, that amount could be more than 350 billion pounds per year, as estimated by the USDA in 1997 (Kantor et al., 1997). Additionally, microbial contamination of food can be introduced with or without amplification by bacterial growth at any step between production and consumption. Consequently, the potential units to be analyzed include separate food plants, individual animals, all process lots, individual servings, and everything in between. Furthermore, while grossly contaminated food may be easily detected by discoloration, odor, or other manifestations of degradation, most types of microbial contamination with infectious agents leave no visual, olfactory, or tactile clues. Recognition of such microbial contaminates requires direct detection or isolation by growth in culture. But the sampling of food for this purpose is highly problematic. The distribution of microbial contaminates in a product is usually uneven, making sampling decisions critical to success. This may be especially true for pathogens introduced intentionally and sporadically into the food supply. Finally, the amount of confidence in the safety of food screened by testing is proportional to the amount of sampling done and inversely proportional to the prevalence of the pathogen. High levels of assurance require high sampling and expense rates, especially for rare pathogens.

TABLE 5-2. Comparison of Food Monitoring and Disease Surveillance.


Comparison of Food Monitoring and Disease Surveillance.

In addition to the limitations imposed by sampling issues, the actual sensitivity of testing methods greatly impacts the usefulness of food monitoring. Under some circumstances microbes may exist in food in very small quantities and still cause significant problems when consumed by a large number of people. For instance, in the 1994 nationwide ice cream-associated outbreak of Salmonella enteritidis, less than six organisms per serving were found in contaminated lots, which is far below the standard infective dose and detection limit of most tests. Nevertheless, the wide distribution of the product caused an estimated 224,000 people to become ill (Hennesey et al., 1996). The sensitivity of laboratory tests can be increased by increased sampling. However, the cost of testing and the cost of food lost to testing increases proportionately. Bacterial amplification occurs in improperly prepared or stored food, and during refrigeration of products contaminated with psychotrophic bacteria such as Listeria monocytogenes. In these instances, bacteria or their toxins may be present in large quantities and are relatively easy to detect. This is not the case with foods contaminated during production, storage, transport, or preparation that have been otherwise handled properly, or with foods contaminated by viruses and parasites, such as Norovirus or Cyclospora cayetanesis, which do not amplify in food. In addition, microbial cells tend to become damaged in food, and many food matrices (a food matrix represents the components of food as consumed) interfere with testing procedures. Damaged cells may readily revive under the perfect culture conditions of the human body, but fare less well under the relatively harsh conditions of in vitro culture. The problems of food testing are exemplified by the 1996 E. coli O157:H7 outbreak in Japan where, despite excellent epidemiological evidence linking 6,000 cases of disease to radish sprouts, pathogens could not be recovered from products after extensive screening (Michino et al., 1999). On the other side of the equation, pathogens found in food may or may not pose a threat to public health. For instance, not all Listeria monocytogenes found in food are likely to cause disease in people (Jacquet et al., 2004), but discovery of nonvirulent microorganisms by food monitoring may lead to unnecessary and expensive regulatory action.

New technology has lowered the threshold for food monitoring utility. Nucleic acid-based assays such as polymerase chain reaction (PCR) can in theory solve in vitro viability issues of microorganisms in food and improve test sensitivity and specificity. However, the technology does not change basic issues related to sampling, low pathogen load, and test interpretation. Also, nucleic acids typically survive microbial inactivation procedures such as pasteurization (Hilfenhaus et al., 1997), and PCR reactions are inhibited by substances in many food matrices, adding new interpretive challenges. Biosensor technology, while still in its infancy, holds potential for real-time monitoring of foods (Anderson and Taitt, 2005). Future developments may improve biosensor detection limits and specificity, which are currently inadequate for most analytes. Although encouraging in terms of potential throughput and per-sample cost, biosensor technology has many of the same theoretical limitations described for nucleic acid and conventional technology.

The sum of sampling and sensitivity issues make food monitoring impractical as a broadly applied tool for protecting the food supply. Mass food testing would be extremely expensive and would have very low predictive values. Lot-by-lot testing may be considered if (1) the risk cannot be reduced by process changes or engineering controls, (2) if a test is available that can detect the contaminant at suitably low levels, (3) if the test turnaround time does not interfere unduly with product requirements, and (4) if the cost to the consumer is justifiable.

Strengths and Limitations of Disease Surveillance

One hundred percent of food consumed by humans is essentially subject to a natural bioassay in the consumers. Each case of disease represents some failure of our food protection systems that can potentially be corrected. Disease surveillance is the collection of information that can, in theory, detect problems anywhere in the food supply chain. The denominator for U.S. disease surveillance is 281 million discrete individuals (2000 census data). Unlike food products, people can announce that they have been contaminated by presenting themselves to their physician or by calling a foodborne disease complaint line. Microbial pathogens tend to be evenly distributed through specimens such as feces due to extended mixing in the digestive tract.

