Universal Newborn Hearing Screening:  Summary of Evidence FREE

Each year approximately 5000 infants are born in the United States with moderate-to-profound, bilateral permanent hearing loss (PHL). Estimates of the prevalence of moderate, severe, and profound congenital PHL among newborns range from 1 in 900 to 1 in 2500.^1- 8 Moderate, severe, and profound congenital PHL is associated with delayed language, learning, and speech development.^9- 13 This delay is measurable before age 3 years^14- 15 and has consequences throughout life. On average, deaf students graduate from high school with language and academic achievement levels below those of fourth-grade students with normal hearing.^16- 17

Diagnosis and treatment are often delayed until ages 1 or 2 years in children with congenital PHL, particularly among those at low risk for PHL.^18- 22 Current theory holds that auditory stimuli during the first 6 months of life are critical to the development of speech and language skills.^23- 25 Advocates of universal newborn hearing screening (UNHS) believe that earlier application of available therapies, such as speech and language therapy, amplification, and family support, could reduce or eliminate the gap in language skills between deaf and hearing children.^26- 27 Screening also identifies infants who have mild and unilateral hearing impairment, but the consequences of delay are not well established in these infants, and most are not candidates for hearing aids and other therapies associated with early identification.^7,20,28- 29

Selective screening of high-risk newborns is an alternative to UNHS. Among infants in a neonatal intensive care unit (NICU), the risk of moderate-to-severe PHL is 10 to 20 times higher than in the general population.³⁰ In addition to NICU admission, the Joint Committee on Infant Hearing high risk guidelines specify 4 other risk factors (BOX).^31- 32 From 10% to 30% of newborns meet these criteria, which can identify 50% to 75% of all cases of moderate-to-profound bilateral hearing loss.²

1. Neonatal intensive care unit admission for 2 or more days
2. Usher syndrome, Waardenburg syndrome, or findings associated with other syndromes known to include hearing loss
3. Family history of hereditary childhood sensorineural hearing loss
4. Congenital infections such as toxoplasmosis, bacterial meningitis, syphilis, rubella, cytomegalovirus, and herpes
5. Craniofacial anomalies, including morphologic abnormalities of the pinna and ear canal

In 1995 the US Preventive Services Task Force (USPSTF) found insufficient evidence to recommend UNHS,³³ based on the low prevalence of PHL and the risk for substantial misclassification. While the evidence for the efficacy of early intervention for patients diagnosed by screening was incomplete, the Task Force endorsed selective screening of high-risk newborns because of their higher prevalence of PHL.

Since 1995, many health care professionals and federal health care agencies have advocated for UNHS, which is now mandated by law in 32 states.^18,31,34 Is this widespread support for UNHS now justified? To update the USPSTF recommendations, we critically reviewed recent evidence to identify strengths, weaknesses, and gaps in the evidence supporting UNHS.

We focused our literature search on key questions underlying the clinical logic of screening for hearing impairment in newborns (Figure 1). The logic assumes that screening tests are accurate; that screening reduces delays in diagnosis and treatment; that earlier treatment results in better language function within the preschool period; and that improvement in early language function will improve educational, occupational, and social function later in life. We reviewed the literature about each assumption, and used the results to construct a mathematical model of the benefits and harms of UNHS.

We searched MEDLINE, CINAHL, and PsychINFO for relevant articles published in English from 1994 to September 2000, using the keywords hearing disorders and infant or newborn combined with terms for screening and relevant treatments, such as early intervention, amplification, and American Sign Language. The search was updated quarterly through August 2001. We also examined reference lists of review articles^7,35- 42 and queried experts. To identify articles published before 1994, we relied on systematic reviews published in 1996 and 1997.^20,33

Two authors reviewed titles and abstracts of 864 articles from the original searches and selected 340 articles as possibly relevant to 1 of the key questions. We then selected the following to include in evidence tables: (1) controlled trials; (2) reports on the accuracy, yield, or harms of screening using otoacoustic emissions (OAEs) or auditory brainstem response (ABR); or (3) reports of the effects of screening, or of early identification and treatment, on language outcomes. Ten studies of the yield of universal screening programs,^26,43- 51 1 study of the accuracy of OAEs and ABR in high-risk infants,⁵² and 8 studies of language outcomes^{14- 15,53- 58} met these inclusion criteria. An additional 110 articles provided supplemental data about the included screening studies, results of selective screening programs, and other background information.

