Objective. To determine whether the risk of unintentional injury requiring emergency department (ED) or inpatient care in children is transiently increased over a 90-day period after injury to a sibling.
Design. Retrospective cohort.
Setting. King County, Washington.
Participants. A total of 41 242 children 0 to 15 years of age continuously enrolled in Medicaid and living in King County during the period October 1, 1992 through September 30, 1993 (27 450 child-years).
Outcome Measures. The outcome was an unintentional injury treated in the ED or inpatient setting. Incidence rates and hazard ratios were calculated for children whose sibling had been injured in the previous 90 days, compared with children without such exposure. Multivariate analysis was used to adjust for age, gender, race, sibling group size, and noninjury ED use.
Results. There were 4921 injuries treated only in the ED and 82 hospital admissions. The incidence of ED treated injury was 305 per 1000 child-years among children whose sibling had been injured in the previous 90 days and 174 per 1000 child-years among children without this exposure (relative risk: 1.75; 95% confidence interval: 1.56–1.95). The incidence of injury-related hospitalization was 1.7 per 1000 child-years among children whose sibling had been injured in the previous 90 days, compared with 3.0 per 1000 child-years among children without this exposure (relative risk: .57; 95% confidence interval: .07–2.12). Injury risk peaked in the period 4 to 10 days after a sibling's injury and returned toward, but did not attain, baseline risk over the subsequent 2½ months. The magnitude of this effect depended on the child's age; the relative risk of injury was higher among older children.
Conclusions. Injuries treated in the ED or inpatient setting appear to cluster within sibling groups over brief periods of time. Shared social or environmental exposures may contribute to this clustering and may be amenable to targeted, time-limited prevention interventions.
- ED =
- emergency department •
- CI =
- confidence interval
Unintentional injury is a leading cause of death and disability in childhood.1 Unlike many other causes of morbidity and mortality among children, injuries are thought to be almost wholly preventable.2 Unfortunately, clinical efforts to prevent childhood injury have met with only modest success. Although educational or motivational strategies can change injury-related knowledge or behavior, few clinical interventions have been shown to reduce injury morbidity unless paired with community mobilization or environmental modification efforts.3 One strategy to improve the efficacy of clinical injury prevention interventions might be to target populations of children known to be at high risk during specific periods of vulnerability.
Many studies have sought to identify factors or conditions that predispose individual children to injury morbidity, including sociodemographic traits,4–9 family attributes,10–13 behavioral factors,5 14–17temperament,18 and exposure to stressful life events.5 ,11 19–22 One of the most striking and robust observations is that occurrence of a medically attended injury is associated with increased risk of reinjury.5 ,16 ,20 ,21 23–26 Similar findings have been observed among adult patients.27–32 However, the concept of accident proneness has been criticized in the pediatric literature because, among other concerns, it is a risk factor that can only be diagnosed after repeated injury, making it a poor trigger for programs that hope to prevent injuries.33–35
We hypothesized that the siblings of an injured child might themselves be at increased risk for injury over some period of time, perhaps because of changes in family supervision associated with the index injury or with preexisting social stresses. If this were the case, a clinical encounter for injury care in 1 child could trigger a time-limited preventive intervention that would benefit all children in the family. For the siblings of the injured child, such an intervention might provide benefit before any injury occurred. To test this hypothesis, we sought to determine the incidence of medically treated injuries among the siblings of children admitted to the hospital or treated for injury in an emergency department (ED) setting.
We conducted a retrospective cohort study of children of 0 to 15 years of age who were continuously enrolled in Medicaid in a single urban county over a 12-month period. The relative risk of injury requiring ED or inpatient care was estimated as a function of the time elapsed since the last medically attended injury in a sibling. For comparison, similar estimates were made of the risk of reinjury associated with time elapsed since a previous medically attended injury in the same child.