Pathogen load is not generally an intrinsic limiting factor in foodborne disease surveillance, as microbial amplification is part of the pathogenesis of most foodborne infections. There are exceptions to this generalization, such as hepatitis A or hemolytic uremic syndrome due to E. coli O157:H7 where presenting symptoms are secondary to primary acute infections, and occur at a point in the disease during which pathogen numbers are declining and may limit detection. Detection is also a problem for many enterotoxins, heavy metals, and other foodborne toxins. Test specificity is less of a problem in disease surveillance than food monitoring as established pathogens detected in ill humans or animals are highly correlated with disease.

Foodborne disease surveillance also has inherent limitations. The time interval between a contamination event and a surveillance signal may be considerable. The process can involve many sequential steps. At a minimum, disease surveillance requires (1) exposure, (2) incubation, and (3) presentation to a physician or other notification. In addition, pathogen specific surveillance may also require (4) collection of specimens, (5) diagnostic laboratory testing, (6) reporting to public health authorities, (7) submission of samples to public health laboratories, (8) further laboratory characterization and reporting, (9) analysis of surveillance data, (10) interview of patients, (11) outbreak investigation, (12) case-control study, and (13) national reporting. As a result, detection and investigation of outbreaks can be a lengthy process. In addition, foodborne disease surveillance cannot by definition prevent initial cases.

Although disease surveillance cannot prevent initial cases, it has been very effective at (1) preventing ongoing transmission, (2) identifying unforeseen problems in the food and water supplies, and (3) identifying trends in foodborne disease that can guide public health policy (Besser et al., 2003). For instance, between 1997 and 2004 PulseNet and associated disease surveillance activities played a prominent role in the recall of millions of pounds of contaminated food withdrawn from U.S. markets. Examples of outbreak investigations that uncovered unsafe practices include the 2003 outbreak due to vacuum-packed blade-tenderized steaks allowed identification of the manufacturing process that rendered the product unsafe under current cooking recommendations (Laine et al., 2005) and the 2005 nationwide outbreak of S. typhimurium associated with the use of uncooked potentially contaminated products in finished ice cream leading to changed use recommendations (Center for Food Safety and Applied Nutrition, 2005). Changes in the U.S. regulatory policy occurred after Listeriosis outbreak investigations highlighted the problems associated with hot dogs (Mead et al., 2006) and turkey deli meat (Gottlieb et al., 2006). Finally, disease surveillance facilitates discovery of significant trends, such as increasing fluoroquinolone resistance in human infections caused by the use of agricultural antibiotics (Smith et al., 1999) or problems associated with increasing importation of fresh produce (Naimi et al., 2003). As much as we may want to prevent initial cases through food monitoring, humans are the best possible culture media and bioassay for detection of human disease agents, and disease surveillance will likely remain our most powerful detection tool for detecting problems in the food supply for years to come.

Types of Foodborne Disease Surveillance Programs

The three most common foodborne disease surveillance strategies, complaint or notification systems, pathogen-specific surveillance, and syndromic surveillance, differ in their application and limitations (see Table 5-3). Complaint and notification systems use a wide variety of factors to link cases of disease. State or local governments, industry, or institutions gather reports of diarrheal illnesses possibly linked to foodborne exposure. The typical scenario is recognition among attendees at a church potluck of group illness or a physician noticing a cluster of cases with some commonality, such as a similar syndrome. Because these systems do not rely on identification of specific pathogens, they can potentially detect a disease cluster caused by any disease agent, including unknown or modified agents, in a relatively short time. This makes complaint systems a valuable adjunct to pathogen-specific surveillance. The information most often used to link cases is personal recognition, which is a useful method for detection of local events but by itself is less valuable for detection of low-level widespread contamination events. However, outbreaks identified through complaint systems may be linked together to detect widespread events. This can be accomplished through disease communication systems such as Epi-X, reporting systems such as eFORS, or through pathogen-specific systems such as PulseNet, once an agent has been identified. All of these mechanisms were used to detect the 1998 international outbreak of Shigella sonnei and enterotoxigenic E. coli caused by contaminated parsley (Naimi et al., 2003).

TABLE 5-3. Comparison of Disease Surveillance Approaches.


Comparison of Disease Surveillance Approaches.

