BACKGROUND: Since 2014, the Colorado Department of Public Health (CDPHE) has conducted both population-based and facility-specific, healthcare-associated infection (HAI) surveillance through the Centers for Disease Control and Prevention’s (CDC) National Healthcare Safety Network (NHSN) for Clostridium difficile (CD). Population based surveillance is conducted among residents of the 5-county Denver metropolitan area (Adams, Arapahoe, Denver, Douglas, and Jefferson counties) as part of CDC’s Emerging Infections Program (EIP), while NHSN surveillance is statewide. CDC annually publishes a toolkit for validation of CD data reported in NHSN. We aimed to assess the accuracy of NHSN-reported CD data and to identify gaps in surveillance using EIP databases and CDC’s toolkit for CD validation.
METHODS: Facilities and medical records were selected using CDC’s recommended guidelines for NHSN validation. Toxin-positive and molecular assay positive line lists from 2015 were obtained from laboratories at selected facilities. Positive tests were matched by specimen collection date, birth date and gender with NHSN’s CD data. These data were matched with CD data from EIP. A case was defined as non-matching if it differed by one or more variables. Non-matching cases were further investigated and a percentage was calculated for each facility. Medical records of randomly selected NHSN CD events were reviewed to identify: non-reported events (met criteria but not reported in NHSN); over-reported events (did not meet criteria but reported in NHSN); criteria misinterpretations; surveillance implementation errors; and other discrepancies.
RESULTS: Twenty facilities were targeted; 13 facilities were also part of EIP surveillance. To date, nine facilities (6 participating in both surveillance approaches) have been audited. Of 6 facilities participating in both surveillance approaches, the percent of unmatched cases between NHSN and EIP ranged from 1-9%. Reasons for discrepancies included data entry errors and cases not meeting NHSN CD event criteria. Six (67%) facilities had data-mining software with built-in NHSN rules and definitions for CD event identification. We identified 8 non-reported and no over-reported events; 7 (88%) non-reported events were from facilities with no data mining software and performed >1 CD test during an episode of care.
CONCLUSIONS: Comparing NHSN data against population-based surveillance provided more credibility to Colorado’s HAI surveillance for CD. Use of electronic data-mining software and restricting CD testing during an episode of care may minimize reporting errors and enhance quality of NHSN-reported data.