Sign In
Individual Sign In
Create an Account
Institutional Sign In
OpenAthens Shibboleth
March 2002

Intracranial Hemorrhage in Children Younger Than 3 YearsPrediction of Intent

Author Affiliations

From the Radiology Department, Children's Hospital of Wisconsin, Milwaukee (Dr Wells); and the Division of Biostatistics, Medical College of Wisconsin, Milwaukee (Dr Laud). Dr Vetter is in private practice in Singapore.


Copyright 2002 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2002

Arch Pediatr Adolesc Med. 2002;156(3):252-257. doi:10.1001/archpedi.156.3.252

Objective  To determine whether certain computed tomographic imaging patterns in infants and young children with intracranial hemorrhage help predict intentional compared with unintentional injuries.

Design  Retrospective consecutive case series over a 10-year period.

Patients  Two hundred ninety-three children younger than 3 years with intracranial hemorrhage.

Main Outcome Measures  The sensitivity and specificity of computed tomographic imaging patterns for intentional head injury.

Setting  Regional pediatric medical center.

Results  Four variables used in the multiple logistic regression analysis for predicting intentional head injury were statistically significant (P<.05): subdural hematoma located over the cerebral convexities, hematoma within the interhemispheric subdural space, hygroma (nonhemic subdural fluid) with intracranial hemorrhage, and absence of a skull fracture with intracranial hemorrhage. The prediction model for the diagnosis of intentional head trauma using combinations of these 4 variables and a .45 probability cutoff point indicated a sensitivity of 84% (95% confidence interval, 78%-90%) and a specificity of 83% (95% confidence interval, 74%-89%).

Conclusion  Computed tomographic imaging patterns of intracranial hemorrhage in children younger than 3 years help predict whether the injury was intentional.

THE DIFFERENTIATION between intentional and unintentional causes of head trauma in infants and young children is a common medical and legal dilemma. Most head trauma is caused by falls that rarely result in a significant intracranial pathologic condition; however, intentional injury is the most common cause of severe traumatic brain injury in infants.1,2 Billmire and Myers3 reported that 64% of infant head injuries of sufficient severity to warrant hospital admission (excluding uncomplicated skull fractures) and 95% of serious intracranial injuries are the result of child abuse. Accurate correlation of the computed tomographic (CT) findings with the reported mechanism of injury is an essential component of the care of the infant or young child with head trauma.

Results of studies based on surgical, radiological, and autopsy data suggest that different types of brain injuries tend to occur with intentional vs unintentional trauma. Most investigators believe this is caused by a predominance of inertial forces with intentional brain injury, resulting in movement of the brain to yield concussion, edema, subdural hematoma, and diffuse axonal injury. Unintentional head trauma more often involves contact forces, which tend to produce focal injuries such as laceration, fracture, contusion, and epidural hematoma.46

We retrospectively reviewed all intracranial hemorrhages detected with CT in children younger than 3 years at our institution over a 10-year period. The main objective was to determine which CT imaging patterns help predict whether the injury was intentional.


We reviewed the diagnostic imaging reports for all patients younger than 3 years who underwent cranial CT examinations at Children's Hospital of Wisconsin, Milwaukee, between February 1, 1991, and May 31, 2001. We selected all patients with radiological studies that demonstrated intracranial hemorrhage. We reviewed (blinded to the CT findings) the medical records of these children and recorded the patient's age and sex, the reported circumstances of the injury, the presence or absence of concurrent injuries, and the results of official child abuse investigations. We excluded children with hemorrhage because of prematurity, birth trauma, medical conditions, or intracranial surgical procedures. Each of the remaining patients was assigned to 1 of 3 mutually exclusive causative categories as follows: intentional injury, unintentional injury, and intent uncertain. We considered an injury to be unintentional if it was witnessed by someone other than the caretaker or there were no discrepancies between the described mechanism and the physical findings. We additionally categorized the unintentional injury group according to whether the injury was caused by a fall, motor vehicle collision, a fall in an infant walker, or by being struck by an object. We considered an injury to be intentional if there was a confession of abuse, the injuries were incompatible with the stated mechanism of injury, or the caretaker offered no explanation for the injuries. Those patients who could not clearly be assigned to the intentional or unintentional injury categories were assigned to the intent uncertain group.