The study cohort was assembled from data that identified children 0 to 15 years of age living in King County, Washington and who were eligible for Medicaid under the Aid for Families with Dependent Children program over the period from October 1, 1992 through September 30, 1993. Analysis was restricted to families in which all eligible children were continuously enrolled in Medicaid during the study period. Families with children born during the study period and continuously enrolled thereafter were included. Over the 12-month study period, 53 725 eligible children (0–15 years of age) were enrolled in the Medicaid program. Of these, 9839 (18.3%) were excluded because they, or their siblings, were not continuously enrolled after initial enrollment. Another 2640 children (4.9%) were enrolled in the cohort but did not contribute follow-up information because they did not complete a 90-day “run-in” period (described below) before conclusion of the study. Four children were excluded because of irreconcilable irregularities in their enrollment information. This left 41 242 children in 23 609 families for analysis.
Medicaid enrollment and claims data included all claims for inpatient and outpatient medical care. These claims were filed before Washington State Medicaid adopted a capitated managed care system in 1994.
Children were linked in the dataset as siblings by using a case number in the eligibility file that identified a common adult guarantor (usually the mother under Aid for Families with Dependent Children). Validation that siblings resided in the same household was not possible. These groups of administratively defined siblings were considered sibling groups.
Children were classified into 1 of 4 mutually exclusive age groups (0 through <1 year of age, 1 through <4 years of age, 4 through <10 years of age, and 10 through <15 years of age). Sibling group size was defined as the number of children under 15 years of age in a sibling group at the conclusion of the study. To adjust for a family's propensity to use the ED for medical care, we created a binary variable that identified the 37% of sibling groups with at least 1 member who used the ED for noninjury care during the study period.
Children were considered to be exposed for 90 days after treatment of 1 or more of their siblings in the ED or hospital for an injury diagnosis. Exposure status was subcategorized according to the amount of time that had elapsed since the last such injury event (1–3 days, 4–10 days, 11–30 days, 31–60 days, and 61–90 days). Thus, a subject's exposure status could change with the passing of time or with new injury events within the sibling group or both. Unexposed children had an exposure to a previous injury >90 days in the past, or no previous injury recorded for the sibling group.
The outcome of interest was ED treatment or admission for an unintentional injury. Injuries treated only in a non-ED ambulatory setting were excluded. Urgent care clinic visits were not included unless the ED was specifically coded as the place of service. Injury visits were identified from claims records by International Classification of Diseases, 9th Revision codes (ICD-9-CM: 800–854, 860–887, 890–897, 900–904, 910–929, and 940–995). A visit was considered injury-related if an injury or poisoning diagnosis appeared on any claim from that date. Visits were not considered injury-related if only a diagnostic code for iatrogenic injury, late effects of injury, or follow-up care was noted. Very few injury claims had E-codes recorded; however, of those that did, visits were not considered injury-related if an E-code for suicide or child abuse was present because of the study's focus on unintentional injuries.
All ED or inpatient claims for a given child listing the same date of service were considered to represent the same clinical encounter. Because 2 claims resulting from the same clinical encounter could show different dates of service attributable to errors in coding or claim form completion, we elected to omit estimates of risk 1 to 3 days after an injury from our reported results. Only 2% of injury outcome events occurred during this period. We assumed that most claims would report a date of service accurate to within 3 days.
We defined a run-in period of 90 days from the beginning of the study or subject enrollment to insure accurate assignment of exposure status. Only injuries occurring after this initial run-in for each sibling group were considered as outcomes in the analysis. This allowed us to know the elapsed time since any sibling group injury (up to 90 days) for all injury outcomes. A 90-day cutoff was chosen based on the distribution of recurrent injury events in this sample. Subjects were considered to be nonexposed after >90 days had elapsed since any injury in the sibling group. A diagram illustrating the run-in period and the assignment of exposure status in 2 hypothetical sibling groups is shown in Fig 1.
Incidence rates and crude rate ratios were calculated for the risk of injury leading to ED or inpatient care over the 90-day period after the exposure of interest.36 Survival analysis employing Cox regression with time dependent covariates was used to estimate the instantaneous risk of injury as a function of time since the last injury within the sibling group, age, gender, race, and sibling group noninjury ED use. Adjustment for sibling group size was made by stratification.
We modeled both the risk of injury over the entire 90-day period after an exposure and the risk of injury over a number of smaller, discrete time intervals. Robust variance estimates were used to correct standard errors for the multiple outcome events possible within individuals and the clustered nature of the data within sibling groups.37Hazard ratios, with 95% confidence intervals (CIs), were obtained from the exponentiated regression coefficients and their standard errors. Statistical analyses were performed using Stata 6.0 (Stata Corp, College Station, TX).