Salmonella surveillance is one of the oldest pathogen-specific surveillance systems. Routine collection of information about S. typhi began in 1912 and was expanded to include all Salmonella in 1942. Serotype-specific Salmonella surveillance began in 1963 (Swaminathan et al., 2006). Refinement of the case definition due to the more specific serotype information causes outbreak cases to stand out from background sporadic cases and strengthens the association between illness and a common source. Over the years Salmonella serotype surveillance has uncovered diverse problems in the food supply that might not have otherwise been discovered, such as persistent low-level contamination of shell eggs (St. Louis et al., 1988), recurring contamination of pasteurized ice cream during transport (Hennesey et al., 1996), and problems with sprout production (CDC, 2002a). In 1997 Bender et al. showed that the benefits of increased specificity of Salmonella serotype surveillance could be extended to E. coli O157:H7 through the routine use of PFGE (Bender et al., 1997). The Centers for Disease Control and Prevention (CDC) expanded this capability nationwide with the creation of PulseNet USA in 1998 (Swaminathan et al., 2001) and globally with PulseNet International. The primary factors used to link cases to each other are disease agent and time rather than personal recognition. This dramatically lowered the number of reported cases needed to detect widespread outbreaks. A good example is the 2003 outbreak of vacuum-packed blade-tenderized steaks, which was initially detected by two cases with unique PFGE subtypes and resulted in the recall of 739,000 pounds of potentially contaminated product distributed in multiple states (Laine et al., 2005). Although very sensitive and specific, by definition pathogen-specific surveillance only works for those agents under surveillance. Acts of bioterrorism or naturally occurring outbreaks caused by agents not under surveillance would not be detected by this mechanism nor would outbreaks due to unknown agents. Eighty-two percent of cases of food-borne illnesses are thought to be due to unrecognized agents (Mead et al., 1999). Although pathogen-specific surveillance samples only a small percentage of total foodborne disease cases, it nevertheless has been one of the most robust indicators of problems in the food supply.

Syndromic surveillance is the third type of disease surveillance that could potentially be applied to detection of problems in the food supply. In recent years, this term has generally applied to broad monitoring of nonspecific health data such as diarrheal illness or markers of illness, such as Imodium sales. This is in contrast to complaint or notification systems described above which, while utilizing nonspecific health data such as diarrheal illness, are narrowly focused on recognized clusters or specific complaints. Several large networks have been established to conduct syndromic surveillance, such as BioSense, the Electronic Surveillance System for Early Notification of Community-based Epidemics (ESSENCE), and National Retail Data Monitor (NRDM). Potential outbreaks are detected by spikes in the incidence of common syndromes or surrogate indicators rather than agent or personal recognition, so in theory these systems should be able to detect problems due to known or unknown agents. The primary problem with syndromic surveillance is an unfavorable signal-to-noise ratio. The number of cases needed to trip the system is inversely proportional to the specificity. Thus, syndromic surveillance based on nonspecific health data is useful for detecting very large local events but is very insensitive to small or widespread events. One of the nation’s largest systems was established in New York City (NYC) in 2001 (Das et al., 2005; Heffernan et al., 2004). In one year of surveillance involving 2.5 million patient records, the NYC surveillance system had 18 spikes in reports of diarrhea and vomiting syndromes clustered in three outbreak periods. One of the spikes was followed by five institutional outbreak investigations (detected by other mechanisms), but it is unclear whether the surveillance data itself would have been of sufficient specificity to identify the outbreaks.

Surveillance for specific rare syndromes such as hemolytic uremic syndrome or Guillain-Barre syndrome resembles pathogen-specific surveillance more than syndromic surveillance in terms of sensitivity and specificity. Surveillance for unexplained death or critical illnesses with possible infectious etiology is a particularly important type of surveillance. Although postmortem diagnosis of illness can be problematic, unexplained death and critical illnesses may be the only clue that a major foodborne or other type of bioevent has occurred. Surprisingly, only three such programs in the United States exist.

Existing Networks

A large array of acronyms are used to describe existing networks that play serve some role in protecting the food supply (see Table 5-1). Included are communication networks (HAN, Epi-X, INFOSAN, RASFF), surveillance data management systems (NETSS, NEDSS), syndromic surveillance projects and tools (Biosense, UNEX, ESSENCE, NRDM, RODS), a surveillance system designed to determine the burden of foodborne disease (FoodNet), outbreak response and reporting networks (eFORS, GOARN), animal surveillance networks (NAHSS, CAHFSE), an enteric disease antibiotic-resistance monitoring network (NARMS), laboratory data handling networks (eLEXNET, PulseNet, Global Salm-Surv, CaliciNet), and laboratory response networks (LRN, FERN, NAHSS, NPDN, eLRN, ICLN). Laboratory response networks such as LRN and FERN have protocols for testing suspect food items but currently have no role in primary detection of problems in the food supply. Current food-monitoring programs are designed to support HACCP programs or monitor for specific pathogens in high-risk foods but play little role in broad monitoring for unexpected introduction of contaminants into the food supply. Outbreak reporting systems such as eFORS and GOARN can potentially detect national or international outbreaks by linking local outbreaks without previously recognized connection. PulseNet and associated foodborne disease surveillance programs are currently the most sensitive methods for detecting unforeseen problems in the food supply.