We reviewed (blinded to the clinical findings) the CT images of the study subjects and recorded the location of intracranial hemorrhage according to the following anatomical categories: epidural, subdural, subarachnoid, intraventricular, and parenchymal. We recorded additional details of the intracranial hemorrhage according to the following subcategories: subdural hematoma located over the cerebral convexities, subdural hematoma located within the interhemispheric fissure, subdural hematoma in other locations, petechial parenchymal hemorrhage, and parenchymal hematoma. All study patients had at least one of these findings. In addition, the presence of subdural hygroma, brain edema, and skull fracture was recorded. For the purposes of categorization, we considered any subdural fluid collection with CT attenuation values less than those of the brain to be a hygroma. The only exceptions were hyperacute subdural hematomas (a mixture of clotted and unclotted blood), which were categorized as subdural hematomas only. Chronic subdural hematomas were included in the hygroma category. The CT imaging categories were not mutually exclusive.


We selected the following variables to test for an association with intentional trauma as the cause of intracranial hemorrhage: lack of an epidural hematoma, interhemispheric subdural hematoma, convexity subdural hematoma, lack of subarachnoid hemorrhage, intraventricular hemorrhage, lack of petechial hemorrhage, parenchymal hematoma, hygroma, edema, lack of a fracture, male sex, and patient age. We treated age as a continuous variable and all others as binary (yes/no) variables. Logistic regression was used to create a prediction rule; variables were studied individually and simultaneously. Our final prediction model included variables that, in the same model, had P values less than .05. We used this model to estimate probabilities of intentional injury based on combinations of these variables. The sensitivity and specificity for the diagnosis of intentional trauma were calculated. We then applied the model to the patients in the intent uncertain category. All calculations were performed using the LogXact software package (CYTEL Software Corp, Cambridge, Mass), which uses exact permutational distributions as opposed to the more commonly available asymptotic likelihood methods.


The patient characteristics are summarized in Table 1. The study population consisted of 293 individuals, 177 male (60.4%) and 116 female (39.6%) patients. The mechanism of injury was intentional trauma in 148 patients (50.5%), unintentional trauma in 109 (37.2%), and intent uncertain in 36 (12.3%). Comparing the children having intentional vs unintentional injuries, the former tended to be younger (mean age, 8.0 months vs 11.7 months, respectively) and were more frequently male (60.8% vs 55.0%, respectively).

Table 1. 
Image not available
Patient Characteristics and Causes of Intracranial Hemorrhage

Four hundred sixty-nine intracranial hemorrhages classified by location were identified in the 293 study subjects. The imaging findings are correlated with the causes and are listed in Table 2. The most common site of hemorrhage was in the subdural spaces, identified in 184 patients (63%) (Figure 1). The most common type of subdural hemorrhage was in the interhemispheric region, which occurred in 143 children (49%). Most (105 patients [73%]) of these children had intentional injuries. Unintentional trauma was responsible for 21 (15%) of the interhemispheric subdural hematomas; all were injuries associated with relatively high force—3 falls of more than 2 m, 12 motor vehicle collisions, 4 falls while in an infant walker, and 2 blows to the head.

Table 2. 
Image not available
Computed Tomographic Imaging Findings Correlated to the Causative Categories of Injury*
Figure 1.
Image not available

Common computed tomographic (CT) findings with intentional trauma. An 18-month-old boy was shaken and thrown to the floor by his mother's boyfriend 3 hours prior to the CT examination. Subsequent CT imaging (not shown) demonstrated worsening edema over the next few days. The child died 5 days later. A, Note the acute convexity subdural hemorrhage adjacent to the right frontal lobe and along the tentorium (arrows). B, A more superior CT image shows a right posterior interhemispheric subdural hematoma (arrow).

Hemorrhage over the cerebral convexities was almost as common as interhemispheric hemorrhage, occurring in 137 children (47%). Convexity and interhemispheric subdural hematomas occurred simultaneously in 96 (52%) of the 184 patients with subdural hematoma. Intentional injury was the most common cause of convexity subdural hematoma, occurring in 99 patients (72%). Unintentional injuries accounted for only 14 (10%) of the convexity subdural hematomas; most were caused by severe injuries, such as motor vehicle collisions, falls from more than 2 m, or infant walker–stair injuries. Only 3 patients with this pattern of hemorrhage fell less than 2 m.

