Table 3 Model performance comparing baseline clinical model (Model 1) and clinical model plus newborn screening analytes for prediction of neonatal sepsis.

From: Using newborn screening analytes to identify cases of neonatal sepsis

Model		c-statistic^a,b	c-statistic adjusted^a,b	Optimism correction^a,b	AIC (lower is better)	IDI^c (95% CI)	NRI^d (95% CI)
Term births (≥37 weeks’ gestation)
Model 1 ^e	Infant sex, gestational age, birth weight, plurality, and TPN	0.579	0.577	0.002	19372	—	—
Model 2	Model 1 variables + relative fetal-to-adult Hb level	0.580	0.577	0.003	19372	0.000075 (0, 0.0002)	0.010446 (−0.0271, 0.048)
Model 3	Model 2 variables + 17 − OHP + TSH	0.705	0.704	0.001	17977	0.063638 (0.0584, 0.0688)	0.55832 (0.521, 0.5957)
Model 4 ^f	Model 3 variables + restricted cubic spline terms for the top five ranked analytes/analyte ratios + remaining analytes/analyte ratios until maximum number of parameters was reached (maximum number of parameters: 294)	0.848	0.848^g	—^g	13788	0.21417 (0.2046, 0.2238)	0.97868 (0.9443, 1.013)
Top 10 analytes/analyte ratios ^h		Tyrosine; Tyrosine:Relative fetal-to-adult Hb level; Malonylcarnitine:Relative fetal-to-adult Hb level; Tetradecenoyl carnitine:Relative fetal-to-adult Hb level; Dodecenoylcarnitine:Phenylalanine; Tetradecenoyl carnitine; Malonylcarnitine; Glutarylcarnitine:Alanine; Carnitine:Dodecenoylcarnitine; Tetradecenoyl carnitine:Alanine
Late preterm births (34–36 weeks’ gestation)
Model 1 ^e	Infant sex, gestational age, birth weight, plurality, and TPN	0.684	0.683	0.001	7722	—	—
Model 2	Model 1 variables + relative fetal-to-adult Hb level	0.687	0.685	0.002	7711	0.00042 (0.0001, 0.0007)	0.10901 (0.04, 0.178)
Model 3	Model 2 variables + 17 − OHP + TSH	0.727	0.725	0.002	7588	0.003943 (0.0024, 0.0055)	0.34868 (0.279, 0.4184)
Model 4 ^f	Model 3 variables + restricted cubic spline terms for the top five ranked analytes/analyte ratios + remaining analytes/analyte ratios until maximum number of parameters was reached (maximum number of parameters: 80)	0.801	0.782	0.019	7086	0.016826 (0.0146, 0.019)	0.69197 (0.6274, 0.7565)
Top 10 analytes/analyte ratios ^h		Dodecanoylcarnitine:Relative fetal-to-adult Hb level; Decanoylcarnitine; Dodecanoylcarnitine; Dodecenoylcarnitine:Relative fetal-to-adult Hb level; Dodecenoylcarnitine:Phenylalanine; Decanoylcarnitine:Relative fetal-to-adult Hb level; Valerylcarnitine:Methylglutarylcarnitine; Malonylcarnitine:17-OHP; Methylglutarylcarnitine; Decanoylcarnitine:Phenylalanine
Early preterm births (<34 weeks’ gestation)
Model 1 ^e	Infant sex, gestational age, birth weight, plurality, and TPN	0.653	0.650	0.003	6696	—	—
Model 2	Model 1 variables + relative fetal-to-adult Hb level	0.653	0.649	0.004	6695	0.000258 (−0.0002, 0.0007)	0.079332 (0.0166, 0.142)
Model 3	Model 2 variables + 17 − OHP + TSH	0.658	0.654	0.004	6678	0.002719 (0.0014, 0.0041)	0.074809 (0.0107, 0.1389)
Model 4 ^f	Model 3 variables + restricted cubic spline terms for the top five ranked analytes/analyte ratios + remaining analytes/analyte ratios until maximum number of parameters was reached (maximum number of parameters: 102)	0.701	0.667	0.034	6559	0.016748 (0.0136, 0.0199)	0.33787 (0.2746, 0.4011)
Top 10 analytes/analyte ratios ^h		Methylglutarylcarnitine; Valerylcarnitine:Methylglutarylcarnitine; Tyrosine:Valine; Methylmalonylcarnitine:Phenylalanine; Dodecanoylcarnitine:Valine; Methylmalonylcarnitine:Valine; Dodecanoylcarnitine:17-OHP; Octadecenoylcarnitine:Arginine; Dodecanoylcarnitine:Phenylalanine; Hexadecanoylcarnitine:Arginine

Abbreviations: AIC: Akaike Information Criterion; CI: confidence interval; Hb: hemoglobin; IDI: Integrated Discrimination Improvement; NRI: Net Reclassification Improvement; 17-OHP: 17-hydroxyprogesterone; TPN: total parenteral nutrition; TSH: thyroid stimulating hormone
^aC-statistic is equivalent to the area under the curve (AUC).
^bAdjusted c-statistic is based on internal validation results using 200 bootstrap samples.
^cIntegrated Discrimination Improvement (IDI) quantifies the impact of additional variables on the average sensitivity of the preceding nested model (e.g., the IDI shown for Model 2 compares Model 2 with Model 1, etc.).
^dNet Reclassification Improvement (NRI) quantifies the net increase/decrease in predicted values for the outcome compared to the preceding nested model (e.g., the NRI shown for Model 2 compares Model 2 with Model 1, etc.).
^eBaseline model containing only clinical variables.
^fFinal fitted model.
^gC-statistic is unadjusted due to non-convergence of optimism-corrected model.
^hRank ordered by absolute value of the regression parameter in Model 4.

Back to article page

Table 3 Model performance comparing baseline clinical model (Model 1) and clinical model plus newborn screening analytes for prediction of neonatal sepsis.

Search

Quick links