Challenges in providing residual risks in carrier testing.

The probability an individual is a carrier for a recessive disorder despite a negative carrier test, referred to as residual risk, has been part of carrier screening for over 2 decades. Residual risks are calculated by subtracting the frequency of carriers of pathogenic variants detected by the test from the carrier frequency in a population, estimated from the incidence of the disease. Estimates of the incidence (and therefore carrier frequency) of many recessive disorders differ among different population groups and are inaccurate or unavailable for many genes on large carrier screening panels for most of the world's populations. The pathogenic variants detected by the test and their frequencies also vary across groups and over time as variants are newly discovered or reclassified, which requires today's residual carrier risks to be continually updated. Even when a residual carrier risk is derived using accurate data obtained in a particular group, it may not apply to many individuals in that group because of misattributed ancestry or unsuspected admixture. Missing or inaccurate data, the challenge of determining meaningful ancestry-specific risks and applying them appropriately, and a lack of evidence they impact management, suggest that patients be counseled that although carrier screening may miss a small fraction of carriers, residual risks with contemporary carrier screening are well below the risk posed by invasive prenatal diagnosis, even if one member of the couple is a carrier, and that efforts to provide precise residual carrier risks are unnecessary.

INTRODUCTION

Residual carrier risk (RCR) after carrier testing is the chance an individual with a negative carrier test could still be a carrier of the disorder. This review will discuss how residual risks after negative carrier screening are determined, the problems with providing them to clinicians and patients, and their limited clinical utility. Examples will be drawn from cystic fibrosis (CF) carrier screening, where the practice of providing RCRs began, but are relevant to all recessive disorders.

THE ORIGIN OF RCRs

Cystic fibrosis carrier screening began over 2 decades ago with a test that screened the cystic fibrosis transmembrane conductance regulator gene (CFTR) for 25 pathogenic variants (PVs) chosen because they were present in greater than 1 in 500 people in the U.S. general population and in certain subpopulations, such as Black or Ashkenazi Jewish individuals., While this strategy identified a majority of carriers in many populations, a RCR remained after a negative carrier screen because the rest of the >900 rarer PVs in CFTR were omitted in order to reduce the complexity and cost of the test and improve its positive rate. Guidelines, therefore, recommended that the providers “… define as accurately as possible, based on current knowledge, the residual risk that the person tested could be a carrier of an untested or unknown mutation.” This recommendation was made out of concern that practitioners and patients alike were unfamiliar with the concept of clinical false negatives in genetic testing and needed to be alerted that false negatives will occur. Because an even greater fraction of the PVs responsible for CF outside the United States and Northern Europe were omitted, an even greater RCR existed in non‐Europeans testing negative for this panel. Therefore, an additional recommendation was made that different RCRs be provided that are tailored to an individual's “racial/ethnic” background without addressing pitfalls in calculating accurate ancestry‐specific RCRs or in determining an individual patient's ancestry.

CALCULATING RCRs

RCR after negative carrier screening is the difference between the frequency of carriers of every PV responsible for a recessive disease in a population, and the frequency of carriers of just those PVs that can be detected by the screening test (See Appendix A). Next‐generation sequencing makes sequencing of the entire CFTR gene (and many others) cost‐effective and reduces false negatives by allowing detection of essentially all previously described PVs as well as many novel variants classifiable as PVs (because of the type of DNA change), a yield far exceeding the ∼25 PVs detected using old technology. Other variants also detected but not reported in carrier screening include variants of uncertain significance (VUSs) that lack definitive classification (although subject to future reinterpretation as pathogenic or benign) and benign variants of no medical significance.

The frequency of carriers of every PV responsible for a rare disease in a population is difficult to measure directly and so is usually approximated as twice the square root of the observed incidence of the disease (Appendix A) using the Hardy‐Weinberg law (Appendix B). The frequency of the carriers of the PVs that can be detected by the test must also be determined, either by directly sequencing large numbers of people in the population and counting the number of carriers or, more commonly, by sequencing a large number of unrelated affected individuals and using the proportion of PVs found to estimate their allele frequencies, and therefore carrier frequencies (Appendix A). The latter approach is well suited to rare disorders because just affected individuals are sequenced rather than a large number of unaffected individuals looking for rare carriers. Similar calculations for X‐linked inheritance are in Appendix A.