This study was approved by the human subjects review committee at the University of Washington and by the human research review board at the State of Washington Department of Health and Human Services.
Over the study period, 41 242 children 0 to 15 years of age contributed 27 450 person-years of follow-up. A total of 5003 incident injury events occurring after the run-in period were identified from claims records (4921 requiring only ED care and 82 requiring inpatient care). These injuries were sustained by 4162 children. Demographic characteristics of the study population and of the subset of injured children are shown in Table 1.
In the sample population, the incidence of ED-treated injury was 179 per 1000 child-years and the incidence of injury hospitalization was 3 per 1000 child-years. Children exposed to injury in a sibling during the preceding 90 days had an incidence of ED-treated injury of 305 per 1000 child-years and an incidence of injury hospitalization of 2 per 1000 child-years. Incidence rate calculations are shown in Table 2.
Compared with injury rates among nonexposed children the unadjusted risk ratio for ED treated injury was 1.75 (95% CI: 1.56,1.95) and for injury hospitalization was .57 (95% CI: .07,2.12). After adjustment for age, gender, race, sibling group size, and sibling group ED use for noninjury diagnoses during the study year, the relative hazard for injury treated in the ED among children whose sibling had been injured in the preceding 90 days was 1.53 (95% CI: 1.34,1.74). After adjustment for the same covariates, the relative hazard for injury requiring hospitalization among children whose sibling had been injured in the preceding 90 days was .46 (95% CI: .13,1.64).
Injury risk was also modeled as a function of time elapsed since the last sibling injury. The adjusted relative hazard of ED treated injury, for each time period after an exposure event, is shown in Fig 2. The relative risk of injury after medically treated injury in a sibling peaked at 2.24 in the second week after the event. Risk was elevated >50% above baseline for the month after the event, then declined toward the baseline by the third month.
Interaction terms representing the effect of gender, race, sibling group size, and sibling group ED use on the association between recent injury and injury risk were nonsignificant. When interaction terms representing the effect of age group on the association between injury risk and recent sibling injury were added to the model, they were found to be statistically significant (P = .006). This suggested that the presence or magnitude of the temporal association between sibling injury and subsequent injury risk depended on the subject's age.
Accordingly, the data were stratified and analyzed separately by age group. The categorical variables modeling time since sibling injury, taken as a whole, remained statistically significant predictors of injury risk in all age groups. The effect of exposure, however, was found to vary according to the age of the subject. Figure 3 demonstrates the modification (by age group) of injury risk after recent exposure to a sibling injury. This exposure seemed protective among infants under 12 months of age. In all other children, exposure was associated with an increased relative injury hazard. The magnitude of this effect increased with increasing age of the subject.
Finally, crude injury incidence for children exposed to a recent sibling injury was compared with a similar risk estimate for children who were themselves injured within the preceding 90 days. These rates are shown in Fig 4. The reinjury risk among injured children was greater than the risk of injury to a sibling at all time points over the 90 days of follow-up. When injury experience during the 90 days after an injury to any member of the sibling group was estimated, the cumulative injury incidence within an exposed sibling group over a 90-day period was 10.4%.
Previous work has shown that the risk of unintentional injury in childhood is related to socioeconomic status, maternal age, maternal parity, and to a number of “host factors,” such as gender, chronic medical conditions, and childhood behavioral problems.4–12 ,15 ,17 ,18 ,38 ,39 The child's previous injury experience typically proves to be a strong predictor of subsequent injury risk.5 ,16 ,20 23–26 It is unclear whether this increase in injury risk is attributable to factors intrinsic to the child or to elements in the social or physical environment that might be mutable.
In this study, among urban children enrolled in Medicaid, an injury treated in the ED or inpatient setting was a significant risk marker for subsequent child injury in the family. The period of increased risk lasted for at least 90 days after the index injury and affected both the child who was injured and also, importantly, his or her siblings. This effect on siblings suggests a potential exposure in the social or physical environment that may affect all children in the family.
