Article Text
PDF
Abstract
Objectives Falls are the leading cause of non-fatal injury among young children. The aim of this study was to identify and quantify the circumstances contributing to medically attended paediatric fall injuries among 0–4 years old.
Methods Cross-sectional data for falls among kids under 5 years recorded between 2012 and 2016 in the National Electronic Injury Surveillance System was obtained. A sample of 4546 narratives was manually coded for: (1) where the child fell from; (2) what the child fell onto; (3) the activities preceding the fall and (4) how the fall occurred. A natural language processing model was developed and subsequently applied to the remaining uncoded data to yield a set of 91 325 cases coded for what the child fell from, fell onto, the activities preceding the fall, and how the fall occurred. Data were descriptively tabulated by age and disposition.
Results Children most often fell from the bed accounting for one-third (33%) of fall injuries in infants, 13% in toddlers and 12% in preschoolers. Children were more likely to be hospitalised if they fell from another person (7.4% vs 2.6% for all other sources; p<0.01). After adjusting for age, the odds of a child being hospitalised following a fall from another person were 2.1 times higher than falling from other surfaces (95% CI 1.6 to 2.7).
Conclusions The prevalence of injuries due to falling off the bed, and the elevated risk of serious injury from falling from another person highlights the need for more robust and effective communication to caregivers on fall injury prevention.
- Fall
- Descriptive Epidemiology
- Child
Data availability statement
Data are available in a public, open access repository. Data from the National Electric Injury Surveillance System are freely accessible to the public through the Consumer Product Safety Commission. Please visit https://www.cpsc.gov/Research--Statistics/NEISS-Injury-Data for details.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Falls are the leading cause of non-fatal injury and the leading cause of traumatic brain injury among children aged 0–14 years. Young children under 5 years are mostly likely to experience fall injuries at home. Despite the prevalence of fall injuries among children, there is limited information on the circumstances contributing to falls, making it challenging to craft meaningful prevention messages.
WHAT THIS STUDY ADDS
This study identifies circumstances of paediatric fall injuries and those most likely to result in hospitalisation. The findings highlight the prevalence of injuries due to falling off the bed, and the risk of serious injury from falling from another person.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
This study identifies opportunities for improving communication to caregivers on fall injury prevention.
Background
Falls are the leading cause of non-fatal injury among children aged 0–14 years in the USA, leading to approximately 2 million emergency room visits and over 30 000 hospitalisations each year. These falls result in over US$7 billion in medical costs and US$16 billion in work loss.1 Among children 0–14 in 2019, the rate of non-fatal paediatric fall injuries was highest among 1–2 years ols (5848 per 100 000) and 3–4 years old (4124 per 100 000), followed by infants less than 1 year (3670 per 100 000).1 Falls are also the leading cause of traumatic brain injury (TBI) among children, accounting for 90% of TBIs in children aged 0–4 years and 42% of TBIs in children 5–14 years old.2 Children under 5 years of age are most likely to experience fall injuries at home, while school age children and adolescents tend to experience fall injuries at school or during sporting activities.3 4
Despite the prevalence of fall injuries among children, there is limited evidence on how to prevent them. A recent review established strong evidence for reducing the use of baby walkers, education about and distribution of stair gates, and mandates for window guards, but the authors did not find research on strategies to prevent falls from furniture.5 The American Academy of Pediatrics recommends adult supervision, using playpens or cribs, and removing sharp-edged furniture and surfaces that offer potential for unintended climbing.6 Recent analyses using data from national samples have described the demographic characteristics of children experiencing fall injuries as well as the consumer products associated with those falls,3 4 but little is known about the circumstances leading up to and surrounding fall events, information essential to designing effective prevention strategies.
The aim of this study was to identify and quantify the circumstances contributing to medically attended paediatric fall injuries among a nationally representative sample of 0–4 years old to inform injury prevention recommendations. The analysis harnessed the power of machine learning techniques and the rich information contained in case narratives within the National Electronic Injury Surveillance System-All Injury Programme (NEISS-AIP) to (1) describe the circumstances in which these injuries occurred and (2) examine the relationship between the fall circumstances and injury severity.
Methods
The NEISS-AIP collects data on non-fatal injuries from a sample of 66 hospital emergency departments in the USA. The data are collected by the Consumer Product Safety Commission (CPSC) in partnership with Center for Disease Control’s (CDC) National Center for Injury Prevention and Control and are weighted to be nationally representative. In addition to demographic characteristics, diagnosis and disposition, mechanism and intent of injury, a narrative field of up to 400 characters captures notes about the reason for the emergency room visit.7–9 As no coded, quantiative variables on the circumstances for the emergency room visit are available in the NEISS data, the narrative was the primary source of information for this analysis.
