Baseline incidence was defined as S. Enteritidis incidence in the year in which an EQAP was adopted in a state or group of states affected by the S. Enteritidis epidemic. We calculated S. Enteritidis incidence for a state or group of states as the number of reported human S. Enteritidis isolates in a year divided by that state's or group of states' population for that year expressed per 100,000 persons. We defined a state affected by the S. Enteritidis epidemic as one for which the S. Enteritidis incidence was >1/100,000 in any year between 1980 and 1999.

States that adopted EQAPs were grouped into state- and industry-sponsored EQAPs. We grouped EQAPs into state-sponsored and industry-sponsored types on the basis of whether the state government was actively involved in third-party monitoring, supervision, provision of technical advice, and procedure of handling houses that are found to be S. Enteritidis positive. In this study, state-sponsored EQAPs were defined as having active state Department of Agriculture and Department of Health involvement in providing technical advice, supervising and monitoring the programs, requiring third-party auditing, testing eggs for contamination with S. Enteritidis if houses were positive, and diverting eggs found to be contaminated with S. Enteritidis to pasteurization and hard cooking. Industry-sponsored EQAPs were defined as lacking state government involvement, recommending but not requiring third-party audits of the program, and recommending immediate extra cleaning of S. Enteritidis–contaminated houses upon depopulation of the houses (19).

We calculated the percentage change in annual S. Enteritidis incidence relative to the baseline (hereafter referred to as the change in S. Enteritidis incidence):

where, I stands for SE incidence; ΔRI for change in SE incidence; i = 1, 2,..., N for state; t = -T,...,-1, 0, 1,..., T for time, t₀ = year of EQAPs introduction

We then divided the change in S. Enteritidis incidence for a given year by the number of years before or after the intervention to get the annualized percentage change in S. Enteritidis incidence (hereafter referred to as the annualized change in S. Enteritidis incidence). We used two methods to examine changes in S. Enteritidis incidence: a simple change-point procedure and regression analysis.

We constructed a graph with a horizontal axis representing time in years and a vertical axis representing S. Enteritidis incidence (Figure 2A). If an intervention is effective, S. Enteritidis incidence should decrease after the baseline year (line b, Figure 2A). If the S. Enteritidis incidence had been increasing before the intervention, a smaller increase in incidence after the baseline would also show evidence of effectiveness (line c, Figure 2A). If the S. Enteritidis incidence had been decreasing before the intervention, we would expect a faster decrease in incidence after the baseline. Similarly, the lack of change in S. Enteritidis incidence observed before, during, and after the intervention would be evidence of lack of effect (line d, Figure 2A), and an increase in S. Enteritidis incidence (line e, Figure 2A) would be evidence that the intervention was associated with an acceleration of the epidemic.

The effectiveness of an intervention can further be assessed by using the annualized change in S. Enteritidis incidence (Figure 2B). To show evidence of an intervention's effectiveness, we would expect the annualized change in S. Enteritidis incidence after intervention to be below the baseline rate (dotted line b, Figure 2B) or the preintervention rate (dotted line b compared to a, Figure 2B). A more modest effect after intervention that shows the epidemic continuing to grow at a diminished annualized rate would find the change in incidence above baseline but below the preintervention level (dotted line c compared to a, Figure 2B). An ineffective intervention would result in an annualized change in S. Enteritidis incidence after intervention that is equal to the rate before intervention (dotted line d compared to a, Figure 2B). An annualized change in S. Enteritidis incidence above the preintervention rate (line e compared to a, Figure 2B) would be evidence that the intervention was associated with an acceleration of the epidemic. If changes in S. Enteritidis incidence continued at about the same rate or were sustained for a number of years, the annualized change in S. Enteritidis incidence would, in time, trend toward the baseline. We calculated Yates corrected chi-square values to verify whether changes in observations of annualized S. Enteritidis incidence were statistically significant.

