NIPT assay performance and PPV
The NIPT assay used by TDL Genetics is VeriSeq NIPT Solution v2, which is manufactured by Illumina and processed at our London laboratory.
To determine the clinical accuracy of the VeriSeq NIPT Solution v2, Illumina evaluated plasma samples from pregnant women with singleton and twin pregnancies. Samples were obtained from deidentified banked plasma samples that were previously processed from peripheral whole blood specimens.
Key outcomes are presented below; the full data set is available from Illumina.
Basic Screen Performance
For the basic screen, anomalies include T21, T18, and T13. A total of 2,243 singleton and twin samples were included in the analysis. All seven twin pregnancies were correctly detected as T21 and are not reported in Table 1.
The assay performance in the basic screen as shown in Table 1 is calculated excluding a subset of 64 samples affected by RAAs, autosomal partial deletions or duplications, or known mosaicism. These 64 samples included eight T21 and three T18 mosaics. Five of these 11 samples were identified as affected with the anomaly detected by the VeriSeq NIPT Assay Software v2.
|Sensitivity||> 99.9% (130/130)||> 99.9% (41/41)||> 99.9% (26/26)|
|2-sided 95% CI||97.1%, 100%||91.4%, 100%||87.1%, 100%|
|Specificity||99.90% (1982/1984)||99.90% (1995/1997)||99.90% (2000/2002)|
|2-sided 95% CI||99.63%, 99.97%||99.64%, 99.97%||99.64%, 99.97%|
The VeriSeq NIPT Solution v2 sex chromosome results were compared to the clinical reference standard outcome and are summarized in the following table. The percent concordance was calculated for each sex chromosome within each clinical reference standard outcome. Percent concordance was calculated as the number of samples in which the VeriSeq NIPT Solution v2 sex chromosome call matched the clinical reference standard classification, divided by the total number of samples with the same clinical reference standard classification.
|Fetal Sex Classification||Phenotype
* Five twin pregnancies were correctly classified as presence of Y. Two pregnancies were correctly classified as no presence of Y.
** Other cytogenetic results were XXXXX and XXYY.
Positive Predictive Value and Negative Predictive Value of the VeriSeq NIPT Solution v2
Positive predictive value (PPV) and negative predictive value (NPV) of the test provide information regarding the ability of the test to inform clinical decisions based on test sensitivity, specificity, and pretest probability that a fetus is trisomy affected (prevalence). Because PPV and NPV depend on prevalence and the prevalence for these aneuploidies can vary across different subject populations, PPV and NPV were calculated for a range of plausible prevalence values based on the sensitivity and specificity values observed in the basic screen (without known mosaics) of the clinical accuracy study.
|Prevalence (%)||PPV (%)||NPV (%)|
|Prevalence (%)||PPV (%)||NPV (%)|
|Prevalence (%)||PPV (%)||NPV (%)|
Due to the low prevalence of trisomy 21, 18, and 13 in twin pregnancies, only a small number of affected twin samples were available for the clinical study. To estimate the performance of the VeriSeq NIPT Solution v2 in twin pregnancies, in silico models based on observations from clinical samples were used to simulate populations of twin pregnancies. This simulation was consistent with the intended use population.
Estimating Trisomy 13, 18, and 21 and Chromosome Y Performance in Twin Pregnancies
To estimate the performance of the VeriSeq NIPT Solution v2 in twin pregnancies, in silico models based on observations from clinical samples were used to simulate populations of twin pregnancies. This simulation was consistent with the intended use population. The distribution of fetal fraction was determined from approximately 4,500 twin samples and compared to the distribution from approximately 120,000 singleton samples. The distribution of fetal fraction conditional on aneuploidy status was determined from singleton putative calls (1,044 trisomy 21, 307 trisomy 18, and 192 trisomy 13). Combining the two distributions allowed for inferences of aneuploidy detection in twins. Sets of dizygotic and monozygotic twins were simulated, and a weighted average representing their prevalence in the intended use population was taken (2 dizygotic: 1 monozygotic) to estimate sensitivity. For specificity, sets of unaffected twins were simulated.
