ABSTRACT

Restaurant inspections seek to identify and correct risk factors for foodborne illness, but restaurant inspection data are not typically used more broadly as a food safety surveillance tool. In 2015, there was an outbreak of Salmonella serotype Newport infections associated with multiple restaurants in a chain (chain A), primarily in Minnesota. The outbreak was associated with tomatoes that were likely contaminated at the point of production. The objective of this study was to demonstrate the potential usefulness of aggregated restaurant inspection data in aiding individual outbreak investigations. Reports of the last inspection for all chain A restaurants that preceded the first reported case meal date in the outbreak were obtained from local health departments and the Minnesota Department of Health. Ordinal logistic regression was used to assess differences in risk factor and good retail practice violation categories and specific violations in restaurants with zero cases (nonoutbreak restaurants) (n = 25), one to two cases (n = 16), and at least three cases (n = 13). For restaurants with a “protection from contamination” violation in the routine inspection that preceded the outbreak, the proportional odds ratio for outbreak level was 4.92 (95% confidence interval: 1.57, 15.39; P = 0.01). These findings suggest that food handling practices in the outbreak restaurants may have increased contamination of foods through cross-contamination, which in turn increased transmission at outbreak restaurants. These data suggest that aggregated data from routine inspection reports can provide useful information to aid in outbreak investigations and other foodborne illness surveillance and prevention activities.

HIGHLIGHTS
  • Restaurant inspection data are not typically used as surveillance data.

  • Food handling practices may have increased pathogen contamination in a 2015 outbreak.

  • Protection from contamination violations were cited more in outbreak restaurants.

  • Aggregated inspection reports can provide useful information to aid in outbreak investigations.

Public health surveillance—the ongoing, systematic collection and analysis of data to prevent and control disease and injury—is the foundation for public health efforts to prevent foodborne illness (16). Surveillance plays a critical role in the detection and control of foodborne illness by driving the iterative cycle of public health prevention (16). Surveillance allows detection of an outbreak via an observed increase in reported cases above a threshold that ideally leads to an investigation. Epidemiologic investigations then seek to identify the source of an outbreak to halt further disease spread and prevent future infections through corrective actions and root-cause determinations.

Restaurants are frequent settings for foodborne illness outbreaks, accounting for 64% of outbreaks in 2017 in the United States (4). Inspection of food service establishments to prevent foodborne illness is a core function of state and local health departments. The goal of risk-based inspections is to identify and correct risk factors (RFs) for foodborne illness. Results of an inspection are shared with food managers and are filed for future reference in subsequent inspections, but they are not typically used as surveillance data. Although inspections represent point-in-time observations, patterns of violations may indicate gaps in an establishment's food safety program.

Previous literature to understand the relationship between restaurant inspection results and risk of foodborne illness have shown mixed results, largely because of failure to account for differing transmission pathways across pathogens (6, 811, 15). In a study looking at routine inspections in chain versus nonchain restaurants, small sample sizes limited the ability to compare data across pathogen types (15). There is evidence that agent, food item pairing, and route of transmission are important for understanding the relationship between restaurant inspections and risk of foodborne illness (11). Salmonella provides a good lens to understand this relationship, because Salmonella can take advantage of all five major RFs for foodborne illness: food from unsafe sources, improper holding temperature, improper cooking temperature, cross-contamination, and poor employee health and hygiene.

The National Environmental Assessment Reporting System (NEARS) is a surveillance system that captures environmental assessment data from foodborne illness outbreak investigations to improve retail food safety programs. Environmental assessments are conducted after the identification of an outbreak and its contributing factors. A NEARS review of 9,788 restaurant-associated outbreaks showed that most commonly reported contributing factors were associated with food handling and preparation practices (1). However, identification of contributing factors in outbreaks is limited by outbreak characteristics, such as whether an etiologic agent or specific food vehicle was identified (2). Review of routine inspection reports may provide additional useful information for identifying the causes of outbreaks and prevention opportunities.

