Objectives. To estimate the effect of early childhood abuse (ie, inflicted injury) on medical costs of head trauma.
Methods. Abstracts of patient records were drawn from the annual 1993–2000 Colorado state-mandated hospital discharge database maintained by the Colorado Hospital Association. The 2 dependent variables were total charges (TC) and length of stay. Our key independent variable was the nature of injury, ie, inflicted or unintentional; other independent variables were age, severity level, death, and trauma designation of the hospital. Comparisons of variables between patients with inflicted and unintentional head trauma were performed using Student’s t tests or χ2 statistics. Ordinary least squares regression was used to estimate the marginal and total effects of inflicted injury on TC and LOS.
Results. Of the 1097 head trauma patients <3 years old, 814 had unintentional and 283 had inflicted head trauma. Head trauma was defined using the Centers for Disease Control definition of traumatic brain injury. Patients with inflicted injuries were younger and had a higher average severity level and overall mortality rate than did patients with unintentional head trauma. The regression models showed that, controlling for age and severity, patients with inflicted head trauma stayed in the hospital 52% longer (2 days), and had a mean total bill 89% higher ($4232 more) than did patients with unintentional head trauma.
Conclusions. The findings from multivariate models of TC and length of stay corroborate the simpler univariate findings of earlier studies. By focusing on the impact of those cases of child abuse that lead to a specific, severe clinical entity (traumatic brain injury), we isolated a significant economic impact of abuse on health care expenditures for injury.
- child abuse
- craniocerebral trauma
- traumatic brain injury
- direct service costs
- health care costs
- health expenditures
- head injuries
- hospital costs
- length of stay
- severity of illness
- nervous system
In 1999, there were an estimated 826 000 victims of substantiated child maltreatment known to child protective services agencies nationwide, or 11.8 victims per 1000 children. Of these, ∼1100 children died, of whom 77% were aged 3 years or younger.1 The most common cause of death from physical abuse is head trauma,2 the most severe forms of which occur in the first 3 years of life.3 More than 95% of serious intracranial injuries during the first year of life and 85% during the first 2 years are the result of abuse.4,5 Approximately 100 000 to 200 000 pediatric head injuries occur per year, with a rate of 193 to 367 per 100 000 children.6
We use the term inflicted to refer to “nonaccidental” injuries or injuries that result from maltreatment and unintentional to refer to “nonabusive” injuries. Beginning with the classic clinical descriptions of Kempe7 and the subsequent literature on child maltreatment, the identifying characteristics of intentional pediatric injuries, and of head injury in particular, have become widely recognized.8
Although recent research compares clinical attributes of inflicted and unintentional head traumas,2,9,10 few systematic studies have examined the differential impact of inflicted versus unintentional etiology of head injury on resource utilization and system outcomes, such as length of stay (LOS) and medical expenditures. Four previous studies have compared resource utilization of inflicted and unintentional injuries in children. A severity-adjusted cost comparison of 13 abused pediatric intensive care unit (PICU) patients and 924 nonabused PICU patients found higher mean daily charges ($5289 vs $1913) for the child abuse cases compared with other PICU patients, but no significant difference in LOS.11 A comparison of Pennsylvania inpatients in 1995, 348 with maltreatment and 1052 with unintentional injuries, included children 0 to 19 years old (45% were older than 4 years) and all forms of maltreatment (22% of injuries were “self-inflicted”).12 Mean and median total charges (TC) between the 2 groups were not significantly different. A retrospective review of injured children 0 to 4 years old in the National Pediatric Trauma Registry (1997 abused; 16 831 unintentional) found a significantly longer LOS among those with confirmed abuse (9.3 vs 3.8 days), although the report did not examine costs.13 A study of 3472 consecutive trauma hospitalizations for children aged 0 to 14 years found that 76 abused patients had a longer LOS (14.4 days vs 6.8 days) and higher mean charges ($20 360 vs $7791) than did nonabused children.14
The purpose of this effort is to build on prior research by comparing inflicted and unintentional injuries in a more homogeneous study group: our patients had more similar pathophysiology (traumatic brain injuries [TBIs] only) and a narrower age range (0 to 36 months). We also advance the literature by conducting multivariate analyses adjusted for client and hospital factors.