Two authors abstracted data for population, test performance, outcomes, and methodological quality from each included study. We classified each study as "good," "fair," or "poor" using prespecified criteria developed by the USPSTF for grading the internal validity of studies.⁵⁹ When necessary, we sought from authors additional information needed to apply the criteria. The entire 13-member Task Force discussed the review, examined and rated the quality of 4 key studies of early intervention,^54,56- 58 and provided overall guidance.

We constructed a mathematical model of the likely benefits and harms of UNHS vs selective screening of a hypothetical cohort of 10 000 newborns, estimating prevalence, sensitivity and specificity, compliance, treatment effect size, and other model parameters from the included studies. Excel 97 (Microsoft Corp, Redmond, Wash) was used for all analyses.

Ten publications, including 1 controlled trial, 4 state-based programs, and 5 hospital-based programs, provided information about the yield of UNHS in actual screening programs (Table 1). In the 2 good-quality studies, screening detected 1 case of bilateral, moderate-to-profound PHL for every 925 to 1422 newborns screened. The yield was 2041 to 2794 low-risk and 86 to 208 high-risk newborns screened to find 1 case.^{29,43- 44,61} Fair-quality and poor-quality studies had higher yields due to inclusion of infants who had mild, unilateral, or unconfirmed hearing loss.

Most programs in Table 1 used a 2-stage screening protocol, in which an infant who fails the initial test (OAE or ABR) is retested in the hospital or as an outpatient within 12 weeks of discharge, and is referred for audiological evaluation if he or she fails the second test. Among infants with positive screening tests, the likelihood that the infant has hearing loss (ie, the positive predictive value [PPV] of the screening test) varied by study. In 1 good-quality study, the overall PPV for the second-stage screening test was 6.7%.⁴³ In the well-baby nursery, the PPV was 2.2%, meaning that 1 of every 45 infants referred for outpatient audiological evaluation eventually proved to have moderate-to-profound bilateral PHL. None of the screening programs measured the sensitivity and specificity of neonatal screening against an independent gold standard, although 3 reported the percentage of cases (6%-15%) missed by screening but eventually diagnosed by other means.^26,28,43

A behavioral test is the appropriate gold standard determination for permanent hearing impairment,⁶² but cannot be performed reliably before 8 to 9 months of age.^63- 64 One large, good-quality study of test performance measured the sensitivity and specificity of OAE and ABR using an independent gold standard, visual reinforcement audiometry, performed at ages 8 to 12 months.⁵² The study found that these screening tests are not sensitive enough to rule out significant hearing loss. The sensitivity of OAE ranged from 80% for moderate hearing loss to 98% for profound hearing loss. For ABR, sensitivity and specificity were 84% and 90%, respectively. The 2-stage protocol missed 11% of affected ears. Overall, neonatal testing resulted in a final diagnosis of bilateral moderate-to-profound PHL among 1 in 230 high-risk and 1 in 2348 low-risk infants.

These estimates of accuracy and yield are lower than expected, but are probably more reliable than those from actual screening programs (Table 1). In those programs, decisions about diagnosis and treatment are made on the basis of a diagnostic ABR performed when the infant is 1 to 6 months old. The use of this intermediate diagnostic standard facilitates earlier intervention, but may overestimate the number of cases of PHL. In the Wessex trial, the first audiological examination was done when the infants were between ages 8 and 12 weeks. Of 158 infants who screened positive, 27 were diagnosed as having PHL; in 2 of these cases (7.4%), however, the diagnosis was wrong, and the infants had normal hearing when reexamined at 4 months or 10 months of age.⁶¹ In another study, 5 (29%) of 17 infants initially diagnosed to have moderate PHL were later found to have only mild hearing loss.²⁸ No other studies listed in Table 1 described follow-up procedures to determine how often the intermediate diagnosis of PHL was incorrect.