FACTORS AFFECTING THE ACCURACY OF RCRs

The accuracy of RCRs depends on how well disease incidence is known. Although estimates of the incidence of CF in the United States and many European Union (EU) countries are accurate because of CF newborn screening and the widespread practice of managing all CF patients in specialized CF clinics, measurements of CF incidence are challenging for most areas of the world lacking these programs. In Latin America, only four countries, Brazil, Uruguay, Argentina, and Chile, had national newborn CF screening programs as of 2020, yet even with newborn screening, estimates of CF incidence in Brazil range 9‐fold, from 1/1,600 to 1/14,000, due to varying degrees of genetic admixture of different socioeconomic groups within Brazil. In countries without CF newborn screening, ascertainment is incomplete and estimates of incidence are probably artifactually low. For example, the published incidence figures for CF In India vary 2.5‐fold (1/40,000 and 1/100,000) and 3.5‐fold in Japan (1/100,000 and 1/350,000). If disease ascertainment is incomplete even for a well‐studied disease like CF, it is even less complete for most other rarer disorders and nationalities.

Compounding the difficulties in measuring disease incidence is the problem of using the Hardy‐Weinberg equation to estimate the frequency of all PVs from disease incidence in a population. The Hardy‐Weinberg equation requires simplifying assumptions including that a population consists of randomly mating individuals (Appendix B). In practice, disease incidence, no matter how accurate, is not measured and reported from an idealized population of randomly mating individuals. Instead, incidence is generally determined in groups defined by a common language, ethnicity, geographic ancestry, or national boundary within which there is still genetic stratification due to non‐random mating patterns.

Stratification and consanguinity inflate the incidence of autosomal recessive disorders beyond what would be expected from the actual carrier frequency under the Hardy‐Weinberg equation and consequently cause overestimation of carrier frequency and, therefore, RCR derived from incidence.,

As an example of stratification, the population of the United Arab Emirates (U.A.E.) consists of subgroups of different nationalities and/or ethnicities, with 25% from India, 12% from Pakistan, 9% indigenous Emirati, 7% Bengali, 5% Filipino, 6% immigrants from Europe, Australia, North America, and Latin America, and Britain, with the remaining ∼30% from other Arab states. Intermarriage between members of many of these groups is limited for reasons of religion or country‐of‐origin, and consanguinity is present., The overall incidence of cystic fibrosis in the U.A.E. is 1/15,876 but differs among these different ancestral subgroups. An average CF carrier frequency calculated from the observed average incidence is 2 × √(1/15,876) = 1/63 in the U.A.E but is likely not applicable to many individuals in any of these subgroups.

Accurate measurements of the carrier frequency of PVs that can be detected by sequencing are also required for accurate RCRs. For example, a CF screening program in Yucatan, Mexico screened 96,071 consecutive newborns by sequencing their CFTR genes and ascertained 7 patients with a firm diagnosis of CF, for an incidence of 1/13,724, and carrier frequency = 1/59 (0.017) (Appendix C). Upon sequencing of the 14 CFTR genes in these 7 patients, 9 of the 14 (64%) PVs were seen, giving a carrier frequency of detectable PVs = 0.64 × 1/59 = 1/92 (0.011). The RCR for a patient from Yucatan testing negative for these nine alleles = 1/164 (95% confidence limit 1/96–1/625). Two missense VUS's in three patients were also seen but could not be classified as PVs and two putative PVs were not found.

This Yucatan example demonstrates many of the problems associated with RCR calculations. First, the frequency of all pathogenic variants is calculated from the incidence by assuming random mating and applying the Hardy‐Weinberg law, which may not be the case here as population genetic studies of indigenous individuals across Mexico suggest significant stratification with excess homozygosity., Second, because genetic testing of affected individuals is limited, especially in understudied demographic groups, sample sizes are small, which means that estimates of allele frequencies will be imprecise, with wide confidence limits, and rarer PVs will be missed until more sequencing of affected individuals identifies more PVs and VUSs, requiring an amended RCR calculation. There may be a delay before sufficient data accumulate to allow the two missense VUSs to be classified as either pathogenic or benign; if they are classified as PVs, RCR falls to 1/409. Finally, additional sequence analysis may ultimately identify the two undetected variants, which may also at first be classified as VUSs but later reclassified to PVs, which would further decrease the RCR. Known variants (benign or VUS) may be reclassified to or from being PVs, thereby altering the set of PVs detected by the test whose carrier frequencies are used to calculate the RCR.

ANCESTRY AFFECTS RCR

Groups differ in disease incidence and PVs responsible for disease., For example the incidence of CF is 1/1353 in Ireland, 1/8500 in Mexico, 1 in 25,000 in Finland and between 1/150,000 and 1/300,000 in Japan., Similarly, the particular PVs responsible for CF and their allele frequencies differ among groups, with some PVs either enriched in, or even exclusive to, one group. For example, the most common CFTR PV in Europe, p.Phe508del, is found in ∼72% of CF patients of non‐Hispanic European ancestry but only 39% of CF patients of Asian ancestry and an even smaller percentage in people of African ancestry. Such differences among groups in disease incidence and in the frequency of PVs detectable by the screening test result in different RCRs. Unfortunately, the two values needed for an RCR, accurate disease incidence data and frequency of PVs detected by the test, are lacking or incomplete for most groups and more epidemiological and molecular data are needed to fill these gaps if RCRs are to be generally reliable.

