Observer error and prediction of outcome--grading of head injury based on computerised tomography

To measure inter-observer error of a recently reported computerised tomography scoring system and to assess the ability of the scoring system to predict outcome in head injury patients.

Two radiologists independently graded all CT scans performed during the admission of all head injured patients. They were blinded to the clinical condition of the patient. Patients were followed up at 12 months and given a Glasgow outcome score. Outcomes were matched to the 2 independent assessments done on the first CT scan for each patient.

A total of 123 head injury patients were studied. For the diffuse injury categories, there were 410 gradings made. Of these, 32% differed by at least one category. Where at least one of the radiologists identified non-evacuated mass lesions there were 148 gradings. Of these, one radiologist reported an un-evacuated mass lesion in 47%, which was not reported by the other. The first CT scan was evaluated on 119 patients. Using the Chi-Squared test, the diffuse injury IV category was the only one to show a strong relationship with outcome as measured by the Glasgow outcome score.

The prediction of outcome for head injury patients based on CT scans has significant shortcomings. In our study, there was significant variation in grading by experienced radiologists. The separate categories were also poor predictors of outcome at 12 months except for diffuse injury IV. The classification of mass lesions needs modification to be useful.