The objective of this study was to evaluate the potential usefulness of routine inspection data to aid in individual foodborne illness investigations using a 2015 outbreak of Salmonella Newport infections as a case study. The Minnesota Department of Health investigated an outbreak of Salmonella Newport infections from August to October 2015 (13). It identified 92 culture-confirmed cases. Of these, 81 (88%) reported that they ate or likely ate at one of multiple restaurants in a chain (chain A) before illness onset. Known meal dates for cases ranged from 1 August to 25 September. Confirmed cases reported eating at 1 of 29 chain A locations in Minnesota. In a case-control study, mild tomato salsa was the only menu item associated with illness in a multivariable analysis, and only tomatoes were associated with illness in an ingredient-specific analysis (13). Mild tomato salsa was the only menu item that contained raw tomatoes. Traceback investigations revealed that all chain A restaurants in the Minnesota distribution chain received tomatoes from the same supplier. In response to the outbreak investigation, chain A switched to a different source for tomatoes. There was only one case with a meal date after the tomato supplier was changed. Given that all chain A locations had a common menu, has standard food handling practices, and likely received the implicated product from the same distributor, this outbreak provides a useful case study to identify how violations cited in routine inspections may be associated with the outbreak.

MATERIALS AND METHODS

Data

At the time of the 2015 Salmonella Newport outbreak at chain A, there were 58 chain A locations in Minnesota. Reports of the last inspection for chain A restaurants that preceded the first reported case meal date in the 2015 outbreak were obtained from local health departments and the Minnesota Department of Health. For four chain A locations, no inspection report was available because the last inspection had occurred before 2014 or the restaurant was newly opened. Violations were abstracted from inspection reports from 13 jurisdictions for the 54 restaurants that had an inspection report that preceded the first case meal date. Because of variation in the inspection form used across jurisdictions, all violations were mapped to the Conference for Food Protection's inspection report structure (5). Violations were categorized as either risk factor (RF) violations or good retail practice (GRP) violations. There were 27 RF violation types, which are most likely to lead to foodborne illness. There were 29 GRP violation types; these reflect preventive measures that if not controlled could lead to foodborne illness. The 27 RF violations were grouped into 11 unique categories, and the 29 GRP violations were grouped into six categories. Restaurants were stratified into three categories based on the number of outbreak cases (at least three cases, one to two cases, and zero cases [nonoutbreak restaurants]), because more cases suggests greater transmission in restaurants.

Statistical analyses

Means and standard errors were calculated for the number of days between the preceding inspection report and the outbreak and for total, RF, and GRP violations for restaurants with at least three cases or one to two cases and for nonoutbreak restaurants. Next, t tests were used to compare means for restaurants with at least three cases or one to two cases to means for nonoutbreak restaurants. Ordinal logistic regression was used to calculate proportional odds ratios (POR) to assess differences in seven RF and five GRP violation categories for which violations were cited in restaurants grouped by number of cases exposed at the restaurant (outbreak level). Outbreak levels included restaurants with at least three cases, one to two cases, and zero cases (nonoutbreak restaurants). This method combines comparisons (restaurants with at least three cases or one to two cases compared with nonoutbreak restaurants and restaurants with at least three cases compared with one to two cases and nonoutbreak restaurants) into a POR that summarizes the overall association. Models were calculated that used violation categories as binary predictors (i.e., having a violation in a given category) and continuous predictors (number of violations in a given category). Ordinal logistic regression was used to assess differences in the specific RF and GRP violations for violation categories that were statistically significant at P < 0.05. Likelihood-ratio tests were used to calculate the proportionality of odds across outbreak levels, and none of the models were found to violate the proportional odds assumption. A subanalysis was conducted to identify patterns among frequently cited violations. Analyses were conducted using Stata version 14.2 (StataCorp LLC, College Station, TX).

RESULTS

Of the 54 chain A restaurants, 13 restaurants had at least three cases, 16 restaurants had one to two cases, and 25 restaurants had no cases (i.e., nonoutbreak restaurants). The number of cases per restaurant for outbreak restaurants ranged from 1 to 10 cases with a median of 2 cases. The mean number of days between the last routine inspection and the first outbreak case meal date was 176.7 ± 29.7 days for restaurants with at least three cases, 149.9 ± 27.5 days for restaurants with one to two cases, and 172.1 ± 20.5 days for nonoutbreak restaurants. There were no significant differences in the mean number of days in which the preceding restaurant inspection was conducted for outbreak restaurants (at least three cases or one to two cases) compared with nonoutbreak restaurants (P values of 0.9 and 0.5, respectively). Restaurants with at least three cases had more total, RF, and GRP violations on average than restaurants with one to two cases and nonoutbreak restaurants, although these differences were not statistically significant (Fig. 1).