PITFALLS IN APPLYING ANCESTRY‐SPECIFIC RCRs TO AN INDIVIDUAL PATIENT

Even assuming an RCR can be accurately determined for a particular ancestral, national or ethnic group, providing the relevant RCR figure to an individual following a negative carrier screen requires the screened individual be correctly assigned to a group for which the RCR was determined. Ancestral group assignment has most often relied on self‐reported ancestry (SRA) or on ethnicity based on a shared language (e.g., Spanish), or history/religion (Ashkenazi Jewish). However, for individuals with mixed genetic ancestry, especially when the mixture includes ancestries from different continents where PV frequencies may differ dramatically, the SRA is not an accurate reflection of either the individual's overall genetic ancestry, but even more importantly, will lack precision regarding the genetic ancestry of the alleles within the gene involved in the disease in question., , There is no clear consensus among geneticists and genetic testing laboratories on how SRA should be defined or used and it is unknown how often inaccurate RCRs based on an incorrect SRA are provided to patients. For example, using SRA (e.g., “Brazilian”) to decide which RCR applies to a Brazilian of mixed ancestry will miss the mark because we do not know the ethnic origin of her CFTR alleles (e.g., African, Native American, or European) and can only offer a weighted average across these ancestral possibilities.

If SRA alone, especially for an admixed individual, does not provide the specificity needed to calculate an RCR used in counseling that individual,, genetic ancestry derived from single nucleotide polymorphisms (SNPs) is an alternative. SNPs are genetic markers whose allele frequencies differ among people whose ancestors historically have lived in distinct, separated geographic locations throughout the world., , , SNP alleles and their frequencies, determined in cohorts of individuals whose ancestors are known to have originated from particular geographical regions, are used to determine the likely ancestral origin of every location in the genome, depending on the number and informativeness of SNPs around that location, how extensive the reference datasets for their allele frequencies are, and when admixture occurred in that individual. These chromosomal ancestry estimates are summed to obtain genetic ancestry averaged across an individual's entire genome.

However, basing an RCR calculation for CF, for example, on average ancestry derived from genome‐wide SNPs will be inaccurate for individuals who differ in the ancestry of the alleles in the region of CFTR, regardless of how similar their average genome‐wide ancestry might be. For example, the degree of admixture of African, Native American, and European ancestry for 96 self‐identified Puerto Rican individuals is shown in Figure 1. All subjects carried a contribution from all three ancestral populations with an average of 66% European, 16% African, and 18% Native American ancestries. However, the proportion of ancestral alleles from each of these three continental ancestries varied among individuals, with African ancestry between ∼10% and ∼55%, European ancestry between ∼30% and ∼80%, and Native American ancestry between ∼5% and ∼20%. Individuals 95 and 96 have nearly identical contributions of alleles of the three continental origins. Yet, if the alleles at the CFTR locus were among the ∼20% of alleles of European origin in individual 95 and among the ∼50% of alleles of African ancestry in individual 96, their RCRs following a negative CF screening test would be different. In contrast, individual 1 far exceeds individual 96 in the relative contribution of European versus African ancestral alleles. Yet, if the alleles at the CFTR locus were among the ∼80% of alleles of European ancestry in individual 1 and among the ∼20% of alleles of European ancestry in individual 96, their RCR following a negative CF screening test would be similar, regardless of how these two individuals self‐identify and regardless of the total proportions of other European and non‐European alleles they carried across their genomes.

FIGURE 1

Graph showing the proportions of alleles of three different continental ancestries, African, Native American, and European, respectively, assessed at >116,000 SNP loci from across the genome in 96 self‐identified Puerto Rican individuals. Each vertical bar represents a single individual, beginning with individual 1 on the left to 96 on the far right, arranged by increasing percentage of alleles of African continental origin carried (Figure from Burchard et al.).

The admixture problem does not only affect communities with a clear historical record of admixture of individuals originating on different continents. Individuals of mixed ancestry are present even within supposedly homogeneous groups such as Europeans whose allele frequencies vary across Europe and the Middle East, and even for Northern Europeans such as the Irish or Finns.

