Study inclusion and exclusion flow diagram.
Horsley T, Clifford T, Barrowman N, Bennett S, Yazdi F, Sampson M, Moher D, Dingwall O, Schachter H, Côté A. Benefits and Harms Associated With the Practice of Bed SharingA Systematic Review. Arch Pediatr Adolesc Med. 2007;161(3):237-245. doi:10.1001/archpedi.161.3.237
To examine evidence of benefits and harms to children associated with bed sharing, factors (eg, smoking) altering bed sharing risk, and effective strategies for reducing harms associated with bed sharing.
MEDLINE, CINAHL, Healthstar, PsycINFO, the Cochrane Library, Turning Research Into Practice, and Allied and Alternative Medicine databases between January 1993 and January 2005.
Published, English-language records investigating the practice of bed sharing (defined as a child sharing a sleep surface with another individual) and associated benefits and harms in children 0 to 2 years of age.
Any reported benefits or harms (risk factors) associated with the practice of bed sharing.
Forty observational studies met our inclusion criteria. Evidence consistently suggests that there may be an association between bed sharing and sudden infant death syndrome (SIDS) among smokers (however defined), but the evidence is not as consistent among nonsmokers. This does not mean that no association between bed sharing and SIDS exists among nonsmokers, but that existing data do not convincingly establish such an association. Data also suggest that bed sharing may be more strongly associated with SIDS in younger infants. A positive association between bed sharing and breastfeeding was identified. Current data could not establish causality. It is possible that women who are most likely to practice prolonged breastfeeding also prefer to bed share.
Well-designed, hypothesis-driven prospective cohort studies are warranted to improve our understanding of the mechanisms underlying the relationship between bed sharing, its benefits, and its harms.
Bed sharing, the practice of adults sleeping on the same surface as children, is a controversial routine about which health care professionals are often asked to advise parents. Bed sharing involving parents and young infants appears to be an increasingly common practice in society, even though there are few detailed reports regarding its true prevalence.1,2
In the 1990s, great interest in reexamining the practice of bed sharing began after reports from New Zealand linked bed sharing and sudden infant death syndrome (SIDS).3,4 Since then there has been a growing body of research on the possible benefits of bed sharing and also the harms, namely the association with SIDS. However, the benefits and harms of bed sharing continue to be debated extensively, without resolution. The American Academy of Pediatrics issued a statement in 1997 advising of the risk of bed sharing under particular circumstances (eg, soft sleep surfaces; situations with a caregiver who smoked or used alcohol or drugs).5 The most recent policy statement of the American Academy of Pediatrics is similar, but it also acknowledges that the topic of bed sharing remains highly controversial.6 Other health authorities have also issued statements on bed sharing and recommended that the safest place for infants to sleep is in a standard crib, in the parents' room.7,8 These official positions have not, however, been based on a thorough review of the literature.
This systematic review was undertaken to provide health care professionals with a thorough review of evidence (eg, case-control and prospective cohort studies) on the harms and benefits of bed sharing. Our objectives were to identify and synthesize evidence of the following: (1) benefits and harms to children associated with bed sharing, (2) factors (eg, smoking) altering bed-sharing risk, and (3) effective strategies for reducing harms associated with bed sharing.
The MEDLINE, CINAHL, Healthstar, PsycINFO, the Cochrane Library, Turning Research Into Practice, and Allied and Alternative Medicine databases were searched for records published in any language between January 1993 and January 2005. Published and unpublished studies of any design were considered. Bed sharing was defined as the practice of sharing a sleep surface between adults and young children. Any report investigating the practice of bed sharing and associated benefits and harms, in children 0 to 2 years of age, was included.
All records identified by our searches were uploaded to a systematic review management software program. Records were screened in duplicate by means of titles and abstracts. Each potentially relevant record was marked and the full-text reports were obtained. Each record was screened independently to achieve consensus by 2 reviewers, and disagreements were resolved through discussion.
Data pertaining to study design, population demographics, study characteristics, risk factors, and exposure (eg, description of bed-sharing environment) were extracted by one reviewer and verified by another for all relevant reports.