Using the 2015 data (most recently available at the time of this study), a randomly selected sample of 4546 case narratives was manually coded for four characteristics: (1) what the child fell from; (2) what the child fell onto; (3) the action leading to the fall (eg, if the child slipped, was dropped) and (4) the precipitating event (eg, what the child was doing immediately before the fall). These four characteristics were selected by the study team as circumstances that help describe the context of fall injuries and contributing factors that may inform prevention. The manually coded sample was randomly selected by year of age to ensure the distribution of ages in the coded sample would match that of the larger sample. A coding dictionary was established through a content analysis the first 250 case narratives. For the fell from and fell onto characteristics, the codes were guided by product codes from CPSC; additional codes for falls from or onto non-product-related items (ie, from another person or from a standing position), and codes for fall action and precipitating event were developed and assigned based on a review of the narratives. The remaining selected narratives were then coded by a single coder according to the coding dictionary. A study coauthor met regularly with the coder and reviewed a sample of the codings to verify consistency with the coding dictionary.
Using the sample of manually coded cases as the ‘gold standard’, a machine learning model was designed to predict the four characteristics of interest. Narratives were preprocessed to optimise the text for natural language processing (NLP) by the spaCy Python library.10 Subsequently, NLP was used to identify the type of falls in the narratives. Because the narratives are centred around a fall event, we trained deep learning event classification pipelines using state-of-the-art transformer models, including Bidirectional Encoder Representations from Transformers, or BERT, which achieve high accuracy on well-established shared tasks.11 In particular, we used the FLAIR text classification Python library,12 with pretrained BERT base model from Hugging Face.13 After fall events were labelled, they were translated and mapped to the coded variables via a postprocessing pipeline in spaCy. After an optimal model was developed, it was applied to the remaining uncoded narratives to yield a final coded set of over 90 000 cases coded for the four characteristics of interest: what the child fell from, what the child fell onto, the fall action and the precipitating event.
Descriptive statistics of child’s age (grouped as: infants <1 year, toddlers 1–2 years and preschoolers 3–4 years), sex, race and disposition (classified as hospitalised or treated and released) were generated for the entire sample. Tabulations for the four circumstance characteristics were generated by age group. Subsequently, we tabulated the circumstance characteristics by disposition, and selected those fall circumstances with at least 4% hospitalised and greater than 5000 cases to examine the relationship between circumstances and injury severity. These two criteria were selected because they represent those circumstances at elevated risk for hospitalisation with sufficient data for statistical comparisons. Relative risk was established with a Rao-Scott χ2 test and odds of hospitalisation was estimated using logistic regression adjusting for age. Statistical analyses were conducted in SAS V.9.4.
Patient and public involvement
No patients or members of the public were involved in the design, conduct, reporting or dissemination plans of our research.
Results
A total of 91 325 falls among children under 5 years were recorded in the NEISS-AIP data between 2012 and 2016, reflecting a weighted estimate of 3.87 million total (95% CI 2.94 million to 4.80 million) and 773 674 annual (95% CI 588 161 to 959 187) fall injuries treated in US emergency departments. The majority of these injuries were treated and released, however, 2.7% resulted in hospitalisation. See table 1 for additional descriptive statistics on sex, age and race.
National estimates of paediatric falls in the USA, NEISS-AIP, 2012–2016
The circumstances leading to fall injuries are described in table 2. Infants (<12 months) most often fell from the bed (33%), primarily onto the floor (97%). Infants also commonly fell from baby equipment (16%), or from another person (11%). Toddlers (1–2 years) and preschoolers (3–4 years) fell from a standing or seated position in 17% and 18% of cases, respectively, primarily during some type of independent movement (51% and 53% respectively). Falls from the bed were also common in these age groups (13% in toddlers and 12% in preschoolers), usually preceded by independent movement (29% and 43%, respectively). The fall action could not be identified in the majority (79%) of cases; of those cases where the action was identified, trips (6.7%) and slips (3.8%) were the most frequent. The precipitating event was identified in 52% of cases, with children most commonly moving independently (25%) such as playing or running immediately before the fall.
Estimates of paediatric fall circumstances by age, 2012–2016
Children were more likely to be hospitalised if they fell from another person (7.4% vs 2.6% for all other sources; p<0.01) (see table 3). These children were most likely dropped (75%) during dependent movement (ie, being carried by an adult, 74%). After adjusting for age, the odds of a child being hospitalised following a fall from another person were 2.1 times higher than falling from other surfaces (95% CI 1.6 to 2.7). Relative to other sources of falls, children falling from baby equipment (4.5% vs 2.6%; p<0.01) and miscellaneous furniture such as tables, desks and dressers (4.7% vs 2.6%; p<0.01) had elevated rates of hospitalisation, although only falls from furniture remained statistically significant after adjusting for age. The odds of being hospitalised due to a fall from a being dropped (vs other fall action) and following dependent movement (vs other precipitating events) were also elevated after adjusting for age (see table 4).
