TERATOLOGY 42:513-519 (1990)

Sensitivity, Specificity, and Positive Predictive Value of Multiple Malformations in lsotretinoin Embryopathy SurveiIlance MICHELE C. LYNBERG, MUIN J. KHOURY, EDWARD J. LAMMER, KIRSTEN 0. WALLER, JOSE F. CORDERO, AND J. DAVID ERICKSON Birth Defects and Genetic Diseases Branch, Division of Birth Defects and Developmental Disabilities, Center for Environmental Health and Injury Control, Centers for Disease Control, Atlanta, Georgia 30333 (M.C.L., M.J.K.,J.F.C., J.D.E.); California Birth Defects Monitoring Program, Emeryville, California 94608 (E.J.L.);Division of Field Services, Epidemiology Program Office, California Department of Health Services, Emeryville, California 94608 (K.O.W.)

ABSTRACT

Isotretinoin causes serious birth defects in about 25%of babies exposed in the first trimester of pregnancy. Despite warnings about the drug's teratogenicity, cases of isotretinoin embryopathy continue to occur; more than 80 such cases have been reported since 1982. The true magnitude of the problem is unknown, however, and case estimates range to more than 1,000. The need for isotretinoin embryopathy (IE) surveillance is therefore great. Sixty-one known cases were evaluated to determine the sensitivity (proportion of cases with a given defect pattern) of various defect combinations. Using data from the Metropolitan Atlanta Congenital Defects Program for the period before isotretinoin was available, we evaluated the specificity (proportion of malformed infants without exposure who do not have the pattern of defects) for the various defect combinations. Ear malformations (microtia, anotia, absence or stricture of auditory canal, missing pinnae) have an associated sensitivity of 70.5% and a specificity of 99.5%.Ear defects combined with central nervous system (CNS) defects (microcephalus, hydrocephalus, reduction deformities of the brain) and cardiovascular (CVS) defects (conotruncal defects, aortic arch abnormalities) have an associated sensitivity of 19.7%and a specificity of 100.0%.The case definition of ear defects combined with either CNS or CVS defects maximizes both specificity (99.9%)and sensitivity (45.9%).The investigators are now evaluating the feasibility of using this pattern of defects to monitor for IE within a national monitoring program.

Isotretinoin causes serious birth defects in about 25% of fetuses exposed in the first trimester of pregnancy (Lammer et al., '88). Despite warnings about the drug's teratogenicity, cases of isotretinoin embryopathy (IE) continue to occur. More than 80 cases have been reported since 1982 when the drug was introduced into the American market (Food and Drug Administration, '89). The true magnitude of the problem is unknown, however, and case estimates range from about 100 to more than 1,000 (Food and Drug Administration, '88). Despite isotretinoin's potent teratogenicity, the Food and Drug Administration (FDA) has allowed the drug to remain on the market largely because of its effectivePUBLISHED 1990 BY WILEY-LISS, INC.

ness in the treatment of recalcitrant cystic acne. Following discussions with the FDA, the manufacturers of isotretinoin sent a series of letters alerting physicians about the drug's dangers, revised product labels, and instituting new guidelines for the use of isotretinoin among women of reproductive age. Public health officials must be able to monitor the continuing occurrence of IE in the United States, particularly in light of these new intervention and prevention strategies. Received April 12, 1990; accepted July 11, 1990. Address reprint requests to Michele C. Lynberg, PhD, MPH, Division of Birth Defects and Developmental Disabilities, Birth Defects and Genetic Diseases Branch, Centers for Disease Control, 1600 Clifton Road, MS F37, Atlanta, GA 30333.

514

M.C. LYNBERG ET AL.