Relevant studies failing to use a contemporaneous comparison (eg, case series or retrospective cohort) were excluded from any analysis. Although it is feasible to provide data from studies without a comparison, analytical “solutions” to such designs do not currently exist. Even though there is no randomization, cohort and case-control studies offer some control over the influence of bias because they incorporate a comparison group and can also adjust for known or suspected confounders in the statistical analysis.
The Newcastle-Ottawa Scale9 was used for quality assessment for prospective cohort and case-control study designs by 1 reviewer (T.A.). For prospective cohort studies, items include assessment of selection (representativeness of samples, ascertainment of exposure, and demonstration that the outcome of interest was not present at the start of study), comparability (control for important factors by either matching and/or adjusting for confounders in the analysis), and outcomes (independent blind assessment, sufficient length of follow-up for outcomes to occur, and adequacy of follow-up). Case-control studies assess items related to selection (adequacy of case definition, representativeness of cases, and selection and definition of controls), comparability (on the basis of design or analysis), and exposure (ascertainment of exposure, method of ascertainment for cases and controls, and nonresponse rate).
Quality assessment was determined solely by what was reported in each study. No attempt was made to contact authors for missing information. Companion reports (subsequent publications using the same sample population) were used to supplement missing data when required.
Suppose ORb(smoker) is the odds ratio (OR) for the association between SIDS and bed sharing among smokers, and ORb(nonsmoker) is the OR for the association between SIDS and bed sharing among nonsmokers. These 2 ORs are directly interpretable estimates of the association between SIDS and bed sharing in the 2 groups (smokers and nonsmokers). The interaction between bed sharing and smoking can be represented by the ratio ORb(smoker)/ORb(nonsmoker). A test of the statistical significance of this interaction can be based on whether this ratio is significantly different from 1. An alternative representation is shown in the following 2 × 2 table:
Note that the reference category in this table is infants who were exposed to neither bed sharing nor smoking, for whom the OR is defined to be 1. The association between SIDS and bed sharing together with smoking (relative to neither bed sharing nor smoking) is ORsb, the association between SIDS and bed sharing in the absence of smoking is ORb, and the association between SIDS and smoking in the absence of bed sharing is ORs. These ORs are related to ORb(smoker) and ORb(nonsmoker) defined in the preceding paragraph, as follows: ORb(smoker) = ORsb/ORs and ORb(nonsmoker) = ORb. Thus, ORb is directly interpretable as the OR for the association between SIDS and bed sharing among nonsmokers, but ORsb and ORs are not directly interpretable in terms of the association between SIDS and bed sharing. Therefore, the interaction ratio can be expressed as ORb(smoker)/ORb(nonsmoker) = ORsb/(ORs × ORb). In other words, the interaction represents the synergistic effect of smoking together with bed sharing compared with the independent effects of bed sharing and of smoking.
Characteristics of the studies, including design, population, and risk factors, were tabulated. Quantitative estimates of the association between bed sharing and harms were extracted by a statistician using a standardized extraction form. The ORs and 95% confidence intervals for bed sharing as a risk factor were extracted. Adjusted ORs were chosen in preference to unadjusted ORs because of concerns about confounding within case-control studies.
Initial searches identified a total of 1218 records from bibliographic sources. After duplicate publications were excluded, titles and abstracts were evaluated; nonrelevant and non-English publications were excluded. A total of 323 articles were retrieved for relevance assessment for which we used the full text of each study. Eighty-three topic-relevant reports were identified for inclusion. We further excluded 43 reports from formal synthesis because they were noncomparison studies (eg, case series). In total, 40 reports (30 case-control design and 10 prospective cohort design) met our final inclusion criteria and constitute the body of evidence for this review (Figure).
