Limits...
Improving the Positive Predictive Value of Non-Invasive Prenatal Screening (NIPS)

View Article: PubMed Central - PubMed

ABSTRACT

We evaluated performance characteristics of a laboratory-developed, non-invasive prenatal screening (NIPS) assay for fetal aneuploidies. This assay employs massively parallel shotgun sequencing with full automation. GC sequencing bias correction and statistical smoothing were performed to enhance discrimination of affected and unaffected pregnancies. Maternal plasma samples from pregnancies with known aneuploidy status were used for assay development, verification, and validation. Assay verification studies using 2,085 known samples (1873 unaffected, 69 trisomy 21, 20 trisomy 18, 17 trisomy 13) demonstrated complete discrimination between autosomal trisomy (Z scores >8) and unaffected (Z scores <4) singleton pregnancies. A validation study using 552 known samples (21 trisomy 21, 10 trisomy 18, 1 trisomy 13) confirmed complete discrimination. Twin pregnancies showed similar results. Follow-up of abnormal results from the first 10,000 clinical samples demonstrated PPVs of 98% (41/42) for trisomy 21, 92% (23/25) for trisomy 18, and 69% (9/13) for trisomy 13. Adjustment for causes of false-positive results identified during clinical testing (eg, maternal duplications) improved PPVs to 100% for trisomy 21 and 96% for trisomy 18. This NIPS test demonstrates excellent discrimination between trisomic and unaffected pregnancies. The PPVs obtained in initial clinical testing are substantially higher than previously reported NIPS methods.

No MeSH data available.


Relationship of fetal fraction estimated using a median of X and Y                            based methods with the fetal fraction estimated from a model of                            autosomal read counts among 1366 male samples that were not included in                            the development of the autosomal model.                        Pearson correlation coefficient (r) = 0.81.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5382935&req=5

pone.0167130.g001: Relationship of fetal fraction estimated using a median of X and Y based methods with the fetal fraction estimated from a model of autosomal read counts among 1366 male samples that were not included in the development of the autosomal model. Pearson correlation coefficient (r) = 0.81.

Mentions: For female fetuses, fetal fraction was estimated using a regularized regression model [27]. Briefly, a training set of 3281 samples from known male fetuses was used to model fetal fraction (estimated as described above) as a function of sample bin counts normalized by the sample total read count but uncorrected for GC content. Bins residing on chromosomes 13, 18, 21, X or Y chromosomes were excluded from the modeling process. The model was a regularized linear regression model implemented with the R package “glmnet” (version 1.9–8). Ten-fold cross-validation using an alpha parameter of 1 was used to select the lambda parameter having the minimum cross-validated error for use in building the final model which is subsequently used to estimate fetal fraction for female fetuses. Fig 1 demonstrates the relationship of fetal fraction estimated using a median of X and Y based methods with the fetal fraction estimated from a model of autosomal read counts among 1366 male samples that were not included in the development of the autosomal model. The Pearson correlation coefficient (r) was 0.81.


Improving the Positive Predictive Value of Non-Invasive Prenatal Screening (NIPS)
Relationship of fetal fraction estimated using a median of X and Y                            based methods with the fetal fraction estimated from a model of                            autosomal read counts among 1366 male samples that were not included in                            the development of the autosomal model.                        Pearson correlation coefficient (r) = 0.81.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5382935&req=5

pone.0167130.g001: Relationship of fetal fraction estimated using a median of X and Y based methods with the fetal fraction estimated from a model of autosomal read counts among 1366 male samples that were not included in the development of the autosomal model. Pearson correlation coefficient (r) = 0.81.
Mentions: For female fetuses, fetal fraction was estimated using a regularized regression model [27]. Briefly, a training set of 3281 samples from known male fetuses was used to model fetal fraction (estimated as described above) as a function of sample bin counts normalized by the sample total read count but uncorrected for GC content. Bins residing on chromosomes 13, 18, 21, X or Y chromosomes were excluded from the modeling process. The model was a regularized linear regression model implemented with the R package “glmnet” (version 1.9–8). Ten-fold cross-validation using an alpha parameter of 1 was used to select the lambda parameter having the minimum cross-validated error for use in building the final model which is subsequently used to estimate fetal fraction for female fetuses. Fig 1 demonstrates the relationship of fetal fraction estimated using a median of X and Y based methods with the fetal fraction estimated from a model of autosomal read counts among 1366 male samples that were not included in the development of the autosomal model. The Pearson correlation coefficient (r) was 0.81.

View Article: PubMed Central - PubMed

ABSTRACT

We evaluated performance characteristics of a laboratory-developed, non-invasive prenatal screening (NIPS) assay for fetal aneuploidies. This assay employs massively parallel shotgun sequencing with full automation. GC sequencing bias correction and statistical smoothing were performed to enhance discrimination of affected and unaffected pregnancies. Maternal plasma samples from pregnancies with known aneuploidy status were used for assay development, verification, and validation. Assay verification studies using 2,085 known samples (1873 unaffected, 69 trisomy 21, 20 trisomy 18, 17 trisomy 13) demonstrated complete discrimination between autosomal trisomy (Z scores >8) and unaffected (Z scores <4) singleton pregnancies. A validation study using 552 known samples (21 trisomy 21, 10 trisomy 18, 1 trisomy 13) confirmed complete discrimination. Twin pregnancies showed similar results. Follow-up of abnormal results from the first 10,000 clinical samples demonstrated PPVs of 98% (41/42) for trisomy 21, 92% (23/25) for trisomy 18, and 69% (9/13) for trisomy 13. Adjustment for causes of false-positive results identified during clinical testing (eg, maternal duplications) improved PPVs to 100% for trisomy 21 and 96% for trisomy 18. This NIPS test demonstrates excellent discrimination between trisomic and unaffected pregnancies. The PPVs obtained in initial clinical testing are substantially higher than previously reported NIPS methods.

No MeSH data available.