Published by Oxford University Press on behalf of the International Epidemiological Association
© The Author 2005; all rights reserved. Advance Access publication 20 April 2005
International Journal of Epidemiology 2005;34:874–887
Infant sleeping position and the sudden
infant death syndrome: systematic review of
observational studies and historical review of
recommendations from 1940 to 2002
Ruth Gilbert,1* Georgia Salanti,2Melissa Harden1and Sarah See1,3
4 April 2005
Background Before the early 1990s, parents were advised to place infants to sleep on their
front contrary to evidence from clinical research.
We systematically reviewed associations between infant sleeping positions and
sudden infant death syndrome (SIDS), explored sources of heterogeneity, and
compared findings with published recommendations.
By 1970, there was a statistically significantly increased risk of SIDS for front
sleeping compared with back (pooled odds ratio (OR) 2.93; 95% confidence
interval (CI) 1.15, 7.47), and by 1986, for front compared with other positions
(five studies, pooled OR 3.00; 1.69–5.31). The OR for front vs the back position
was reduced as the prevalence of the front position in controls increased. The
pooled OR for studies conducted before advice changed to avoid front sleeping
was 2.95 (95% CI 1.69–5.15), and after was 6.91 (4.63–10.32). Sleeping on the
front was recommended in books between 1943 and 1988 based on extrapolation
from untested theory
Conclusions Advice to put infants to sleep on the front for nearly a half century was contrary
to evidence available from 1970 that this was likely to be harmful. Systematic
review of preventable risk factors for SIDS from 1970 would have led to earlier
recognition of the risks of sleeping on the front and might have prevented over
10 000 infant deaths in the UK and at least 50 000 in Europe, the USA, and
Australasia. Attenuation of the observed harm with increased adoption of the
front position probably reflects a ‘healthy adopter’ phenomenon in that families
at low risk of SIDS were more likely to adhere to prevailing health advice. This
phenomenon is likely to be a general problem in the use of observational studies
for assessing the safety of health promotion.
Sudden infant death, review, meta-analysis
1Centre for Evidence-based Child Health, Institute of Child Health, London, UK.
2MRC Biostatistics Unit, Institute of Public Health, Forvie Site, Robinson
Way, Cambridge, UK.
3Present address: Waltham Forest Primary Care Trust (PCT),
* Corresponding author. Centre for Evidence-based Child Health, Centre for
Paediatric Epidemiology and Biostatistics, Institute of Child Health, 30,
Guilford Street, London WC1N 1EH, UK. E-mail: email@example.com
the 20th century. At the start of the 20th century, such deaths
were attributed to overlying, particularly by drunken mothers.1
By the 1940s, as more deaths were investigated by autopsy,
pathologists realized that few deaths were due to maternal
overlying, and alternative mechanisms for ‘accidental
mechanical suffocation’ were sought. In 1944, Abramson, a
pathologist in New York State, noted that two-thirds of infants
dying from mechanical suffocation were found face down,
contrary to the usual sleeping position for infants at the time.2
His observations, which were corroborated by reports in the UK
Sudden unexpected unexplained infant death, now known as
sudden infant death syndrome (SIDS), was recognized as a
major cause of infant death in the UK and USA throughout
by guest on May 14, 2011
SLEEPING POSITION AND SIDS
and Australia3,4led to a health promotion campaign that
recommended avoidance of the front position.5
The campaign was short-lived. In 1945, a paediatrician,
Woolley, rejected Abramson’s hypothesis of suffocation on the
front based on experiments in which he had covered babies’
faces with layers of blankets.6He reported that the oxygen
content of the air breathed by the babies only fell when they
were covered with a rubber sheet and that babies moved if
breathing was obstructed. He also criticized the explanation of
suffocation because it ‘instilled guilt and self-incrimination in
Emergence of alternative explanations for death, such as
unrecognized infection4,7,8inhalation of vomit9and hyper-
sensitivity reaction to inhaled milk,10further strengthened the
argument against the suffocation hypothesis and highlighted
the need for data on risk factors. The first published
case–control study was started in 1956 in the USA,11and in
1958, a similar study in the UK was the first to measure infant
sleeping position in SIDS victims and live control babies.12At
around the same time, it became increasingly common to
advocate sleeping on the front. We now know that front
sleeping is a major cause of SIDS. We wanted to know whether
systematic review of the evidence could have reversed this
harmful advice sooner or whether variation in the association
between sleeping on the front and SIDS was consistent with
recommendations at the time. We did a systematic review and
meta-analysis of the effect of front and side sleeping on the risk
of SIDS, and an historical review of recommendations on infant
sleeping position in books and pamphlets on infant care
available in the UK between 1940 and 2002. We focussed on
how the strength of the evidence for a harmful effect of front
sleeping changed before and after advice changed in favour of
avoidance of the front position. We hypothesized that the effect
of the front position on SIDS might differ depending on
whether health advice favoured front or not as families that
adopt health advice are likely to be at lower risk of SIDS.