Many of the reports included companion (or subsequent) publications, and thus quality assessment was conducted with the Newcastle-Ottawa Scale only for the publications that were deemed to be the “primary” publication (eg, containing the most relevant information per question). Companion publications were used to supplement data when necessary. We identified 17 case-control studies3,4,10- 37 and 10 prospective cohort studies38- 47 (Table 1 and Table 2). Comparability (a category of the scoring scale) was heterogeneous within these studies. Of the case-control studies, only 11 of the 17 publications met all criteria for maximum scoring for this category (2 points) at this level.10,12,14- 18,21,23- 25 The remaining studies received only 1 point.3,11,13,19,20,22 This means that the cases and controls who were studied were not necessarily sufficiently similar to reduce the likelihood of bias influencing the study results. Put another way, these studies did not plan a priori to match their cases with their controls on certain variables of interest (ie, age or sex), nor did they adjust for these variables in their statistical analysis (eg, control for confounding). The exposure assessment category was consistently the lowest-rated section within the included studies.
Unlike the case-control studies, the prospective cohort studies fared best in the outcome evaluation (5 of 10 receiving the maximum 3 stars38- 40,42,44) but performed poorly within the selection category (4 of 10 receiving the maximum 4 stars41,42,45,47). This raises concerns that, if the populations being examined are inadequately selected or are subject to bias, it is difficult to be confident that the outcomes reported are true representations of the cohorts examined.
The consistency across studies of associations between bed sharing and harms or benefits was examined. We decided early in the process that no attempt would be made to pool estimates of the association between bed sharing and harms or benefits across studies. This was motivated by a number of concerns. Differences in how confounding was controlled would make pooling of estimates questionable. Inconsistencies in how interactions were examined and reported (eg, incomplete data) were also problematic. For example, when an individual study did not find an interaction to be statistically significant, further detail on the interaction was typically not provided. From the perspective of potential pooling, this selective reporting posed a problem akin to publication bias, in which statistically nonsignificant results may not be available. Pooling only the available results could lead to bias. Finally, varying definitions of exposure and overlapping data sets make pooling problematic.
The included studies reported on a total of 17 different data sets for 19 publications (all case-control studies) (Table 3). The studies were conducted in 10 countries (England, Germany, Ireland, Japan, New Zealand, Norway, Russia, Scotland, the Netherlands, and the United States). In addition, 1 study grouped data from 20 regions of Europe and included data from other included publications. Most of the studies included infants aged up to 1 year. The majority of populations included in these studies were white. One study reported data collected solely on indigenous people (Northern Plains Indians) in the United States,16 and 1 other study examined a predominantly African American population.15 The New Zealand data sets were collected to reflect adequate representation for Maoris and Pacific Islanders.
The studies we reviewed were consistently aimed at identifying the prevalence of known or potential risk factors for SIDS. Seven publications4,11,15,18,20,22,28 were more specifically aimed at investigating bed sharing and SIDS, although the studies were not originally designed as such. Two publications did not report data on bed sharing, although the data were collected.14,25 One reported data solely for the cases,22 and another reported only the prevalence of bed sharing in cases and controls without any further analysis.19
Definitions of sleeping location (bed sharing or non–bed sharing) were heterogeneous. Nevertheless, the studies can be classified broadly into 2 subgroups: those reporting routine sleep location (5 studies)10,12,16- 18 and those reporting bed sharing on a particular night (last sleep for cases and reference sleep for controls) (5 studies).13,15,19,21,23 Three studies4,11,20 reported data on both routine bed sharing and bed sharing on a particular night. For 3 studies, the definition of sleep location was not clearly reported14 or no data were available.22,25
Overall, there were 11 publications reporting on different data sets for which ORs (and 95% confidence intervals) were provided for bed sharing data.3,10- 13,15,16,18,20,21,23 The results were grouped as follows: 5 studies reported a nonsignificant OR, 1 after univariate analysis only16 and 4 after multivariate analysis.10- 12,18 Four of these used routine practice as their definition of bed sharing.10,12,16,18 One study15 reported a nonsignificant OR when parents were bed sharing (last sleep) but a significant OR for any bed sharing (with anyone including siblings). Five studies reported a significant OR for bed sharing, which ranged from 2.02 (Mitchell et al3) to 16.47 (McGarvey et al20), and 4 of these 5 studies defined bed sharing in terms of last sleep or reference sleep.3,13,20,23
For the outcome of SIDS, the most frequently investigated interaction with bed sharing was smoking (most commonly by the mother either during pregnancy or postpartum). For the purpose of reporting interactions, we have listed the primary publication (which was defined as the publication containing the most relevant data for our report) (Table 4).