Successes, Problems, and Promises of PulseNet and Associated Foodborne Disease Programs

Since its inception in 1998, PulseNet has been a key tool in the detection of a wide range of problems in the food supply, such as Listeria monocytogenes in luncheon meat; Shigella sonnei in imported parsley; E. coli O157:H7 in meats, unpasteurized juices, and produce; and Salmonella in almonds and custom ice cream. Estimating cost and benefits of surveillance programs is difficult and rare, but it is becoming clear that prevention benefits of PulseNet and associated activities far outweigh its costs. Bell and colleagues estimated that 800 cases of disease were prevented by the recall of 250,000 hamburgers after detection of the 1993 E. coli O157:H7 outbreak on the West Coast of the United States, which occurred prior to development of PulseNet (Bell et al., 1994). Since that time, PulseNet findings have directly or indirectly led to the recall of many times that amount of contaminated products and enabled continuing contamination problems to be detected and rectified. In 1997, PulseNet enabled the Colorado Department of Health to detect an outbreak caused by potentially contaminated beef that led to the recall of 25,000,000 pounds of beef, and 18,600,000 pounds in 2002 (CDC, 2002b). Elbasha et al. estimated that the Colorado PulseNet system would recover all costs if only five cases per year were prevented (Elbasha et al., 2000). A single case of hemolytic uremic syndrome caused by E. coli O157:H7 can cost up to $453,675 in medical costs alone (Buzby et al., 1996).

The PulseNet system is currently used to track nine diseases by use of standardized PFGE protocols, but potentially can be used to track any infectious disease confirmed by detection of a specific microorganism. It increases the inherent sensitivity of disease surveillance by refining the case definition and provides a platform for rapid communication and comparison of results. More specific case definitions improve the signal-to-noise ratio, and permits small, yet significant events to be detected. PulseNet has recently bridged the gap between food monitoring and disease surveillance data by incorporating data from the USDA and FDA monitoring and investigation programs. An outbreak of Salmonella Kiambu associated with beef jerky in 2003 was detected by linking PFGE patterns from food monitoring to disease surveillance patterns, resulting in the recall of 22,000 pounds of contaminated product. PulseNet International has expanded the network worldwide and has already had early successes at tracing an international outbreak of shigellosis associated with airline meals (Gaynor et al., 2004) and an outbreak associated with ground beef from a U.S. military installation (CDC, 2004). In spite of its history of successes and awards, PulseNet operates unevenly around the country, with most activity associated with a small number of states. Surveys by the Association of Public Health Laboratories and Council for State and Territorial Epidemiologists found important deficits in the nation’s capacity to detect, investigate, and respond to food safety problems (Association of Public Health Laboratories, 2003; Council of State and Territorial Epidemiologists, 2004). Most laboratory and epidemiology programs are understaffed and underfunded, leaving many food safety problems undetected and unresolved. As a result of these issues, PulseNet operates at only a fraction of its potential, with its greatest promise yet to be exploited.

Other Potential Systems

Although PulseNet is the most developed system for detection of problems in the food supply, it is not the only useful approach. Clusters of notifiable disease can potentially be detected through analysis of data from the Public Health Laboratory Information System, and widespread events have been detected using the CDC’s Electronic Outbreak Reporting System (eFORS) and Epi-X postings. Because the etiology of outbreaks reported through eFORS or Epi-X can be known or unknown, these systems can capture complaint data not associated with a nationally notifiable pathogen. Like PulseNet, eFORS is not being used to its full potential. Resource limitations at the state and local level limit detection and reporting of outbreaks. Completion of the National Electronic Disease Surveillance System (NEDSS) will provide a backbone for national surveillance, but the lack of standard forms for interviewing cases will continue to limit outbreak detection in the absence of specific agent information.

Surveillance for unexplained death and serious illness caused by possible infectious etiology has significant potential for use in detecting serious problems in the food supply, such as intentional tampering. As much as we would like to prevent death, it is not possible to anticipate every possible mode by which pathogens could be intentionally or unintentionally introduced into the food supply. It seems that at a minimum we should be able to know that such an event is occurring so that appropriate control measures can be instituted to prevent additional cases. Investigation of unexplained death possibly caused by infectious causes is expensive and difficult. Nevertheless, national resources dedicated for this purpose is very limited and includes only small CDC-funded programs in California, Minnesota, and Connecticut.