We reviewed recommendations on infant sleeping position in
books or pamphlets available in the UK from 1940 to 2002.
We chose 1940 to include a period before the front position
was widely advocated. We searched the Modern Medicine
Collection at the Wellcome Trust library, and, because of a lack
of more recent texts, the British Medical Association library
from 1965 to 2002. We included any book or pamphlet that
referred to the care of normal term infants aged ?6 months,
and mentioned infant sleeping position. Searches used the
library indexing system for books on infant care and we also
searched electronically using terms for paediatric, parent, and
baby (details of search strategy available from authors).
One reviewer (S.S. or M.H.), assessed whether texts met the
inclusion criteria and prepared a hard copy file with the extract
and book title but not the date of publication. A second
reviewer (R.G.) categorized the recommendation as favouring
front, back, side, or non-front position(s), or neutral if all or
none were implicitly or explicitly favoured. A second reviewer
(S.S.), independently categorized one-third of the texts and
there was complete agreement with the first reviewer.
We included any case–control or cohort study that compared
the risk of SIDS in infants sleeping on their front, side, or back.
Studies had to be based on SIDS infants and live healthy control
infants from the same community. We searched for any
comparative study of infant sleeping position and SIDS in
MEDLINE (1966–2002) and EMBASE (1980–2002), using a
detailed search strategy (available from the authors), and
reference lists of review articles, a PhD thesis on the history of
SIDS,13and included studies. Abstracts were scanned by one
reviewer (S.S., M.H., or R.G.), and full texts of potentially
eligible studies retrieved. R.G. and S.P. jointly extracted data
from included studies.
We used data on the position in which the infant was placed to
sleep before death or interview, or if lacking, data on usual
position, or position found. If usual position was measured at
multiple ages, we used results closest to 3 months of age. We
recorded the method of selection of cases and controls,
matching criteria, if any, and whether data collection methods
differed in cases and controls.
Our primary aim was to compare the risk of SIDS in infants
sleeping front and back. As some studies did not separately
report side and back positions, we also compared front with
non-front positions. However, grouping side with back will
attenuate the observed risk associated with the front position if
the side position is also harmful. We therefore calculated odds
ratios (ORs) for SIDS associated with sleeping front vs back,
front vs non-front, and side vs back.
To avoid confounding, we used the unadjusted matched OR if
reported. Otherwise we calculated the unmatched OR.14Because
studies differed in their design, populations, and methods, we
used a random effects model in which it is assumed that the
observed ORs are sampled from a common distribution around
a mean effect with variance measured by the heterogeneity
parameter. We estimated 95% confidence intervals (CIs) and
considered a P-value ?0.05 as statistically significant. Hetero-
geneity in the OR for SIDS was assessed by the chi-squared test
(Q-test) and quantified using I2which reflects the proportion of
variation that is not due to sampling error.15The possibility of
publication bias was evaluated using funnel plots and the Egger
and Begg tests.16,17
We determined the year at which there was a statistically
significant association between front or side sleeping positions
and SIDS by using a cumulative meta-analysis based on year of
publication as described by Lau.18The overall heterogeneity
was used in the calculation of the CIs for the cumulative OR at
every step using a random effects model. We applied recursive
cumulative meta-analysis to examine the direction and
magnitude of the relative changes in the cumulative evidence
as a function of the cumulative sample size.19,20At the end
of every information period j, the ratio (cumulative ORj)/
(cumulative ORj ? 1) was assessed and compared with unity. If
larger than one, this was interpreted as a ‘move’ of the evidence
towards defining the front position as more harmful than in the
previous information period.
by guest on May 14, 2011
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
To explore potential sources of heterogeneity we initially used
conventional meta-regression to determine an association with
variables previously suggested.21In a univariate model, we first
determined the effect of the position recorded in cases (before
death, usual, or after death), year of publication, recruitment year
(measured as the mid-point between start and end of
recruitment), matching criteria for controls and cases, and country
and continent of study. The combined effect on heterogeneity of
the variables found to be significant in the univariate analysis was
estimated in a multivariate meta-regression model. We extended
the meta-regression analyses to examine the hypothesis that the
prevalence of front sleeping in control infants is associated with
heterogeneity. This is because parents who put their babies to
sleep in the front position when advised not to, might have a
different risk of SIDS than parents who do so when front sleeping
is the norm (similarly for the side position).
