Approximately one to three per 1000 children are born with at least moderate, bilateral hearing disorders 1234. Children with congenital hearing impairment benefit from early detection and treatment of their hearing loss 5. The neurological development of hearing abilities requires acoustic stimulation in the first 18 months of life. Deficits due to lack of acoustic stimulation within the first two years are not or not easily recovered by later rehabilitation. The consequences include delayed development of speech and other cognitive and social functions. This delay is already measurable in the first 3 years of life 6. If disorders are detected and treated in time, either by the use of a hearing device or cochlea implant, most of the children develop normally and do not need additional speech therapy 78. Early diagnosis and treatment within the sensitive time frames are therefore essential. The German consensus conference on neonatal hearing screening proposed diagnosis in the first 3 months and the start of treatment in the first 6 months of life.

Various tests and test combinations with acceptable sensitivity and specificity are available. Transient evoked oto-acoustic emissions (TEOAE) or brainstem evoked responses (BERA) can be measured. One common strategy consists of a two-step TEOAE with the first measurement within the first days of life and a second test a few days later 9 (chapter 2.2.7).

While modern screening techniques for early detection are available there is still a gap between the consensus on detection as early as possible and the current situation. Only 50% to 60% of children with permanent hearing impairment are diagnosed before their second birthday with traditional health care services 2. While congenital hearing loss is a serious health problem, there is little evidence to support the use of routine universal screening because of the following factors:

• as the prevalence is very low, the positive predictive value of the tests is low

• screening technologies are still in development

• possible costs and consequences are not sufficiently known

• benefits of early intervention are frequently expressed in qualitative terms without presenting unbiased measures of outcome.

In 1995 the US Preventive Services Task Force found insufficient evidence in favour of universal neonatal hearing screening (UNHS) 10. The Task Force proposed selective screening of new-borns with risk factors to improve the predictive value of the test. In the UK a national neonatal hearing screening programme aimed at detecting bilateral moderate to severe hearing impairments has been recommended 11 and partially implemented within a pilot project 12. Several studies have modelled the outcomes of UNHS versus risk factor screening 131415. Keren et al 16 presented a cost-effectiveness analysis based on a decision tree model which reported short-term effectiveness of UNHS compared to risk factor screening.

However, there is presently no model explicitly quantifying the effectiveness of UNHS versus other programme alternatives in terms of early diagnosis, nor has it been taken into account that diagnosis of hearing impairment within the first few months of life is more "valuable" for the child's further development than diagnosis later in life. Children identified within the developmentally sensitive time frame should therefore be given more weight compared to children with delayed diagnosis.

The objective of this clinical decision analysis was to systematically compare two screening strategies for the early detection of new-born hearing disorders: UNHS and risk factor screening; with the option of no systematic screening regarding their influence on early diagnosis.

Our specific objectives were to show differences between strategies expressed as the number of quality weighed detected child months (QCM), the number of true positive cases at the age of 6 and 12 months, and the number of false positive cases. We also wanted to investigate which parameters had the most influence on the reported differences and how likely these differences were.

We developed a clinical decision model to compare two different screening strategies with the option of no systematic screening. Parameters were extracted from the literature, empirically derived from a representative patient survey, and estimated by experts. Univariate and multivariate sensitivity analyses were performed on all relevant parameters.

The decision model was used to predict absolute and incremental effectiveness of two new-born hearing strategies compared with the option of no screening in new-born infants. For the modelling of effectiveness the recommendations of the Panel on Cost-Effectiveness in Health and Medicine were followed 17.

Three possible strategies for neonatal hearing screening (NHS) were evaluated:

- Universal neonatal hearing screening (UNHS): Every hospital-born baby is screened during the first days of life.

- Risk factor screening (RS): The prevalence of hearing disorders is estimated to be higher at risk groups such as children with a positive family history, with congenital infections, cranofacial abnormalities, low APGAR-Score or low birth weight. This strategy screens all children with one or more risk factors for hearing disorders.

- No systematic screening (NS): Children undergo the usual distraction test when presented to the paediatric service during the routine visit at the age of 12 weeks. Some neonates are screened in the hospital, but not in a systematic way. This reflects the present situation in Germany. UNHS is performed in some maternity hospitals and outpatient paediatric services with pilot screening projects in several regions.

The target population of this analysis was all newborn infants. Health effects are expressed as quality weighed number of detected child months (QCM), and as true positive and false positive cases at certain developmentally important ages (6 and 12 months); for example, if a hearing impairment was diagnosed briefly after birth, the infant contributed six QCM at the age of six months. If the infant's hearing loss was diagnosed at the age of five months, the infant added only one detected child month at the age of six months. The term 'quality' is to reflect the idea that the early detection of impairment is a better and desired outcome, although there is no data on quality of life gained by this early detection. QCM, true positives and false positives are reported at the age of 6 and 12 months and with a time horizon of 120 months. Child months which were added up until the age of 6 months were multiplied with a weight of 1, child months added after the age of 6 months were multiplied with decreasing weighting. The derivation of this weight index is described in the section 'data and assumptions' below. We assumed that all children with hearing impairment would be detected before the age of 72 months, the age of school entry, regardless of the kind of screening strategy. In order to give outcomes in the present more weighting, compared to outcomes in the future, future effects have been discounted at an annual rate of 3%. Discounting reflects the higher value of money spent now as opposed to in the future. Similarly, discounting also weights outcomes experienced now (e.g., being diagnosed as true positive) more heavily than those experienced in the future. Table 1 gives the model parameters and their references.