The most important indicators of the benefit of UNHS are the number of additional cases of significant hearing impairment that are diagnosed and treated early.^18,29 In a British trial comparing UNHS with no newborn screening, UNHS increased rates of referral to an audiologist by age 6 months for infants with moderate-to-profound PHL (an increase of 51 per 100 000 infants; 95% confidence interval, 7.4-94.0 per 100 000; P = .03), but not confirmation of diagnosis (P = .22) or initiation of management within 10 months (P = .08).⁴³ Among those infants with moderate or severe hearing loss, however, screening led to highly significant increases in confirmation and management by age 10 months. With UNHS, 13 of 23 (57%) children with moderate or severe impairment were diagnosed by 10 months, compared with 2 of 13 (14%) without UNHS.⁴³ Use of UNHS did not reduce the rate of diagnosis after 18 months, either overall (5/27 for UNHS vs 6/26 for the control group) or in the moderate-to-severe subgroup.

In the best-quality US study,²⁹ ages of diagnosis for mild-to-moderate hearing losses were 3.5 months (mean) and 2.5 months (median), and those for severe PHL were 6.3 months (mean) and 3.8 months (median). In about 40% of cases, the diagnosis of PHL was delayed until ages 1 to 2 years because of infant illness, developmental delays, noncompliant parents,²⁹ or transient conductive hearing losses.^28,52

How much of the overall effect of UNHS can be attributed to screening low-risk infants? Uncertainty about the proficiency of selective screening makes it difficult to determine how much UNHS can add. The most frequently cited studies of selective screening were performed over 10 years ago,^65- 68 and they did not report the number of infants diagnosed before ages 6 or 10 months. In recent good-quality and fair-quality studies of UNHS, 19% to 42% of infants with PHL had no risk factors.^29,43,51,67 Only 1 study reported results in sufficient detail to calculate how often low-risk infants with bilateral moderate-to-profound PHL are diagnosed early by UNHS. For every 7692 low-risk infants screened, 1 additional case of PHL was diagnosed before age 10 months.

No prospective, controlled study directly examined whether newborn hearing screening results in improved speech, language, or educational development. None of the state-based programs described in Table 1 reported the outcomes of treatment for infants identified as having hearing impairment.

Table 2a summarizes methodological aspects and results of 8 recent cohort studies from 3 intervention programs that compared early- and late-identified children with impaired hearing. All of these studies used standardized receptive and expressive tests to evaluate speech and language skills in preschool children,^69- 70 and all reported statistically significant associations between the age at the time of diagnosis and language development at ages 2 to 5 years.

Six of these 8 studies reported speech and language results for children enrolled in the Colorado Home Intervention Program (CHIP).^{14- 15,53- 56,71- 72} One of these compared language performance of hearing-impaired children born in hospitals with UNHS programs to that for children born in hospitals with no UNHS program (Table 2).⁵⁵ It found that mean scores for expressive, receptive, and total language were within normal ranges for the screened group and 18 to 21 points higher (P<.001) than for the unscreened group. A 20-point gap is more than 1 SD lower than normal for age, which would indicate that a child with average intellect would have the language abilities of a child who had an IQ of 80. Children identified as having hearing impairment prior to age 6 months (whether in the screened or unscreened group) had a smaller gap between language development and cognitive ability than children identified after 6 months. Language development was within the normal range for 56% of the screened group compared with 24% of the unscreened group.

While this study used relevant, validated measures of language outcomes and controlled for several important potential confounders, study design issues limited the conclusions that could be drawn. Eligibility for the screened group was determined by the availability of an assessment of language outcomes at ages 2 to 4 years. Because the groups were drawn from different hospitals and time periods, factors other than exposure to UNHS might have influenced outcomes. Selection of subjects and assessment of outcome were unblinded, and neither the number of excluded subjects, nor the reasons for exclusion, are reported.

