BACKGROUND: Mold infections such as aspergillosis and mucormycosis are highly fatal and have caused high-profile outbreaks recently, yet little is known about their epidemiology because no U.S. surveillance has been conducted in >20 years. To fill this need, we designed a pilot mold infection surveillance system to operate within the Georgia Emerging Infections Program (EIP), with potential for future expansion to other EIP sites. Because up to half of mold infections are not detected by culture, the surveillance system was designed to capture both culture and pathologic specimens. This presents challenges because EIP surveillance has traditionally relied on microbiology-based case identification, and reviewing narrative text-based pathology reports is challenging and time-intensive. Therefore, we developed a mold-specific automated strategy to identify mold infections from pathology reports.
METHODS: We screened pathology reports from June 1–December 1, 2016, from Emory University Hospital and affiliated anatomic pathology laboratories, in order to design and test a text search algorithm to identify specimens from patients with mold infections. Using the Cerner CoPath electronic record query function, we identified pathology reports with a final diagnosis or addendum field containing any of the terms “fung*”, “hypha*”, “Aspergillus”, “Mucor*”, and “mold”. From remaining reports, we excluded those with final diagnosis or addendum fields containing any of the terms “no fung*”, “no evidence of fung*”, “fungoides”, “candida”, and “onychomycosis”. We manually reviewed remaining records and excluded reports with nail fungus, yeast, and those that specified no fungi were detected. Specimen type was not included in automated searches or review.
RESULTS: Of total 46,154 pathology reports of specimens collected during the 6-month timeframe, we identified 1,416 (3%) with final diagnosis or addendum fields containing the first set of fungal search terms. Of these, 1,139 (80%) contained ≥1 of the excluded terms. Of the remaining 277 (20%) reports of possible mold-containing specimens, 82 (6%) cases meeting our case definition were identified by manual review.
CONCLUSIONS: A text-based search strategy showed that mold is rare among pathology specimen reports, and only 20% of resulting reports required review. Despite efforts to use automated searches to identify mold infection specimens exclusively, manual review was required to overcome limitations in searching for complex phrases and can be modified to further improve specificity (e.g., excluding non-invasive mold colonization). Based on these results, use of pathology reports appears feasible for conducting public health mold surveillance, and the pilot could be expanded to additional hospitals for population-based surveillance.