FoodNet is a collaborative program among the Centers for Disease Control and Prevention, 10 participating state health departments, the US Department of Agriculture’s Food Safety and Inspection Service, and the US Food and Drug Administration. FoodNet began in 1996 with 5 sites (California, Connecticut, Georgia, Minnesota, and Oregon) and by 2004 had grown to 10 sites (with the addition of Colorado, Maryland, New Mexico, New York, and Tennessee). FoodNet sites conduct active, population-based surveillance for pediatric HUS (HUS in children <18 years of age) among residents of the FoodNet catchment areas. Until 2007, cases of HUS without a history of diarrhea preceding the illness were also included in surveillance. Cases of thrombotic thrombocytopenic purpura (TTP) were also included because TTP diagnosed after a diarrheal illness is usually caused by infection with STEC O157 or another STEC, and thus likely represents HUS [8].

A total of 627 physician-diagnosed pediatric HUS cases were reported to the FoodNet HUS surveillance system during 2000–2007. An average of 78 cases was reported each year (range, 54–110). Most cases (66%; 416) were in children <5 years old; of these, 64% were in children <2 years old. Fifty-six percent (350) were female. The overall incidence rate was 0.78 per 100 000 children. Incidence was highest in children <5 years old (1.9 per 100 000 persons). The highest incidence rates were in Oregon (1.3), Minnesota (1.1), and Tennessee (1.0); the lowest rates were in Georgia and Maryland (0.4).

Of all reported cases, 326 (52%) were classified as confirmed, 53 (8%) as probable, and 248 (40%) as suspected (Table 1). Confirmed, probable, and suspected cases were reported every year. The proportion of cases classified as confirmed in each site ranged from 40% to 67%, the proportion of cases classified as probable ranged from 3% to 30%, and the proportion classified as suspected ranged from 29% to 49%. Among the 248 suspected cases, evidence of microangiopathic changes was the criterion most commonly not met or missing (50% not met, 5% missing), followed by evidence of renal injury, (37% not met, 9% missing), or anemia (31% not met, 8% missing). Sixty-nine (28%) suspected cases did not report diarrhea in the 3 weeks before HUS diagnosis and an additional 5 were missing information on this criterion. Among suspected cases, there were 176 (71%) with 1 criterion not met or missing, 49 (20%) with 2 criteria not met or missing, 2 (8%) with 3 criteria not met or missing, and 3 (1%) with 4 criteria not met or missing.

We compared the crude incidence rates by year for confirmed cases, confirmed and probable cases, and confirmed, probable, and suspected cases (Figure 1A). Regardless of the case definition used, the year-to-year pattern of incidence appeared similar. When both confirmed and probable cases were included, the annual incidence rate was an average of 16% (range, <1%–32%) higher than the rate when only confirmed cases were used. However, the annual incidence rate when all cases (confirmed, probable, and suspected) were included was an average of 92% (range, 81%–119%) higher than the rate calculated with only confirmed cases. Inclusion of both probable and suspected cases did not change the rank order of incidence among sites or in children <5 years.

Two FoodNet sites (Connecticut and Minnesota) had state restrictions to medical record access that limited their ability to conduct HDD review, and therefore, 123 (20%) cases were excluded from our comparison of ascertainment methods. Among the 504 cases for which a method of ascertainment was verified, 76% were captured by provider-based surveillance and 68% were captured by HDD review (Figure 2). Two hundred forty-nine (49%) were identified independently by both provider-based surveillance and HDD review, 136 (27%) were found only by provider-based surveillance, and 96 (19%) were found only by HDD review (Figure 2). An additional 23 cases were found by other means (ie, during an outbreak or during routine STEC surveillance). The proportion of cases identified by provider-based surveillance that were confirmed or probable (55%) was slightly higher than the proportion identified by HDD review (49%), but this difference was not statistically significant (Table 1). At 4 sites (Colorado, Georgia, Maryland, and New York), provider-based surveillance identified a disproportionately greater number of confirmed and probable cases, while in one (Oregon), HDD review identified more cases (Figure 3). The percentage of cases found by both methods ranged from 19% in Colorado to 89% in Oregon. When the crude incidence rates of cases captured by provider-based surveillance and HDD review were compared, the year-to-year pattern appeared similar from 2000 to 2006. In 2007, there was a higher rate among cases captured by HDD review than provider-based surveillance (Figure 1B).

Of the 615 reported HUS cases with complete information on dates of illness onset and case ascertainment, the median duration from the date of illness onset to the entry of data into the surveillance system was 2 months. There were 470 (76%) cases entered into the system within 1 year of diagnosis and 535 (85%) within 2 years. Among cases with a verified method of ascertainment, 79% of those identified by HDD review were entered into the surveillance system within 2 years, compared with 95% of those captured by provider-based surveillance. Most (315, 82%) cases captured by provider-based surveillance were entered into the HUS surveillance system within 6 months of diagnosis.