The most widely cited CHIP study compared 72 hearing-impaired children identified prior to age 6 months with 78 hearing-impaired children identified after 6 months.⁵⁴ After adjustment for cognitive function, children identified before age 6 months had language scores at or near their cognitive test scores, whereas children identified after age 6 months performed, on average, 20 points lower on language scores than on cognitive scores. Children with low cognitive abilities (cognitive quotient <80) experienced a smaller improvement in total language, but no statistically significant improvement in receptive and expressive language abilities.

This study had limitations as well. Late-identified children were more likely to be cognitively impaired, to have severe or worse hearing loss, to use sign language, and to have mothers with lower educational achievement. The statistical method used in the analysis did not simultaneously adjust for more than 2 factors and may not have removed the influence of these differences. Additionally, the study did not provide data on dropout rates in the 2 groups, and outcome assessments were not masked.

All of the studies in Table 2 had several important limitations. The study populations were composed of convenience samples. None of the studies had clear criteria for inclusion, none had blinded assessments, and all selected children for inclusion based on the availability of a language assessment between ages 2 to 5 years. This could introduce bias, because early identified children who remained in the program may have had better results than early identified children who were not available for follow-up. None of the studies provide information on attrition or follow-up rates. All but 1 compared children who were identified early and late by means other than UNHS, rather than children whose age at identification and enrollment was determined primarily by whether or not they were screened. Other factors, such as family involvement (an important contributor to language development),⁵⁷ the degree of other disability, or the quality of pediatric care, might have influenced the time of identification and the language outcome. The task force rated the strength of evidence linking early treatment with improved language function "inconclusive" and the quality of evidence as "fair/poor."

Screening. Potential adverse effects of false-positive screening tests include misdiagnosis, parental misunderstanding and anxiety, and unfavorable labeling. As noted earlier, the intermediate diagnostic standard determining PHL is imperfect; in expert hands, as many as 7% of infants diagnosed as having permanent PHL may eventually prove to have normal hearing. The frequency of misdiagnosis in everyday practice settings has not been studied.

In the only controlled trial,⁴³ parents whose infants were screened had anxiety and attitudes similar to parents in the unscreened group. Three other small studies found false-positive screening results produced significant or lasting anxiety in 3.5% to 14% of parents.^50,73- 74 Additionally, 8% of mothers said they treated their child differently (eg, spoke louder or clapped their hands).⁵⁰ No study attempted to assess the effect of parental anxiety or changes in parental behavior on infants' development or on the parent-infant relationship.

Treatment. The harms of early intervention have not been adequately studied, and differing ethical and philosophical attitudes about deaf awareness and culture have led to controversy about the content of early interventions.³³ Treatment strategies for hearing loss in children include hearing aids or other amplification, American Sign Language and/or English instruction, speech and language therapy, and family education and support. Different experts advocate substantially different approaches based on competing theories of language acquisition and communication. Treatment strategies vary widely, even among programs in states or hospitals with established screening programs.⁷⁵ This variation reflects uncertainty about the efficacy of the interventions, which have not been evaluated in randomized trials or in population-based cohort studies.^13,20,33

The argument for early intervention is based on the prevailing theory of language development, which holds that early auditory input is an important precursor of language development. An opposing viewpoint suggests that, during infancy, nonverbal communication, joint attention, shared experiences, and mutual understanding are more important precursors of language development than are hearing speech and forming sounds. Proponents of this view theorize that early intervention could harm infants because it leads parents to focus on "means of communication the child has the least prerequisites for" and on the child's disability instead of his or her competencies.⁷⁶ Because there are no randomized trials of different management strategies, it is impossible to assess the merits of these concerns.

