Draft Genetic Test Review

---

Cystic Fibrosis
Clinical Validity

(436KB)

---

CLINICAL VALIDITY

Question 18: How often is the test positive when the disorder is present?
Question 19: How often is the test negative when the disorder is not present?
Question 20: Are there methods to resolve clinical false positive results in a timely manner?
Question 21: What is the prevalence of the disorder in this setting?
Question 22: Has the test been adequately validated on all populations to which it may be offered?
Question 23: What are the positive and negative predictive values?
Question 24: What are the genotype/phenotype relationships?
Question 25: What are the genetic, environmental or other modifiers?

---

CLINICAL VALIDITY

Question 22: Has the test been adequately validated on all populations to which it may be offered?

The DNA testing utilized for prenatal screening is aimed at identifying specific mutations that cause cystic fibrosis. The test is designed to identify these mutations in any DNA sample regardless of the characteristics of the individual being tested (e.g., race or ethnicity). Although the prevalence of cystic fibrosis and the mix of mutations responsible for the disorder may vary by race, the test should reliably identify the target mutation. One exception to this might occur if the presence and/or frequency of unknown polymorphisms would vary by race/ethnicity (or some other factor). In reality, however, it would be difficult for laboratories to thoroughly examine this possibility in all populations to which testing may be offered.

Gap in Knowledge: Polymorphisms by race/ethnicity.

Variation in polymorphism frequency by race/ethnicity has not been well described in the literature. Laboratories should make efforts to report in the literature all polymorphisms in the context of the racial/ethnic background being tested.

---

CLINICAL VALIDITY

Question 23: What are the positive and negative predictive values?

Positive predictive values
There are three possible definitions for positive predictive value. All are based on the principles shown earlier in this section (Question 18, Table 3-2).

• Given screen positive couples, what proportion are actually carrier couples
• Given screen positive couples, what proportion of their pregnancies will be affected
• Given a positive fetal diagnostic test, what proportion of fetuses would eventually develop the cystic fibrosis phenotype

Each will be discussed in more detail in the following sections. 

The positive predictive values for being a carrier couple given that both partners have an identified mutation are dependent on the

• birth prevalence of cystic fibrosis (varies from 1:2,500 to 1:31,000 depending on race/ethnicity)
• proportion of cystic fibrosis mutations identified (varies from 40 to 95 percent depending on race/ethnicity)
• analytic sensitivity (expected to be constant at about 97.9 percent)
• analytic specificity and the subsequent performance of confirmatory testing to identify false positive results. This rate is not well established. For the purposes of the table, the final analytic specificity (after all confirmatory testing has been performed) will be modeled at rates between 99,900 and 99,999 per 100,000 tests (i.e., false positive rates between 1 and 100 per 100,000).
• Screening model used. Table 3-27 is appropriate for the one-step (sequential) and two-step (couple) models only. The expanded two-step (concurrent) model will have approximately twice the number of false positive couples identified because it identifies twice the number of couples in which one is a true positive (since all samples are tested). To compute the positive predictive value for the concurrent model, divide the odds by 2 and recompute the positive predictive value. For example, the first row in Table 3-27 shows a positive predictive value of 97.1 percent (odds of 33:1 or 33/34). For the concurrent model, the corresponding number would be 94.3 percent (odds of 33/2:1 or 16.5/17.5). 

Taking the above factors into account, Table 3-27 shows the corresponding positive predictive values. These values do not vary much with the changing proportions of mutations detected within the range of values provided (Column 2). The positive predictive values are, however, strongly dependent on both the prevalence and the false positive rate. In viewing the positive predictive values, it is important to recognize that the number of couples with positive screening results is initially quite low; 1 in 625, 1 in 2,500 and 1 in 7500, respectively, for the three prevalences shown in the table. Prenatal testing identifies a group with risks several thousand times higher. 

