Detection of Pelvic Inflammatory Disease: Development of an Automated Case-Finding Algorithm Using Administrative Data
Table 3
Performance statistics for PID case identification using an algorithm applied to sample administrative data from women with an ICD-9 code* related to PID.
(a) Accuracy of PID case-finding algorithm: GH development dataset.
New algorithm classification
Total
Not PID
PID
Chart-confirmed Diagnosis
Not PID
34
40
74
PID
10
265
275
Total
44
305
349
(b) Accuracy of PID case-finding algorithm: KPCO validation dataset.
New algorithm classification
Total
Not PID
PID
Chart-confirmed Diagnosis
Not PID
34
58
92
PID
34
315
349
Total
68
373
441
(c)
Performance Statistics (95% CI)
GH development dataset
KPCO validation dataset
PID case identification using ICD-9 codes* alone
PPV
78.8% (74.1–83.0%)
79.1% (75.0–82.8%)
PID case identification using algorithm
Sensitivity
96.4% (93.4–98.2%)
90.3% (86.7–93.2%)
Specificity
45.9% (34.3–57.9%)
37.0% (27.1–47.7%)
PPV
86.9% (82.6–90.5%)
84.5% (80.4–88.0%)
NPV
77.3% (62.2–88.5%)
50.0% (37.6–62.4%)
Proportion of potential cases misclassified
14.3% (10.8–18.5%)
20.9% (17.2–25.0%)
*ICD-9 codes shown in Table 1. Only potential cases with complete chart-review information are included. GH: Group Health, KPCO: Kaiser Permanente Colorado, PID: pelvic inflammatory disease, CI: confidence interval, PPV: positive predictive value, NPV: negative predictive value.