This paper reports estimates of the marginal, ie, incremental, effects of intentional injury on costs and hospital LOS for TBI. This allows an understanding of the unique contribution of abuse to the treatment cost of a given type of injury—TBI—and provides an empirical basis for determining what should be spent on the prevention of abuse based on a narrow definition of direct medical expenditures.
Abstracts of patient records were drawn from the annual 1993–2000 Colorado state-mandated hospital discharge database maintained by the Colorado Hospital Association (CHA). This administrative database provides comprehensive, uniform, case-based data for each anonymous inpatient hospitalized in a nonmilitary, acute-care hospital in the state. Each discharge abstract included demographic information, hospitalization dates, up to 25 International Classification of Diseases Coding Manual, Volume 9 (ICD-9) diagnosis codes, TC, and an All Patient-Refined Diagnosis-Related Groups (APR-DRG) severity index (based on 3Ms APR-DRG Version 12.0; the software is a patient classification system that uses hospital patient discharge data and computer-based logic to assign patients to severity of illness and risk of mortality classes so they can be accurately compared in terms of length of stay, resource consumption, and outcomes).15 The study population for this paper consisted of patients <3 years of age who had an admission with TBI, using the Centers for Disease Control and Prevention’s ICD-9-based definition. The Centers for Disease Control and Prevention defines TBI by ICD-9 codes 800.0–801.9 (fracture of the vault or base of the skull), 803.0–804.9 (other and unqualified and multiple fractures of the skull), and 850.0–854.116 Cases with missing values were listwise deleted from the multivariate analyses.
Drawn from the CHA data, the 2 main dependent variables were TC and LOS for inpatient admissions under study. Total hospital charges were reported by the member hospitals and rounded to the nearest dollar. TC included lab, radiology, pharmacy, routine, and other charges. TC was probably higher than actual costs because of reimbursement rates being lower than submitted charges, but varied across cases such that this variable was likely to be an unbiased estimate of costs. TC entered the model after being logarithm-transformed attributable to positive skewness of the data. LOS was calculated using the discharge date minus the admission date and was set to one if the difference was 0 days. All cases were hospital admissions.
The key independent variable was “inflicted,” an indicator variable designating cases that had any of the 25 diagnosis fields with ICD-9 codes 995.5 (child maltreatment syndrome), E967 × (child battering and other maltreatment), or E968 × (assault by other and unspecified means, including nonaccidental violence). Our variable definition for “inflicted” injuries does not include ICD-9 codes for self-inflicted injury.
Age in months was calculated from the admission date minus the birth date. The age variable entered the model after being logarithm-transformed to account for a nonlinear relationship with the dependent variables.
The APR-DRG severity of illness designation was used to measure each patient’s clinical acuity. The APR-DRG classification system is a commercially designed patient classification scheme that extends the hospital DRG system into severity of illness classifications based on complications and comorbidities (CC) for a given disease category.17 We separated “severity” into 4 indicator variables, ranging from the value 1 if a case was given a severity level of 1 (no/minor CC) to 4 (extreme CC). Published research has found that dependent variables like those used here (TC and LOS) were more precisely estimated with refined severity grouping like the APR-DRG than with unrefined diagnostic categories.18 Forty of the 1097 (3.6%) patients had missing severity level data; 36 of these were cases of inflicted injury. In cases like these, methods such as the expectation maximization (EM) algorithm could be used to impute missing data to avoid listwise deletion of cases.19 We used SPSS (SPSS, Chicago, IL) to implement the EM algorithm and estimated missing severity levels so these clients could be included in the multivariate analyses.20 Findings were robust to the imputation or deletion. We reported models with listwise deletion because they were the more conservative estimates, but note that when imputation allowed for the inclusion of the former missing values, the total effects of inflicted injury on LOS and TC increased.
We included a variable indicating whether the patient’s treating hospital was designated as a pediatric trauma center, specifically, a trauma center with pediatric commitment or a regional pediatric trauma center. Such a designation implies a number of characteristics: the presence of an emergency department and PICU equipped and staffed to care for pediatric trauma patients; a pediatric trauma program including pediatric surgeons, orthopedists, neurosurgeons, pediatric intensivists, emergency physicians, radiologists, nurses, and a trauma coordinator; and an affiliation with a pediatric rehabilitation service.21 This variable was to account for hospital characteristics that could have affected resource utilization, independent of client characteristics, because children in this study were seen in hospitals across the state.