Table 3-27. Estimates of the Positive Predictive Value for Being a Carrier Couple when the One-step (Sequential) or Two-step (Couple) Screening Models are Employed at Three Birth Prevalences and Three False Positive Rates

Birth Prevalence	Mutations Detected (%)	False Positive Rate (per 100,000)	Positive Predictive Value
			(%)	Odds (n:1)
1:2,500	75-95	100	97.1	33
		10	99.7	340
		1	99.9	3,400
1:10,000	60-80	100	92.9	13
		10	99.3	140
		1	99.9	1,400
1:30,000	40-60	100	82.9	4.8
		10	98.2	56
		1	99.8	570

 

Actual data to confirm this modeling are scarce. A preliminary estimate of the positive predictive value can be made, based on information collected as part of ongoing prenatal cystic fibrosis diagnostic testing. As described in an earlier section, a major prenatal diagnostic referral laboratory in the United States requires that carrier couples submit new blood samples and the parental genotypes along with the amniotic fluid. The referral laboratory has documented false positive carrier classification on more than one occasion (Heim R, personal communication, 2001). This preliminary observation suggests that the false positive rate (after confirmatory testing is completed) is likely to be between 1 and 10 per 100,000 couples tested.

The positive predictive values for having an affected fetus given a carrier couple are dependent on the proportion of carrier couples correctly identified (positive predictive value for carrier couples from Table 3-27). Were all carrier couples to be correctly classified, their risk for an affected fetus would be 1 in 4 (odds 1:3), assuming that all mutations are highly penetrant (see Questions 20 and 24 for a further discussion). The risks change only slightly when the positive predictive value for carrier couples is reduced to as low as 95 percent (risk of 23.8 percent, odds 1:3.2). Were the positive predictive value for carrier couples to be as low as 80 percent, the risk would be appreciably lower (20 percent, odds 1:4). Given the likely positive predictive values for carrier couples, the approximate risk of 25 percent (odds of 1:3) is appropriate for counseling purposes when the penetrance of the mutations identified are known to be high.

The positive predictive values for having an offspring with the cystic fibrosis phenotype given a positive fetal diagnostic test are mainly dependent on the genotype/phenotype relationship and the error rate for diagnostic testing. Error could occur during fetal diagnostic testing because of maternal cell contamination. For that reason, it is important that laboratories performing fetal diagnostic testing collect parental genotypes. Currently, there is little information available concerning the reliability of cystic fibrosis testing of fetal cells. Nearly all fetuses with two identifiable mutations will eventually develop the cystic fibrosis phenotype, but a small proportion will be less severely affected. The relationship between genotype and phenotype (Question 24) and the impact of the environment and other genes on the phenotype (Question 25) are discussed in more detail in the following sections. Also, a few of the uncommon mutations are often not always associated with the classic phenotype. This is discussed in another section (Question 20).

Gap in Knowledge: The Performance of Cystic Fibrosis Mutation Analysis as a Prenatal Diagnostic Test
The analytic sensitivity and specificity of cystic fibrosis testing in fetal cells obtained by amniocentesis or chorionic villus sampling is not well documented. Maternal cell contamination might rarely contribute to false positive results, especially if the result is based on uncultured cells.

Negative predictive values
The negative predictive value is defined in this section as the probability of a couple with a negative test results not having a child with cystic fibrosis and is based on the principles shown earlier in this section (Question 18, Table 3-2). As discussed earlier, one of three screening models could be employed, two-step (or sequential), one-step (or couple) and an expanded one-step (concurrent). In the two step model, there are two types of negative results: the woman tests negative and the partner is not tested (N/NT), and the woman tests positive and the partner tests negative (P/N). In the one-step model, all couples are reported as negative, unless both partners are identified as carriers (CNP – couple not positive). In the expanded one-step model, all couples' samples are tested and two types of negative test results are possible; one partner tests positive and the other negative (P/N), and both partners test negative (N/N). When necessary, the following section will provide negative predictive values stratified by model and couple test results.

The negative predictive values are dependent on

• the screening model and combination of test results in the couple
• the birth prevalence of cystic fibrosis (varies from about 1:2,500 to 1:31,000 depending on race/ethnicity)
• the proportion of cystic fibrosis mutations detected (varies from about 40 to 95 percent depending on race/ethnicity)
• analytic sensitivity (expected to be constant at about 97.9 percent) 

Analytic specificity has little impact on the negative predictive value, even when confirmatory testing to identify false positive results is taken into account. For that reason, it is not included in the present calculations.