Targeted microbial monitoring of selected high-risk foods and processes plays an important role in protecting the food supply. Expansion of monitoring activities may be justified in situations where initial cases cannot be tolerated. However, broadly applied food-monitoring programs are likely to be costly and insensitive due to intrinsic sampling and testing limitations. Foodborne disease surveillance programs cannot prevent initial cases, but are nevertheless the most sensitive method for detecting unrecognized problems in the food supply. Current foodborne disease surveillance programs, such as PulseNet, are operating at only a fraction of their potential, largely because of underfunding. Disease surveillance programs have a high benefit-to-cost ratio and represent a basic function of public health. Real-time foodborne disease surveillance at the state, federal, and international levels is an achievable, relatively inexpensive goal that would have an immediate, positive impact on food safety.


CDR Kimberly Elenberg, M.S., R.N.2 and Artur Dubrawski, Ph.D.3

The Consumer Complaint Monitoring System (CCMS), owned by the Food Safety and Inspection Service (FSIS) of the USDA, is a database used to record, evaluate, and track all consumer food complaints involving meat, poultry, and egg products. It has assisted public health professionals with evaluating approximately 4,000 consumer complaints since January 2001.

Consumer complaints are received through phone calls to the USDA field compliance officers, and through the toll-free phone number of the Meat and Poultry Hotline. Adverse food event reports can also be received and managed for imported products and school lunch products distributed through USDA’s Food and Nutrition Service. Complaints typically involve reports of illness, injury, foreign objects, contamination (including chemical contamination), allergic reactions, and improper labeling.

To further improve food safety and security, FSIS’ Office of Public Health Science developed the Consumer Complaint Monitoring System II (CCMS II). CCMS II complements the existing foodborne threat surveillance systems Pulse-Net and FoodNet (see Summary and Assessment and Tauxe in Chapter 3). Although PulseNet and FoodNet actively track pathogen strains isolated from humans with foodborne disease, CCMS employs a form of passive surveillance to provide the earliest possible warning of a wide variety of foodborne threats. This system rapidly organizes and analyzes the incoming complex, multivariate data from consumer complaints concerning such adverse events as food-associated symptoms or foreign objects present in food.

As other contributors to this workshop have noted, the possible sources of foodborne contamination are extremely vast. Although using adverse food reports to assess possible foodborne threats presents the considerable challenge of discerning significant trends against a background of random noise, this approach also offers several benefits. Unlike FoodNet and PulseNet, CCMS is not limited to assessing known pathogens, which account for a minority of foodborne illness; unknown agents account for 81 percent of U.S. foodborne illnesses and hospitalizations and 64 percent of deaths (Mead et al., 1999). In addition to tracking possible emerging foodborne pathogens, including intentionally modified organisms, CCMS examines chemical and foreign object contaminants in food (the latter is the subject of 80 percent of all adverse food reports received by USDA).

Finding Patterns in Consumer Complaints

Surveillance activities involving CCMS are conducted by both the FSIS Office of Public Health Science and the FSIS Office of Food Defense and Emergency Response. The analytical component of CCMS, called Emerging Patterns in Food Complaints (EPFC), developed in collaboration with Artur Dubrawski’s group at Carnegie Mellon University, employs computational methods, including multiple-attribute algorithms and Bayesian probability models, to detect patterns in complaints received by CCMS—data that are too voluminous and complex for the human mind to grasp. Public health analysts examine the resulting patterns and assess the likelihood that they represent material foodborne threats. CCMS tracks several variables associated with each report, including the date, time, and location of the event; associated symptoms and their onset times or descriptions of foreign objects; and the type and purchase origin of the suspected product.

When two or more similar complaints are received within a reasonably close time span (the particular range of interest varies depending on the shelf life of the involved food), the EPFC statistically quantifies their relatedness. It compares the group of variables that makes up each consumer complaint with a database of approximately 4,000 past cases and also with a series of 42 causal models we have developed based on passive surveillance records of adverse food events. The results that public health analysts receive indicate the most similar past cases, the likeliest causal scenarios, and the most probable case-cause combinations matching the complaints under investigation. For example, EFPC comparisons might find matches between the symptoms and locations associated with two contemporary cases, but little similarity between the product sources; when these results are in turn compared with the range of causal models, the best match is with a community illness not related to a specific product. If CCMS continued to receive similar complaints, we would alert the appropriate state health department to investigate likely causes of community illness, such as water contamination. By contrast, a series of closely timed complaints with high similarity and a common causal scenario would signify a foodborne outbreak; in that situation, investigators would pursue this hypothesis by testing the suspected product(s) and comparing the results with the PulseNet database. To date, the system has recognized several outbreaks: of Salmonella, Listeria monocytogenes, and E. coli O157:H7.

