Discussion

According to these findings from Connecticut, culture-based surveillance and coded hospital discharge data show substantial differences in trends of case counts of MRSA BSIs and MRSA septicemia hospitalizations from 2010 to 2018. The discrepancies identified in this analysis are consistent with results of other analyses showing different trends in MRSA BSI incidence when administrative data are used.^{Reference Kourtis, Hatfield and Baggs1,Reference Dantes, Mu and Belflower2,Reference David, Medvedev, Hohmann, Ewigman and Daum5–Reference Klein, Mojica and Jiang7} Reasons suggested by the authors of these publications include limitations of administrative data and differences in the areas under surveillance.^{Reference David, Medvedev, Hohmann, Ewigman and Daum5–Reference Klein, Mojica and Jiang7} Recently published papers seem to hold code-based surveillance methodology as a benchmark for assessing progress in clinical outcomes related to MRSA infection, as long as the limitations are well described.^{Reference Klein, Mojica and Jiang7} We have shown that even when using data from the same set of hospitals, administrative data do not approximate results from an audited active laboratory-based surveillance system. When assessing MRSA BSI case counts, there is no consistent trend among hospital discharge data that mirrors trends seen in EIP, both in this analysis and overall.^{Reference Kourtis, Hatfield and Baggs1,Reference Dantes, Mu and Belflower2,Reference David, Medvedev, Hohmann, Ewigman and Daum5–Reference Klein, Mojica and Jiang7} As noted in the results, case counts and trends appear to be more similar between MRSA septicemia coding and blood cultures in later years, but this does not appear to be directly aligned with the ICD-10-CM transition (see the Supplemental Material online). Ongoing monitoring of recent trends will be important prior to assuming that measuring trends with administrative codes is appropriate.

The EIP surveillance and administrative data sources may not capture the same infections. This discrepancy could be related to differences in the criteria used by administrative coders to define septicemia and the EIP surveillance definition for MRSA bloodstream infection (ie, positive blood cultures).^{Reference Klein, Mojica and Jiang7} There are clinical differences between septicemia and BSIs; however, researchers commonly use septicemia codes in administrative data as a surrogate for BSIs.^{Reference Inagaki, Lucar, Blackshear and Hobbs8} However, if non-BSI MRSA positive cultures were included as MRSA septicemia in the administrative data, we would expect administrative data to estimate more MRSA BSIs than EIP surveillance, opposite of our studyʼs findings. Additional potential causes for the discrepancy in the cases identified include that positive cultures are obtained in only 50% of sepsis cases and that coding for sepsis has increased over time, even when sepsis incidence measured by objective clinical data has not.^{Reference Novosad, Sapiano and Grigg9,Reference Rhee, Dantes and Epstein10}

Our analysis has several limitations. Connecticut hospital discharge data only capture the first ten discharge diagnosis codes. However, sensitivity analyses indicate that the lack of diagnosis codes after the tenth position does not meaningfully explain the discrepancies in case counts and results seen because <5% of MRSA septicemia codes occurred after the tenth position (see the Supplemental Material online).

Our comparison again highlights concerns regarding the accuracy and credibility of using administrative data to track changes in MRSA BSI incidence over this 9-year period. Validations of S. aureus administrative coding in both the ICD-9-CM and ICD-10-CM eras are needed to evaluate the appropriateness of using administrative codes in epidemiological research for other purposes. For example, a recent study published using administrative data alone found significant associations with increased in-hospital mortality, length of stay, and 30-day readmission for patients with MRSA bacteremia compared with MSSA bacteremia, citing Klein et al as precedent for using ICD-9-CM codes for identification.^{Reference Klein, Mojica and Jiang7,Reference Inagaki, Lucar, Blackshear and Hobbs8} The impact that potential misclassification of S. aureus identification has on studies evaluating clinical outcomes is unclear. The concerning discrepancy between trends in case counts from Connecticut suggest that administrative data do not correctly assess historical trends in MRSA BSIs, and validation of codes should be conducted before future use in studies of MRSA epidemiology.