Table 3-28 shows negative predictive values under a variety of circumstances. These values are strongly dependent on the screening model and combination of the couple's test results, the prevalence of cystic fibrosis and the proportion of mutations detected. In viewing the negative predictive values, it is important to recognize that nearly all couples do not include two carrier. When the birth prevalence is relatively high (i.e., 1:2,500), 624 of every 625 couples do not include two carriers. Therefore, those couples will not have a child with cystic fibrosis. According to the table, the higher the mutation detection rate, the lower the risk in those couples with negative test results. This is because higher mutation detection rates will be associated with the identification of more true carrier couples, and those couples are not included in this table. 

Not addressed here are the more complicated scenarios where the partners are of differing ethnic/racial backgrounds. When this occurs, the negative predictive values will differ from those in the table and are even dependent on which partner is tested first (unless the expanded one-step (concurrent) model is employed).

Table 3-28. Negative Predictive Value by Test Model and the Couple's Test Result at Three Birth Prevalences and Three Proportions of Mutations Identified

CF Birth Prevalence	Proportion of Mutations Identified (%)	Negative Predictive Value (Odds of 1:n)1
		Two-Step		One-Step	Expanded One-Step
		N/NT	P/N	CNP	N/N	P/N
1:2,500	95	99.99	99.93	99.99	99.99	99.93
		(34,000)	(1,400)	(18,000)	(470,000)	(1,400)
	85	99.99	99.83	99.99	99.99	99.83
		(14,000)	(580)	(8,100)	(83,000)	(580)
	75	99.99	99.73	99.98	99.99	99.73
		(9,100)	(360)	(5,400)	(33,000)	(360)
1:10,000	80	99.99	99.89	99.99	99.99	99.89
		(45,000)	(910)	(26,000)	(210,000)	(910)
	70	99.99	99.84	99.99	99.99	99.84
		(31,000)	(630)	(19,000)	(98,000)	(630)
	60	99.99	99.79	99.99	99.99	99.79
		(24,000)	(480)	(15,000)	(57,000)	(480)
1:31,000	50	99.99	99.85	99.99	99.99	99.85
		(60,000)	(690)	(41,000)	(120,000)	(690)
	40	99.99	99.83	99.99	99.99	99.83
		(51,000)	(580)	(37,000)	(83,000)	(580)
	30	99.99	99.80	99.99	99.99	99.80
		(44,000)	(500)	(33,000)	(62,000)	(500)

¹ Negative predictive value (expressed as a percentage) is the proportion of negative test results associated with a non-cystic fibrosis fetal genotype. The accompanying odds are often referred to as the ‘residual risk' and are the odds for having an affected fetus given a negative test result.

N/NT - one partner negative, the other was not tested

P/N - one partner positive, the other was negative

CNP – the couple was not screen positive

N/N – both partners were negative

Prenatal screening models and initial positive rates  
The initial aim of prenatal screening may be defined as identifying carrier couples (Table 3-2), but that is often not done in a single step. Three screening models have been employed in pilot trials (Question 33) and will discussed in more detail, later.

• The two-step (or sequential) model first tests the woman's sample. If a mutation is identified, the woman and her partner are contacted and the partner's sample is collected and tested.
• The one-step (or couple) model calls for samples to be collected from both partners at the outset. The woman's sample is usually tested first. When a mutation is identified, the partner's sample can be tested without re-contacting the couple. Unless both partners are carriers, the test is considered negative.
• The expanded one-step (or concurrent) model also requires that samples be collected from both partners, but all samples are tested, and all carriers are notified of their status.

