Abstract: TH-PO659
Combination of Different Urinary Omics Traits Improves the Prediction of Postnatal Renal Outcome in Fetuses with Posterior Urethral Valves (PUV)
Session Information
- Pediatric Nephrology
November 02, 2017 | Location: Hall H, Morial Convention Center
Abstract Time: 10:00 AM - 10:00 AM
Category: Developmental Biology and Inherited Kidney Diseases
- 403 Pediatric Nephrology
Authors
- Buffin-Meyer, Benedicte, Inserm U1048, Toulouse, France
- Klein, Julie, Inserm U1048, Toulouse, France
- Muller, Francoise, Inserm U1048, Toulouse, France
- Breuil, Benjamin, Inserm U1048, Toulouse, France
- Moulos, Panagiotis, Inserm U1048, Toulouse, France
- Nadia, Lounis, Inserm U1048, Toulouse, France
- Bascands, Jean-loup, Inserm U1048, Toulouse, France
- Decramer, Stéphane, Inserm U1048, Toulouse, France
- Schanstra, Joost, Inserm U1048, Toulouse, France
Background
Urinary omics-based strategies are promising tools in medicine as they have already led to the design of multimarker models for the assessment of complex diseases. Nevertheless, advances still need to be made since most models using single omics traits are unable to reach a 100% accuracy and display a so-called “gray zone” defined by the uncertainty of the prediction. Here we verified the hypothesis whether a combination of urinary fetal peptides and metabolites provides an improved prediction of postnatal renal function in fetuses with PUV compared to the individual omics traits.
Methods
Using capillary electrophoresis coupled to mass spectrometry, we explored the urinary metabolome from 13 PUV fetuses with early ESRD and 12 PUV fetuses without postnatal ESRD at 2 years.
Results
This allowed the identification of 24 differentially abundant fetal urinary metabolites which were modelled into a svm classifier, alone or in combination with 12 peptides predictive of disease progression (Klein et al, PMID:23946195). The predictive capacities of models composed of metabolites (24m model), peptides (12p model) or association of both (24m_12p model) were compared in a separate independent validation cohort of 35 fetuses with PUV. The gray zone was generated as the range of svm scores for which the negative likelihood ratio (LR) was >0.05 and the positive LR was <20. Sensitivity, specificity (excluding patients in gray zone) as well as area under the ROC curve (AUC) and net reclassification improvements (NRI) of PUV patients were evaluated (Table 1).
Conclusion
While the individual metabolome- and peptidome-based models already display high accuracy for identification of the disease classes, the discriminative power can be significantly improved by combination of omics traits. This supports the general concept that multi-omics approaches can improve the clinical assessment of disease.
Table 1
Model | 24m | 12p | 24m_12p |
Sensitivity (%) | 50.0 [25 - 74] | 81.3 [62 - 100] | 87.5 [71 - 100]* |
Specificity (%) | 68.4 [47 - 89] | 78.9 [61 - 97] | 84.2 [68 - 100] |
AUC | 0.90 [0.79-1] | 0.97 [0.91-1] | 0.99 [0.96-1] |
NRI versus 24m (%) | 36.5 [4-69]§ | ||
NRI versus 12p (%) | 11.5 [0-39] |
*p<0.05 vs 24m, §p<0.05 vs 0.
Funding
- Private Foundation Support