New Algorithms Could be Useful Detecting C Difficile Infections

March 22, 2022
Kenny Walter

Kenny Walter is an editor with HCPLive. Prior to joining MJH Life Sciences in 2019, he worked as a digital reporter covering nanotechnology, life sciences, material science and more with R&D Magazine. He graduated with a degree in journalism from Temple University in 2008 and began his career as a local reporter for a chain of weekly newspapers based on the Jersey shore. When not working, he enjoys going to the beach and enjoying the shore in the summer and watching North Carolina Tar Heel basketball in the winter.

The investigators assessed and validated 3 rule-based algorithms for C difficile infection detection.

New research has validated the value in using algorithms to detect new hospital-onset Clostridioides difficile infection (CDI).

A team, led by Suzanne Desirée van der Werff, PhD, Postdoctoral Researcher at Karolinksa Institutet, developed a new algorithm for healthcare-onset C difficile infections, which could help hospitals and healthcare facilities react quicker to potential infection outbreaks.

“Effective surveillance is important to register adverse events, to enable quick response to outbreaks, and to evaluate control measures,” the authors wrote. “However, most surveillance is based on time-consuming and resource-intensive manual review of patient records, which is also prone to subjective interpretation and surveillance bias.”

The Dataset

In the observational study, the investigators evaluated a set of fully automated rule-based surveillance algorithms that included free-text analysis, for healthcare-onset C difficile infections in hospitalized patients using electronic health records from the Karolinska University Hospital.

The investigators developed and assessed a trio of rule-based algorithms, based on ICD-10 code A04.7 (enterocolitis due to C difficile) during admission, positive stool sample with C difficile toxin or toxin-producing C difficile present during admission, and the results of algorithm 2 and CDI symptoms.

Overall, they used 561 positive stool samples between January 2011 and December 2011, while manually annotating 14,107 free-text medical notes ±3 days around 682 positive stool samples to identify all possible ways to describe diarrhea stool and pseudomembranous colitis.

The investigators analyzed all stool samples during admission for C difficile as a potential CDI episode. Each admission without stool samples also counted as 1 potential CDI episode.

The team defined positive stool samples as when C difficile toxin A and/or B or a toxin-producing C difficile organisms was detected, based on cell cytotoxicity assay and toxigenic culture.

They also used a seven-day infection window period for the presence of CDI symptoms.

Validating the Data

The algorithms were assessed against a validation data set of 750 randomly selected admissions between 2012-2013. The validation group included samples from patients admitted with at least 1 positive stool sample for C difficile (n = 287), patients admitted with only negative stool samples for C difficile (n = 213), and patients admitted with no stool samples analyzed for C difficile (n = 250).

Of the 750 patients included in the validation cohort, 641 were adults and 35.2% (n = 253) had a C difficile infection. In addition, 35.1% (n = 225) of the adults had a CDI and 35.9% (n = 28) of the pediatric patients had a CDI.

Patients with a CDI were older (median age, 69 vs 58 years), had a longer length of stay (median days, 17 vs 5), had a higher Charlson comorbidity index (median, 2 vs 1), and had a higher in-hospital mortality rate (9.5%; n = 24 versus 1.9%; n = 9; all P < .001) compared to the cohort of patients without a C difficile infection.

A manual record review of 983 potential CDI episodes confirmed 351 episodes, 26.4% (n = 260) were deemed new, 6.1% (n = 60) were ongoing, and 3.2% (n = 31) were recurrent.

However, the investigators only used new episodes for the algorithm performance evaluation.

Sensitivity and Specificity

The first algorithm had a sensitivity of 0.442 (95% CI, 0.381–0.504) in correctly identifying CDI episode, compared to sensitivities of 1.000 (95% CI, 0.999–1.000) and 0.992 (95% CI, 0.980–1.000) in the second and third algorithm, respectively. Both the second and third algorithm had a specificity of 1.000 (95% CI, 0.999-1.000).

Using algorithm 2, 12 patients in the validation set were misclassified as positive for CDI, compared to 6 patients using algorithm 3. In addition, 3 patients were misclassified as negative using algorithm 3.

“Fully automated algorithms based on microbiological data with or without free-text analysis of symptoms performed well for surveillance purposes in detecting CDI, whereas an algorithm based on ICD-10 code had insufficient sensitivity,” the authors wrote. “This inadequacy was related to poor recording of the CDI ICD-10 code despite positive stool tests and symptoms for CDI.”

Conclusions

However, overall the study shows the algorithms could have some value in a hospital setting.

“In conclusion, algorithms based on microbiological test results only are likely to perform well in hospitals with symptom indications for C. difficile testing, while free-text analysis of medical notes could improve surveillance algorithm performance if more liberal indications for C. difficile testing are used,” the authors wrote.

The study, “The accuracy of fully automated algorithms for surveillance of healthcare-onset Clostridioides difficile infections in hospitalized patients,” was published online in Antimicrobial Stewardship & Healthcare Epidemiology.


x