Predicting Intracranial Traumatic Findings on Computed Tomography in Patients with Minor Head Injury: The CHIP Prediction Rule

Context Available prediction rules to guide selective use of computed tomography in patients with minor head injuries were developed for use in patients with a history of loss of consciousness. Contribution This prospective study included 3181 adults with minor head injury with or without loss of consciousness. A prediction rule based on risk factors (such as age; Glasgow Coma Scale score; skull fracture; and posttraumatic vomiting, amnesia, or seizure) successfully identified patients who had intracranial computed tomography findings (sensitivity, approximately 95%) or neurosurgical intervention (sensitivity, 100%). Caution External validation in different populations is needed before widespread application of the rule. The Editors Minor head injury is one of the most common injuries seen in western emergency departments, with an estimated incidence of 100 to 300 per 100000 people (1). Patients with minor head injury include those with blunt injury to the head who have a normal or minimally altered level of consciousness on presentation in the emergency department, that is, a Glasgow Coma Scale (GCS) score of 13 to 15, and a maximum loss of consciousness of 15 minutes, posttraumatic amnesia for 60 minutes, or both (2). Intracranial complications after minor head injury are infrequent but commonly require in-hospital observation and occasionally require neurosurgical intervention (3, 4). The imaging procedure of choice for reliable, rapid diagnosis of intracranial complications is computed tomography (CT) (5, 6). Because most patients with minor head injury do not show traumatic abnormalities on CT, it seems inefficient to scan all patients with minor head injury to exclude intracranial complications. Of the published prediction rules for the selective use of CT in patients with minor head injury, the New Orleans Criteria (NOC) and the Canadian CT Head Rule (CCHR) have been externally validated (79). Researchers in internal and external validation studies have shown that both rules identify 100% of patients requiring neurosurgical intervention and most patients with traumatic intracranial findings on CT (3, 1012). The external validation studies, however, yielded lower specificities than the development studies (10, 12). The originally reported specificities were probably too optimistic because of their partial derivation from data sets that were also used for the model development (13). Also, in both studies researchers included only a subset of patients with minor head injury. Most notably, researchers developed the NOC and the CCHR for patients with minor head injury who have a history of loss of consciousness or amnesia, which many of these patients presenting to emergency departments do not have. Generalizability of the NOC and the CCHR is therefore limited. We aimed to develop a widely applicable and easy-to-implement prediction rule for the selective use of CT in all patients with minor head injury with or without a history of loss of consciousness. To avoid optimism for the model's performance, we used penalty factors and internal validation by using bootstrapping procedures to attain more realistic predictions of the model's performance in an external patient population (13). Methods Patients We prospectively collected data on consecutive patients in 4 university hospitals in the Netherlands that were participating in the CT in Head Injury Patients (CHIP) study (Figure 1) (14). Inclusion criteria included initial presentation within 24 hours of blunt injury to the head, a minimum age of 16 years, and a Glasgow Coma Scale (GCS) score of 13 to 14 or a GCS score of 15, with at least 1 of the following risk factors: history of loss of consciousness, short-term memory deficit, amnesia for the traumatic event, posttraumatic seizure, vomiting, severe headache, clinical evidence of intoxication with alcohol or drugs, use of anticoagulants or history of coagulopathy, external evidence of injury above the clavicles, and neurologic deficit. Exclusion criteria were transfer from another hospital, contraindications for CT, or concurrent injuries precluding a head CT at presentation. Figure 1. Study flow diagram. The number of patients presenting with head injury is an estimate based on the proportion of patients included out of the total number of trauma patients seen by a neurologist or neurologist-in-training in the emergency department of the participating center that included most patients. CT = computed tomography. A neurologist or a neurologist-in-training under telephone supervision of a neurologist examined patients, after which a head CT was performed as soon as possible, in accordance with the current Dutch guidelines (15). We performed head CT according to a routine trauma protocol, with a maximum slice thickness of 5 mm infratentorially and 8 mm supratentorially, without intravenous contrast administration. A neuroradiologist or a trauma