Table 2.
Prediction of autonomic failure through machine learning and single ABPM parameters
| Accuracy | AUC | Specificity | Sensitivity | PPV | NPV | |
|---|---|---|---|---|---|---|
| Machine learning (all ABPM variables) | ||||||
| LDA | 91% (83–96) | / | 98% (90–100) | 79% (60–92) | 96% (77–99) | 89% (80–94) |
| ABPM predictive variables (categorical) | ||||||
| ≥ 3 Hypo-epΔ15/24h | 82% (71–90) | / | 100% (93–100) | 52% (33–71) | 100% | 77% (70–83) |
| Reverse dipping | 79% (68–87) | / | 90% (79–97) | 59% (39–76) | 77% (58–89) | 79% (71–86) |
| Hypo-awΔ15/24h | 74% (63–84) | / | 92% (80–98) | 45% (26–64) | 76% (54–90) | 74% (67–80) |
| SD d-SBP (> 16 mmHg) | 71% (60–81) | / | 92% (81–98) | 35% (18–54) | 71% (46–88) | 71% (65–77) |
| w-BPV (> 11 mmHg) | 57% (46–68) | / | 51% (37–65) | 69% (49–84) | 44% (36–54) | 74% (61–84) |
| PPH | 56% (44–67) | / | 54% (39–68) | 59% (39–76) | 43% (33–53) | 69% (58–79) |
| ABPM predictive variables (continuous) | ||||||
| aDiurnal SBP | / | 0.62 (0.49–0.75) | / | / | / | / |
| aDiurnal MBP | / | 0.55 (0.42–0.69) | / | / | / | / |
| aDiurnal DBP | / | 0.58 (0.46–0.71) | / | / | / | / |
| Nocturnal SBP (cut-off 123 mmHg) | 74% (67–80) | 0.72 (0.60–0.84) | 90% (79–97) | 45% (26–64) | 72% (51–87) | 74% (67–80) |
| Nocturnal MBP (cut-off 95 mmHg) | 75% (64–84) | 0.73 (0.61–0.85) | 96% (87–99) | 38% (21–58) | 85% (57–96) | 73% (67–78) |
| Nocturnal DBP (cut-off 75 mmHg) | 74% (63–83) | 0.67 (0.54–0.80) | 92% (81–98) | 41% (24–61) | 75% (52–89) | 73% (67–79) |
Autonomic failure (AF+) was used as the outcome. The predictive power of each ABPM variable was calculated through a 2 × 2 contingency table for dichotomous variables (Hypo-awΔ15/24h, ≥ 3 Hypo-epΔ15/24h, postprandial hypotension, reverse dipping pattern, high weighted blood pressure variability) and through the ROC curve for continuous variables (diurnal and nocturnal blood pressure values). The accuracy of the continuous variables refers to the cut-point of the ROC curve with the best sensitivity–specificity compromise (123 mmHg for SBP, 95 mmHg for MBP, 75 mmHg for DBP)
LDA: linear discriminant analysis; ABPM: ambulatory blood pressure monitoring; Hypo-awΔ15/24h: awakening hypotension; Hypo-epΔ15/24h: hypotensive episodes; SD d-SBP: standard deviation of diurnal systolic blood pressure; w-BPV: weighted blood pressure variability; PPH: post-prandial hypotension; AUC: area under the curve; PPV: positive predictive value; NPV: negative predictive value; SBP: systolic blood pressure; MBP: mean blood pressure; DBP: diastolic blood pressure
aThe ROC-curve output with diurnal BP value was obtained by inverting the outcome (AF−) to have an AUC greater than 0.5; accuracy metrics have not been reported being not significant