Indicator variables for each year marked the year of discharge to control for fixed effects that were not measured but are correlated with time; 1993 was omitted as the reference year. The “death” variable indicated whether the patient’s discharge status was “expired.” Ordinary least squares regression was used to estimate the effect of abuse and other variables on LOS and TC. Analyses were conducted using SAS 8.01 (SAS Institute, Cary, NC). The study was reviewed by the Institutional Review Board of the University of Colorado Health Sciences Center and granted an exemption based on anonymity of subjects.
Of the 1097 patients <3 years old, 814 had unintentional and 283 had inflicted head trauma. Table 1 provides descriptive statistics comparing inflicted and unintentional cases. In these univariate comparisons, there were significant differences (P < .05) between groups, with longer LOS and higher TC for inflicted TBI patients. Patients with inflicted TBI had a lower mean age than patients with unintentional head trauma. (P < .05). Average severity level and overall mortality rate were higher among patients with inflicted injuries as well (P < .05).
Other variables of interest were gender, race/ethnicity, and insurance type. There was a slight male preponderance in both groups, but not a significant one. The distribution of race/ethnicity was not significantly different between the groups but type of insurance was, with significantly more children with Medicaid insurance in the inflicted injury group (P < .05). However, none of these entered significantly in the multivariate models and were not included in the final estimations.
Model selection was conducted with forward selection by comparing nested models and inspection of “inflicted” as a confounder through interactions. Some variables were included based on theory (such as “inflicted,” age, and severity), and some retained if significant (hospital characteristic trauma designation).
Results of regression analyses of LOS and TC are provided in Table 2. The main effect variable, “inflicted,” was significantly and independently related to both LOS (P < .05) and TC (P < .01). The “inflicted” factor entered the models in multiple variables via interaction terms with age and severity levels. To calculate the total unique effect of inflicted injury on LOS and TC, we evaluated the parameter values at the sample means, resulting in the values reported at the end of the table. Inflicted cases were, on average, 52% higher in LOS and 89% higher in TC compared with unintentional traumas, controlling for age and severity. Thus, inflicted injury had significant and positive independent effects on both LOS and TC, controlling for the variables that should explain resource utilization (eg, age and severity).
The age variable and age interacted with “inflicted” were statistically significant. The interaction was included to address the significantly younger ages of the children whose injuries were inflicted. Figure 1 shows the age distribution for patients with inflicted and unintentional injuries.
To allow nonlinear relationships between severity levels and the dependent variables, the 3 highest severity levels entered the models independently. In both the LOS and TC models, the effects of the severity levels were positive and significant (P < .01), with parameter values increasing from severity levels 2 to 4. The models indicated significant relationships between severity level and the “inflicted” variable. Figure 2 displays the percentage of inflicted and unintentional cases by severity level. The distribution of unintentional injuries peaked at the lowest severity level (67% of cases) and declined quickly. In contrast, inflicted injuries were most prevalent at severity level 2 (representing moderate CC) with 36% of cases. Severity level 3 (major CC) represented 30% of cases, and 20% were classified as extreme CC.
Figure 3 displays how severity level and nature of injury interact with mean LOS. For inflicted injuries, LOS was higher at severity levels 1 to 3. For severity level 4, this relationship was reversed. This distribution is not altered by eliminating deaths from the analysis. Figure 4 displays median costs for each severity level, which echoes the LOS relationship.
The variable for mortality was negative and significant in both models, indicating that injured children who survived had significantly higher treatment days and dollars. Cases treated at hospitals that had pediatric trauma designations showed significantly higher LOS and TC. In a separate analysis, an additional variable indicating the number of rehabilitation services in the child’s county of residence had the opposite effect, lowering days and dollars significantly. (The subset of children for whom the county of residence variables was not available, and thus were excluded from the analysis including rehabilitation services, represented significantly higher proportions of children with inflicted injuries. Other model-variable effects were robust to this specification in direction but not magnitude. This model was not reported, but tables are available from the authors.
