Article Text

Download PDFPDF

Rear-facing versus forward-facing child restraints: an updated assessment
  1. Timothy L McMurry1,
  2. Kristy B Arbogast2,
  3. Christopher P Sherwood3,
  4. Federico Vaca4,
  5. Marilyn Bull5,
  6. Jeff R Crandall6,
  7. Richard W Kent6
  1. 1 Department of Public Health Sciences, University of Virginia, Charlottesville, Virginia, USA
  2. 2 Center for Injury Research and Prevention, Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania, USA
  3. 3 Biocore, Charlottesville, Virginia, USA
  4. 4 Department of Emergency Medicine, Yale School of Medicine, New Haven, Connecticut, USA
  5. 5 Department of Pediatrics, Indiana University School of Medicine, Indianapolis, Indiana, USA
  6. 6 Center for Applied Biomechanics, University of Virginia, Charlottesville, Virginia, USA
  1. Correspondence to Professor Timothy L McMurry, Department of Public Health Sciences, University of Virginia, Charlottesville, Virginia, USA; tmcmurry{at}


Objectives The National Highway Traffic Safety Administration and the American Academy of Pediatrics recommend children be placed in rear-facing child restraint systems (RFCRS) until at least age 2. These recommendations are based on laboratory biomechanical tests and field data analyses. Due to concerns raised by an independent researcher, we re-evaluated the field evidence in favour of RFCRS using the National Automotive Sampling System Crashworthiness Data System (NASS-CDS) database.

Methods Children aged 0 or 1 year old (0–23 months) riding in either rear-facing or forward-facing child restraint systems (FFCRS) were selected from the NASS-CDS database, and injury rates were compared by seat orientation using survey-weighted χ2 tests. In order to compare with previous work, we analysed NASS-CDS years 1988–2003, and then updated the analyses to include all available data using NASS-CDS years 1988–2015.

Results Years 1988–2015 of NASS-CDS contained 1107 children aged 0 or 1 year old meeting inclusion criteria, with 47 of these children sustaining injuries with Injury Severity Score of at least 9. Both 0-year-old and 1-year-old children in RFCRS had lower rates of injury than children in FFCRS, but the available sample size was too small for reasonable statistical power or to allow meaningful regression controlling for covariates.

Conclusions Non-US field data and laboratory tests support the recommendation that children be kept in RFCRS for as long as possible, but the US NASS-CDS field data are too limited to serve as a strong statistical basis for these recommendations.

  • motor vehicle occupant
  • epidemiology
  • child survival
  • child
  • behavior
View Full Text

Statistics from


The recommendations for infants and young toddlers to ride rear-facing in their child restraint system (CRS) have been broadened in the USA in the last decade to extend beyond the traditional 1 year of age. The National Highway Traffic Safety Administration (NHTSA) currently recommends that all children less than 1 year old should be seated rear-facing and that those aged 1–3 years should ride rear-facing for as long as possible, until reaching the top height or weight limit allowed by the child restraint manufacturer.1 The American Academy of Pediatrics recommends that ‘All infants and toddlers should ride in a rear-facing seat until they are at least two years of age or, preferably, until they reach the highest weight or height allowed by their car seat manufacturer’.2 These recommendations reflect the biomechanical need to support a young child’s torso, neck and head, and distribute crash forces over the entire body. The weight of a child’s head and the anatomy of the developing spine, including incomplete vertebral ossification and increased ligamentous laxity, increase a young child’s risk for spinal cord injury and excessive head excursion during a crash. A rear-facing child restraint system (RFCRS) provides support to the head and spine that significantly reduces neck loading in crashes having a frontal component.3

