Table 1.
Overall performance of the A-stage classifiers used in the algorithm.
| Test | UPCR-based A-Stage Classifier |
DSP-based (Scale 1) A-Stage Classifier |
DSP-based (Scale 2) A-Stage Classifier |
|---|---|---|---|
| n = 19,099 paired measurements | n = 12,185 paired measurements | n = 43,486 paired measurements | |
| Squared error | 0.219 (0.213, 0.224) | 0.256 (0.251, 0.261) | 0.235 (0.23, 0.241) |
| Accuracy (95% CI) | |||
| A1 | 86.7% (86.4%, 87.0%) | 80.9% (80.5%, 81.3%) | 82.2% (81.8%, 82.5%) |
| A2 | 80.0% (79.7%, 80.3%) | 76.0% (75.6%, 76.4%) | 78.5% (78.1%, 79.0%) |
| A3 | 92.3% (92.0%, 92.6%) | 94.3% (94.1%, 94.4%) | 95.3% (95.1%, 95.5%) |
| Sensitivity (95% CI) | |||
| A1 | 86.2% (85.2%, 87.1%) | 90.9% (89.9%, 91.9%) | 93.2% (93.0%, 93.4%) |
| A2 | 63.5% (62.4%, 64.5%) | 41.4% (40.1%, 42.8%) | 35.7% (34.0%, 37.4%) |
| A3 | 86.7% (85.9%, 87.5%) | 83.3% (82.1%, 84.4%) | 81.0% (80.0%, 82.0%) |
| Specificity (95% CI) | |||
| A1 | 87.1% (86.4%, 87.7%) | 69.6% (68.6%, 70.5%) | 62.1% (61.1%, 63.1%) |
| A2 | 87.1% (86.6%, 87.5%) | 89.5% (88.7%, 90.2%) | 92.0% (91.9%, 92.2%) |
| A3 | 94.8% (94.5%, 95.1%) | 96.8% (96.5%, 97.1%) | 97.1% (97.0%, 97.3%) |
The 95% confidence intervals were calculated based on 10-fold cross-validation; DSP Scale 1: reported as Negative, Trace, 1 + , 2 + , 3 + , 4 + ; DSP Scale 2: reported as Negative, Trace, 10, 30, 100, 300, or ≥ 300; only the performance of the final pooled classifiers across Columbia University, University of Minnesota, and Vanderbilt University are summarized, for detailed breakdown of site-specific performance see Tables S1–S5, and for additional validation studies and comparisons with recently published methods by Sumida et al.24, see Table S6.