In addition to providing investigators with valuable leads for tracing the origin of suspected and confirmed contaminants, CCMS can further aid such management decisions as the identification of an emergency response that is appropriate to the nature, location, and extent of a threat. CCMS data can also be used to alert food producers to more general food safety issues associated with specific processes and/or plants. For example, a large number of apparently diverse complaints associated with products from a single company might reflect on overall condition, such as substandard sanitation. When many complaints point to a single causal model (for example, a metal foreign object in a specific product), EPFC analysis can pinpoint the source of the problem and lead to its solution (routinely inspect the product with a metal detector). This process can occur rapidly. For example, within 30 minutes of a report of students being injured with glass in their school lunches, CCMS can identify and investigate the implicated food commodities and, if necessary, communicate with the affected schools to ensure that foods containing the product are removed from cafeteria lines immediately.

Future Expansion of CCMS

Based on the demonstrated ability of the analytical component of CCMS to detect interesting, low-amplitude signals in sparse, noisy data, further expansion of the system is warranted. Although currently operated on a state-by-state basis, CCMS will soon permit state public health officials to track pathogens and contaminants beyond their borders and to view a national “snapshot” of adverse food events in real time. Data from earlier points in the farm-to-table continuum, which would be especially valuable in identifying patterns associated with zoonotic disease, could be analyzed in conjunction with consumer complaints as part of an integrated national biosurveillance system. As CCMS grows in importance and visibility, it will be important to establish surge capacity to deal with the expected high volume of complaints that could arise during a foodborne outbreak.

Early detection is the key to mitigating the consequences of foodborne illness. Improving the timeliness of data collection, pattern detection, and communication of information about adverse food events using CCMS and related passive surveillance technologies requires better tools as well as changes in the way these events are reported. Finally, because tools such as CCMS are only as effective as the humans who use them, we must educate everyone who plays a role in food safety, and encourage collaboration at all levels.