Acute intracranial hemorrhage with coexistent low attenuation subdural hygromas occurred in 43 patients (15% of the study population). Associated hemorrhages included 1 epidural hemorrhage, 29 interhemispheric subdural hematomas, 37 convexity subdural hematomas, 1 subarachnoid hemorrhage, 1 petechial hemorrhage, and 1 intraventricular hemorrhage. All patients with hygromas were in the intentional injury or intent uncertain categories.

Epidural hematomas occurred in 80 patients (27%). Epidural hematomas were most frequently associated with unintentional injuries (62 patients [78%]) and were uncommon with intentional injury (14 patients [18%]). Thirty-nine epidural hematomas (63%) related to unintentional injury were caused by a fall. Most of the epidural hematomas were small, and not associated with other injuries; 69 (86%) were less than 1 cm thick and were isolated (Figure 2).

Figure 2.
Image not available

Common computed tomographic (CT) appearance of unintentional head trauma. A small (probably venous) epidural hematoma is present on the right side (arrow) in a 16-day-old boy who had fallen from his mother's arms onto a concrete floor. Bone windows (not shown) demonstrated a skull fracture. This hematoma resolved without surgical evacuation, and the child made a complete clinical recovery.

Subarachnoid hemorrhage occurred in 36 patients (12%); a coexisting intracranial injury (edema or other area of hemorrhage) was identified in 27 (75%) of these 36 patients. Unintentional injuries were more common than intentional injuries in patients with subarachnoid hemorrhage (61% vs 33%). Nine (41%) of the 22 patients with subarachnoid hemorrhage who had unintentional injuries had fallen; 10 (45%) were involved in motor vehicle collisions.

Petechial parenchymal hemorrhage was identified in 45 (15%) of the 293 patients. Twenty-seven patients (60%) with petechial hemorrhage had unintentional trauma and 15 (33%) had intentional trauma. Thirty-two patients (71%) with petechial hemorrhage had a skull fracture.

Edema accompanied intracranial hemorrhage in 85 patients (29%). In 80 patients (94%) the edema occurred in conjunction with an interhemispheric subdural hematoma or a convexity subdural hematoma. Sixty-six patients (78%) with edema were classified in the intentional injury category.

A skull fracture was identified in 140 patients (48%). Seventy-nine (99%) of the 80 patients with an epidural hematoma had a skull fracture. Forty-three (29%) of the 148 patients having a diagnosis of intentional trauma had a skull fracture.


The logistic regression analysis results are given in Table 3. Evaluation of the variables individually showed P values less than .05 for all variables except intraventricular hemorrhage, parenchymal hematoma, and male sex. However, in multiple regression, only 4 variables were statistically significant: convexity subdural hematoma, interhemispheric subdural hematoma, lack of a skull fracture, and hygroma. The final model is summarized in Table 4. Note that the hygroma and no fracture variables include the requirement for coexistent acute intracranial hemorrhage. The model is based on the imaging findings at the time of clinical presentation; therefore, patients with intracranial hemorrhage who subsequently develop hygromas should not be considered to have a hygroma when applying the prediction rule.

Table 3. 
Image not available
Logistic Regression Analysis of Variables Studied Individually for a Diagnosis of Intentional Trauma
Table 4. 
Image not available
Multiple Logistic Regression Analysis for a Diagnosis of Intentional Trauma

The model-based estimates of the probabilities of intentional trauma according to combinations of the 4 selected variables are listed in Table 5. To maximize both sensitivity and specificity, we classified patients as having intentional injuries when their predicted probability of intentional injury in the logistic model was greater than .45. Using this cutoff point, the sensitivity of the prediction model for the diagnosis of intentional trauma is 84% (95% confidence interval, 78%-90%) and the specificity is 83% (95% confidence interval, 74%-89%). The cutoff point was selected to provide a reasonable balance between the sensitivity and specificity. Application of the prediction rule to the patients in the intent uncertain category estimates that 25 (69%) of the 36 patients had intentional trauma.