A total of 10 publications provided data on interactions for smoking,3,10- 13,15,16,18,20,21 but complete data were available in only 43,11,13,21 (Table 4). Owing to varying definitions of exposure (eg, maternal smoking during pregnancy or postpartum), a total of 15 interactions were summarized. Reported interaction ratios were all greater than 1 (range, 1.60-29.23), suggesting that the association between bed sharing and SIDS is greater among smokers than nonsmokers. Of the 15 publications reporting interaction ratios, 6 interactions (in 5 reports) were statistically significant,3,10,12,13,20 6 (in 4 reports) were not statistically significant,3,15,16,18 and 3 (in 2 reports) were not clear.11,21 Interaction ratios were not available for a number of studies,12,15,16,18 primarily those in which the interaction was reported to be statistically nonsignificant, and thus the ratios may be substantially lower in these cases. Because of the way results were reported in the studies, the confidence interval was not consistently available for the OR among smokers. In total, 6 such confidence intervals (in 4 reports) were not available.3,11,13,21 Two confidence intervals (in 1 report) were available for the OR among smokers, and both were statistically significant.3 Of 8 ORs (in 4 reports) among nonsmokers,3,11,13,21 only one3 was statistically significant.
A large number of other factors (20, not including smoking) were also reported, and the data are presented in Table 5. For 18 factors, only 1 estimated interaction was available; for 2 factors, an estimate was available in 3 different publications for a total of 24 interactions. Of these 25 interactions, 6 were statistically significant,11,13,20,31 17 were not statistically significant,3,4,10,16,18,20,24 and 2 were unclear.11,13 Of the 6 estimated interactions that were statistically significant, it is of interest that 3 of them11,13,20 were for age of infant and were all in the same direction, namely, indicating a decreased association between bed sharing and SIDS with increasing age. The other 3 statistically significant interactions were as follows: daytime31 (indicating a decreased association between bed sharing and SIDS during daytime sleeping), breastfeeding initiated at birth20 (indicating a decreased association between bed sharing and SIDS for mothers who initiated breastfeeding at birth), and history of illness since birth20 (indicating an increased association between bed sharing and SIDS for infants who had a history of illness since birth). For the interaction between history of illness since birth and bed sharing, the authors of the report comment that this raises the question of whether some infants are taken into the parental bed specifically because of illness, and they speculate that it may be the illness rather than bed sharing per se that is associated with death.
We focused our investigations on 3 purported child-related benefits of bed sharing: breastfeeding, parent-child bonding, and sleep-related issues. Our searches identified 3 studies (in 4 publications) that examined the effect of bed sharing on the practice of breastfeeding.41,43,45,47 All were prospective cohort studies and were published between 1999 and 2004. They were conducted in England,47 the United States,41,43 and New Zealand45 and reported on various follow-up intervals including 3 months,47 6 months,41 12 months,45 and the longest interval, 18 years.43 The ethnicities of the study populations were similar, with white subjects being preponderant, while 1 study included black non-Hispanic, Hispanic, and Asian participants.41
Breastfeeding was most likely to be defined in general terms, eg, any breastfeeding43,45; however, 1 study did define it as “present if it occurred within the previous 24 hours.”41 A subset of studies examined bed sharing and its influence on the duration of breastfeeding in mother-infant pairs.43,45,47
These studies suggest a positive association between bed sharing and breastfeeding (ie, increased duration of breastfeeding); however, the data cannot clarify the issue of causality (eg, whether bed sharing promotes breastfeeding and/or whether breastfeeding promotes bed sharing). It is possible that these data reflect the propensity for women who are most likely to practice prolonged breastfeeding to also prefer to bed share.
Our searches did not identify relevant studies, with a contemporaneous comparison, examining the effect of bed sharing in relation to bonding. The association between attachment and bed sharing has not been studied, to our knowledge.