The broadened recommendations were motivated by a series of research studies that evaluated the benefits of RFCRS. These included sled tests evaluating the biomechanical advantage of rear-facing3–6 and epidemiological data from the USA and Europe that documented the improved protection provided by RFCRS for these young occupants compared with other restraint conditions, including forward-facing child restraint systems (FFCRS), seat belts or no restraint at all. Swedish researchers and public health authorities have long been strong advocates of extended rear-facing. Children in that country remain rear-facing up to 4 years of age and transition directly from the RFCRS to a booster seat. Their policies cite Swedish data that quantify a reduction in the risk of Abbreviated Injury Scale (AIS) 2+ injuries of 90% relative to unrestrained children.7 8 Using US data, Hertz9 documented a reduction in fatal injury by 71% for infants younger than 1 year of age in passenger cars and by 58% in light trucks when using a CRS compared with similar-age children who were unrestrained. Note that these studies did not compare RFCRS with FFCRS directly, driven in part by the relatively high percentage of unrestrained occupants in the Hertz study and the lack of FFCRS as a restraint option in Sweden.

An effort to quantify the relative effectiveness of RFCRS compared with FFCRS using real-world US data was conducted by Henary et al.10 National Automotive Sampling System Crashworthiness Data System (NASS-CDS) data from 1988 to 2003 for crash occupants 0–23 months restrained in an RFCRS or FFCRS were analysed. Across all crash types, children in FFCRS were 76% more likely to be seriously injured (Injury Severity Score or ISS ≥9) than children restrained in RFCRS. When those 12–23 months were analysed separately, rear-facing children were more than five times less likely to be seriously injured than when restrained in FFCRS.

Recently, Henary et al were asked to verify their conclusions based on feedback from an independent researcher that the original results could not be replicated. On review we have determined that the NASS-CDS survey weights were improperly handled in the analysis, using SAS PROC LOGISTIC with a weight statement instead of PROC SURVEYLOGISTIC, which caused the apparent sample size to be larger than the actual sample size. This allowed regression adjustments that are not feasible and resulted in inflated statistical significance. In addition, the methods in the Henary et al paper were not documented well enough to replicate the study inclusion criteria, although we hypothesise the original study might have included children in booster seats, which are not recommended for 0-year-olds or 1-year-olds. The following represents a correction of the original analysis and results using appropriate survey analysis techniques, as well as a more detailed description of the methods. In addition, the analyses have been updated to include all currently available data.


The present study used all available NASS-CDS data (years 1988–2015) to compare injury outcomes and child seat orientation. Two analyses are presented, the first based on data from NASS years 1988–2003 to match the time period presented by Henary et al, and the second based on data from years 1988–2015.

Children were included in the analyses if they were coded as 0 or 1 year old (equivalently, 0–11 months or 12–23 months); travelling in a passenger vehicle (body type <50); restrained in a rear-facing child seat designed to be rear-facing (chorient=1) or a forward-facing seat designed to be forward-facing (chorient=12); restrained in an infant, toddler or convertible child seat (chtype=1, 2 or 3); not involved in a rollover (rollover=0 or missing) or fire (fire=0 or missing); and not exposed to air bag deployment. For NASS-CDS data for years 2009 onward, the child needed to be in a vehicle less than 10 years old since occupant outcomes in older vehicles were not thoroughly investigated.

Children were coded as injured if they received an ISS (NASS-CDS variable ISS) of 9 or greater or were fatally injured. An ISS of 9 or greater is equivalent to having at least one injury of severity 3 or greater using the AIS,11 or AIS 2 injuries to at least two body regions plus at least one AIS 1 injury to a third body region. While fatally injured children would typically have an ISS greater than 9, their injuries are sometimes less well documented, so these cases were therefore explicitly included regardless of their AIS-coded injuries. We similarly coded children as having MAIS injuries of 2 or more if the NASS-CDS variable MAIS had a value in the range of 2–6, or if the child was fatally injured. MAIS injuries of 7 (unspecified) were recoded to AIS severity 1 (n=9).

In an effort to describe the types of injuries children sustained in car crashes, we also examined AIS 2+ injuries by body region. NASS-CDS years 1992–1988 use AIS 1985 codes, which have a different division of body regions from the more recent versions of AIS, so this analysis was restricted to only years 1993 and later.

Data are described as unweighted counts to communicate the available sample sizes, together with survey-weighted percentages, means, SD and P values. Overall, we found a very low number of injuries, which precluded the use of regression to control for occupant and crash characteristics.12 The analyses were therefore restricted to bivariate comparisons of injury and seat orientation.