radiologist (9 in total, not blinded to the patients' history and clinical findings) interpreted scans in brain and bone window settings. The institutional review board waived patient informed consent after review of our study protocol because current Dutch guidelines and European Federation of Neurological Societies' guidelines recommend routine head CT for patients meeting our inclusion criteria (15, 16). Definitions We considered a patient to have a history of loss of consciousness when a witness or the patient reported it. We defined short-term memory deficit as persistent anterograde amnesia. We deemed amnesia present for the traumatic event if the patient could not recall the entire traumatic event. We defined posttraumatic seizure as a seizure witnessed or suspected after the injury and vomiting as an episode of emesis after the traumatic event. We classified headache as being either diffuse or localized. We evaluated the presence and severity of intoxication clinically by evidence of slurred speech, alcoholic fetor, or nystagmus; we did not perform routine blood toxicology tests. Anticoagulant treatment included only coumarin derivatives. We scored the use of platelet aggregation inhibitors (for example, aspirin and clopidrogel), but we did not consider it to be a risk factor. We assessed noniatrogenic coagulopathy, which we considered a risk factor, by patient history, but we did not perform routine blood coagulation tests. We defined external evidence of injury as clinically significant discontinuity of the skin or extensive bruising. We classified injury suspect of a fracture as clinical signs of fracture, whereas we classified other injuries, such as contusions, lacerations, or abrasions, as contusion. We defined focal neurologic deficit as any abnormality on routine clinical neurologic examination that indicated a focal cerebral lesion. Data Collection We collected data on patient and trauma characteristics, symptoms, and risk factors; physical and neurologic examination; CT findings; and neurosurgical intervention. Examining physicians entered data on patient history and examination into a database (OpenSDE, Erasmus MCUniversity Medical Center Rotterdam, Rotterdam, the Netherlands) before the patient underwent CT. If this interfered with their clinical workflow, they entered the data after the CT (17). The reading radiologist added the CT findings. We collected data on neurosurgical intervention, additional CT scans performed, and the clinical outcomes of patients by searching the hospital's patient information system. Outcome Measures Our primary outcome measure for this analysis was any intracranial traumatic finding on CT, which included all neurocranial traumatic findings except for isolated linear skull fractures. A secondary outcome measure was neurosurgical intervention contingent to initial CT. We defined neurosurgical intervention as any neurosurgical procedure (craniotomy, intracranial pressure monitoring, elevation of skull fracture, or ventricular drainage) within 30 days of the traumatic event. Risk Factors We selected all of the risk factors from the NOC and the CCHR (7, 9): age, headache, vomiting, intoxication, persistent anterograde amnesia, retrograde amnesia more than 30 minutes, injury above the clavicles (including clinical signs of skull or basal skull fracture), GCS score less than 15 at 2 hours postinjury, and dangerous trauma mechanism (pedestrian vs. vehicle, fall from height, and ejected from motor vehicle). We tested other risk factors from clinical guidelines for the use of CT in minor head injury (15, 16, 1821) for additional effects. We combined the variables cyclist versus vehicle and pedestrian versus vehicle into 1 variable (pedestrian or cyclist vs. vehicle) for statistical analysis because they are similar trauma mechanisms. Statistical Analysis We based sample size on an estimated 25 variables for multivariable logistic regression analysis. For reliable analysis, we required at least 10 events of the primary outcome measure per variable, that is, 250 events for 25 variables (22). Given an incidence of traumatic findings on CT of 8% to 10%, we needed to include 3125 patients. We assumed that missing data were missing at random, and we imputed them on the basis of the available data means to avoid bias (2327). The proportion of imputed data was 3.8%, which included items documented as unknown and items that were not documented. Of all cases, 1956 (62%) were complete. Loss of consciousness and posttraumatic amnesia had the highest proportion of missing or unknown data (18% and 10%, respectively). We imputed both as present on the basis of the available variable means and as consistent with clinical practice. We used the entire data set, after missing value imputation, for all analyses. We evaluated the study sample for demographic characteristics, mechanism of injury, traumatic findings, neurosurgical intervention, GCS scores, and the presence of risk factors. We tested associations of each risk factor with the primary o