Response to the correspondence on “Performance of single‐gene noninvasive prenatal testing for autosomal recessive conditions in a general population setting”

We appreciate the opportunity to respond to the correspondence by Benn et al. on Wynn et al., a study of the performance of carrier screening with sgNIPT on a large, general population of its intended use. We feel the authors expressed unsubstantiated concerns about the need for and impact of the assay and methodology of the study. Benn et al. incorrectly claim that the need for carrier screening assessing fetal risk without the partner sample relies on “dubious estimates regarding partner testing difficulty.” There are multiple publications that have independently demonstrated partner uptake of less than 50% in various US populations for both targeted and expanded carrier panels.1-3 The authors contend that implementing this workflow will increase counseling needs. However, with sgNIPT assessing fetal risk directly, only 2%–3% of carriers will be identified to have an increased risk for an affected fetus. This means that approximately 97% of carriers, all of whom would require counseling in the traditional workflow, would receive a reassuring low-risk result, reducing the counseling burden on our already stretched genetics workforce. Finally, it is important to note that contrary to the authors' comments, the methodology of the study is well-aligned with published guidance on establishing clinical validity4, 5 and is consistent with the methodology of multiple other studies of NIPT clinical performance.6, 7 Benn et al. provide a miscalculation of the proportion of outcomes collected in our study by including the entire sample, of which, the vast majority (>80%) were not identified as carriers. In order to obtain an accurate assessment of the performance of sgNIPT, we solicited fetal/neonatal outcomes on more than 40% of patients who were identified as carriers and obtained outcomes on over 30% of the high-risk results. The outcomes cohort was enriched for cases identified via sgNIPT as high risk, which is necessary to ensure representation of affected cases for rare conditions and allow for sensitivity analysis, a methodology utilized in other studies of NIPT.6, 7 Positive predictive value (PPV) represents the proportion of pregnancies identified as high risk that are affected with the condition in question, and as such the calculation is not impacted by low risk calls as Benn et al., incorrectly states.8 Rather, PPV is falsely inflated when a study population is enriched for a priori high-risk cases; to prevent this bias we excluded known high-risk couples (HRC) from the analysis. We agree with Benn et al. that specificity and sensitivity are impacted by non-random sampling and note this in the limitations. This non-random sampling would artificially deflate the reported specificity and negative predictive value (NPV); therefore, the results represent a conservative estimate of specificity and NPV. To account for the non-random sampling and aid the reader in interpreting these numbers, we included a statistical model to estimate overall sensitivity and specificity of carrier screening with sgNIPT for the full cohort. Benn et al. caution that we do not report out condition-specific test performance, however, the data for specific conditions is included in the Supporting Information of the main publication. Moreover, we note that the assay is designed for the general population, where overall performance is more relevant. In addition, we demonstrate that our condition-specific personalized fetal risk estimates correlate strongly with newborn outcomes. The authors also indicated that the relationship between fetal fraction and result was not discussed. In fact, the relationship between fetal fraction, risk assessment, and outcome is illustrated in Figure S1, which demonstrates the accuracy of the assay is not correlated with fetal fraction, and our prior publications have demonstrated excellent performance of sgNIPT even at low fetal fractions.9 The authors expressed concern that carrier screening with reflex to sgNIPT has been developed for four conditions. Importantly, these conditions align with American College of Obstetricians and Gynocologists recommendations for carrier screening for the general pregnant population.10 The authors claim that the rate at which HRCs would be identified with this testing is lower than that of expanded carrier screening (ECS). However, the authors' comparison relied on perfect use of ECS (meaning paternal testing is completed and paternity is assured). This assumption ignores a growing body of literature that illustrates a less than 50% rate of paternal carrier screening uptake in the US.1-3 We acknowledge that as some laboratories take the approach of continuing to expand carrier screening panels further and further they will detect an increasing number of HRCs despite the real-world limitations described above. However, the utility of adding conditions that are relatively common but mild or are incompletely penetrant is questionable. In a perfect world where (i) all patients undergo carrier screening prior to becoming pregnant, (ii) paternity is always assured, and (iii) paternal carrier screening is always completed, ECS will detect more HRCs. However, the vast majority of the US population does not meet these criteria, and the clinical utility of traditional carrier screening is severely diminished for these patients.1-3, 11, 12 We are in agreement with the authors that “Non-availability of a partner can result in risk uncertainty, unnecessary invasive tests, and additional expense” and alternatives to the traditional carrier screening model such as carrier screening with sgNIPT are needed for more patients to have access to a complete fetal risk assessment for the most common autosomal recessive conditions. None. Juli Wynn, Shannon Rego O'Rourke, and Jennifer Hoskovec are employees of BillionToOne and hold options to hold stock in the company. Sriram C. Perni is a paid consulting Prenatal Medical Director at BillionToOne. Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.