Figure 3-12 graphically displays the sequence of testing for each of these models in a population of 100,000 non-Hispanic Caucasians (the derivation of these numbers can be found in Appendix F). All three models identify the same 118 carrier couples. Overall, 30 of the 40 fetuses are detectable, yielding a clinical sensitivity of about 75 percent. Among the remaining couples in which both partners are not carriers, the number of ‘positive' test results varies considerably by screening model chosen. The number of ‘false positives' is relatively high in this modeling, as the observed analytic specificity of 0.005 is used. If used routinely, confirmatory testing is likely to considerably reduce this rate. 

Clinical performance estimates when the endpoint of screening is considered to be the diagnosis of an affected fetus rather than the identification of a carrier couple From the public health or epidemiologic viewpoint, identifying carrier couples is an intermediate stage of the screening process. Neither of the partners will have health problems due to their carrier status. The final stage of prenatal cystic fibrosis screening (Figure 3-12) is to identify fetuses with two mutations, thereby allowing couples to make decisions about planning for the birth of an affected child or considering the option of pregnancy termination. All three screening models identify the same affected fetuses. The differences are in the numbers of couples who will be made aware of their carrier status and counseled. That rate is 3.9 percent in the two-step model, 0.12 percent for the one-step model, and 7.7 percent for the expanded one-step model. Both false positive and false negative results will occur, but the rates are highly dependent on individual laboratory performance and whether confirmatory testing is performed routinely. A detailed derivation of the data in Figure 3-12 is contained in Appendix F.

Figure 3-12. Prenatal Cystic Fibrosis Clinical Screening Performance in 100,000 non-Hispanic Caucasians According to Three Models

Expected prenatal screening performance in Hispanic Caucasian couples: Figure 3-13 shows the numbers of initially positive test results, along with the numbers of carrier couples identified for each of the three screening models applied to a population of 100,000 Hispanic Caucasians. The same assumptions are used here as for Figure 3-12, and the same three screening models are examined. Here, however, the prevalence of cystic fibrosis is set to 1:13,500, and 72 percent of the mutations are assumed to be identifiable by the panel. All three screening models identify the same four affected fetuses. The differences are in the numbers of individuals who will be made aware of their carrier status and counseled. That rate is 1.7 percent in the two-step model, 0.02 percent for the one-step model, and 3.3 percent for the expanded one-step model. Both false positive and false negative results will occur, but the rates are highly dependent on individual laboratory performance and whether confirmatory testing is performed routinely.

Figure 3-13. Prenatal Cystic Fibrosis Screening Performance in Population of 100,000 Hispanic Caucasians According to Three Models

Expected prenatal screening performance in African American couples: Figure 3-14 shows the numbers of initially positive test results, along with the numbers of carrier couples identified for each of the three screening models. The same assumptions used for Figure 3-12 are used in this example, and the same three screening models are examined. The prevalence of cystic fibrosis, however, is set to 1:15,000, and 65 percent of the mutations are assumed to be identifiable by the panel. All three screening models identify the same three affected fetuses. The differences are in the numbers of individuals who will be made aware of their carrier status and counseled. That rate is 1.5 percent in the two-step model, 0.01 percent for the one-step model, and 3.0 percent for the expanded one-step model. Both false positive and false negative results will occur, but the rates are highly dependent on individual laboratory performance and whether confirmatory testing is performed routinely.

Figure 3-14. Prenatal Cystic Fibrosis Screening Performance in a Population of 100,000 African Americans According to Three Models

Expected prenatal screening performance in Ashkenazi Jewish couples: Figure 3-15 shows the numbers of initially positive test results, along with the numbers of carrier couples identified for each of the three screening models. The same assumptions used for Figure 11 are used for this example, and the same three screening models are examined. The prevalence of cystic fibrosis, however, is set to 1:2,300, and 94 percent of the mutations are assumed to be identifiable by the panel. All three screening models identify the same 37 affected fetuses. The differences are in the numbers of individuals who will be made aware of their carrier status and counseled. That rate is 4.3 percent in the two-step model, 0.15 percent for the one-step model, and 8.4 percent for the expanded one-step model. Both false positive and false negative results will occur, but the rates are highly dependent on individual laboratory performance and whether confirmatory testing is performed routinely.