Our searches identified 5 studies that examined bed sharing and sleep-related issues.38- 40,43,46 Four studies38- 40,43 examined infant sleep-wake patterns or problems (ie, night awakenings) and 2 examined infant sleep physiology.39,46
The 3 studies that examined infant sleep-wake patterns38- 40 were published between 1996 and 2004 and represented populations with varied socioeconomic status (SES) from the United States39,40 and New Zealand.38 All were case-control in design and included infants aged 5 weeks to 48 months. All studies showed that infants who bed share have an increased number of awakenings when compared with solitary-sleeping infants; however, 2 of the studies38,40 showed that individual awakenings were shorter in the bed sharers than in the solitary sleepers. It is interesting that in 1 study39 the proportion of bed-sharing children with night awakening occurring 3 or more times per week was approximately double that of the non–bed-sharing children. The difference was significant for lower-SES whites (75% vs 29%; P = .006) and higher-SES African Americans (46% vs 21%; P = .008), but among higher-SES whites it was not statistically significant, perhaps because of small numbers of regular bed sharers in this group (50% vs 25%; P = .2).
No primary studies (that included a comparison) examining strategies to reduce child-related harms associated with bed sharing were identified through our literature search.
Our review highlights 3 general difficulties with the studies: (1) few of the studies specifically investigated the risks or benefits of bed sharing; (2) definitions used for bed sharing, especially in the harm studies, were too heterogeneous to compare across studies; and (3) incomplete reporting of interactions hampered synthesis.
Studies we reviewed concerning the harms associated with bed sharing were, for the most part, derived from population-based case-control studies undertaken in the mid-1990s. The objective of most included studies was not to evaluate the risks and/or harms of bed sharing directly, but rather to study a variety of potential risk factors for SIDS. Typically, in the larger national-based studies, comprehensive questionnaires were administered to parents whose children succumbed to sudden unexpected death and to parents of matched controls. Embedded within these questionnaires were items soliciting information on bed sharing. After completion of data collection in these large epidemiologic studies, multiple articles—each relying on the same data set, but focusing on a different aspect of a risk factor—were published. More specifically, much of the data analyses was exploratory, with bed sharing being just one of many variables examined as possible risk factors. It is extremely difficult to draw conclusions from these reports because they were, more often than not, intended to generate, rather than test, hypotheses by exploring a multitude of potential risk factors and by performing multiple tests of statistical significance. The definition of risk exposure (bed sharing) varied considerably between studies, as did the definition of smoking status. Any attempt to compare results across these studies was therefore extremely difficult.
When there is significant interaction between a risk factor and SIDS (eg, smoking), an association between bed sharing and SIDS might not be meaningful unless the specific factor is taken into account. While the investigation of interactions was one of our primary objectives, it was difficult to glean the information because reports of interactions frequently lacked sufficient detail for our purposes.
The evidence does suggest that there may be an association between bed sharing and SIDS among smokers (however smoking status is defined) but that this association may not be present among nonsmokers. This does not mean that no association between bed sharing and SIDS exists among nonsmokers, but simply that existing evidence does not convincingly establish such an association. The evidence also suggests that bed sharing may be more strongly associated with SIDS for younger infants. It should be noted that, for the 3 reports11,13,20 that showed this association, a portion of the data of 2 of them11,20 were included in the third.13 This finding for younger infants is also supported by recent publications by Tappin et al48 from Scotland and McGarvey et al49 from Ireland that reported increased risk of SIDS for bed-sharing infants younger than 11 and 10 weeks, respectively (these 2 studies were published beyond our search dates for the systematic review). This latest study from Ireland49 is an 8-year study (1994-2001) and is composed of some data (1994-1998) from an earlier publication by the same authors.20
Our findings concerning infants of nonsmoking parents and younger infants need to be qualified. Differences between study designs, data and reporting, and the limited attention paid to the control of confounders, in some studies, preclude the drawing of definitive conclusions.
Evidence suggests that there is a positive association between bed sharing and an increase in the rate and duration of breastfeeding; however, the data cannot clarify the issue of causality. It is possible that the data reflect the propensity for women who are most likely to practice prolonged breastfeeding to also prefer to bed share.