The fraction of each simulated sample affected by the trisomy (ie, the affected fraction) was calculated differently for each sample category:
- For monozygotic twins, the affected fraction of each sample was set to 1.0 because, in this situation, the trisomy affects both twins.
- For dizygotic twins, it was assumed that only one twin was affected (to have both dizygotic twins be affected is extremely rare). Affected fraction values were simulated using the known distribution of fetal fraction ratios as determined from sex discordant clinical twin samples. A conservative approach was taken whereby it was assumed that the affected twin always had the lowest fetal fraction of the two twins. A correction factor was applied for fetal fractions being on average lower in trisomy 13 and 18 pregnancies.
- For unaffected twins, the affected fraction of each sample was set to zero.
For twins affected by either trisomy 18 or 13, the fetal fraction corresponding to the affected fraction of the sample was reduced. The reduction was proportional to the average reduction in fetal fraction observed in clinical data in trisomy 18 or 13 singletons versus euploid singletons.
Both the overall fetal fraction and the affected fraction of each simulated sample were then used to calculate an aneuploidy score using the standard VeriSeq NIPT Solution v2 algorithm. Sensitivity was calculated by determining how often the aneuploidy scores for the simulated affected twins were above the corresponding aneuploidy cutoff. Correspondingly, specificity was calculated by determining how often the aneuploidy scores for the simulated unaffected twins were below the corresponding aneuploidy cutoff (Table 6). 95% confidence intervals were estimated based on the number of real clinical twin samples in the original data set, which were classified as either affected or unaffected by the relevant trisomy.
To estimate chromosome Y sensitivity in twin samples, sets of XY/XY and XX/XY twins were simulated. A weighted average representing their prevalence in the intended use population was taken (1 XY/XY: 1 XX/XY). To estimate chromosome Y specificity in twins, a set of XX/XX twins was simulated. The overall fetal fraction values were simulated according to the known distribution of fetal fraction in clinical twin samples.
For XY/XY and XX/XY twins, corresponding chromosome Y scores were estimated using the known relationship between fetal fraction and chromosome Y scores in clinical singleton samples classified as male. For XX/XY twins only, affected (ie, male) fetal fraction values were simulated using the known distribution of fetal fraction ratios observed between twins from the same pregnancy, as determined from sex discordant clinical twin samples. A conservative approach was taken whereby the affected fraction was selected such that it corresponded to the smaller of the two twins. For each simulated XX/XY sample, the chromosome Y score was multiplied by the affected fraction.
For XX/XX twins, chromosome Y scores were sampled from those scores observed in clinical singleton samples classified as female. The chromosome Y score and the overall fetal fraction were then used to classify each simulated sample as chromosome Y present or chromosome Y absent using the standard VeriSeq NIPT Solution v2 algorithm.
Sensitivity was calculated by determining how often the simulated XY/XY or XX/XY twins were correctly classified as chromosome Y present. Specificity was calculated by determining how often the simulated XX/XX twins were correctly classified as chromosome Y absent. 95% confidence intervals were estimated based on the number of real clinical twin samples in the original data set that were classified as either chromosome Y present or chromosome Y absent.
Table 18 provides point estimates and estimated 95% confidence intervals for the sensitivity and specificity of VeriSeq NIPT Solution v2 to detect trisomy 21, 18, 13, and the presence of Y in a simulated population of twin pregnancies consistent with the intended use population. Confidence intervals were estimated based on the number of QC passing clinical twin samples classified as either affected or unaffected by the relevant trisomy. The sensitivity calculation assumes that two thirds of affected twin pregnancies are dizygotic with one affected twin, while one third of affected twin pregnancies are monozygotic with both twins affected.
The estimates listed in Table 6 pertain to twin pregnancies only. Due to even lower prevalence, data for higher- order pregnancies (triplets or higher) were insufficient to establish appropriate statistical models to estimate accuracy of aneuploidy detection.