The OR for front vs any other position can be written as using
logOR ? logitP(front|case) ? logitP(front|control) and the
prevalence of front sleeping is estimated in the controls as
P(front) ? P(front|control). Consequently, regression of logOR
to P(front) will be biased by regression to the mean.
To overcome this we fitted a hierarchical model similar to
that described by Thompson et al. to model background risk
in randomized controlled trials.22,23As the studies were case-
control rather than trials, we made some modifications to the
methods (see Appendix).
We retained in the model any factors that were statistically
significantly associated with heterogeneity in the conventional
meta-regression, and assessed the extent to which the factors
included in the model explained the variation between studies
by measuring the change in the heterogeneity parameter. If
factors included in the model explained heterogeneity, the
heterogeneity parameter (variance in the random effects)
would be expected to get smaller. This model was fitted using
Markov chain Monte Carlo methods within a Bayesian
framework. The analysis was conducted using Intercooled Stata
8.2 (Stata Corp., College Station, TX), R 1.9.1 (R Foundation for
statistical computing, Vienna) and Winbugs 1.4.1.
Table 1 summarizes the recommendations made in 83 texts that
met the inclusion criteria (details available from the authors).
From 1940 to the mid-1950s, texts favoured the back or side
positions and only one, in 1943, recommended the front
position. From 1954 until 1988, a substantial proportion of texts
consistently favoured front sleeping, although many also
favoured the side and back. The sudden shift in favour of
front sleeping is best illustrated by ‘Baby and Child Care’ by
Dr Benjamin Spock who recommended the back position in his
1955 edition, and the front position in 1956.24In his 1958
edition, he argued ‘If he vomits, he’s more likely to choke on
the vomitus. Also, he tends to keep his head turned to the same
side—usually toward the centre of the room. This may flatten
the side of his head.’ Many authors repeated these arguments.
Others argued that front sleeping reduced wind,25,26coughing
due to mucus,27and made respiration easier.26Suffocation was
considered to be possible only if the baby was very weak.26
These views were not universal. In editions of his textbook in
1945, 1950, and 1959, Nelson stated that ‘position during sleep
Table 1 Recommended infant sleeping position in books on infant care
No. textsFrontSideBackNon-front Neutral
* Books written by Dr Benjamin Spock.
by guest on May 14, 2011
SLEEPING POSITION AND SIDS
is relatively unimportant, but should be changed often to
prevent moulding of the cranium’.28–30Others were less
equivocal. One author recommended in 1953, ‘Sleeping on his
abdomen never should be permitted because of the danger of
suffocating’.31In 1966, another warned ‘Very small babies
should never be left alone lying on their tummies. This is an
American fashion to strengthen the back, but we think the
dangers of suffocation are not sufficiently remote to justify it.’32
No texts favoured the front position after 1988. From the mid-
1950s to 1990, many texts continued to recommend the side
position, but few advocated sleeping on the back. In the early
1990s, most texts recommended the side position or simply
advised against front sleeping, but apart from one text in 1990,
the back position was not consistently advocated until 1995.
Of the 2897 abstracts scanned, and the 206 full text articles
retrieved, 40 studies met the inclusion criteria (Figure 1 and
Table 2). Four further studies were excluded (Figure 1). No
randomized controlled trials were found. All 40 included studies
provided data on front vs non-front positions, but only 24
studies separately recorded back and side positions. Of the 40
studies, 23 (and 15/24 reporting side and back positions)
included some degree of matching of controls with cases. Of
these, unadjusted matched ORs were available for 9/23 studies
(and for 7/15 reporting side and back positions).33,33–44For
one study, we derived pooled ORs from data reported for
separate ethnic groups.37All studies were case–control except
for one cohort study reported in two stages. This resulted in data
for 2 years of the study (15 SIDS victims) being included twice
in the cumulative meta-analyses.45,46Repeated use of the same
data was avoided for all the other studies except for Mitchell 2
1999 (details in Table 2). No substantial evidence was found for
publication bias for any of the sleeping position comparisons
either by examining the funnel plots or applying the Egger or
Begg tests (lowest P-value ? 0.103).
There was a statistically significantly higher risk of death
associated with the front position whether compared with the
back (Figure 2a) or non-front positions (Figure 2c). There was a
weak association between the side position and the risk of SIDS,
which was marginally worse than back (Figure 2e).
The cumulative meta-analyses showed that the association
between death and the front position compared with back had
become statistically significant by 1970, after the first two
case–control studies (cumulated OR 2.93; 95%CI 1.15–7.47;
Figure 2b). When front was compared with non-front, the
association was not statistically significant until 1986, after
inclusion of five studies (cumulated OR 3.00; 1.69–5.31;
(1966 - Jan 2003) n= 2384
(1980 - Jan 2003) n= 1521
(of SIDS reviews/studies) n= 128
Total papers identified
3 studies: SIDS and controls from different populations109-111
1 study, controls were ‘near-miss’ SIDS 110
1 study, no data presented for controls67
Full papers retrieved for
Potentially eligible studies n=44
Sleep position compared in SIDS vs.
