Table 3.
Performance of AutoML in the validation cohort, Changhai prospective cohorts, and Zhongda prospective cohorts at 95% sensitivity
| Cohort | Biopsy result | Total | Performance, % | |
|---|---|---|---|---|
| csPCa | Benign+nsPCa | |||
| Validation internal cohort, cut-off value= 23.8% | ||||
| AutoML probability > cut point | 195 | 353 | 548 | Sensitivity= 95.1 |
| AutoML probability <= cut point | 10 | 250 | 260 | Specificity= 41.5 |
| Total | 205 | 603 | 808 | PPV= 35.6, NPV= 96.2 |
| csPCa biopsy prevalence % | 25.4 | Fraction predicted negative | 32.2 | Missing= 4.9 |
| Changhai prospective cohort, cut-off value= 21.5% | ||||
| AutoML probability > cut point | 168 | 203 | 371 | Sensitivity= 95.5 |
| AutoML probability <= cut point | 8 | 71 | 79 | Specificity= 25.9 |
| Total | 176 | 274 | 450 | PPV=45.3, NPV=89.9 |
| csPCa biopsy prevalence % | 39.1 | Fraction predicted negative | 17.6 | Missing=4.5 |
| Zhongda prospective cohort, cut-off value= 24.3% | ||||
| AutoML probability > cut point | 93 | 98 | 191 | Sensitivity=95.9 |
| AutoML probability <= cut point | 4 | 64 | 68 | Specificity=39.5 |
| Total | 97 | 162 | 259 | PPV=48.7, NPV=94.1 |
| csPCa biopsy prevalence % | 37.5 | Fraction predicted negative | 26.3 | Missing=4.1 |
nsPCa Non-significant prostate cancer, csPCa Clinically significant prostate cancer, AutoML Automated machine learning, LR Logistic regression, NLR Negative likelihood ratio, NPV Negative predictive value, PLR Positive likelihood ratio, PPV Positive predictive value