FIGURE 1

Mean number of total, risk factor (RF), and good retail practice (GRP) violations by outbreak case category, Minnesota. Restaurants with at least three cases (black bars), restaurants with one to two cases (dark gray bars), and nonoutbreak restaurants (light gray bars).

FIGURE 1

Mean number of total, risk factor (RF), and good retail practice (GRP) violations by outbreak case category, Minnesota. Restaurants with at least three cases (black bars), restaurants with one to two cases (dark gray bars), and nonoutbreak restaurants (light gray bars).

One RF violation category was more likely to be associated with outbreak restaurants compared with nonoutbreak restaurants: protection from contamination (Table 1). This category consisted of three specific violations: “food separated and protected,” “food contact surfaces: cleaned and sanitized,” and “proper disposition of returned, previously served, reconditioned, and unsafe food.” Eight restaurants with at least three cases had at least one violation in the protection from contamination category, which represented 62% of these restaurants. By comparison, only 38% (n = 6) of restaurants with one to two cases and 20% (n = 5) of nonoutbreak restaurants had a violation in this category. The POR for restaurants with a protection for contamination violation was 4.92 (95% confidence interval (95% CI): 1.57, 15.39; P = 0.01). There were two GRP violation categories more frequently associated with outbreak level: food temperature control (POR: 4.97; 95% CI: 1.54, 16.03; P = 0.01) and physical facilities (POR: 6.60; 95% CI: 2.11, 20.64; P < 0.01).

TABLE 1

Risk factor and good retail practice violation categories cited in the inspection preceding a 2015 chain A Salmonella Newport outbreak for restaurants with at least three cases, one to two cases, and zero cases (nonoutbreak restaurants), Minnesota

Risk factor and good retail practice violation categories cited in the inspection preceding a 2015 chain A Salmonella Newport outbreak for restaurants with at least three cases, one to two cases, and zero cases (nonoutbreak restaurants), Minnesota
Risk factor and good retail practice violation categories cited in the inspection preceding a 2015 chain A Salmonella Newport outbreak for restaurants with at least three cases, one to two cases, and zero cases (nonoutbreak restaurants), Minnesota

Two of three violations in the protection from contamination violation category were reported in inspection reports preceding this outbreak: “food separated and protected” and “food contact surfaces: cleaned and sanitized” (Table 2). Of these, “food contact surfaces: cleaned and sanitized” was cited more frequently in restaurants with at least three cases (54%) compared with restaurants with one to two cases (19%) and nonoutbreak restaurants (12%). An example of this violation includes sanitizer that does not contain the proper sanitary or chlorine solution. The POR for having a violation for “food contact surfaces: cleaned and sanitized” was 5.37 (95% CI: 1.53, 18.90; P = 0.01). Two of four food temperature-control violations were cited; however, there were no significant differences between restaurant categories. Of nine physical facilities violations, seven were cited in inspections preceding this outbreak. Of these, two were associated with outbreak level: “plumbing installed: proper backflow devices” (POR: 8.56; 95% CI: 2.08, 35.21; P < 0.01) and “physical facilities installed, maintained, and clean” (POR: 3.41; 95% CI: 1.19, 9.83; P = 0.02). Examples of “physical facilities installed, maintained, and clean” violations include disrepair or a lack of cleanliness of the physical facilities, such as the presence of mold, missing ceiling panels, and the need for recaulking. Among the 10 outbreak restaurants with a “food contact surfaces: cleaned and sanitized” violation, 7 (70%) also had a violation for “wiping clothes properly used and stored.”

