Introduction

Top Abstract Introduction Participants and methods Results Discussion References

Although there is wide acceptance that all newborn babies should be screened for abnormalities,¹ there is no consensus on how this should be done. Biochemical screening for phenylketonuria and congenital hypothyroidism is effective, but clinical examination for defects in hips, vision, and hearing and other congenital abnormalities is less well founded on scientific evidence.² Up to 12% of babies may have some detectable abnormality³ but not all will impact on health or require action.⁴

For babies born in hospital, clinical neonatal screening is usually carried out twice before discharge, once within 24 hours of birth and again a few days later. The rationale for the first examination is to detect abnormalities that may require early action. The second aims to detect those which may have been missed at the first screening and to detect others which may have become apparent latersuch as cardiac defects as the fetal circulation adapts to extrauterine life.⁵

The need for a second examination, however, has been questioned. ^{4 6 7} We therefore compared the policies for performing one rather than two hospital neonatal screening examinations as judged by their effectiveness in detecting congenital abnormalities and the consequent use of hospital and community resources.

	Participants and methods

Top Abstract Introduction Participants and methods Results Discussion References

The trial was approved by the Grampian Health Board and University of Aberdeen joint ethics committee. All babies delivered at Aberdeen Maternity Hospital between March 1993 and February 1995 were eligible except those discharged home directly from the labour ward (domino births) and those transferred to the neonatal unit within 6 hours of birth. Eligible babies were randomised to one of two policies: one screen policyone neonatal screening examination on day 3 or on the day before expected discharge if earlier; or two screen policyone screening examination within 36 hours of birth and a second on the day of discharge or on the day before expected discharge if after day 3. The two screen policy was current practice in the hospital before the trial. Babies allocated to the one screen policy were examined at the latest on day 3 even if they stayed in for longer; babies allocated to the two screen policy and who stayed in hospital for more than 3 days had their second examination on a later day.

The trial adopted a controlled switchback design,⁸ which is in essence an extended crossover design. This allows the comparison of the screening policies to be unaffected by external changes over time. Babies were allocated to one or other policy depending on their postnatal ward and the calendar month. Each month half the wards in the hospital operated the one screen policy and the rest the two screen policy. Recruitment continued for 2 years. The initial month was allocated at random with crossover to the opposite policy on the first day of each subsequent month. The screening policy to which a ward was assigned could not therefore be influenced by patients or staff. Also, neither the mother nor doctor could choose which ward a baby would be in. Babies were examined by NHS staff, including an associate specialist (JAR), community medical officers, and paediatric senior house officers. Training in and execution of routine neonatal examinations remained unaltered throughout the trial period.

Babies were followed up for their first year of life. The main outcome measures were congenital conditions coded at discharge from hospital; results of the community health assessments at 8 weeks and 8 months; use of general practioner services in the first year of life (for a randomly selected 10% subsample only); outpatient referrals; and inpatient admissions that involved congenital conditions.

Hospital examinations
Conditions diagnosed in hospital were identified from the routinely collected and computerised Scottish Morbidity Records (SMR11 and SMR11(E)) by using the ICD-9 (international classification of diseases, ninth revision) codes which referred to congenital conditions likely to be detected by neonatal screening. They were then matched with hand extracted data that described the results of examinations.

Community resources
The results of the 8 week and 8 month assessments were linked by using the community health index number as a unique identifier. The general practitioners of a 10% subsample of babies, who were selected at random each month, were sent a questionnaire requesting details of all contacts during the baby's first year of life. General practitioners were blind to the screening policy each baby had received.

Hospital resources
Babies referred to outpatient departments were tracked by using the computerised Scottish outpatient record (SMR0). Inpatient admissions in the first year of life were found by using the Scottish inpatient and day case record summary sheet (SMR1). This includes details of type of admission, specialty, and conditions diagnosed (with ICD-9 codes). These data collecting systems tracked babies admitted to hospitals throughout Scotland.

Hip and cardiac anomalies
All babies suspected of having a hip anomaly were referred to a central orthopaedic clinic. Subsequent management of these babies during their first year of life was described with the clinic's dedicated database and linked to hospital discharge findings with the unique hospital number. Similarly, all babies with cardiac problems were seen centrally (by PB). Data for babies confirmed as having a problem were linked to initial findings by using the hospital number.

Statistical analyses
As each baby could have more than one condition the results were calculated as the number of diagnoses per hundred babies. Analysis was by intention to treat. Comparisons of rates of abnormality and referral are presented as 95% confidence intervals for the difference with the normal approximation to the difference between two Poisson variables.⁹ We also calculated 95% confidence intervals for the difference between two proportions for all main comparisons.