Table 5. 
Image not available
Model-Based Estimates of Probabilities of Intentional Trauma*

We propose a prediction rule for the diagnosis of intentional trauma in children younger than 3 years with intracranial hemorrhage. The model requires the exclusion on clinical grounds of patients with hemorrhage due to prematurity, birth trauma, surgery, or nontraumatic medical causes. No other clinical information is assumed. We thus attempted to reproduce the common clinical scenario at the time of presentation of the infant or young child with head trauma, when the available objective clinical information is incomplete and the reported history of the injury may be viewed with some skepticism.

The 4 variables that exhibited statistical significance in multiple logistic regression are compatible with the work of other investigators and theoretical considerations. The common occurrence of subdural hemorrhages and lack of skull fractures support the concept that acceleration-deceleration forces predominate with intentional head injuries. Unintentional injuries more often involve translational forces, thereby favoring skull fractures and epidural hematomas. The association of hygroma with the diagnosis of intentional trauma supports the medical dictum that acute intracranial hemorrhage in combination with nonhemic subdural fluid (eg, acute and chronic subdural hematomas) indicates that at least 2 significant episodes of trauma have occurred, unless there is a clear history of a preexisting medical condition that resulted in a subdural effusion or hygroma.

To our knowledge, this article reports the largest case series in the medical literature to investigate the correlation between CT imaging findings and causes of intracranial hemorrhage in infants and young children. There are, however, several limitations of this study. A validation sample to test the predication rule is lacking. The outcome event, intentional trauma, is subject to variations in medical, legal, and societal definitions. The creation of the prediction rule assumes that the predictive findings were defined properly and accurately assessed, that is, the CT images were properly interpreted. Clinical usefulness of the model requires others using the prediction rule to use the same definitions of the significant variables.

The separation of the assessment of outcome and clinical prediction is imperfect in this study. We were blinded to the CT imaging findings while reviewing the medical records and assigning a diagnosis (the outcome). However, the clinicians and social service professionals who investigated the circumstances of the injuries were aware of, and likely influenced by, the CT imaging findings. This potential error is mitigated by the use of objective factors to assure a correct diagnosis of intentional injury, such as a confession by the perpetrator, or the presence of suspicious skeletal injuries, retinal hemorrhages, or soft tissue injuries. Other limitations are (1) the lack of uniformity in the veracity of investigation among the subjects in this retrospective study, and (2) the potential influence of the CT findings on the clinical decision regarding initiation of a formal child abuse investigation.

Specific confirmatory information in the medical literature for our findings is lacking; we could find no reports proposing a prediction rule for neuroimaging findings in the diagnosis of intentional injury. Several studies evaluating head trauma in abused children have shown a high incidence of subdural hematoma.79 Zimmerman et al9 first reported the important correlation between child abuse and subdural hemorrhage in the interhemispheric space; this has been confirmed by other investigators.10 There is wide variation in the reported incidence of chronic nonhemic subdural fluid collections with intentional head trauma, although the correlation is widely accepted.1113

There are a few series in the medical literature that compare head injuries in children having intentional vs unintentional trauma that include statistical analysis of the results. Ewing-Cobbs et al14 evaluated 60 patients between 0 and 6 years of age who had brain injuries and found significantly (P<.05) more subdural hematomas and hygromas in the patients with intentional injuries than in those with unintentional injuries. Duhaime et al15 studied 100 children 24 months of age and younger who were admitted to the hospital with head trauma and found a statistically significant (P<.001) relationship between intentional injury and the presence of intradural (subdural, subarachnoid, contusion) hemorrhage. Hymel et al10 found interhemispheric hemorrhage, subdural hemorrhage, "large (nonacute) extra-axial fluid," and basal ganglia edema to occur significantly more frequently (P≤.05) in 39 children having a diagnosis of intentional trauma than in the control group of 39 patients with unintentional injuries. Reece and Sege16 evaluated 195 children younger than 3 years who had head trauma and found that hemorrhage in the subdural and subarachnoid spaces occurred with a significantly (P = .001) higher incidence in patients with intentional injuries.