Table 3 summarizes the benefits and harms of UNHS and selective screening in a hypothetical cohort of 10 000 newborns. We used our literature review to estimate prevalence, sensitivity and specificity, compliance, and the likelihood of being diagnosed and treated before age 10 months. There are no reliable data to estimate how often selective screening misses patients whose risk factors were not detected during hospitalization. We assumed that in a selective screening program, 20% of high-risk infants are never tested in the hospital, vs 10% for UNHS. There are also no reliable data by which to estimate the probability that a low-risk infant will be diagnosed by 10 months without newborn screening; we estimate this to be 35% in our base case.

With UNHS, an additional 7800 screening tests would be done, resulting in the diagnosis of 6 additional cases of moderate-to-profound hearing loss diagnosed before age 10 months. Of these, 3 additional cases would be treated before age 10 months. Thus, the number needed to screen (NNS) to detect 1 additional case before age 10 months would be 1441 and the NNS to treat 1 additional case before 10 months would be 2401. With UNHS, 254 newborns would be referred for audiological evaluation because of false-positive second-stage screening test results (vs 48 for selective screening), and 1 of these would also be falsely diagnosed to have PHL at the first posthospital visit to an audiologist.

Of the 6 additional early diagnosed, low-risk newborns, how many would actually benefit from early treatment? The data needed to estimate this—the probabilities of a poor language outcome with and without early treatment—are not known. To use a hypothetical example, if 50% of newborns with PHL would have poor language ability if diagnosed after age 10 months, and early intervention reduced this by 50%, then the NNS to prevent 1 additional case of delayed language acquisition would be 6771.

Table 4 summarizes the evidence for each of the major assumptions underlying the case for UNHS. Several gaps in information about UNHS effectiveness remain. Modern screening tests for hearing impairment can improve identification of newborns with PHL, but as many as 10% of newborns with normal hearing will require a second screening test. From 1% to 3% of newborns will be referred for audiological assessment; over 90% of those referred are false-positives. The consequences of false-positives have not been adequately evaluated, nor has the reliability of audiological and behavioral assessment⁶⁴—the standard used to make decisions about treatment—been established. Moreover, the false-negative rate is higher than previously thought, probably 20% to 30% in most programs. This new finding calls into question the assumption that a newborn who passes a screening test has normal hearing. It also suggests that stricter pass criteria, which have been promoted as a way to reduce false-positives, also reduce the effectiveness of screening.

A clearer picture of the consequences of delayed diagnosis in low-risk newborns would strengthen the case for universal screening. While diagnosis of hearing impairment is often delayed in children with congenital hearing impairment, the risk of delayed language development in otherwise healthy infants diagnosed at ages 1 or 2 years is not known. Because the frequency and severity of poor language outcomes in this group is uncertain, only adequately controlled trials or cohort studies can establish the efficacy of early intervention.

Several cohort studies show that, by ages 2 to 4 years, children with hearing aids and other therapy in the first 6 months of life had better language skills than those treated later. None of these studies compared an inception cohort of newborns offered UNHS with infants managed by usual care (including selective screening). Additionally, these studies had unclear criteria for selecting subjects, making it impossible to exclude baseline differences in the compared groups as a cause for the association of early identification with language development. The hypothesis that early intervention is a predictor of language acquisition is plausible, but the studies do not establish that screening low-risk newborns is the important factor. None of the studies attempted to link short-term improvements to better function later in life.

As use of UNHS rapidly increases, it is important to conduct longitudinal studies of UNHS to address these knowledge gaps. Further randomized trials of UNHS seem unlikely to be conducted in the United States. Although it would be possible to compare states with and without UNHS, such studies would be prone to uncontrollable confounding due to differences among states. However, better evidence about the effectiveness of UNHS is needed and could be obtained via population-based studies that begin with inception cohorts and carefully report outcomes in all possible patients, as well as rates of loss to follow-up. Speech, language, and scholastic achievement of deaf and hard-of-hearing children should be followed over time. States that have UNHS should conduct such population-based studies to evaluate whether the long-term language outcomes of deaf children improve as the age of identification decreases.