	Results

Top Abstract Introduction Participants and methods Results Discussion References

Of the 10 835 babies born alive during the 2 year recruitment period, 9712 (89.6%) were eligible for randomisation; 4835 were allocated to the one screen policy and 4877 to the two screen policy (figure). We could not link data on 1.6% of babies; the numbers lost to follow up because of death (n=7) or moving out of the area were similar in both groups. The groups' baseline characteristics were similar in respect of sex, mode of delivery, birth order, and weight (table 1). Length of stay was the same in both groups (median (interquartile range) 4 (2 to 6) days).

View larger version (32K): [in this window] [in a new window]   Inclusion of babies in study

In practice 459 (9.5%) of the one screen policy group were actually examined twice, and 766 (15.7%) of the two screen policy group were examined once. Reasons for failure to adhere to protocol included detection of serious physical abnormality within the first 24 hours, failure to change from one policy to the other at the start of a month, and babies going home early. Babies allocated to the one screen policy were less likely ever to be examined by a senior member of staff (56%) compared with 85% examined at least once by a senior doctor under the two screen policy (difference 28%; 95% confidence interval 26% to 30%). Of the two screen policy babies, 32% were examined on day 4 or later compared with 3% in the one screen policy group.

Significantly fewer conditions were diagnosed in hospital among one screen policy babies than two screen policy babies (8.3 v 9.9; difference 1.6; 0.3 to 2.7; table 2). This was largely because of fewer suspected musculoskeletal problems, especially suspected hip anomalies (2.8 v 3.6; difference 0.8; 0.1 to 0.5).

Of the babies for whom we could link data, 37 (0.38%) failed to have an assessment at 8 weeks. There was no evidence of an excess of abnormal findings between the groups at either the 8 week or the 8 month community assessment (table 3). Nor were there more followed up in primary care or referred to secondary care from this community screening programme, both overall and among the 10% subsample.

Although the difference in the proportions of babies attending outpatient clinics in their first year of life was not significant (18.5% v 19.9%; difference 1.4%; 2.9% to 0.1%; table 4), the observed difference was largely explained by more attendances at the orthopaedic outpatient clinic (6.0% v 7.0%; difference 1%; 1.9% to 0.02%).

In their first year of life 1471 (15%) babies were admitted as inpatients at least once; of these, 369 (3.8%) were admitted more than once (table 4). There were no apparent differences between the two groups in the proportion of admissions, the type (whether planned or emergency), or the specialty.

There were no clear differences between the groups in the number with a primary diagnosis of a congenital condition at their first admission (1.0 v 0.7 diagnoses per 100 babies; difference 0.3; 0.1 to 0.7) or any admission (1.6 v 1.4 diagnoses per 100 babies; difference 0.2; 0.3 to 0.7).

The larger number of hip anomalies suspected in hospital under the two screen policy (see table 2) was reflected in more babies being seen at the orthopaedic clinic (125 v 176; table 5). There was no difference, however, in the proportion who received active management (outpatient splinting or surgical reduction 0.3% v 0.2%; difference 0.1%; 0.1% to 0.3%; table 5). For babies who had been judged normal in hospital (on the basis of a negative Ortolani-Barlow manoeuvre¹⁰) there was no clear difference in the proportion subsequently referred nor in those who then required active management (0.2% v 0.1%; difference 0.1%; 0.1% to 0.2%; table 5). This applied also to those referred because of a family history (table 5). There was no difference in the proportion of babies confirmed to have a cardiac anomaly between the two groups irrespective of whether or not they were diagnosed in hospital (table 5).

	Discussion

Top Abstract Introduction Participants and methods Results Discussion References

Most (90%) babies born during the study period were included in the trial, representing all those for whom routine neonatal screening was appropriate. The lower than expected rate of diagnoses in the trial babies (10% rather than 12%) probably reflects the relatively higher risk among babies who were ineligible because they were admitted directly to the neonatal unit.

Methodological issues
The trial was a pragmatic comparison of two policies as they might be used in hospitals. It was therefore expected that some of those allocated one examination would actually have a second and that some allocated two examinations would have only one. As we based the analyses on "intention to treat," however, we should have avoided the introduction of any bias. Possible bias introduced in assessing outcome was also minimised by using routinely collected data and blinding of providers of data to policy allocation.