We examined observations of pre-EQAP annualized rates of change in S. Enteritidis incidence for a period of up to 5 years and for a variable period of up to 8 years of post-EQAP observations. The time periods selected were considered to be long enough to include any relevant lag effects and short enough to exclude confounding influence of other interventions, such as those that require refrigerating eggs. To analyze the timing of the decrease in S. Enteritidis incidence, we grouped states on the basis of duration of postintervention follow-up and then type of EQAP and compared the annualized changes in S. Enteritidis incidence for each of the 5 years before adopting EQAPs and up to 5 years after adopting EQAPs. For example, states that adopted EQAPs in 1996 had 3 years of common experience with the intervention from 1996 to 1999 and formed a group based on this common length of time. The annualized changes in S. Enteritidis provided a measure for comparing incidence of S. Enteritidis before, during, and after EQAP adoption.

We also examined the percentage change in S. Enteritidis incidence by using a pooled regression model. The pooling method can be used to combine cross-section and time-series data. This technique allows for the error terms to have equal variance on the chosen values of the explanatory variables within a state, but unequal variance between states (20–25), which results in efficient and unbiased parameter estimates. We estimated a pooled regression equation for five cross-sectional states (Connecticut, Louisiana, Indiana, Pennsylvania, and California) for 5 years post-EQAP by using SHAZAM (20). The pooled regression equation was:

We included Louisiana in the regression model, although it was unaffected by the epidemic, to improve the degrees of freedom for the model. Only four of the states that were affected by the S. Enteritidis epidemic had 5 years of post-EQAP experience. Also, we included a dummy variable in the model to control for states that were not affected by the S. Enteritidis epidemic.

Independent variables (and types) were percentage of eggs produced in participating farms (continuous), type of EQAP (binary: state- or industry-sponsored: yes/no), number of United States Department of Agriculture (USDA) S. Enteritidis outbreak traceback investigations (continuous), proportion of population at high risk for S. Enteritidis (children <5 years and seniors >65 years) (continuous), and whether the state was in the northeast geographic region (Connecticut, Pennsylvania, New York) (binary: yes/no). The proportion of eggs produced under EQAPs by state and year was the index to measure participation in EQAPs.

We sent a detailed questionnaire to state veterinarians and public health officials in all states that were involved with S. Enteritidis control and prevention efforts. We also asked state officials to share the questionnaire widely with stakeholders (e.g., state departments of agriculture, laboratory workers, and egg industry officials) in state S. Enteritidis mitigation efforts.

The questionnaire collected data on whether egg producers in the state had adopted an EQAP, and if so, the type of EQAP to which most producers in the state adhered (e.g., industry-or state-sponsored), year of EQAP initiation, estimated proportion of total commercial layer-flock participation in the EQAP by year, and elements of the EQAP to which participants were required to adhere. The annual number of S. Enteritidis cases was obtained from reports by state and local health departments to the National Salmonella Surveillance System (1). Estimates of the annual population data for states were obtained from the Bureau of the Census of the U.S. Department of Commerce (26).

The annual numbers of eggs produced by state from 1972 to 1999 were obtained from USDA's National Agricultural Statistics Service (NASS) (27). To calculate the proportion of eggs produced under an EQAP for each state, we assumed no difference in egg production per layer between layers raised under an EQAP and those raised under no EQAP. We then calculated the proportion of eggs produced under each EQAP as a product of the proportion of total layer flocks that participated in the EQAP and the annual total number of eggs produced by each state. This calculation may overestimate the annual total number of eggs produced for human consumption. The category "table eggs" would provide a closer estimate of eggs produced for human consumption. However, due to confidentiality concerns, NASS does not publish complete information on table egg production.

To estimate a proxy for the proportion of the state's population at high risk for S. Enteritidis, we used the resident population <5 years of age and >65 years of age and total resident population. Data for estimates of the resident population by age and state for 1989–1999 were obtained from the U.S. Census Bureau (26). We obtained the number of successful S. Enteritidis outbreak traceback investigations (investigation to establish origin of S. Enteritidis–contaminated eggs) by state from USDA's S. Enteritidis Task Force Status Reports for 1990 to 1993 (8). Similar information was not available for tracebacks from 1996 to 1999, when the Food and Drug Administration was responsible for tracebacks. The typical procedure when a traceback leads to farms is for the regulatory body to take environmental samples of manure areas, egg belts and escalators, fans, and feed. If the environment tests positive for S. Enteritidis, the farmer can either divert the eggs to pasteurization or hard cooking for the lifetime of the flock, divert the eggs until they test negative for S. Enteritidis, or depopulate the flock.

