Table 1.
Performance Characteristics of 3 Rule-Based Algorithms for Classifying Healthcare-Onset Clostridioides difficile Infection According to the ECDC Definition
| Algorithm | Among 983 Potential CDI Episodes in 750 Admissions in the Validation Data Set, No. | Extrapolated Results to All Hospitalized Patients During the Validation Period (CDI-Episodes, N = 180,715) |
|||||||
|---|---|---|---|---|---|---|---|---|---|
| True Positives | False Positives | False Negatives | True Negatives | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) | AUC (95% CI) | |
| Algorithm 1 a | 115 | 69 | 145 | 654 | 0.442 (0.381–0.504) | 0.999 (0.998–0.999) | 0.531 (0.397–0.682) | 0.998 (0.998–0.998) | 0.720 (0.701–0.739) |
| Algorithm 2 b | 260 | 12 | 0 | 711 | 1.000 (0.999–1.000) | 1.000 (0.999–1.000) | 0.956 (0.932–0.978) | 1.000 (0.999–1.000) | 0.999 (0.999–1.000) |
| Algorithm 3 c | 258 | 6 | 2 | 717 | 0.992 (0.980–1.000) | 1.000 (0.999–1.000) | 0.978 (0.958–0.992) | 1.000 (0.999–1.000) | 0.996 (0.993–1.000) |
Note. ECDC, European Center for Disease Control; AUC, area under the receiver operating characteristic (ROC) curve; NPV, negative predictive value; PPV, positive predictive value. CDI-episode: all stool samples analyzed for C. difficile during admission were regarded as potential CDI-episodes, and each admission without stool samples analyzed for C. difficile counted as one potential CDI-episode. The extrapolated results of the algorithms from the validation data set to the validation period cohort (2012–2013) were based on the sampling proportion of potential CDI episodes from the 3 different groups: (1) admissions with a positive stool sample for C. difficile; (2) admissions with only negative stool samples for C. difficile and; (5) admissions without stool samples analyzed for C. difficile.
Algorithm 1: ICD-10 code A04.7.
Algorithm 2: Positive stool sample with C. difficile toxin or toxin-producing C. difficile: 12 false-positive cases due to having positive stool sample, but no symptoms (n = 8) and misclassification of CDI by annotator being true-positive cases (n = 4).
Algorithm 3: Positive stool sample with C. difficile toxin or toxin-producing C. difficile and CDI symptoms: 6 false positives due to negation rule for symptoms not working (n = 1), wrong detection of symptom by regex (n = 2) and misclassification of CDI by annotator being true-positive cases (n = 3); 2 false-negative cases due to symptoms missed by algorithm (n = 1) and misclassification of CDI by annotator being true-negative case (n = 1).
Algorithm 1 could only differentiate between (new) CDI or no CDI and classified 184 (18.7%) as CDI episodes. Algorithms 2 and 3 classified, 272 (27.7%) and 264 (26.9%) as new CDI episodes, 61 (6.2%) and 60 (6.1%) as ongoing CDI episodes, and 33 (3.4%) and 32 (3.3%) as recurrent CDI episodes, respectively. The results in the table are based only on new CDI episodes.