Congenital diagnoses
The trial has shown clearly that examining babies in hospital twice rather than once resulted in more congenital abnormalities being suspected at the time of discharge (see table 2). The excess may have resulted from more babies being examined by experienced staff, because a second examiner might detect something missed at first, or because of new conditions which developed over time. These extra "diagnoses," however, did not lead to any detectable increase in interventions that might improve infant health nor did the infants from one group make extra use of emergency services, as might have been expected if important conditions had been missed in hospital. Thus there was no evidence that one examination was less effective than two in identifying babies who required medical attention.

Hip anomalies
The larger number of congenital abnormalities diagnosed at birth in the group examined twice was primarily attributable to an excess in suspected hip anomalies. This resulted in extra referrals to outpatient departments (2.6% v 3.4%, see table 5). These extra visits did not lead to more active management, and similar numbers underwent splinting or operation in both groups. A second examination did not there- fore seem to improve sensitivity but did reduce specificity, which led to apparently unnecessary intensive surveillance.

Heart anomalies
Most cases of serious cardiac anomaly first present with clinical symptoms¹⁴ and therefore the value of routine neonatal screening has been questioned.⁷ In this trial there was no clear evidence that a policy of more intensive screening led to a difference in the number of babies suspected of having serious cardiac problems at discharge.

Community screening
The trial confirmed the need for further surveillance after a baby leaves hospital. Of the 44 babies who received active management for congenital dislocation of the hip, 13 (30%) had negative findings on the Ortolani-Barlow manoeuvre in hospital and were later detected in the community. Of the 10 babies who required surgery, six (60%) were not detected in hospital.

Conclusion
Despite more suspected abnormalities among babies allocated a policy of two rather than one hospital neonatal examination the trial did not show any net health gain from this policy. A two screen policy does, however, carry additional resource implications for hospital services and extra anxiety for parents whose children are wrongly suspected of having abnormalities.

	Acknowledgments

We thank the ward staff who made the study possible and the general practitioners who responded to the surveys. Ian Russell was an original grant holder and contributed to initial study design, Adrian Grant gave valuable comments on the analyses and draft reports, and James McLauchlan provided orthopaedic clinic data. The views expressed, however, are those of the authors.

Contributors: All authors were members of the Neonatal Examination and Screening Trial (NEST) Steering Group and contributed to study design and implementation, analysis and interpretation of results, and writing. In addition, CG designed and piloted the study and is the guarantor for the work. CR and MC carried out the statistical design and analysis, AM implemented daily administration and case note and data extraction, coding and entry, JAR carried out routine neonatal screening, PB provided extra cardiac clinic data, and DL provided extraction of data from case notes.

Funding: The Health Services Research Unit is funded by the Chief Scientist Office of the Scottish Office Department of Health, which also funded the study through a project grant.

Competing interests: None declared.

	References

Top Abstract Introduction Participants and methods Results Discussion References

1. Maternity Services Advisory Committee to the Secretaries of State for Social Services and for Wales. Maternity care in action part III: care of the mother and baby (postnatal and neonatal care). London: HMSO , 1985.
2. Robinson R. Effective screening in child health. BMJ 1998; 316: 1-2[Free Full Text].
3. Hall DMB. Health for all children. A programme for child health surveillance. Oxford: Oxford University Press , 1989.
4. Moss GD, Cartlidge PHT, Speidel BD, Chambers TL. Routine examination in the neonatal period. BMJ 1991; 302: 878-879.
5. Manning DJ. One or two neonatal examinations? BMJ 1991; 302: 1209[Free Full Text].
6. Hughes AP, Stoker AJ, Milligan DWA. One or two neonatal examinations? BMJ 1991; 302: 1209.
7. Cartlidge PHT. Routine discharge examination of babies: is it necessary? Arch Dis Child 1992; 67: 1421-1422[Free Full Text].
8. Cochran WG, Cox GM. Experimental designs. New York: Wiley , 1957.
9. Daly LE, Bourke GJ, McGilvray J. Interpretation and uses of medical statistics. Oxford: Blackwell Scientific Publication , 1991.
10. Barlow TG. Early diagnosis and treatment of congenital dislocation of the hip. J Bone Joint Surg 1962; 44B: 292-301.
11. Kalamchi A, MacEwen GD. Avascular necrosis following treatment of congenital dislocation of the hip. J Bone Joint Surg Am 1980; 62: 876-888[Abstract/Free Full Text].
12. Bradley J, Wetherill M, Benson MK. Splintage for congenital dislocation of the hip. Is it safe and reliable? J Bone Joint Surg Br 1987; 69: 257-263.
13. Godward S, Dezateux C. Surgery for congenital dislocation of the hip in the UK as a measure of outcome of screening. Lancet 1998; 351: 1149-1152[Medline].
14. Abu-Harb M, Wyllie J, Hey E, Richmond S, Wren C. Presentation of obstructive left heart malformations in infancy. Arch Dis Child Fetal Neonatal Edition 1994; 71: 179-183.

