Table 5.
Predictive validity of the miRNA signaturea for identification of infants with NOWS requiring pharmacologic treatment or prolonged hospitalization.
| Indices of validity2 | Model 1 | Model 2 | Model 3 | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Firth Model1 | Bias-adjusted | Firth Model1 | Bias-adjusted | Firth Model1 | Bias-adjusted | ||||
| Estimate | 95% CI | Estimate | Estimate | 95% CI | Estimate | Estimate | 95% CI | Estimate | |
| Need for pharmacologic treatment | |||||||||
| Accuracy, % correct of total | 93.1 | 82.8 | 93.1 | 86.2 | 93.1 | 84.5 | |||
| Sensitivity, % | 88.2 | (63.6; 98.5) | 70.6 | 94.1 | (71.3; 99.9) | 82.4 | 88.2 | (63.6; 98.5) | 70.6 |
| Specificity, % | 95.1 | (83.5; 99.4) | 87.8 | 92.7 | (80.1; 98.5) | 87.8 | 95.1 | (83.5; 99.4) | 90.2 |
| PPV, % | 88.2 | (63.6; 98.5) | 70.6 | 84.2 | (60.4; 96.6) | 73.7 | 88.2 | (63.6; 98.5) | 75.0 |
| NPV, % | 95.1 | (83.5; 99.4) | 87.8 | 97.4 | (86.5; 99.9) | 92.3 | 95.1 | (83.5; 99.4) | 88.1 |
| Prolonged hospitalization | |||||||||
| Accuracy, % correct of total | 96.6 | 86.2 | 98.3 | 94.8 | 96.6 | 91.4 | |||
| Sensitivity, % | 93.3 | (68.1; 99.8) | 73.3 | 100.0 | (78.2; 100.0) | 93.3 | 100.0 | (78.2; 100.0) | 86.7 |
| Specificity, % | 97.7 | (87.7; 99.9) | 90.7 | 97.7 | (87.7; 99.9) | 95.3 | 95.3 | (84.2; 99.4) | 93.0 |
| PPV, % | 93.3 | (68.1; 99.8) | 73.3 | 93.8 | (69.8; 99.8) | 87.5 | 88.2 | (63.6; 98.5) | 81.3 |
| NPV, % | 97.7 | (87.7,99.9) | 90.7 | 100.0 | (91.6; 100.0) | 97.6 | 100.0 | (91.3; 100.0) | 95.2 |
Model 1 includes signature miRNAs.
Model 2 includes signature miRNAs and gestational age.
Model 3 includes signature miRNAs, gestational age, and the type maternal MOUD.
CI confidence interval, MOUD medication for opioid use disorder, NPV negative predictive value, PPV positive predictive value.
aIncludes let-7d-5p, miR-128-3p, miR-30c-5p, miR-421, miR-584-5p for need for pharmacologic treatment and let-7b-5p, miR-10b-5p, miR-128-3p, miR-30c-5p, and miR-421 for prolonged hospitalization. For prolonged hospitalization, miR-10b-5p was dropped from Model 2 and miR-10b-5p, let-7b-5p, and miR-128-3p were dropped from Model 3 due to lack of convergence.
1Maximum likelihood estimates adjusted for small sample size using Firth's penalized likelihood approach.
2Based on using prediction probability point specified for each model (pharmacological treatment: Models 1 and 2: ≥ 0.46, Model 3: ≥ 0.60; prolonged hospitalization: Model 1: ≥ 0.55, Model 2: ≥ 0.45, Model 3: ≥ 0.40).