Birth defects monitoring programs usually examine trends in rates of specific single defects, but they do not concurrently monitor for defects that occur in combination with other specific anomalies. However, many known teratogenic substances are associated with a spectrum of birth defect combinations (Sever and Brent, '86; Shepard, '86; Schardein, '85; Jones, '88). The benefits of using combinations of two and three defects for monitoring trends in birth defects have been previously discussed (Khoury et al., '87). Depending on the particular congenital malformations caused by a teratogen, a n increase in the rate of birth defects due to the exposure may be detected more rapidly by monitoring rates of defect combinations rather than by monitoring the rates of individual defects. Monitoring of multiple malformations is more effective if affected infants tend to have combinations of defects rather than isolated defects. A constellation of characteristic defects occurs in fetuseslinfants exposed to isotretinoin in the first trimester (Lammer et al., '85; Lammer, '87; Lammer et al., '88). Craniofacial, central nervous system (CNS), cardiovascular system (CVS), and thymic malformations are the most common defects. Ear abnormalities include microtia, anotia, low set ears, and deafness. CNS defects include microcephalus, hydrocephalus, and reduction malformations of the brain such as agenesis of the corpus cerebellar vermis. CVS defects most commonly include conotruncal heart defects (e.g. tetralogy of Fallot, transposition of the great vessels), aortic arch abnormalities, and ventricular septa1 defects. Other defects, including micrognathia, retinal and optic-nerve abnormalities, and cleft palate are less prevalent among exposed fetuseslinfants. To determine which associated defects might be useful in monitoring for IE, we evaluated the sensitivity, specificity, and positive predictive value of patterns of defects. The results of this evaluation are being used to investigate the feasibility of using this pattern of defects to monitor for IE within the national Birth Defects Monitoring Program. METHODS

Study population For this analysis, we studied the first 61 malformed fetuses and infants identified by the Longitudinal Study of Infants Exposed

to Isotretinoin in Utero. These 61 cases were accumulated from 2 study populations that differed by their methods of ascertainment. The first population, a case series, is comprised of malformed isotretinoin-exposed subjects who were ascertained retrospectively, that is, after delivery and diagnosis. The second population, a n exposure cohort, is comprised of malformed fetuses and infants who were delivered of prospectively ascertained group of women who used isotretinoin after conception. For both of these populations, the malformed fetuseslinfants and exposed pregnancies were initially identified through reports to the isotretinoin manufacturer, FDA, CDC, or the investigators of the Longitudinal Study of Infants Exposed to Isotretinoin in Utero. The reports came from physicians, genetic counsellors, teratology information services, and patients. The occurrence of birth defects among these fetuseslinfants has been described in detail (Lammer et al., '85; Lammer, '87; Lammer et al., '88). For this investigation, records were reviewed and defects were classified according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). As previously mentioned, the most common organ systems involved were the ear, CVS, and CNS. The ICD-9-CM codes included in this analysis are shown in Table 1. Specific ear malformations included within ICD-9-CM code 744.0 were complete absence of the ear, absence or stricture of auditory canal, and absence of auricle. The ear malformation included within ICD-9-CM 744.2 was microtia.

Metropolitan Atlanta Congenital Defects Program The Metropolitan Atlanta Congenital Defects Program (MACDP) is sponsored by the Centers for Disease Control (CDC), the Georgia Mental Health Institute Laboratory of Human and Behavioral Genetics, and Emory University School of Medicine (Edmonds et al., '811. Conducted in cooperation with hospitals and physicians in metropolitan Atlanta, MACDP began surveillance activities in 1967 and monitors regularly and systematically the occurrence of malformed births among residents within the five-county metropolitan Atlanta area. About 35,000 births are monitored each

ISOTRETINOIN EMBRYOPATHY SURVEILLANCE TABLE 1 . Categorization of birth defects observed in isotretinoin-exDosed fetuseslinfants MACDP (8digit code) BDMP (ICD-9-CM codes) Orean svstem Defect Code Defect Ear Absence or stricture of auditory canal 744.000-744.009 Anomalies of the ear causing hearing impairment Absence of auricle 744.010-744.019 Microtia 744.210-744.219 Other specified anomalies CNS Microcephalus 742.100-742.109 Microcephalus Reduction deformities of brain 742.200-742.299 Reduction deformities of brain 742.300-742.399 Congenital hydrocephalus Congenital hydrocephalus cvs Common truncus 745.000-745.019 Common truncus 745.100-745.199 Transposition of great vessels Transposition of great vessels 745.200-745.209 Tetralogy of Fallot Tetralogy of Fallot 745.300-745.309 Common ventricle Common ventricle 747.100-747.199 Coarctation of the aorta Coarctation of the aorta/ aortic arch abnormalities 747.200-747.299 Other anomalies of the aorta Other anomalies of the aorta