Figure 3-15. Prenatal Cystic Fibrosis Screening Performance in 100,000 Ashkenazi Jewish Couples According to Three Models

Expected prenatal screening performance in Asian American couples: Figure 3-16 shows the numbers of initially positive test results, along with the numbers of carrier couples identified for each of the three screening models. The same assumptions used for Figure 3-12 are used in this analysis, and the same three screening models are examined. The prevalence of cystic fibrosis, however, is set to 1:31,000, and 49 percent of the mutations are assumed to be identifiable by the panel. All three screening models identify the same affected fetus. The differences are in the numbers of individuals who will be made aware of their carrier status and counseled. That rate is 1.0 percent in the two-step model, <0.01 percent for the one-step model, and 2.0 percent for the expanded one-step model. Both false positive and false negative results will occur, but the rates are highly dependent on individual laboratory performance and whether confirmatory testing is performed routinely.

Figure 3-16. Prenatal Cystic Fibrosis Screening Performance According to Chosen Model in 100,000 Asian American Couples

Appendix F. Computation of Screening Performance for the Three Prenatal Models in non-Hispanic Caucasians 

Clinical sensitivity among 160 carrier couples

• For the two-step (sequential) model, 138 (160 * 0.88 * 0.979), carrier women are initially detected and 119 (137.6 *.0.88 * 0.979) of the corresponding partner-carriers are also identified. Among the 119 carrier couples, about 30 (119/4) fetuses affected with cystic fibrosis are expected.
• For the one-step (couple) model, only the 119 carrier couples are identified as having a positive test result and the same 30 fetuses are identified.
• For the expanded one-step (concurrent) model, all but the 3 couples with two unidentifiable mutations (160 * (1-0.88 * 0.979)²) will have at least one partner with a positive test result. However, the same 119 carrier couples and 30 affected fetuses will be identified.

Clinical specificity among 99,840 non “carrier couples”.
Assuming a carrier frequency of 1/25, 7,680 couples will consist of one true carrier and one true non-carrier partner. In the remaining 92,160 couples, both will be true non-carriers. All analyses assume that the ‘false positive rate' is 0.005 (0.5 percent) (e.g., analytic specificity of 99.5%).

• Applying the two-step model to the 7,680 couples yields 3,308 women with true positive tests (7,680 / 2 * 0.88 * 0.979). Among the remaining 4,372 women tested, 22 false positive results will occur (4,379 * 0.005). Among the 3,330 (3,308+22) partners tested, no true positives and an estimated 17 false positive results will occur (3,330 * 0.005). No affected fetuses will be identified among these 17 false positive couples. Applying the two-step model to the 92,160 couples in whom none are carriers will yield initial false positive results among 461 women (92,160 * 0.005). Among their partners, two will have a positive test (461 * 0.005). Overall, 3,791 women will initially be identified as being positive (3,308 true positives and 22 + 461 false positives). Up to 19 false positive couples will be reported; none will have an affected fetus identified.
• Applying the one-step model to the same groups will yield the same 19 false positive couples, but all remaining couples would receive negative test results, as the remaining women with false positive results will have partners with no mutation identified. Among the 19 false positive couples, none will have an affected fetus identified.

Applying the expanded one-step model to the 7,680 couples yields 6,616 (7,680 * 0.88 * 0.979) couples where one partner is a true positive. Of these, 6,583 (6616 * 0.995) will find one partner to be a true negative, but in 33 (6,616 * 0.005) the other partner will be a false positive and the couple will be incorrectly reported as a positive couple. Among the remaining 1,064 couples, five (1,064 * 0.005) will be incorrectly reported as one positive and one negative. The remaining 1,059 (1064 – 5) couples will be found to be both negative. Applying the expanded one-step model to the 92,160 couples in whom none are carriers will initially yield 461 false positive test results and 2 (461 * 0.005) of these will result in a false positive carrier couples. In the remaining 91,699 partner tests, an additional 458 (91,699*.005) false positive tests will occur yielding a total of 917 (459+458) couples with one positive and one negative test result. The remaining 91,241 couples will receive a correct negative/negative report.