  1. Anderson GP, Taitt CR. Biosensor-based detection of foodborne pathogens. In: Fratamico PM, Bhunia A, Smith JL, editors. Foodborne Pathogens: Microbiology and Molecular Biology. Wymondham, UK: Academic Press; 2005.
  2. APHL (Association of Public Health Laboratories). A Recipe for Stronger Food Safety Programs. Washington, D.C: Association of Public Health Laboratories; 2003.
  3. Bell BP, Goldoft M, Griffin PM, Davis MA, Gordon DC, Tarr PI, Bartleson CA, Lewis JH, Barrett TJ, Wells JG, Baron R, Kobayashi J. A multiple outbreak of Escherichia coli O157:H7-associated bloody diarrhea and hemolytic uremic syndrome from hamburgers: The Washington State experience. Journal of the American Medical Association. 1994;272(17):1349–1353. [PubMed: 7933395]
  4. Bender JB, Hedberg CW, Besser JM, Boxrud DJ, MacDonald KL, Osterholm MT. Surveillance for Escherichia coli O157:H7 infections in Minnesota by molecular subtyping. New England Journal of Medicine. 1997;337(6):388–394. [PubMed: 9241128]
  5. Besser J. Session IV: What Are the Tools and Technologies for “Real Time” Surveillance of the Food Supply for Conventional and Unconventional Adulterants?; Presentation at the Forum on Microbial Threats Workshop Meeting, Foodborne Threats to Health: The Policies and Practices of Surveillance, Prevention, Outbreak Investigations, and International Coordination; Washington, DC. Institute of Medicine, Forum on Microbial Threats; Oct 25, 2005.
  6. Besser J, Beebe J, Swaminathan B. Investigation of foodborne disease outbreaks. In: Pfaller MA, editor. Manual of Clinical Microbiology. 8th ed. Washington, D.C: American Society for Microbiology Press; 2003.
  7. Buzby JC, Roberts T, Lin JC-T, MacDonald JM. Bacterial Foodborne Disease: Medical Costs and Productivity Losses. Washington, D.C: Department of Agriculture; 1996. Agricultural Economics Report No. 741.
  8. CDC (Centers for Disease Control and Prevention). Escherichia coli O157:H7 infections associated with eating a nationally distributed commercial brand of frozen ground beef patties and burgers—Colorado, 1997. Morbidity and Mortality Weekly Report. 1997;46(33):777–778. [PubMed: 9272586]
  9. CDC. Outbreak of Salmonella serotype kottbus infections associated with eating alfalfa sprouts—Arizona, California, Colorado, and New Mexico, February–April 2001. Morbidity and Mortality Weekly Report. 2002;51(1):7–9. [PubMed: 11831433]
  10. CDC. Multistate outbreak of Escherichia coli O157:H7 infections associated with eating ground beef—United States, June–July 2002. Morbidity and Mortality Weekly Report. 2002;51(29):637–639. [PubMed: 12186220]
  11. CDC. Escherichia coli O157:H7 infections associated with ground beef from a U.S. military installation—Okinawa, Japan, February 2004. Morbidity and Mortality Weekly Report. 2004;54(2):40–42. [PubMed: 15660018]
  12. Center for Food Safety and Applied Nutrition. Bulletin to the Food Service and Retail Food Store Industry Regarding Cake Batter Ice Cream and Similar Products. Aug 19, 2005. [accessed April 10, 2006]. [Online]. Available: http://www​.cfsan.fda​.gov/~ear/ret-batt.html.
  13. Council of State and Territorial Epidemiologists. National Assessment of Epidemiological Capacity: Findings and Recommendations. Washington, D.C: 2004.
  14. Das D, Metzger K, Heffernan R, Balter S, Weiss D, Mostashari F. New York City Department of Health and Mental Hygiene. Monitoring over-the-counter medication sales for early detection of disease outbreaks—New York City. Morbidity and Mortality Weekly Report. 2005;54(Suppl):41–46. [PubMed: 16177692]
  15. Elbasha EH, Fitzsimmons TD, Meltzer MI. Costs and benefits of a subtype-specific surveillance system for identifying Escherichia coli O157:H7 outbreaks. Emerging Infectious Diseases. 2000;6(3):293–297. [PMC free article: PMC2640869] [PubMed: 10827120]
  16. FAO (Food and Agriculture Organization of the United Nations). International Cooperation on Food Contamination and Foodborne Disease Surveillance, Second FAO/WHO global forum of food safety regulators [Online]. 2004. [accessed February 1, 2006]. Available: http://www​.fao.org/docrep​/meeting/008/ae163e.htm.
  17. GAO (General Accountability Office). Bioterrorism. A Threat to Agriculture and the Food Supply. 2003. [accessed April 10, 2006]. [Online]. Available: http://www​.gao.gov/new.items/d04259t.pdf.
  18. GAO. Emerging Infectious Diseases: Review of State and Federal Disease Surveillance Efforts. 2004. [accessed April 10, 2006]. [Online]. Available: http://www​.gao.gov/new.items/d04877.pdf.
  19. Gaynor K, Kanenaka B, Colindres R, Mintz E, Kalluri P, Kitsutani P, Nakata M, Wedel S, Boxrud D, Jennings D, Yoshida H, Tosaka N, Park SY, Effler PV. International Outbreak of Shigella sonnei among Airline Passengers—Honolulu, Hawaii. Presentation abstract, Session C, 54th Annual Epidemic Intelligence (EIS) Conference; April 11–15; Atlanta, GA. Aug, 2004.