Another novel finding of this study was the temporal variation in relative risk. Other studies have demonstrated that, among adults, the risk of certain injury outcomes is elevated for a limited period after a sentinel event or clinical marker,27 ,28 ,40 but this type of temporal variation in risk has not, to our knowledge, been explored in childhood injuries. We showed that injury risk varied over the 90 days after the index event, peaking in the second week after a sibling injury and returning toward baseline at 90 days. This suggests that many high-risk families may, in fact, be transiently passing through high-risk periods, rather than existing at a permanently elevated risk.
The mechanism underlying such high-risk periods is unclear. Accident proneness is a concept that has been generally discounted among pediatric injury prevention practitioners.33–35 Critics argue that accident proneness is defined based on an unusual injury history and is of little use as a concept in injury prevention. They also charge that labeling a child as accident prone blames the victim for persistent injury liability when, in fact, there may be elements of the social or physical environment that account for the child's persistently increased risk. Our results confirm that an injury event is associated with a high risk of reinjury in the same child. This risk is highest immediately after the event, falling off over the subsequent 3 months. Although some of this increase in risk may be attributed to characteristics intrinsic to the child, the temporal variation over a relatively short period and the associated increased risk among siblings suggest that a social or environmental exposure of limited duration may mediate at least some of these injuries.
What might this exposure be? Previous studies have demonstrated that various life events, including marriage, divorce, maternal psychiatric illness, relocation, the birth of a child, or a death in the family are associated with increased medical morbidity, including unintentional injuries.5 11–13 ,19 ,20 41–43 In our study, the index injury event might have been, in itself, a life stress that predisposed toward subsequent injury in all family members. Increased injury risk might have been mediated through changes in family supervision of siblings while caring for the injured child.44Alternatively, both the index injury and any subsequent injuries might have been secondary to the effects of other underlying life stresses. The data available in this study were not adequate to characterize the nature of any such exposure.
Stratification of the sample by age group demonstrated that variation in injury risk following injury to a sibling is modified by the child's age. There was an apparent reduction in injury risk among infants <12 months of age after exposure to a sibling's injury. Older children had an elevated risk in the month after a sibling's injury with risk returning toward baseline over the next 2 months. The interpretation of these observations is difficult without detailed social and injury event histories; 1 potential explanation is that the mechanisms underlying the increased injury risk involve exposures not typically encountered during infancy.
There are a number of limitations to this study. Our definition of siblings was based on administrative data linking children to a common adult guarantor. We could not verify that these siblings shared a common household. However, if misclassification of this type occurred, we would expect it would have led to a more conservative estimate of the risk of exposure to injury in a sibling.
Additional misclassification may result from miscoding of dates. Two or more claims for the same injury-related visit, if submitted with a different date of service or incorrectly entered with different dates, would be treated as 2 or more distinct injuries in our analysis. This could overstate the association between previous and subsequent injury events. We could not verify the accuracy of claims coding in this study. We chose, however, to exclude from our analysis relative risk estimates for injury occurring 1 to 3 days after an index event. We speculate that most claims would be filed with a date of service accurate to within 3 days.
Exclusion of clinic visits for injury was done primarily to reduce the problems posed by multiple encounters for the same injury. Children seen in the ED for an injury may be asked to return for follow-up wound or burn care. Coding such visits as unique injuries would overstate the magnitude of an association between index injury and subsequent injury events in the same child. Typically, however, such children would be appointed to a clinic rather than to the ED for care. We believe that by restricting outcomes to injuries treated in the ED and by excluding the (relatively few) events coded explicitly as follow-up care we have minimized the effect of this problem. Moreover, our data looking at injury outcomes after exposure to a sibling injury should be immune to this bias and yet demonstrate a significant temporal association. To the extent that some significant acute injuries were treated exclusively in a non-ED clinic setting, our ascertainment of incident injuries was incomplete.
Because many injuries do not require medical care, variation in the propensity of families to seek care could influence our results if that propensity were influenced by recent family injury events. We attempted to minimize this bias by restricting exposure and outcome events to injuries treated in the ED or requiring hospitalization. Although we recognize that some families use the ED as a source of primary care, we believed that the restrictions described would reduce the number of discretionary medical care visits analyzed. Nevertheless, it is possible that recent experience with an injury that required ED care might predispose a family to seek ED treatment for any subsequently injured children, with less regard to the severity of the injury. This could result in an overestimation of the reported association. We further sought to minimize the potential confounding caused by family patterns of ED use by adjusting for the sibling group's history of ED use for noninjury complaints.
