Evaluation of Surveillance Methods for Staphylococcal Toxic Shock Syndrome

We compared passive surveillance and International Classification of Diseases, 9th Revision, codes for completeness of staphylococcal toxic shock syndrome (TSS) surveillance in the Minneapolis–St. Paul area, Minnesota, USA. TSS-specific codes identified 55% of cases compared with 30% by passive surveillance and were more sensitive (p = 0.0005, McNemar χ2 12.25).

ratory that tests staphylococcal isolates from TSS cases increased during 2000-2003 (8). To determine the incidence of TSS, active surveillance was initiated at all MSP area hospitals using International Classifi cation of Diseases, 9th Revision (ICD-9), codes assigned at hospital discharge. We compared passive surveillance reports with ICD-9 codes to determine an effective and effi cient surveillance method for TSS.

The Study
The MSP area is composed of 7 counties with a population of 2,642,056 (2000 US Census) and 24 acute-care hospitals. Requests were sent to medical record departments of these hospitals for data on inpatients discharged from hospitals from January 1, 2000, through December 31, 2003, whose medical records indicated >1 of the select ICD-9 study codes. Medical records from all hospitalizations receiving the TSS-specifi c code (040.82 or 040.89) were reviewed ( Figure), and a 20% random sample of medical records from hospitalizations that received >1 nonspecifi c TSS study code ( Of 7,414 hospitalizations with >1 study code, 116 (1.6%) were assigned the TSS-specifi c code and were reviewed ( Figure). Of the remaining 7,298 hospitalizations assigned >1 nonspecifi c TSS code, 1,575 (21.6%) randomly selected hospitalizations were reviewed. Of these 1,691 hospitalizations, 55 had 5 or 6 criteria for TSS, of which 12 (22%) met the CDC case defi nition for streptococcal TSS, and 7 were non-MSP residents. The remaining 36 cases were probable or confi rmed TSS. No cases from UNEX or death certifi cate searches met the TSS case defi nition. Of the 36 TSS cases, 17 (47%) were reported to MDH by passive surveillance. Thirty-one (86%) cases were found by using TSS-specifi c ICD-9 codes. Five cases were found by using non-TSS-specifi c ICD-9 codes. After adjusting for 20% random sampling for cases identifi ed by non-TSSspecifi c codes, we identifi ed the estimated number of cases by using non-TSS-specifi c codes to be 25. This analysis resulted in 56 estimated TSS cases identifi ed in the surveillance area during 2000-2003 by using ICD-9 codes.
The TSS-specifi c ICD-9 code search identifi ed 31 of the 56 estimated TSS cases (sensitivity 55%, specifi city

Conclusions
Surveillance for TSS is challenging given the lack of a diagnostic test and a case defi nition with multiple components. Under the current passive surveillance system, between one third and half of potential TSS cases were identifi ed. Discrepancies were found in reporting, with menstruation-associated cases more likely to be reported to MDH than nonmenstrual-associated cases. This discrepancy was observed with prior active surveillance efforts (10).
Using ICD-9 codes, we found 12 TSS cases that were of streptococcal etiology. Accuracy may be improved by developing separate ICD-9 codes specifi c for staphylococ- cal, streptococcal, or unidentifi ed TSS. In addition to the TSS-specifi c ICD-9 code, we selected 5 other ICD-9 codes on the basis of previous studies to address the concern that TSS cases may be classifi ed under a staphylococcal infection or sepsis code, but not the TSS-specifi c code (8-10). These 5 additional non-TSS-specifi c ICD-9 codes required reviewing 1,575 medical records; only 5 (0.3%) additional TSS cases were identifi ed. The non-TSS-specifi c ICD-9 codes detected 25 estimated cases. However, this detection required 8 trained staff and substantial resources with ≈40 minutes required per medical record review. Passive surveillance requires fewer public health resources because it relies on clinicians to report cases. Active surveillance involves public health resources in identifying cases. The disadvantage of passive surveillance is the potential for missed cases. Despite possible inaccuracies associated with the assignment of ICD-9 codes, these codes represent a standardized data source that may be readily available. In the absence of a specifi c diagnostic test, ICD-9 codes represent an effi cient method for surveillance and following trends.
Medical record abstraction per hospitalization was labor-and resource-intensive and is not feasible for most health departments. With increasing use of automated electronic reporting for disease surveillance (11), querying hospital discharge data for the TSS-specifi c ICD-9 code is a feasible adjunct to passive surveillance to detect TSS trends over time. Consequently, it is imperative that clinicians and coders are thorough to ensure that ICD-9 codes are accurate.
We found it useful to add regular ICD-9 code searches for TSS-specifi c codes as an active surveillance adjunct to our passive surveillance system. This addition increases sensitivity of TSS surveillance with a minimal increase in resources. Use of this more sensitive system increases the ability to detect trends in TSS, which may develop be-cause of changes in bacterial virulence characteristics, host characteristics such as the use of new devices or products, changes in human behavior, or changes in host susceptibility. Evaluation of this approach in other areas to assess sensitivity of TSS surveillance would be useful because coding practices may differ.