All analyses were performed using the R statistical language (V.3.4.1, R Core Team, Vienna, Austria, 2017), along with the survey package,13 14 and survey weighting was used for all analyses except case counts. Primary sampling unit strata are necessary for survey-weighted analysis, but are not in the publicly released 1988–1992 NASS-CDS data sets; NHTSA gave us this information.15 The code used to produce the analyses is included as part of an online supplementary appendix.

Supplementary appendix


In data years 1988–2003, 710 children representing a survey-weighted 417 155 occupants met the inclusion criteria, with 353 (weighted 216 412) in RFCRS and 357 (weighted 200 743) in FFCRS. In data years 1988–2015, 1107 children representing 595 194 weighted occupants met the inclusion criteria, with 679 (weighted 374 681) in RFCRS and 428 (weighted 220 513) in FFCRS. Population and crash characteristics are shown in table 1A (data years 1988–2003) and table 1B (data years 1988–2015) overall and stratified by CRS orientation, with survey-weighted P values comparing forward-facing and rear-facing children. One-year-olds were significantly more likely to be forward-facing than 0-year-olds (P<0.001), and relatedly children in FFCRS tended to be taller and heavier. The difference in height was not statistically significant in the 1988–2015 analysis, which appears to be because in more recent years a much higher percentage of 1-year-olds were in RFCRS. Children in FFCRS sustained MAIS 2+ injuries in side crashes at a borderline significantly higher rate than children in RFCRS (P=0.046), while in frontal crashes children in FFCRS sustained higher rates of MAIS 2+ injuries, but the difference was borderline insignificant (P=0.052) using all years of NASS-CDS.

Table 1A

Crash and demographic characteristics, years 1988–2003, presented as unweighted n (survey-weighted %) for discrete and ordinal variables and survey-weighted mean (SD) for continuous variables, along with survey-weighted P values comparing occupants in RFCRS with occupants in FFCRS

Table 1B

Crash and demographic characteristics, years 1988–2015, presented as unweighted n (survey-weighted %) for discrete and ordinal variables and survey-weighted mean (SD) for continuous variables, along with survey-weighted P values comparing occupants in RFCRS with occupants in FFCRS

Across all years of NASS-CDS, 47 children sustained ISS 9+ injuries and 17 of these were 1-year-olds. Three of the injured 1-year-olds were in RFCRS and the other 14 were in FFCRS (table 2). Table 2 compares the rates of injury by seat orientation across all 0-year-olds and 1-year-olds for the two different time periods studied. Overall, children in RFCRS had a slightly higher rate of ISS 9+ injuries (survey-weighted 0.6%) than children in FFCRS (0.5%), but this result is confounded by age because 0-year-olds were more likely to be injured and more likely to be rear-facing. When stratified by age, children in RFCRS consistently had lower rates of injury than children in FFCRS, although no differences in injury rate were statistically significant at either the ISS 9+ or MAIS 2+ level.

Table 2

Number of injured and uninjured occupants by age and seat orientation along with survey-weighted percentages and P values comparing the injury rates between RFCRS and FFCRS

Child seat designs evolve over time. In order to provide an analysis using only more contemporary data, we also repeated the analysis using NASS-CDS years 2000–2015. In this time period, 0.35% of 1-year-olds in RFCRS sustained ISS 9+ injuries, compared with 0.43% of 1-year-olds in FFCRS; the difference was not statistically significant.

Table 3 presents the AIS 2+ injury rates by age, seat orientation and body region. One-year-olds in FFCRS have noticeably more injuries across almost all body regions, but the MAIS 2+ injury rate difference between RFCRS and FFCRS was not statistically significant for either 0-year-olds (P=0.080) or 1-year-olds (P=0.203).