CLINICAL UTILITY OF RCRs

Given that RCRs are imprecise and difficult to apply in an ancestry‐specific manner, does knowing the approximate values of even imprecise RCRs impact management? The utility in knowing an RCR depends on whether it leads to a residual risk for a disorder in a pregnancy high enough to change clinical management. In the preconception setting, it is difficult to predict how any particular value of risk to a prospective offspring would affect the reproductive plans different couples make (e.g., pregnancy with invasive prenatal diagnosis [IPD], sperm or egg donation, in vitro fertilization with preimplantation diagnosis, and adoption). In the prenatal setting, however, we can benchmark offering IPD as the actionability threshold, using the past ACOG recommendation that IPD be discussed with all pregnant women regardless of personal or family history or advanced maternal age based on the chance of chromosomal aneuploidy (≥∼1/500) outweighing the procedural risk (1/1,000–1/333)., Applying a similar argument, when one member of a couple is known to be a carrier and the other has tested negative, the chance of there being an affected fetus is one‐fourth × RCR; RCR would need to exceed 1/125 to make the risk of the disorder in the pregnancy exceed the 1/500 actionability threshold. Given modern sequencing methods, even imprecise estimates of RCRs place nearly all of them well below levels that would trigger actionability. When both members of the couple are carrier screen negative, an RCR would need to exceed ∼1/11 to make the risk of an affected pregnancy (one‐fourth × [RCR]2) exceed a 1/500 threshold, an RCR that is never seen after negative carrier screening when the entire coding region of a gene is examined (Table 1). Determining precisely how far an RCR is below the level that would trigger a management change makes no additional contribution to clinical decision making.

TABLE 1

Residual Carrier Risks in a couple that exceed a threshold for clinical actionability

		RCR in individuals with negative carrier test that would cause risk to offspring to exceed Actionability Thresholda
		Carrier screen results in members of couple
	Actionability threshold	Both test negative	One is a carrier Other tests negative
Prenatal setting	1/500	RCR > 1/11	RCR > 1/125

Abbreviation: RCR, residual carrier risk.

When one member of the couple is of unknown carrier status, his/her chance of being a carrier is between the RCR of a person with a negative test (if he/she were to be tested and tests negative) and 1 (if he/she were to be tested and tests positive). The RCR needed to exceed threshold for actionability would be between the RCRs for when both test negative and the RCR for when one tests positive, the other negative.

CONCLUSION AND RECOMMENDATIONS

Counseling patients about their carrier risk and offering carrier screening to identify couples at risk so they can receive counseling and learn about reproductive options are essential components of preconception and prenatal care. Couples in which one or both members have negative carrier screens benefit from learning their risk is reduced. However, for all the reasons discussed in this review, the current practice of trying to precisely quantify that reduction in carrier risk, with or without ancestry specificity, is both flawed and, furthermore, of little utility. There are a number of options for dealing with the current unsatisfactory situation.

One is to only continue to provide RCRs when there are ancestry‐specific incidence and PV data of sufficient quality and quantity to justify their use, including confidence limits or ranges on the estimate, and the ancestry of the alleles for the disorder in the individual screened, and not overall ancestral proportions, is well known from SNP‐based ancestry studies. In this case, patients of diverse ancestries would receive different ancestry‐specific RCRs for differing sets of genes, but this approach is difficult for clinicians and testing laboratories to execute and is problematic as social policy.

A second would be to accept that RCR estimates are uncertain and assigning patient ancestry is challenging, so, using the best data available, provide all patients with a range of “ancestry‐agnostic” RCRs, with confidence limits, for only the most common disorders with the greatest RCRs and not deliver RCRs based on difficult‐to‐assign “‘racial/ethnic' background of the tested individual.” Although not taking ancestry into account means using average RCRs that could be discrepant from their true values for a given individual, errors in assigning ancestry are likely to result in just as many erroneous RCR assignments. There is precedent for not taking ancestry into account when providing genetic screening. The American College of Obstetricians and Gynecologists (ACOG) initially considered it appropriate to only screen individuals for CF if there were a high carrier frequency of CF in the ancestral group to which the individuals belong. However, increasing recognition of the difficulty of assigning an individual accurately to a single ancestry led ACOG to retract the previous guideline and recommend that practitioners should “offer CF carrier screening to all patients.”,

Finally, RCRs estimates could be eliminated from most carrier testing. Despite inaccuracies, the vast majority of RCRs do fall below an actionability threshold for all but the most common disorders even in the most challenging situation, when the other member of the couple is known to be carrier; attempts at precision or ancestry‐matching will not increase their actionability. Patients should therefore be counseled honestly that no genetic test is perfectly sensitive, carrier screening will miss a small fraction of carriers, but contemporary carrier testing reduces RCRs well below the risk posed by invasive prenatal diagnosis, even if one member of the couple is a carrier.

RCRs were a well‐intentioned but now vestigial instructional tool from 20 years ago when individuals were screened for only a limited number of variants. More complete sequencing, inaccurate or unobtainable population data, a more sophisticated appreciation of the genetic complexity of ancestry, and the recognition of their limited utility together render them unnecessary. Professional guidelines should be revised with respect to the recommendation to offer RCRs as part of clinical care.