  20. Gottlieb SL, Newbern EC, Griffin PM, Graves LM, Hoekstra RM, Baker NL, Hunter SB, Holt KG, Ramsey F, Head M, Levine P, Johnson G, Schoonmaker-Bopp D, Reddy V, Kornstein L, Gerwel M, Nsubuga J, Edwards L, Stonecipher S, Hurd S, Austin D, Jefferson MA, Young SD, Hise K, Chernak ED, Sobel J. Listeriosis Outbreak Working Group. Multistate outbreak of list-eriosis linked to turkey deli meat and subsequent changes in US regulatory policy. Clinical Infectious Diseases. 2006;42(1):29–36. [PubMed: 16323088]
  21. Heffernan R, Mostashari F, Das D, Besculides M, Rodriguez C, Greenko J, Steiner-Sichel L, Balter S, Karpati A, Thomas P, Phillips M, Ackelsberg J, Lee E, Leng J, Hartman J, Metzger K, Rosselli R, Weiss D. New York City syndromic surveillance systems. Morbidity and Mortality Weekly Report. 2004;53(Suppl):23–27. [PubMed: 15714622]
  22. Hennessy TW, Hedberg CW, Slutsker L, White KE, Besser-Wiek JM, Moen ME, Feldman J, Coleman WW, Edmonson LM, MacDonald KL, Osterholm MT. A national outbreak of Salmonella enteritidis infections from ice cream. The Investigation Team. New England Journal of Medicine. 1996;334(20):1281–1286. [PubMed: 8609944]
  23. Hilfenhaus J, Groner A, Nowak T, Weimer T. Analysis of human plasma products: Polymerase chain reaction does not discriminate between live and inactivated viruses. Transfusion. 1997;37(9):935–940. [PubMed: 9308641]
  24. Jacquet C, Doumith M, Gordon JI, Martin PM, Cossart P, Lecuit M. A molecular marker for evaluating the pathogenic potential of foodborne Listeria monocytogenes. Journal of Infectious Diseases. 2004;189(11):2094–2100. [PubMed: 15143478]
  25. Kantor LS, Lipton K, Manchester A, Oliveira V. Estimating and addressing America’s food losses, Food Review. 1997. [accessed April 10, 2006]. pp. 2–12. [Online]. Available: www​.ers.usda.gov/publications​/foodreview/jan1997/jan97a.pdf.
  26. Laine ES, Scheftel JM, Boxrud DJ, Vought KJ, Danila RN, Elfering KM, Smith KE. Outbreak of Escherichia coli O157:H7 infections associated with nonintact blade-tenderized frozen steaks sold by door-to-door vendors. Journal of Food Protection. 2005;68(6):1198–1202. [PubMed: 15954707]
  27. Mead PS, Slutsker L, Dietz V, McCaig LF, Bresee JS, Shapiro C, Griffin PM, Tauxe RV. Food-related illness and death in the United States. Emerging Infectious Diseases. 1999;5(5):607–625. [PMC free article: PMC2627714] [PubMed: 10511517]
  28. Mead PS, Dunne EF, Graves L, Wiedmann M, Patrick M, Hunter S, Salehi E, Mostashari F, Craig A, Mshar P, Bannerman T, Sauders BD, Hayes P, Dewitt W, Sparling P, Griffin P, Morse D, Slutsker L, Swaminathan B. Nationwide outbreak of listeriosis due to contaminated meat. Epidemiology and Infection. 2006;134(4):744–751. Epub December 1 2005, 1–8. [PMC free article: PMC2870438] [PubMed: 16318652]
  29. Michino H, Araki K, Minami S, Takaya S, Sakai N, Miyazaki M, Ono A, Yanagawa H. Massive outbreak of Escherichia coli O157:H7 infection in schoolchildren in Sakai City, Japan, associated with consumption of white radish sprouts. American Journal of Epidemiology. 1999;150(8):787–796. [PubMed: 10522649]
  30. Naimi TS, Wicklund JH, Olsen SJ, Krause G, Wells JG, Bartkus JM, Boxrud DJ, Sullivan M, Kassenborg H, Besser JM, Mintz ED, Osterholm MT, Hedberg CW. Concurrent outbreaks of Shigella sonnei and enterotoxigenic Escherichia coli infections associated with parsley: Implications for surveillance and control of foodborne illness. Journal of Food Protection. 2003;66(4):535–541. [PubMed: 12696674]
  31. Smith KE, Besser JM, Hedberg CW, Leano FT, Bender JB, Wicklund JH, Johnson BP, Moore KA, Osterholm MT. Quinolone-resistant Campylobacter jejuni infections in Minnesota, 1992–1998. Investigation Team. New England Journal of Medicine. 1999;340(20):1525–1532. [PubMed: 10332013]
  32. St. Louis ME, Morse DL, Potter ME, DeMelfi TM, Guzewich JJ, Tauxe RV, Blake PA. The emergence of grade A eggs as a major source of Salmonella enteritidis infections. New implications for the control of salmonellosis. Journal of the American Medical Association. 1988;259(14):2103–2107. [PubMed: 3279240]
  33. Swaminathan B, Barrett TJ, Hunter SB, Tauxe RV. CDC PulseNet Task Force. PulseNet: The molecular subtyping network for foodborne bacterial disease surveillance, United States. Emerging Infectious Diseases. 2001;7(3):382–389. [PMC free article: PMC2631779] [PubMed: 11384513]
  34. Swaminathan B, Barrett TJ, Fields P. Surveillance for human Salmonella infections in the United States. Journal of AOAC International. 2006;89(2):553–559. [PubMed: 16640306]
  35. Wein LM, Liu Y. Analyzing a bioterror attack on the food supply: The case of botulinum toxin in milk. Proceedings of the National Academy of Sciences. 2005;102(28):9984–9989. [PMC free article: PMC1161865] [PubMed: 15985558]



Clinical Laboratory Manager, Public Health Laboratory.


Food Safety Inspection Service, United States Department of Agriculture; Office of Food Defense and Emergency Response until June 2006, currently Commander, Director of Medical Readiness, USPHS.


Carnegie Mellon University; Auton Lab, The Robotics Institute.