year. The system uses multiple-source case ascertainment to obtain information on any live-born or stillborn infant with a major structural abnormality present a t birth, diagnosed within the first year of life, and ascertained before the infant is 6 years of age. Congenital malformations are classified using a detailed 6-digit code. While this code was originally based on the ICD-9-CM coding scheme, its specificity allows a more detailed classification of birth defects. For example, classification for the ear defects commonly seen in IE may be selected directly from the 6-digit code rather than from the larger ICD-9-CM coding scheme, which includes numerous other unrelated ear malformations. Specificity and predictive value analyses were performed using the 6-digit MACDP codes shown in Table 1. However, because the national birth defects monitoring program is based on the ICD-9-CM7we originally evaluated sensitivity of defect combinations by selecting records with ICD9-CM codes shown in Table 1. After manually reviewing the actual defects, we deleted records which did not fit the case definition. For example, an infant with ICD-9-CM 744.0 recorded on the discharge summary, who had anomalies of the ear ossicle, would not fit the case definition and would be excluded manually. Only babies with absence or stricture of the auditory canal and/or absence of the auricle were considered to have appropriate ear defects. Similarly, we excluded inappropriate CNS and CVS defects, defects which were coded within ICD-9-CM codes selected but which were not actually observed among the isotretinoin embryopathy case series.

515

Code 744.0 744.2 742.1 742.3 742.3 745.0 745.1 745.2 745.3 747.1 747.2

Definitions of Sensitivity, specificity, and positive predictive values

The sensitivity of a pattern of defects can be defined as the proportion of malformed infants with the teratogenic exposure who have the defect pattern (Khoury et al., '89). In this study, the sensitivity of a given defect pattern is defined as the proportion of malformed isotretinoin-exposed fetuses/ infants who have the given defect pattern. For example, if all 61 fetuseslinfants had anotia, the sensitivity of anotia would be 100%. The sensitivity of a defect pattern is an estimate of the proportion of all malformed isotretinoin-exposed fetuseslinfants who would be ascertained if that defect pattern was used in monitoring. A sensitivity of 100% is optimal, indicating ascertainment of all fetuseshnfants with IE. A pattern of defects may also occur without the teratogenic exposure. Specificity of a given defect pattern can be defined as the proportion of malformed fetuseslinfants without the teratogenic exposure who do not have the pattern of defects (Khoury et al., '89). We determined specificities by evaluating the occurrence of defect patterns among MACDP births that occurred between 1968 and 1982, before isotretinoin's introduction into the American market. Achieving 100% specificity for a given pattern of defects would require the complete absence of the defect pattern within the monitoring program during this time period. A specificity of 100%would be ideal, in that the defect pattern would not occur unless the fetuslinfant had been exposed to isotretinoin.

516

M.C. LYNBERG ET AL.

TABLE 2. Sensitivity of defect patterns in isotretinoin-exposedcase series ( N = 61) Number with Percent Defectk) defects with defects EAR 43 70.5 30 49.2 CNS cvs 20 32.8 24 EAR + CNS 39.3 15 24.6 EAR + CVS CNS + CVS 14 23.0 EAR + CNS + CVS 11 18.0 EAR + (CNS or CVS) 28 45.9

Sensitivity and specificity provide a n incomplete picture of the defect pattern's ability to be used as a predictor for in utero exposure to isotretinoin. Also important is positive predictive value (PPV), the probability that a fetuslinfant has a specific teratogenic exposure given that the infant has a n observed pattern of defects. The importance of PPV in birth defects monitoring as well as its relationship with sensitivity and specificity has been previously discussed (Khoury et al., '89). In this study, positive predictive value (PPV) is the proportion of fetuseslinfants with the given defect pattern who are expected to have isotretinoin embryopathy [P(E(D)]. P(EID) is derived from the following formula:

TABLE 3. Specificity of defect patterns, metropolitan Atlanta, 1968-82 fN = 12,224) Number Percent without without Defect(s) defect(s) defects EAR 12,159 99.47 CNS 11.627 95.12 cvs 11;728 95.94 EAR + CNS 12.219 99.96 ,EAR + CVS 12.221 99 98 CNS + CVS i2,204 99.84 EAR + CNS + CVS 12,224 100.0 EAR+ (CNS or CVS) 12.216 99.93 ~