TABLE 2

Risk factor and good retail practice violations for select categories cited in the inspection preceding the 2015 chain A Salmonella Newport outbreak for restaurants with at least three cases, one to two cases, and zero cases (nonoutbreak restaurants), Minnesota

Risk factor and good retail practice violations for select categories cited in the inspection preceding the 2015 chain A Salmonella Newport outbreak for restaurants with at least three cases, one to two cases, and zero cases (nonoutbreak restaurants), Minnesota
Risk factor and good retail practice violations for select categories cited in the inspection preceding the 2015 chain A Salmonella Newport outbreak for restaurants with at least three cases, one to two cases, and zero cases (nonoutbreak restaurants), Minnesota

DISCUSSION

This outbreak was associated with multiple restaurants in one chain with standard operating practices that received the implicated food vehicle from one distributor and provided a unique opportunity to explore the relationship between illnesses and restaurant inspection results. All chain A locations received tomatoes from the same distributor; however, only about half of the restaurants had outbreak-associated cases. Salmonella cross-contamination and recontamination episodes have been linked to poor sanitation practices, poor equipment design, and deficient control of ingredients (3). At the time of the outbreak, tomatoes were chopped at chain A restaurants.

The results demonstrate relationships between food safety violations cited during routine inspections and foodborne illness. Having a protection from contamination violation, specifically “food contact surfaces: cleaned and sanitized” was cited more frequently in outbreak restaurants and most frequently in restaurants with at least three cases. Among the 10 outbreak restaurants with a “food contact surfaces: cleaned and sanitized” violation, 7 (70%) also had a violation for “wiping clothes properly used and stored.” Salmonella has been shown to survive on both whole and chopped tomatoes (12) and on food contact surfaces (14). These findings suggest that food handling practices in restaurants may have increased pathogen contamination through cross-contamination during this outbreak.

Six routine inspections unrelated to the outbreak occurred during the outbreak window. Three of these routine inspections were in restaurants that had outbreak-associated cases. One of these restaurants, which had four cases associated with it, had nine RF violations cited during the routine inspection that occurred during the outbreak, eight of which were repeat violations. These violations included two protection from contamination violations and three potentially hazardous food time-temperature violations. The two other outbreak restaurants with routine inspections that occurred during the outbreak had “food contact surfaces: cleaned and sanitized” violations that were repeat violations. By contrast, only one of the nonoutbreak restaurants that had an inspection during the outbreak had a protection from contamination violation. Repeat violations indicate a pattern of unsafe food handling, and this finding supports the hypothesis that food handling practices may have increased contamination.

This study had some notable limitations. First, because of the small sample size related to the number of jurisdictions involved, we did not have the power to assess agency characteristics that might have affected the findings. However, no single agency accounted for more than 12 inspections. Second, it is possible that not all tomatoes received by chain A restaurants were contaminated, which could explain why some restaurants had outbreak cases while others did not. However, misclassification of a restaurant's exposure status to contaminated tomatoes should have reduced the likelihood of finding an association with food safety violations. Finally, there is the possibility of misclassification of restaurants by case counts. Although most cases were found through routine surveillance, some cases were tested at the Minnesota Department of Health as a part of an active case finding early in the investigation. As noted earlier, misclassification of restaurants likely would have led to a bias toward the null.

Reducing the burden of foodborne illness will pose an increasingly complicated challenge because of disruptive technologies that threaten the loss of pathogen subtyping information, limiting our ability to detect outbreaks and identify the causes of foodborne illness. As such, there is an increasing need to identify novel methods to use existing data to improve surveillance activities. Traditionally, restaurant inspection data have not been used as a hazard surveillance tool to inform prevention activities. This study suggests that information can be gleaned from routine inspection reports as surveillance data to help understand why outbreaks occur. If analyzed during an outbreak investigation, these routine surveillance data could be useful for identifying possible routes of transmission, such as the cross-contamination identified here. Furthermore, this work highlights the need for improved data systems. Unlike other surveillance data, there was no existing database from which to obtain the inspection data used in this study. Restaurant inspection reports were collected from 13 jurisdictions and were often obtained as scanned documents. The data were then manually abstracted from these documents to create an analytic data set. Because of the considerable time needed to create an analyzable database, inspection reports could not be used to inform an outbreak in real time. Improving public health infrastructure, including improving data sources and availability, is a Healthy People 2020 objective (7), and this research shows the value of improving restaurant inspection data systems as a tool to increase surveillance capacity to reduce the burden of foodborne illness.