Did not meet inclusion criteria
Studies to be included in
Figure 1 Flow diagram to show results of searches for the systematic review of comparative studies of infant sleeping position and SIDS
by guest on May 14, 2011
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Table 2 Characteristics of included studies
Study period, selection of cases and controls, and method of data
1958–1961. SIDS: Referred to coroner in 12 London boroughs (aged
2 weeks – 2 years, 6 of 100 SIDS victims were aged ?12 months).
Position found recorded by coroner.
Controls: matched for age, sex and community from register of Medical
Officer of Health. Sleeping position recorded by health visitors.
2 45 11183
1965–1967. SIDS: Consecutive cases in northern Ireland.
Controls: matched for age, sex, and administrative area.
Data collection by home interviews in both cases and controls
Beal 1 198662
1970–1984. SIDS: South Australia interviewed within weeks of death by Beal.
Controls: postal survey of 200 consecutive birth registrations in August 1984.
Tonkin 1 198661
1972–1982. SIDS: position found routinely recorded in Auckland, New Zealand in
1972, 1973, and 1982.
Controls: Plunket nurses (health visitors) in Auckland noted sleeping
position of 50 babies most recently seen (10 nurses in 1972, 15 nurses in 1973).
In 1982 all nurses noted sleeping position of 2 week old babies during a 3 month period.
1980–1982. SIDS: within the Melbourne statistical division, Australia.
Controls: matched by age and same hospital of birth.
Data collection by home interviews in both groups.
1984–1985. SIDS: in the Departement d’Ille et Vilaine, France.
No details on how data were collected.
Controls: Infants born in 7 maternity hospitals in Brittany attending
routine post-natal surveillance. Questionnaire completed by doctor at consultation.
1 85 20
1976–1979. SIDS: UK multicentre study. SIDS resident within local areas.
Controls: matched for age and area.
Data collection by home interviews in both groups.
1980–1986. SIDS: notified by coroners in Tasmania.
Controls: matched for age, sex, and hospital of birth.
Data collection at home interviews in both groups.
Beal 2 199194
1985–1989. SIDS: in South Australia interviewed within weeks of death by Beal.
Controls: postal survey of 200 consecutive birth registrations in August in 1988.
1980–1986. SIDS: deaths in The Netherlands.
Data collected by home interview Controls: Infants at 17 well-baby clinics.
Parents interviewed about sleeping position at 2–4 months and 5–7 months.
Tonkin 2 198964,65
1981–1985. SIDS: deaths in Auckland, New Zealand.
Data collected at interview. Controls: surveyed by plunket nurses in Auckland in
1983 aged 1–4 months. Results used for usual position at 3 months.
1986–1987. SIDS: prospective surveillance of all SIDS deaths in Hong Kong.
Data collected at home interview
Controls: age and sex matched, one from hospital and one from community.
No details given on data collection.
2 5644 16
2 946 32
Fleming 1 199034
1987–1989. SIDS: all SIDS in Avon, UK, interviewed at home within days of death.
Controls: matched by age and area based on same health visitor list as SIDS victim.
Data collection at home interview for both groups.
Study period not stated. SIDS: study in France, reported only in conference proceedings.
Controls: no details available on selection or data collection.
1978–1979. SIDS: All SIDS in six geographically defined areas in the USA.
Controls: matched for age, and study centre (second controls matched for ethnic group and
birth weight not used in this review). Both SIDS and control parents were interviewed
about usual sleeping position in the 2 weeks before death or interview.
1985–1987. SIDS: all deaths in The Netherlands.
Data collection by parent-completed postal questionnaire after telephone contact.
Controls: randomly selected from municipal registers. Data collection by postal
questionnaire asking about usual sleeping position in each of months 1–6. Data for month
3 used in analyses.
24 59 105
Mitchell 1 199197
1987–1988. SIDS: deaths within areas covering 80% of births in New Zealand.
Controls: randomly selected in proportion to hospital births in same areas and frequency
matched for predicted age and season of cases. Home interviews for both groups measuring
position placed at nominated sleep.
Dwyer 1 199145
1988–1990. SIDS: prospective cohort study of births in highest scoring quintile of risk score
for SIDS in Tasmania.
Controls: whole cohort excluding SIDS victims. Usual sleeping position prospectively
recorded in SIDS and controls at 1 month of age.
20 60 15
by guest on May 14, 2011