We developed a decision-analytic Markov model for the evaluation of the effectiveness of different new-born hearing screening strategies. In our model, UNHS identified 72% of all detectable QCM at the age of 6 months, RS identified 44% QCM, the control group with no systematic screening identified 17% QCM. UNHS shows higher effectiveness even under a wide range of additional relevant parameters. QCM was introduced as a dynamic time-to-event measure which takes into account that the age of confirmation of congenital hearing impairments is important for further language development. The model results were not sensitive to test the accuracy of parameters within the assumed range but varied with the prevalence of hearing impairment. We have shown that UNHS leads to an earlier age of confirmed diagnosis compared to selective RS.

The cost effectiveness of UNHS has already been modelled along secondary data 162324. This model is the first to establish a time-dependent and quality-weighted outcome, to introduce empirical data of the natural history of discovery and to present the results within a probabilistic framework. The strength of the presented model is that detected child months are multiplied by a weighted function which adds more benefit per month to children that were diagnosed before 6 months compared to those with late detection. In the existing literature the most interesting outcome, the proportion of children detected early enough, has not been modelled 16.

Our findings are consistent with study results and other projections of effectiveness. The Wessex Universal Neonatal Hearing Screening Trial Group found that 71 more babies with moderate or severe hearing loss per 100 000 target population were diagnosed before the age of 6 months during periods with neonatal screening than during periods without 25. We found that UNHS would yield 112 true positive cases per 100 000 as compared to 38 without systematic screening, which is a difference of 74 babies. In our model UNHS identified 28% more children in time compared to RS. Thompson et al. estimated this difference to be between 19% and 42% 22 and stated that 77% of hearing impaired children would be identified before 10 months. Keren et al. assumed that UNHS detects 77% and RS 52% of hearing impaired children at the age of 6 months 16. Any differences may be due to the fact that our function of discovery without systematic screening is more pessimistic.

Our study has several limitations. Firstly, it does not differentiate between bilateral and unilateral hearing loss or moderate and profound impairment. We believe, however, that even with a more refined model the differences in effectiveness would not be substantial. Variations in the degree of impairment would result in variations in sensitivity of the test, and the model was rather insensitive to changes of test parameters. Secondly, even though QCM has been weighted, it is a surrogate parameter for the actual burden of disease for the child. Preference-based utilities have not been measured and the weighting for the impact of early or late identification of hearing loss used in our analysis were estimated by experts. Sensitivity analyses, however, revealed that without weighing, the difference in effectiveness would still be substantial. Thirdly, effectiveness was measured as a function of time to diagnosis. In routine health care, adequate treatment does not necessarily start immediately after the diagnosis is made. However, as knowledge on the consequences of delayed intervention is limited, including time to intervention as another variable in the model would have resulted in decreased precision 22. Fourthly, our empirical data do not differentiate between congenital and acquired hearing impairment. The rather pessimistic estimation of the detection rate without screening, which might be due to the lack of differentiation between congenital and acquired hearing impairment, may bias the results in favour of UNHS. Similar data, however, have been published (226, presenting a median age at diagnosis 18 months). Our estimate of discovery rate in a setting without screening might be biased by regional differences but can be easily replaced by a constant or a different rate function adapted to other local settings. Fifthly, the impact of late diagnosis on delayed language development, is not yet sufficiently known 22. The economic effects on society in terms of lost productivity 24 and the quality-of-life effects on individuals have been discussed 27. There is a lack of evidence concerning health care utilization due to hearing loss and the proportion of children following a regular school and professional career after timely fitting of a hearing aid. Further research should be undertaken to investigate the effect of the age of diagnosis and intervention, on the development of hearing impaired children and on their quality of life.

The value of this modelling exercise on effectiveness, lies in the facilitation and provision of information to decision makers- by quantitatively projecting available data, making explicit and transparent statements about assumptions and the degree of uncertainty involved in this area. The probability of timely intervention increases with UNHS. UNHS can reduce age of confirmation to a much greater extent than RS. In further studies, our model can be used to predict costs of real life situations to evaluate whether programme implementation costs would surpass cost-effectiveness thresholds. Policy makers can also base their decisions on the incremental effectiveness of UHNS, by introducing a screening programme. The model shows how likely an outcome is under the assumption of parameter uncertainty.

In our model UNHS showed higher clinical effectiveness compared to RS and NS. The strength of our model lies in the naturalistic and generic structure, which makes it useful as a model for further evaluation. The model gives explicit, transparent and quantitative information about the effectiveness of different screening strategies for policy makers who have to decide on the potential impact of neonatal screening. The model presented here is easily adaptable to different settings. It will and should be verified and tested with longitudinal data of ongoing trials and model projects.

This can be one of the first steps towards the needed transparency concerning universal new-born hearing screening.

The author(s) declare that they have no competing interests.

EG developed the decision model, carried out the statistical analyses and drafted the manuscript. FH designed and coordinated the study and participated in the development of the decision model and in the writing process. US participated in the statistical analyses, in the development of the decision model and in the writing process. PS-I, SK and AN performed the systematic review and data extraction. AN participated in the design of the study. JW conceived of the study and participated in its design and coordination. All authors revised the manuscript and read and approved the final version.