~

I---

ascertained in a high quality monitoring system such as MACDP. Specificities of the same defect patterns are shown in Table 3. In contrast to sensitivity, specificity is high and increases with combinations of two or three defects. The most specific defect pattern is the combination of ear, CNS, and CVS defects, with a n associated specificity of 100.0%.This particular combination did not occur within MACDP during the given time period. The combination of ear and either CNS or CVS defects has a n associated specificity of 99.93%. PPV of three selected defect patterns is shown in Figure 1. Ear defects combined with CNS and CVS defects, the most specific PIEID) = A x Se x P(E)/[IA x Se x P(E)) + (p0 x pattern, is the most predictive of isotretin(1 - Sp) x (1 - PIE))] oin exposure. At all exposure frequencies, where A = attack rate, Se = sensitivity, the occurrence of these three groups of deP(E) = probability of exposure in the popu- fects in combination is 100% predictive of lation, p o = the probability of any defect in IE. All fetuseslinfants with this constellaan unexposed population, and S p = speci- tion of defects would, theoretically, be IE fetuseslin fants. ficity. On the other hand, ear defects alone are RESULTS less predictive of IE. The PPV of ear defects On the basis of defects most commonly in combination with either CNS or CVS deseen among fetuseslinfants exposed to fects is intermediate between the PPV of ear isotretinoin in utero, sensitivities of various defects alone and the PPV of ear defects defect patterns were calculated. Sensitivi- combined with CNS and CVS defects. If the ties of selected combinations of one, two, exposure rate among pregnant women is inand three defect categories are shown in Ta- deed 1 in 1,000, as estimated by the FDA ble 2. These defect patterns are not mutu- (Food and Drug Administration, '881, the ally exclusive. For example, a baby with combination of ear defects with either CNS ear, CNS, and CVS defects would be in- or CVS defects has a n associated PPV of cluded in all groups. Defect combinations 85%. Most suspected IE cases would be acare less sensitive than single defects. Fe- tual IE cases. tuseslinfants are much more likely to have a DISCUSSION CNS defect than they are to have both a CNS and a CVS defect. None of the defect The constellation of defects caused by expatterns is particularly sensitive. Depend- posure to isotretinoin in utero, when ing on the defect pattern used, between 18.0 grouped in combinations of two and three and 70.5% of all IE fetuseslinfants might be defects, has varying sensitivities and posi-

517

ISOTRETINOIN EMBRYOPATHY SURVEILLANCE

0.0

0.1

0.2

0.3

0.5

0.4

0.6

EXPOSURE FREQUENCY ~

EAR

.......

EAR+(CNS/CVS)

0.7

0 8

-----

EAR+CNS+CVS

0.9

1.0

( x )

Fig. 1. Predictive value (%) for isotretinoin exposure of selected defect combinations by population exposure frequency.

tive predictive values for exposure to isotretinoin in utero. The association of decreasing sensitivity with increasing number of defect combinations has been illustrated for other teratogens (Khoury et al., '89).The most sensitive single defect related to in utero exposure to isotretinoin is an ear defect, whereas the most predictive defect pattern is the combination of ear, CNS, and CVS defects. PPV is important because it measures our ability to distinguish between malformations caused by isotretinoin and malformations resulting from other causes. If a given defect pattern is used to screen for IE, then optimally, that defect pattern would have a high predictive value. Most of the suspected IE fetuseslinfants would be actual IE fetuseslinfants. PPV depends on the prevalence of exposure in the population. With widespread teratogenic exposure, the positive predictive value for single defects tends to increase markedly. With relatively rare exposures, such as with isotretinoin, however, the ability to delineate a teratogenic syndrome requires the presence of combination of defects in affected infants. In addition to depending on the prevalence of exposure, PPV also depends on the specificity of the defect pattern. Even a small decrease in specificity has a large impact on the PPV of that defect pattern.