Application of the prediction rule to the study patients in the intent uncertain group estimates that 69% of these children had had intentional injuries. This is in keeping with the findings of other investigators that child abuse is significantly underdiagnosed.17


Intentional trauma is the most common cause of intracranial hemorrhage in children younger than 3 years. The subdural space is the most common location of intracranial hemorrhage in children of this age group. Although all severe head injuries in infants and young children must be viewed with a high index of suspicion for intentional trauma, the CT imaging findings that are most useful in differentiating intentional from unintentional causes in those children with intracranial hemorrhage are convexity subdural hematoma, interhemispheric subdural hematoma, nonhemic subdural fluid (hygroma), and lack of a skull fracture.

Back to top
Article Information

Accepted for publication October 22, 2001.

What This Study Adds

The neuroimaging findings for infants and young children with traumatic intracranial hemorrhage provide important clues to the mechanism of the injury. Several studies of children having intentional trauma have shown a high incidence of subdural hematoma; however, the sensitivity and specificity of this and other specific patterns of intracranial injury for the diagnosis of intentional trauma is unknown. In this study, we evaluated the CT imaging findings in 293 children younger than 3 years who had intracranial hemorrhage. We propose a prediction model for the diagnosis of intentional trauma based on 4 CT imaging patterns.

Corresponding author and reprints: Robert G. Wells, MD, Radiology Department, MS 721, Children's Hospital of Wisconsin, 9000 W Wisconsin Ave, PO Box 1997, Milwaukee, WI 53201 (e-mail:

Not Available, Study of the National Incidence and Prevalence of Child Abuse and Neglect: 1988.  Washington, DC National Center on Child Abuse and Neglect1988;
DiScala  CSege  RLi  GReece  RM Child abuse and unintentional injuries: a 10-year retrospective. Arch Pediatr Adolesc Med. 2000;15416- 22
Billmire  MEMyers  PA Serious head injury in infants: accident or abuse? Pediatrics. 1985;75340- 342
Ewing-Cobbs  LKramer  LPrasad  M  et al.  Neuroimaging, physical, and developmental findings after inflicted and noninflicted traumatic brain injury in young children. Pediatrics. 1998;102 ((pt 1)) 300- 307Article
Duhaime  ACGennarelli  TAThibault  LEBruce  DAMargulies  SSWiser  R The shaken baby syndrome: a clinical, pathological, and biomechanical study. J Neurosurg. 1987;66409- 415Article
Gennarelli  TAThibault  LE Biomechanics of acute subdural hematoma. J Trauma. 1982;22680- 686Article
Tsai  FYZee  CSApthorp  JSDixon  GH Computed tomography in child abuse head trauma. J Comput Tomogr. 1980;4277- 286Article
Merten  DFOsborne  DRRadkowski  MALeonidas  JC Craniocerebral trauma in the child abuse syndrome: radiological observations. Pediatr Radiol. 1984;14272- 277Article
Zimmerman  RABilaniuk  LTBruce  DSchut  LUzzell  BGoldberg  HI Computed tomography of craniocerebral injury in the abused child. Radiology. 1979;130687- 690
Hymel  KPRumack  CMHay  TCStrain  JDJenny  C Comparison of intracranial computed tomographic (CT) findings in pediatric abusive and accidental head trauma. Pediatr Radiol. 1997;27743- 747Article
Duhaime  ACChristian  CWRorke  LBZimmerman  RA Nonaccidental head injury in infants—the "shaken-baby syndrome" [review]. N Engl J Med. 1998;3381822- 1829Article
Zimmerman  RABilaniuk  LT Pediatric head trauma [review]. Neuroimaging Clin N Am. 1994;4349- 366
Aoki  N Chronic subdural hematoma in infancy: clinical analysis of 30 cases in the CT era. J Neurosurg. 1990;73201- 205Article
Ewing-Cobbs  LPrasad  MKramer  L  et al.  Acute neuroradiologic findings in young children with inflicted or noninflicted traumatic brain injury. Childs Nerv Syst. 2000;1625- 34Article
Duhaime  ACAlario  AJLewander  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 ((pt 1)) 179- 185
Reece  RMSege  R Childhood head injuries: accidental or inflicted? Arch Pediatr Adolesc Med. 2000;15411- 15
Jenny  CHymel  KPRitzen  AReinert  SEHay  TC Analysis of missed cases of abusive head trauma. JAMA. 1999;281621- 626Article