Because most claims lacked an E-code, we could not accurately assign a mechanism of injury in most cases. This limited our ability to assert that all injuries in our population were unintentional. The inclusion of undiagnosed or unrecognized intentional injuries in our sample would not detract from the significance of the results but would affect their interpretation and any potential associated interventions.
The generalizability of these results to a non-Medicaid or nonurban population is unknown. However, the results suggest that within this population, interventions based on injury history may be warranted and more efficient than universally applied programs. We believe that a clinical encounter for injury care in the ED or inpatient setting could be used as a marker to identify sibling groups moving through periods of high injury risk. The period of increased risk after an initial injury event seems to be of relatively brief duration, perhaps reducing the need for costly long-term interventions. The predictive value of such a marker should be assessed prospectively, along with attempts to identify injury mechanisms or patient characteristics that identify subpopulations at even greater risk. If the value of such a strategy can be confirmed, interventions triggered at the time of an index injury event could be considered.
The nature of the intervention proposed would necessarily depend on the mechanism thought to underlie clustered injuries. We speculate that the transient increase in risk that follows an index injury event and affects all children in a sibling group might be explained by variations in the social environment of the children.19 ,21 ,41 ,42 Interventions to address this type of risk would necessarily extend beyond the in-office counseling that clinicians usually associate with injury prevention activities. A brief but intensive social support intervention might be an appropriate vehicle with which to address the needs of families moving through a period of high risk triggered by the index injury or by other stressful life events.
Previous authors have concluded that no risk factor or sociodemographic variable predicted injury well enough to allow targeted intervention.45 Instead, they have advocated a universal approach wherein all children receive injury prevention interventions. In our sample, over 10% of sibling groups identified based on index injury experienced a second injury treated in the ED or admitted to the hospital within 90 days. This suggests that clinical contact for injury care could be used as a risk marker to identify a population at relatively high risk of injury over a limited period. Targeted, time-limited interventions, triggered by clinical contact for injury care, may prove to be an efficient way in which deliver effective injury prevention to a here-to-fore unrecognized high-risk population. This strategy could be used to complement existing clinic-based injury prevention activities that target the population as a whole.
Dr Johnston was supported by the Robert Wood Johnson Clinical Scholars Program.
We thank Rich Boyesen, from the Medical Assistance Administration, Department of Social and Health Services, State of Washington, for his help in procuring and formatting claims data; and Allen Cheadle, PhD, for his assistance in data cleaning and initial data manipulation.
- Received December 15, 1998.
- Accepted June 3, 1999.
Reprint requests to (B.D.J.) Harborview Injury Prevention and Research Center, Box 359960, 325 9th Ave, Seattle, WA 91804-2499. E-mail:
The views expressed herein are those of the authors and do not necessarily represent the views of the Robert Wood Johnson Foundation.
- Rivara FP,
- Grossman DC
- Bass JL,
- Christoffel KK,
- Widome M,
- et al.
- Braddock M,
- Lapidus G,
- Gregorio D,
- Kapp M,
- Banco L
- Nersesian WS,
- Petit MR,
- Shaper R,
- Lemieux D,
- Naor E
- Overpeck MD,
- Jones DH,
- Trumble AC,
- Scheidt PC,
- Bijur PE
- Christoffel KK,
- Donovan M,
- Schofer J,
- Wills K,
- Lavigne JV
- ↵Nyman G. Infant temperament, childhood accidents, and hospitalization. Clin Pediatr (Phila). 1987;26:398–404
- Beautrais AL,
- Fergusson DM,
- Shannon FT
- Padilla ER,
- Rohsenow DJ,
- Bergman AB
- Sibert R
- Bijur PE,
- Golding J,
- Haslum M
- Eminson CJ,
- Jones H,
- Goldacre M
- Langley JD
- ↵Stata Corp. Stata Reference Manual: Release 6.0, III. College Station, TX: Stata Press; 1999:415–416
- Martikainen P,
- Valkonen T
- Husband P,
- Hinton P
- Kendrick D,
- Marsh P
- Copyright © 2000 American Academy of Pediatrics