The relationship between sensitivity and positive predictive value requires increases in one to be traded for decreases in the other. This results in a dilemma for monitoring systems: should researchers use the most sensitive pattern of defects to obtain the largest number of potentially exposed cases or should they use a less sensitive but more specific case definition that underascertains cases but allows more certainty about the exposure status of cases ascertained? To monitor most effectively, as many cases as possible should be ascertained with as little resources as possible. False positive cases result from decreased specificity, which is a cost of attaining high sensitivities and high ascertainment. The IE case series indicates ear defects have the highest sensitivity whereas the combination of ear, CNS, and CVS defects have the lowest sensitivity. Even if you monitor ear malformations, you will ascertain only 70% of all cases, while increasing the overall number of cases which require review. Our analysis of MACDP data indicates that the combination of ear, CNS, and CVS defects is the most specific combination, whereas CNS defects alone are the least specific. As previously mentioned, as many as 1in 1,000 pregnant women have been exposed to isotretinoin. Because isotretinoin exposure

518

M.C. LYNBERG ET AL.

is relatively rare, a case definition th at maximizes sensitivity leads to a very low predictive value and thus inefficient surveillance. Among those defect patterns with the greatest specificity (greater than 98%), the combination of ear with either CNS or CVS defects is the most sensitive (45.9%). This combination will capture nearly 50% of cases and is reasonably specific. To illustrate the impact that sensitivity and PPV have on monitoring, let us assume that if 260 IE births occur each year and a monitoring program monitors one-third of all U.S. births, then 87 IE births would occur within the monitoring system. With a sensitivity of nearly 50%, about 50 births would be ascertained. With a PPV of 85%, about 47 records would require review to determine those infants who were exposed to isotretinoin. Although ventricular septa1 defect (VSD) is observed frequently in fetuseslinfants exposed to isotretinoin, it has not been included in the case definition. This decision was based on the markedly decreased specificity of heart malformations when VSD was included. VSD is the most common heart malformation and few would be attributed to isotretinoin exposure. Because VSD would be of little use in monitoring for IE it was excluded from the heart malformations used for IE screening. In addition, “low set ears” was mentioned in some of IE records. The defect was considered too vague for this analysis, however, because this minor anomaly is likely to be poorly reported in birth defects surveillance systems. Coarctation of the aorta and interrupted aortic arch also merit discussion. ICD-9-CM code 747.1 includes both defects. Although two cases were originally diagnosed with coarctation of the aorta, upon further diagnostic assessment both were determined to be aortic arch interruption Type B, rather than coarctation. The appropriate ED-9-CM code would be 747.2. Because we felt that (1) other IE fetuseslinfants may be inappropriately diagnosed with coarctation, (2) other fetuseslinfants were diagnosed with interrupted aortic arch, and (3) ICD-9-CM 747.1 includes both interrupted arch and coarctation, we decided to retain both defects in the case definition. Based on the analysis of the case series and MACDP data, the combination of ear and either CNS or CVS defects seems to be

the most useful for IE monitoring. This is the combination currently being evaluated within the national Birth Defects Monitoring Program (BDMP). BDMP was established at CDC in 1974 (Edmonds e t al., ’81). BDMP’s primary objective is to serve as a n early warning system to detect increases or unusual trends in congenital anomaly rates in the United States. Data for BDMP are obtained from two health data processing organizations: the Commission on Professional and Hospital Activities (CPHA) and the McDonnell Douglas Health Information System (MDHIS). These two systems operate similarly as passive surveillance systems; discharge abstracts are coded by hospital records personnel and submitted regularly to CPHA or MDHIS for processing. Information is forwarded quarterly to the Division of Birth Defects and Developmental Disabilities (DBDDD) where birth defects information obtained from the newborn period is processed and analyzed. BDMP is the largest source of data available on malformed newborns delivered in the United States. In 1987,823,366 (21.6%)U.S. births were monitored. One limitation of applying the results of this investigation to the national system, BDMP, is the important differences between BDMP and MACDP. MACDP generally has a much higher level of ascertainment than BDMP (Edmonds et al., ’81).The passive surveillance of BDMP as well as the limited ascertainment of birth defects within the newborn period may result in underascertainment of birth defects associated with isotretinoin embryopathy. Additionally, the BDMP system uses ICD-9-CM codes obtained from the discharge summary, whereas MACDP uses a more detailed modification of the ICD-9-CM code. The increased detail within MACDP allows investigators to select more specific defects, which in tu rn affects the specificity of defect combinations used in monitoring. These factors may affect the ability to monitor for IE at the national level, using BDMP. BDMP remains as the only source of national birth defects data, however. Because of these concerns, we are also initiating isotretinoin surveillance activities within a number of state-based birth defects surveillance systems similar to the MACDP. Another limitation of this investigation is the ascertainment bias of some of the mal-