Although a decline in prevalence of S. Enteritidis in layer-flock eggs might indicate effectiveness of EQAPs in mitigating S. Enteritidis (4,6,10,12,16,28), a connection with reductions of S. Enteritidis infections in humans is necessary to indicate effectiveness of the programs in mitigating human illness. Our simple change-point procedure showed a connection between the introduction of EQAPs at the state level and significant reductions in S. Enteritidis incidence in humans. The regression analysis found that increasing the quantity of eggs produced under EQAPs was associated with reducing S. Enteritidis incidence.

Several factors limited this study. Whether an EQAP was introduced at the beginning of the year or at the end of the year might make a difference, and defining a baseline year might introduce error in the analysis. However, data about the month in which EQAPs were introduced were lacking for most states that adopted these programs. Although some EQAPs are similar in that they were designed through close collaboration among states, they vary in practice and motivation, which limits generalizations about all EQAPs, whether state or industry sponsored. We found verifying the exact practices of each EQAP to be difficult because EQAPs range from self-certification programs, like the 5-Star United Egg Producers program (19) practiced in Indiana and Oregon that does not require microbiologic testing for chicks, pullets, layers, and eggs, to the more structured, regulated, rigorous, and costly Pennsylvania Egg Quality Assurance Program (14). Eleven of 15 states with EQAPs reported that they required periodic sampling and testing of layer environments, layers, and eggs for S. Enteritidis, but 4 (Oregon, Louisiana, Indiana, Georgia) did not. All states that required microbiologic testing, except for Florida, involved their state governments in setting up and monitoring their EQAP programs.

We did not study interactions in the regression model because of few data points and degrees of freedom, which limited the robustness of its results. We were more interested in the direction (positive or negative) of the estimate of the percentage of eggs produced under EQAPs than the magnitude. Also, our results were based on unverified respondent estimates of the proportion of eggs produced under EQAPs, information about the type of EQAP, and when the EQAP was instituted. Further, because accurately estimating prevalence of diabetes, cancer, HIV/AIDS, and pregnancy at the state level was difficult, we used the population of children <5 years of age and seniors >65 years of age for each state to represent the population at high risk for S. Enteritidis.

We assumed that eggs produced in a state are applied to meet the consumption needs of that state, and changes in S. Enteritidis incidence within the state would reflect the effect of the state's EQAP, but this assumption may not be accurate. Eggs in the United States are distributed widely across the nation through a dynamic system that makes it difficult to track the source and destination of eggs by state. Although data about the source and destination of eggs and egg products are desirable, they are not currently available (29).

Not all egg producers immediately join EQAPs, and the percentage of eggs produced in a state under an EQAP varies as producers adopt or leave EQAPs. The simple change-point analysis did not account for these variations and assumed that EQAPs were homogeneous within and among states. The regression model allowed EQAPs to be homogeneous within states and heterogeneous among states.

Our model could not estimate unreported cases in a meaningful way, although these cases constitute most cases of salmonellosis (14). The larger proportion of S. Enteritidis cases goes unreported (30). Other factors may have affected S. Enteritidis incidence in humans that we did not account for in this model because of lack of specific data, such as improvements in egg refrigeration during distribution and handling, traceback investigations from 1996 to 1999, and use of pasteurized eggs. However, these measures were not implemented in tandem with the EQAPs within or among states. Therefore, the close temporal association between implementing EQAPs and decreasing rates of S. Enteritidis infection indicate the importance of EQAPs as a control strategy.

The results of our study indicate that flock-based interventions have had a positive effect on health by reducing S. Enteritidis incidence in humans. These data further indicate that EQAPs probably played a major role in reducing S. Enteritidis illness in the United States. Considering that as of 1999, less than half of shell eggs in the United States were produced under EQAPs (Table 1), and that the number of cases and relative rate of S. Enteritidis have not shown significant decline since 1999, adopting EQAPS by producers and states would likely improve the public's health and prevent reemergence of egg-based Salmonella.