Inflicted head injuries account for a substantial portion of early childhood mortality. A number of studies have explored the clinical difference between inflicted and unintentional head injury and, as discussed above, a few included analyses, albeit simplistic ones, of the financial differences.2,12–15 Our study corroborated the general clinical findings of most earlier reports: children with inflicted injuries had a lower mean age2,12–15 and higher mortality rates2,12,14,15 and severity levels12,14,15 than did the patients with unintentional head injuries. We found the male predominance that Reece and Sege2 and DiScala et al13 did among the group with unintentional injuries, and the male predominance that DiScala et al13 (but not Reece and Sege2) showed among the group with inflicted injuries. We found no racial differences between the 2 groups, unlike Forjouh.12 We detected a higher percentage of Medicaid insurance among children with inflicted injuries, as did Forjouh.12
Overall, patients with inflicted head trauma, TBI, stayed in the hospital 52% longer (2 days) and had a mean total bill 89% higher ($4232 more) than did patients with unintentional head trauma. These findings corroborate those of Irazuzta et al11 and Peclet et al.14 Our study strengthens these findings because of our large sample size, multivariate analysis with controls for age, severity, and other factors, and the greater similarity in pathophysiology (TBI) between our inflicted and unintentional trauma comparison groups.
One limitation of this study is that children who were “dead on arrival” were excluded from the sample because Colorado prehospital providers are not required to bring children into the hospital to pronounce a time of death. Consequently, the most severe head injuries, ie, those resulting in death, were excluded from the hospital admissions we analyzed. One signal that this happened was that among the highest severity (APR-DRG severity level 4) patients, those with inflicted TBI had shorter LOS than did those with unintentional TBI. The one study we reviewed that included children who were “dead on arrival” in their analyses found the costs and mortality rates were uniformly higher for victims of inflicted injuries than for all other injuries in children.15 Another possible explanation is that for survivors of unintentional injury who are very severe, families may call for heroic measures that drive up resource utilization.
An important limitation that our study shares with earlier research is the focus on inpatient, direct costs. The costs of an illness or injury can be thought of as direct and indirect. Direct costs are those that accrue to a person or systems in the course of care, and indirect costs are those resulting from the illness or injury but that are outside the scope of actual treatment. Examples of direct costs would be physician and hospital care expenditures. Examples of indirect costs would be reduced earning and lowered productivity attributable to resultant job limitations for the injured person, or from a household perspective, parents’ lost wages from time spent caring for an injured child. Such issues have been considered, but not estimated, in past literature on pediatric22–26 and adult27–31 head injuries. To begin to determine the economic costs of childhood physical abuse, estimates of the marginal costs of abuse are needed to understand the unique contribution of the nature of an injury to its costs.
The fact that we found considerable marginal costs of inflicted trauma despite this narrow focus suggests that, other variables notwithstanding, the abused child requires more medical resources. It would be reasonable to expect that patients with inflicted head injuries would require greater, and hence, more costly, ongoing psychosocial resources.32–36 To estimate total direct costs attributable to inflicted injury would require a longitudinal, multifaceted analysis including these elements; such implications cannot be made from cross-sectional designs as were used here.
Another limitation of this study was our definition of inflicted injuries. The reliance on coder-generated diagnosis data was less than ideal because we almost surely missed cases of inflicted injury, as do physicians everyday. The miscategorization of inflicted injuries as unintentional worked to strengthen our findings because it biased our results against the hypothesis that inflicted injuries were more costly in resource use.
Such a large and significant difference in resource utilization, despite the limit to acute, inpatient, and direct medical costs underscores 2 important points. First, there is additional morbidity inherent in the intentionality of inflicted head trauma. This phenomenon could be thought of as the residual effect of severity, other things equal, and captured in the nature of abuse variable. This notion is consonant with an assumption that recovery and rehabilitation can be negatively affected by the inability or unwillingness of caregivers to provide consistent and loving care to survivors of abusive or neglectful trauma. Also, studies of abused children confirm that they are more likely to have a history of preinjury medical problems.14,37 This may be related to poor prehospitalization nutritional and health status, or it may be related to the physiologic impact of psychosocial stress. Extensive literature has related medical costs to psychological stress; several reports focus specifically on posttraumatic costs of child abuse.29,30,33,38 The greater marginal cost among abused patients may also be related to prolonged LOS awaiting social services evaluation and placement.