ISOTRETINOIN EMBRYOPATHY SURVEILLANCE

formed fetuseslinfants used to define the pattern of isotretinoin-induced defects. Malformed exposed fetuseslinfants who were retrospectively identified are likely to over represent the true proportion of severely affected fetuseslinfants. Therefore, a potential overestimation of sensitivity may result. Although a prospectively followed cohort has been thoroughly evaluated in terms of determining adverse effects of in utero exposure to isotretinoin, major defects are ascertained about equally well in both the retrospective case series and the prospective cohort. However, minor defects and developmental disabilities, outcomes generally unsuitable for monitoring purposes, are more thoroughly ascertained in the prospective cohort. The number of fetusedinfants with major malformations identified through the prospectively followed cohort of exposed pregnancies is too small ( N = 11) to define an unbiased population for use in this study; thus, the population we used, although potentially biased, represents the only current opportunity for such an analysis. We must be able to monitor the continuing occurrence of isotretinoin embryopathy if we are t o understand whether the new intervention and prevention strategies are effective in reducing these preventable birth defects. The combination of ear defects with either CNS or CVS defects promises to be useful in monitoring isotretinoin embryopathy. An evaluation of the proposed case definition for IE surveillance is currently under way within the nationwide Birth Defects Monitoring Program.

519

LITERATURE CITED Edmonds, L.D., P.M. Layde, L.M. James, J.W. Flynt, Jr., J.D. Erickson, and G.P. Oakley, Jr. (1981) Congenital malformations surveillance: Two American systems. Int. J . Epidemiol., 10:247-252. Food and Drug Administration, Center for Drug and Evaluation Research. Dermatologic Drugs Advisory Committee open public hearing on Accutane (isotretinoin) capsules. April 26, 1988. NDA-18-662. Food and Drug Administration. Center for Drug and Evaluation Research. Dermatologic Drugs Advisory Committee open public hearing on Accutane (isotretinoin) capsules. May 8, 1989. F89-285-65. Jones, K.L. (1988) Smith's Recognizable Patterns of Human Malformation, 4th ed. W.B. Saunders Company, Philadelphia. Khoury, M.J., M.M. Adams, P. Rhodes, and J.D. Erickson (1987) Monitoring for multiole malformations in the detection of epidemics of birth defects. Teratology, 36,345-354. ~.~ ~~

~~~

Khoury, M.J., J.E. Becerra, J.F. Cordero, and J.D. Erickson (1989) Clinical-epidemiologic assessment of patterns of birth defects associated with human teratogens: Applications to diabetic embryopathy. Pediatrics, 843358-665. Lammer, E.J. (1987) Patterns of malformation among fetuses and infants exposed to retinoic acid (isotretinoin). In: Banbury Report 26: Developmental Toxicology: Mechanisms and Risk, Cold Spring Harbor Laboratory, Cold Spring Harbor, New York. Lammer, E.J., D.T. Chen, R.M. Hoar, et al. (1985) Retinoic acid embryopathy. New Engl. J . Med., 313: 837-841. Lammer, E.J., A.M. Hayes, A. Schunior, and L.B. Holmes (1988) Unusually high risk for adverse outcomes of pregnancy following fetal isotretinoin exposure. Am. J. Hum. Genet., 43:A58 (Abstract). Schardein, J.L. (1985) Chemically Induced Birth Defects. Marcel Dekker, New York. Sever, J.L., and R.L. Brent, eds. (1986) Teratogen Update: Environmentally Induced Birth Defects Risks. Alan R. Liss, New York. Shepard, T.H. (1986) Catalog of Teratogenic Agents, 5th ed. Johns Hopkins University press, Baltimore.

Sensitivity, specificity, and positive predictive value of multiple malformations in isotretinoin embryopathy surveillance.

Isotretinoin causes serious birth defects in about 25% of babies exposed in the first trimester of pregnancy. Despite warnings about the drug's terato...
613KB Sizes 0 Downloads 0 Views