Table 3

Number of occupants injured at the AIS 2+ level by age, seat orientation and body region, NASS-CDS years 1993–2015, presented as unweighted n (survey-weighted %)


The results presented herein reanalysed MVC field data from a large US crash surveillance system (NASS-CDS) to evaluate the relative protection provided by RFCRS compared with FFCRS for children up to 2 years of age. Two analyses were conducted: (1) with data from 1988 to 2003 to correct previous work using appropriate survey analysis techniques, and (2) using all currently available NASS-CDS data (1988–2015) to provide an updated assessment of the research question using a larger and more contemporary data set. In 2009, the NASS-CDS programme stopped investigating crashes in vehicles older than 10 years. As a result, our sample is representative of all children in crashes in years prior to 2009, but only of those in newer vehicles in the years 2009–2015.

There were only 47 injured children in the expanded data set, which reflects the overall excellent protection provided by child restraints and the associated low numbers of injured children in the NASS-CDS data set. Sample size calculations indicate that approximately 12 000 children would be required to detect a difference between 0.2% and 0.5% injury rate (the injury rates of RFCRS and FFCRS for 1-year-olds from the 1988–2015 data) with 80% power. All 28 years of NASS-CDS data contain only 1107 children younger than 2 years meeting our inclusion criteria. As a result, comparisons of RFCRS and FFCRS for developing policy and best practice recommendations must consider additional assessment tools, including controlled laboratory experiments, computational and analytical simulations, and complementary field injury data sets.

The relative injury rates found in the current study do not contradict the biomechanically justified assertion that RFCRS provides a safety benefit compared with FFCRS. Both 0-year-olds and 1-year-olds in all data year groupings experienced lower (but not statistically different) rates of injury when restrained in RFCRS compared with FFCRS. The confounding between age and CRS orientation indicates that it is inappropriate to compare relative protection in the age-aggregated data. Rather, the relative protection of RFCRS and FFCRS is better reflected by injury rates in the stratified age samples.

In order to reach a goal of zero deaths and serious injuries, contemporary questions in child passenger safety involve injury risk assessments in which the underlying injury rate may be rather low, such as the RFCRS to FFCRS comparison studied herein. This study’s findings reflect the limitations of NASS-CDS for evaluating child occupant crash safety and illustrate the need for a nationally representative resource for child passenger safety data that are sufficient in numbers and depth of child-specific data to support evidence-based policy. Such a data set is necessary for government, industry and auto safety researchers to jointly improve motor vehicle safety for children and youth.


NASS-CDS data indicate an extremely low injury rate in children up to 2 years of age in both RFCRS and FFCRS. The estimates found herein suggest that both 0-year-olds and 1-year-olds have a lower rate of injury in RFCRS than in FFCRS, but the NASS-CDS data are insufficient to differentiate these injury rates to a 0.05 level of statistical significance. In the absence of a sufficient US field injury data set, biomechanical studies3–6 and the field experience in Sweden,7 8 where most children in this age range and beyond are rear-facing, remain the key justifications for the recommendation to keep children rear-facing as long as possible, up to the height or weight limit of the CRS.

What is already known on the subject

  • The National Highway Traffic Safety Administration and the American Academy of Pediatrics recommend that children be kept in rear-facing child restraint systems (RFCRS) for as long as possible.

  • Child seat recommendations are based on biomechanical and US-based and European-based field data analyses.

What this study adds

  • This study updates and corrects previous field data analyses using data from the US National Automotive Sampling System Crashworthiness Data System database.

  • While this study could not statistically differentiate between the protection afforded by RFCRS and FFCRS, other evidence supports the recommendations that small children be placed in RFCRS.


The authors thank Jeya Padmanaban for requesting reanalysis and update of the Henary et al study, and Dennis Durbin, MD, MSCE, for his contributions to the analytical approach.


View Abstract


  • Contributors TLM, KBA, CPS, FV, MB, JRC and RWK each contributed to the manuscript. TLM contributed to the planning, conduct and reporting of the work, and is the guarantor of the overall content. KBA contributed to the planning, conduct and reporting of the work. CPS contributed to the planning and reporting of the work. FV contributed to the reporting of the work. MB contributed to the reporting of the work. JRC contributed to the planning of the work. RWK contributed to the conduct and reporting of the work.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement All data are publicly available from NHTSA:

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.