Second, inflicted head trauma is costly to our society in medical resource use, and by implication, in human terms.39 Both the financial and human costs highlight the importance of increased funding to develop and implement child abuse prevention programs.40–42 Numerous studies have shown that preventive strategies do help reduce child abuse.3 Despite this, funding for research abuse through the National Center on Child Abuse and Neglect, now the Office on Child Abuse and Neglect, remained flat throughout the study period.45,46 Leaders in the field of child abuse decry that 40 years after Kempe’s landmark article,8 public dialogue about child maltreatment is largely limited to the occasional high-profile case.45 The medical and psychological toll of child abuse has not focused community resources on the issue; perhaps the economic toll of child abuse will.
This work was supported by William T. Grant Foundation Faculty Scholars Award (to Dr Libby).
We thank American Indian and Alaskan Native Programs Writing seminar participants; Donald Bross, JD, PhD, from the Kempe Children’s Center; and Jill G. Joseph, MD, PhD, from the Children’s National Medical Center for helpful reviews.
- Received July 1, 2002.
- Accepted November 12, 2002.
- Address correspondence to Anne M. Libby, PhD, Departments of Psychiatry and Pediatrics, University of Colorado Health Sciences Center, Box 6508, Mail Stop F800, Aurora, CO 80045. E-mail:
- ↵US Department of Heath and Human Services, Administration on Children, Youth and Families. Child Maltreatment 1999. Washington, DC: US Government Printing Office; 2001
- ↵Johnson CF. Child abuse. In: Behrman RE, Kliegman RM, Jenson HB, eds. Nelson’s Textbook of Pediatrics. 16th ed. Philadelphia, PA: WB Saunders Company; 2000:113
- ↵Berkowitz CD. Pediatric abuse. Emerg Clin North Am.1995;13 :321
- ↵Gedeit R. Head injury. Pediatr Rev.2001;22 :118– 124
- ↵Billmire ME, Myers PA. Serious head injury in infants: accident or abuse? Pediatrics.1985;75 :340– 342
- ↵Duhaime AC, Alario AJ, Lewander WJ, et al. Head injury in very young children: mechanisms, injury types, and ophthalmologic findings in 100 hospitalized patients younger than 2 years of age. Pediatrics.1992;90 :179– 185
- ↵Thurman DJ, Sniezek JE, Johnson D, et al. Guidelines for Surveillance of Central Nervous System Injury. Atlanta, GA: Centers for Disease Control and Prevention; 1995
- ↵Norusis MJ. SPSS 9.0 for Windows User’s Guide. Chicago, IL: SPSS, Inc; 1999
- ↵State of Colorado Board of Health. Rules and regulations pertaining to the statewide trauma system; designation of trauma facilities. In: Statewide Trauma System Rules. State of Colorado, Colorado Revised Statutes Title 25, Article 3.5 Sections 703(8,9), 704(f); 1999:30. Available at: www.cdphe.state.co.us/tp/tphom.html
- ↵National Pediatric Trauma Registry. Children and Adolescents With Disability Due to Traumatic Injury: A Data Book. Boston, MA: New England Medical Center; 1996:3–7
- ↵Runge JW. The cost of injury. Adv Trauma.1993;11 :241– 253
- ↵Summers CL, Molyneaux EM. Suspected child abuse: cost in medical time and finance. Arch Dis Child.1992;67 :905– 910
- ↵Prevent Child Abuse America. Total estimated cost of child abuse and neglect in the United States. 2001. Available at: http://www.preventchildabuse.org/research_ctr/cost_analysis.pdf. Accessed February 13, 2002
- ↵Olds DL, Henderson CR, Chamberlin R, Tatelbaum R. Preventing child abuse and neglect: a randomized trial of nurse home visitation. Pediatrics.1986;78 :65– 78
- Olds DL, Henderson CR, Kitzman H, Cole R. Effects of prenatal and infancy nurse home visitation on surveillance of child maltreatment. Pediatrics.1995;95 :365– 372
- ↵Krugman RD. Unimaginable images: seeing is believing. Pediatrics.1997;100 :890
- Copyright © 2003 by the American Academy of Pediatrics