Paediatricfall circumstances by disposition
Risk of hospitalisation for select circumstances of falls
Discussion
An estimated 773 000 fall injuries to children 4 and under were treated in US emergency departments annually between 2012 and 2016, resulting in over 20 000 hospitalisations per year. This study identified falls from beds as a leading contributor to medically attended fall injuries, particularly among infants. In addition, a large share of falls from beds in toddlers and preschoolers were preceded by independent movement such as playing or jumping. These cases likely represent avoidable injuries. Our results are consistent with other analyses of fall injuries. Another study using NEISS data estimated that beds were associated with over 26% of medically attended infant falls and 10% of falls among 1–4 years old.3 And data from paediatric trauma registries have identified falls from the bed as a leading source of falls requiring hospitalisation.14 15 While it may not be possible to eliminate children jumping and playing on beds, prevention messages should emphasise the associated risks, recommend close supervision and educate caregivers about the potential for soft surfaces around beds to offer some modicum of protection.
This analysis also highlights the elevated risk of serious injury for a child falling from another person such as a caregiver. In our analysis, children who fell from another person were twice as likely to be hospitalised than children who fell from any other source. Analysis of data from a trauma registries also identified falls from a caregiver’s arms as a leading source of falls leading to hospitalisation.14 15 These falls occur most often among infants; given the nature of child development, infants spend more time being held and carried by a caregiver. In our analysis, even after adjusting for the child’s age, falls from another person were more likely to result in serious injury. This underscores the need for messaging caution when holding young children. We do not have data as to where these fall injuries occurred, such as on stairs, although falls on stairs are known to be particularly hazardous for children as well as adults.16 Nevertheless, our data strongly suggest that prevention messages regarding the risks of falling while carrying a baby around the home are needed. To our knowledge, this information is not routinely shared with parents in the most commonly available fall prevention literature. Prevention strategies may include attention to footwear, keeping one hand free for holding onto handrails on stairs, and having soft surfacing where possible.
Our analysis shows an increase in the share of falls due to independent movement as children age. Increased independent movement is expected as children develop but there is a need for parents to understand their child’s limits, to establish safe spaces for the child to play, to instal soft surfacing where possible, and to be vigilant with supervision behaviours. Current recommendations advise parents to stop using a crib between 18 and 36 months to prevent the child from climbing and falling out of the crib. This logic can also be applied to the use of playpens and stairgates, but more attention is needed to understand how to help families create a home environment that is safe for young children.
Large investments have been made into national data sources such as NEISS and National Emergency Department Sample to collect data on medically attended injuries, however, the potential for these data to directly inform injury prevention relies on having sufficient context about the events leading up to the injury. Based on our experience in this study, we believe that a new emphasis on including prevention-related information in the database is urgently needed. An additional field dedicated to capturing the precipitating event would increase the utility of these systems for prevention. In our analysis, the precipitating event was extracted from the narrative using NLP, however, that information was unspecified in nearly half of the case narratives. Case narratives have been traditionally used as a catch-all to describe the injury event as well as a description of what was happening when the injury occurred.8 A data field dedicated to capturing precipitating events separate from the injury diagnoses could yield information about the sequence of events leading up to the injury with the richness needed to identify opportunities for intervention and prevention.
This analysis relied on narratives contained within NEISS-AIP which are limited to injuries treated in the ED. These data do not reflect falls treated in other medical settings such as primary and urgent care. The utility of this analysis is also limited by the level of detail contained within the narrative field; over 20% of narratives made no mention of what the child fell from, and the fall action was unspecified in the majority of cases. Additionally, no information on the height of the fall were available. However, an important strength of this analysis is the use of NLP to code a large volume of data for these fall circumstances, allowing us to produce national estimates of substantive contextual information relevant to prevention. To our knowledge, analysis of narrative text from NEISS has been limited to manually coded fields, which are labour intensive, or simple text searches, which have limited utility.17–19 This analysis demonstrates the power of machine learning to inform injury prevention and the potential for NEISS to be used in that capacity.
Current fall prevention recommendations advise parents to supervise small children closely in the home, discourage climbing on furniture, and to use safety features of baby equipment. The prevalence of injuries due to falling off the bed, and the elevated risk of serious injury from falling from another person highlights the need for more robust and effective communication to caregivers on fall injury prevention.
Data availability statement
Data are available in a public, open access repository. Data from the National Electric Injury Surveillance System are freely accessible to the public through the Consumer Product Safety Commission. Please visit https://www.cpsc.gov/Research--Statistics/NEISS-Injury-Data for details.
Ethics statements
Patient consent for publication
Ethics approval
The Johns Hopkins School of Public Health IRB exempted this study from review and oversight for the use of anonymous, publicly available data.
Footnotes
Twitter @eliseomaki
Contributors EO, WS and AG conceptualised the study, obtained funding, provided overall oversight and leadership of this work; MR constructed the natural language programming models; PD-B and RS reviewed the narratives, established the coding framework and labelled narratives accordingly. EO was responsible for the overall content, accepts full responsibility for the finished work, had access to the data, and controlled the decision to publish.
Funding This work was supported by a grant from the National Institute for Child Health and Development (grant number 1R21HD099513) and from the American Public Health Association's Injury and Violence Prevention Data Science Demonstration Programme.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.