ACKNOWLEDGMENTS

The authors thank Sarah Boneske for contributing to preliminary analyses for this project and Logan Ebeling and Nicklaus Koreen for technical assistance.

REFERENCES

REFERENCES
1.
Angelo,
K. M.,
Nisler
A. L.,
Hall
A. J.,
Brown
L. G.,
and
Gould
L. H.
2017
.
Epidemiology of restaurant-associated foodborne disease outbreaks, United States, 1998–2013
.
Epidemiol. Infect
.
145
:
523
534
.
2.
Brown,
L. G.,
Hoover
E. R.,
Selman
C. A.,
Coleman
E. W.,
and
Schurz Rogers
H.
2017
.
Outbreak characteristics associated with identification of contributing factors to foodborne illness outbreaks
.
Epidemiol. Infect
.
145
:
2254
2262
.
3.
Carrasco,
E.,
Morales-Rueda
A.,
and
García-Gimeno
R. M.
2012
.
Cross-contamination and recontamination by Salmonella in foods: a review
.
Food Res. Int
.
45
:
545
556
.
4.
Centers for Disease Control and Prevention
.
2019
.
Surveillance for foodborne disease outbreaks, United States, 2017, annual report
.
U.S. Department of Health and Human Services
,
Centers for Disease Control and Prevention, Atlanta
.
5.
Conference for Food Protection
.
2020
.
Food establishment inspection report
.
Available at: http://www.foodprotect.org/guides-documents/food-inspection/. Accessed 11 November 2019.
6.
Cruz,
M. A.,
Katz
D. J.,
and
Suarez
J. A.
2001
.
An assessment of the ability of routine restaurant inspections to predict foodborne outbreaks in Miami-Dade County, Florida
.
Am. J. Public Health
91
:
821
823
.
7.
Healthy People 2020
.
2019
.
2020 topics and objectives
.
8.
Hedberg,
C. W.,
Smith
S. J.,
Kirkland
E.,
Radke
V.,
Jones
T. F.,
Selman
C. A.,
and
the EHS-NET Working Group
.
2006
.
Systematic environmental evaluations to identify food safety differences between outbreak and nonoutbreak restaurants
.
J. Food Prot
.
69
:
2697
2702
.
9.
Irwin,
K.,
Ballard
J.,
Grendon
J.,
and
Kobayashi
J.
1989
.
Results of routine restaurant inspections can predict outbreaks of foodborne illness: the Seattle–King County experience
.
Am. J. Public Health
79
:
586
590
.
10.
Jones,
T. F.,
Pavlin
B. I.,
LaFleur
B. J.,
Ingram
A.,
and
Schaffner
W.
2004
.
Restaurant inspection scores and foodborne disease
.
Emerg. Infect. Dis
.
10
:
688
692
.
11.
Lee,
P.,
and
Hedberg
C. W.
2016
.
Understanding the relationships between inspection results and risk of foodborne illness in restaurants
.
Foodborne Pathog. Dis
.
13
:
582
586
.
12.
Ma,
L.,
Zhang
G.,
Gerner-Smidt
P.,
Tauxe
R. V.,
and
Doyle
M. P.
2010
.
Survival and growth of Salmonella in salsa and related ingredients
.
J. Food Prot
.
73
:
434
444
.
13.
Minnesota Department of Health
.
2016
.
Minnesota Department of Health 2015 foodborne, waterborne, and animal contact outbreaks and gastrointestinal outbreaks due to other transmission routes
.
St. Paul
.
14.
Moore,
G.,
Blair
I. S.,
and
McDowell
D. A.
2007
.
Recovery and transfer of Salmonella Typhimurium from four different domestic food contact surfaces
.
J. Food Prot
.
70
:
2273
2280
.
15.
Petran,
R. L.,
White
B. W.,
and
Hedberg
C. W.
2012
.
Health department inspection criteria more likely to be associated with outbreak restaurants in Minnesota
.
J. Food Prot
.
75
:
2007
2015
.
16.
Tauxe,
R. V.
2002
.
Surveillance and investigation of foodborne diseases: roles for public health in meeting objectives for food safety
.
Food Control
13
:
363
369
.