We compared demographics, illness severity, and laboratory test findings by case definition and method of ascertainment (Table 2). Compared with suspected cases, a higher proportion of confirmed cases were in children <5 years old (71% versus 59%, P < .01) and occurred in the summer months (48% versus 33%, P < .01). Confirmed cases were also more likely than suspected cases to undergo dialysis (57% versus 34%, P < .0001) and to have a stool specimen obtained (97% versus 79%, P < .0001) or one cultured for STEC O157 (93% versus 72%, P = .04). Cases identified by provider-based surveillance were more likely than those identified by HDD review to occur during the summer months (48% versus 41%, P = .04), to have had bloody diarrhea (77% versus 66%, P < .01), to have any stool specimen obtained (94% versus 87%, P < .01), and to have a stool specimen that was tested for Shiga toxin (40% versus 31%, P < .01).

The findings of the FoodNet HUS surveillance system were affected by key characteristics of the surveillance system, including the methods for case ascertainment and classification. However, year-to-year patterns were similar for all methods examined, suggesting that although the sensitivity and specificity of the various approaches for HUS surveillance vary, several methods can be used to track trends. For FoodNet, a central goal of HUS surveillance is to provide accurate information on trends over time to monitor progress toward meeting public health goals. Selection of the optimal surveillance strategy for HUS depends on balancing the best possible sensitivity and positive predictive value with available resources. Because surveillance for syndromes such as HUS is challenging because of the lack of a single diagnostic test, detailed clinical and laboratory information is needed to validate the reported diagnosis. Additionally, the performance characteristics of the case definition will vary based on the level of clinical and laboratory detail that are available. Thus, it is important to evaluate the various factors that affect the findings of the system by using different surveillance methods to find cases.

The magnitude of the incidence rate reported by the FoodNet HUS surveillance system varied by as much as 92% depending on how cases were classified. Although all persons with HUS were identified through a provider network or were found using hospital discharge codes for HUS and related conditions, almost half were classified as suspected using our definitions. However, regardless of the case definition, the interpretation of trends remained the same, suggesting that the system would yield similar incidence patterns independent of ascertainment methods or case definition. The use of a more sensitive, less specific case definition for HUS that did not require laboratory evidence of HUS likely led to the inclusion of some persons without HUS, as well as many with less severe HUS, as evidenced by the considerably lower dialysis requirements of children meeting only the suspected case definition. Conversely, the use of a less sensitive, more specific case definition (confirmed cases only) would likely result in a failure to include some true HUS cases. All case definitions showed similar changes in HUS rates at FoodNet sites over time.

Provider-based surveillance was the more sensitive method of case ascertainment, contributing 28% more cases than HDD review alone. However, using HDD review to supplement active surveillance increased the total number of cases by 19%. In a perfect world, all cases of HUS would be identified by both methods. However, we observed considerable site-to-site variation in the proportion of cases found by each method, illustrating potential gaps in surveillance that require improvement. Using multiple methods of case ascertainment periodically can also help to validate the effectiveness of surveillance.

Cases found by HDD review were less likely to be confirmed than those found by provider-based surveillance. This was likely due, in part, to the use of broad discharge codes to identify potential HUS cases for surveillance purposes. Because the FoodNet form did not collect information on the specific codes that were used to identify each case, it was not possible to look at the sensitivity of each code by the level of case classification. However, most of the other case characteristics were similar, regardless of the mode of ascertainment, suggesting that either method will capture a similar cross-section of cases. Differences in the proportion of cases where stool was tested for Shiga toxin indicate that providers should be informed about the importance of testing for Shiga toxin.

Several factors determine the optimal strategy for HUS surveillance. While provider-based surveillance is timelier than HDD review, extensive resources are necessary to establish and maintain relationships with healthcare providers. Although less timely, HDD review can be a valuable surveillance tool for illnesses that frequently result in hospitalization, are rare, or have severe outcomes [14–16]. HDD review can be done only once annually, although issues of restricted access to medical records can delay the process. In the case of some FoodNet sites, this resulted in delays of >2 years until a case was reported, or not having access to the patient-level clinical data needed to confirm the HUS diagnosis because some sites were only allowed access to aggregate data on HUS cases that were identified through HDD review. These issues might make HDD review prohibitive for other states that are considering strengthening or expanding HUS surveillance.

Because we limited our analysis only to the surveillance techniques used by the FoodNet surveillance system, our findings might not be comparable to other studies that found cases of HUS using other case definitions [17–21]. Despite the use of 2 surveillance techniques, it is likely that we missed cases and that other surveillance mechanisms could further enhance case finding. At some sites, the majority of cases were identified by only one method, suggesting that there may have been other cases that were missed. Surveillance for HUS is challenging, ideally requiring the use of multiple methods to detect trends over time and ascertain all cases. Health departments should consider available resources and data needs when choosing the optimal surveillance strategy.