Background
Methods
Patient cohorts
The machine learning algorithms
Simple logistic regression
Multitask temporal logistic regression (MTLR)
Patient specific survival prediction modeling (PSSP)
Data preparation and model building
Data preparation
Predictor variables
Data splitting
Hyperparameters optimization
Cost function optimization
Model validation
Results
Variables | IDI cohort (n = 484) | EFV cohort (n = 233) |
---|---|---|
WT/kg (median [IQR]) | 55.0 [48.0–61.0] | 51.0 [47.0–58.0] |
AGE/ years (median [IQR]) | 35.0 [30.0–41.0] | 33.0 [30.0–40.0] |
CD4 / cell per ml (median [IQR]) | 100.0 [29.7–166.0] | 109.0 [46.0–179] |
VL*1000 copies per ml (median [IQR]) | 349 [116.5–595.2] | 123.7 [42.7–253.7] |
SEX (male %) | 30.4 | 44.6 |
TB / %(n) | 7 | 57.5 |
REGIMEN 1 d4T/3TC/NVP-30 (%) | 49.5 | 0 |
REGIMEN 2 d4T/3TC/NVP-40 (%) | 24.7 | 0 |
REGIMEN 3 AZT/3TC/EFV (%) | 25.6 | 100 |
MODEL | AUROC(SE) | AUPRC | F1 | % ACCURACY | % TP | % TN | % FP | % FN | BRIER |
---|---|---|---|---|---|---|---|---|---|
MTLR | 0.9204 (0.0186) | 0.8706 | 0.9194 | 93.76 | 50.66 | 43.14 | 5.98 | 0.24 | 0.0814 |
PSSP | 0.75 (0.027) | 0.6584 | 0.7684 | 81.4 | 46.52 | 34.86 | 18.12 | 0.5 | 0.1974 |
SLR | 0.538 (0.1042) | 0.8752 | 0.937 | 57.94 | 49.54 | 8.4 | 3.58 | 38.44 | 0.1072 |
MODEL | AUROC | AUPRC | F1 | % Accuracy | % TP | % TN | % FP | % FN |
---|---|---|---|---|---|---|---|---|
MTLR | 0.878 (0.016) | 0.892 | 0.93 | 92.9 | 66.1 | 26.8 | 6.9 | 0.2 |
PSSP | 0.824 (0.02) | 0.817 | 0.921 | 92.3 | 66.3 | 26.0 | 7.7 | 0 |
SLR | 0.497 (0.09) | 0.938 | 0.971 | 24.6 | 21.5 | 3.1 | 2.6 | 72.8 |