Commentary: "Switchback" allocationdangerous bends ahead!

Jonathan J Deeks, medical statistician. 

NHS/ICRF Centre for Statistics in Medicine, Institute of Health Sciences, Oxford OX3 7LF

j.deeks{at}icrf.icnet.uk

Switchback or reversal trials are an extension of the classic crossover trial design with participants "switching back" to their initial treatment in an additional period added at the end of the trial.¹ Switchback designs have useful statistical properties when there are temporal trends in outcome that vary between subjects. They have been applied, for example, in cattle feeding and lactation experiments where the outcome (milk yield) is expected to decrease naturally throughout the trial.² Occasionally they have been used in medicine for nutritional experiments.³

But Glazener and colleagues' rationales for using an extended switchback design relate more to ease of trial organisation than statistical efficiency. While the switchback design would control for ward specific temporal trends in rates of abnormality, it is difficult to conceive of a plausible mechanism by which such trends could arise. Given that it is the wards and not the neonates which switch (the neonates each receive only one intervention) it is perhaps unhelpful to think of this study as having a crossover form at all. In fact the data are analysed and presented as if they originate from a simple two group parallel study without any crossover or clustering features, by assuming that for each screening policy the underlying distribution of detected abnormalities is the same in every ward.⁴ As there is no reason to suspect non-random clustering of cases within wards and as just one team undertook the screening throughout the hospital this simplification is probably justified.

There are attractive benefits of allocating screening policies to wards rather than individuals. As all participants allotted to each of the screening policies would always be located together trial execution will have been simplified, the risk of contamination reduced, and the bureaucracy of organising individual allocations avoided. The issues concerning consent to be randomised also differ in cluster randomised trials, which will have impacted on the ease of recruiting large numbers.

Simplifications of trial design, however, rarely come without jeopardising internal validity. To ensure proper random allocation the allocation mechanism must be truly random and the allocations concealed at the time they are assigned.⁵ The random element aims to prevent the misfortune of allocation patterns coinciding with, or being influenced by, a factor related to the outcome. Concealment of the allocation prevents the possibility of conscious or subconscious manipulation of individual assignments. There is empirical evidence that unconcealed randomisation leads to overestimation of treatment effects,⁶ manipulation possibly occurring by participants' registrations being delayed until they would receive the preferred treatment allocation or exclusion of eligible subjects whose allotted allocation would be considered unfavourable.

In this trial, despite the use of a random mechanism to assign the intervention switching policies to the wards, the allocation of a mother to a ward was by the standard haphazard (not really random) process of hospital bed allocation. While such a mechanism may seem difficult to influence it certainly is not concealed. As these criteria are not met the authors cannot guarantee that they have allocated the participants to the two groups in an unbiased manner. For the allocation to have been seriously biased, however, it would be necessary for assignments to have been made with some knowledge of each individual's likely outcome, so that the allocation of some mothers carrying babies of higher (or lower) risk could be manipulated in favour of a particular policy. While such manipulation is a reality in poorly randomised treatment trials, there may be situations in trials of preventive and screening interventions when no risk factors can be identified at the time of assignment to intervention and biased allocation is theoretically impossible. The interpretation of this trial relies on just such an assumption.

	References

1. Jones B, Kenward MG. Design and analysis of cross-over trials. London: Chapman and Hall, 1989:203-204.
2. Cochran WG, Cox GM. Experimental designs. 2nd ed. New York: Wiley, 1957:141-142.
3. Meng QH, Pajukanta P, Valsta L, Aro A, Pietinen P, Tikkanen MJ. Influence of apolipoprotein A-1 promotor polymorphism on lipid levels and responses to dietary change in Finnish adults. J Intern Med 1997; 241: 373-378[Medline].
4. Campbell MK, Grimshaw JM. Cluster randomised trials: time for improvement. BMJ 1998; 317: 1171-1172[Free Full Text].
5. Schulz KF. Randomised trials, human nature, and reporting guidelines. Lancet 1996; 348: 596-598[Medline].
6. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias. Dimensions of quality associated with estimates of treatment effects in controlled trials. JAMA 1995; 273: 408-412[Abstract/Free Full Text].