The evidence also suggests that infants who bed share have an increased number of awakenings during the evening as compared with solitary-sleeping infants. Although speculative, it has been suggested that these awakenings are potentially protective against SIDS, which may relate to the infants' ability to rouse.38- 40
The primary strength of our systematic review is that it provides a thorough review of the evidence as a starting point for those wishing to develop clinical practice guidelines. This is important because most recommendations and guidelines for bed sharing are based on nonsystematic samples of evidence. Our review also summarizes data from studies that included a contemporaneous comparison group. In studies without a comparison, there is no adequate way to assess the influence of bias. In such circumstances, it is pragmatic and scientifically prudent to limit systematic reviews to primary studies that have a comparison group.
Our review was limited to English-language literature. Although this is not an atypical practice for systematic reviews, there may be published, relevant non-English reports available that were not identified in our report. Second, our reporting and assessment of each study were limited to published data because no attempts were made to contact authors for additional information or missing data.
Finally, the scope of the review does not focus on risk factors independently associated with the benefits or harms of bed sharing. For some risk factors, there may have been a statistically significant association between the risk factor and SIDS, in which case the adjustment could have a substantial effect. However, for the same risk factor, the association between bed sharing and SIDS may not vary with the risk factor, in which case there would not be a significant interaction between the risk factor and bed sharing as a risk factor for SIDS. Failure to find a significant interaction with bed sharing does not preclude the existence of a statistically significant association between a risk factor and SIDS.
The issue of bed sharing and sudden death demands reevaluation, with hypothesis-driven studies using a prospective design and a standardized definition of bed sharing. Most of the studies we reviewed were undertaken more than a decade ago. The prevalence of bed sharing may have changed as a result of public health statements, guideline recommendations, or societal factors. Because an increasing proportion of deaths attributed to SIDS occur in families with low SES,50,51 efforts are needed to include data from these families in future studies. The exact sleep environment of those families, as well as other potential confounders, remains unknown. Without these data, it is impossible to determine primary risk factors associated with harms.52
In addition, the validity of smoking exposure was not verified in any of the studies despite knowledge that smoking is underreported.53- 55 Because exposure to smoking has such a strong association with SIDS,56 it is extremely important to include biological markers to verify smoking exposure in future studies.57 Smoking exposure of the infant from other smokers in the household may be significant as well, even if the mother is a nonsmoker. Evidence suggests that the risk of SIDS has been shown to increase with the number of smokers in the household.52 This should also be taken into account in future studies.
Correspondence: Tanya Horsley, PhD, Chalmers Research Group at the CHEO Research Institute, 401 Smyth Rd, Room 238, Ottawa, Ontario, Canada K1H 8L1 (firstname.lastname@example.org).
Accepted for Publication: October 4, 2006.
Author Contributions:Study concept and design: Horsley, Clifford, Sampson, Moher, Schachter, and Côté. Acquisition of data: Horsley, Clifford, Yazdi, Dingwall, and Côté. Analysis and interpretation of data: Horsley, Clifford, Barrowman, Bennett, Moher, and Côté. Drafting of the manuscript: Horsley, Bennett, Yazdi, Sampson, Dingwall, and Côté. Critical revision of the manuscript for important intellectual content: Horsley, Clifford, Barrowman, Bennett, Sampson, Moher, Dingwall, Schachter, and Côté. Statistical analysis: Clifford and Barrowman. Obtained funding: Moher, Schachter, and Côté. Administrative, technical, and material support: Horsley, Clifford, Bennett, Yazdi, and Dingwall. Study supervision: Horsley, Schachter, and Côté. Search expertise: Sampson and Dingwall.
Financial Disclosure: None reported.
Funding/Support: This study was supported by the City of Toronto, Toronto Public Health.
Disclaimer: The views expressed herein do not necessarily represent the views of Toronto Public Health.
Acknowledgment: We thank Raymond Daniel for his contribution to bibliographic database management and article acquisition.
This article was corrected for typographical errors on 4/11/2007.