Development and internal validation of a risk prediction calculator for minor spinal cord injury in CT-negative blunt trauma

Background Clinical calculators are used to determine which trauma patients require computed tomography (CT) scans of the spine. However, mild traumatic spinal cord injury (TSCI) may be present despite a negative CT scan. Therefore, the present study sought to internally validate the use of two calculators (whole spine and cervical-only) to identify such patients. Methods The Spinal Cord Injury Model System Program was used to conduct this retrospective cohort study. Patients at least 15 years old with an American Spinal Injury Association (ASIA) grade D injury due to blunt trauma were included. Patients with and without concurrent vertebral injury were considered CT-evident and CT-occult, respectively. A Firth’s regression was used to establish β coefficients, which were converted into points to predict CT-occult TSCI. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated at the optimal point threshold. Results This study included 589 patients (mean age = 54.1 ± 16.7), of whom 182 (31%) were CT-occult. In the all-injury level calculator, ages 30–70 added three points; ages 70+ added one point; fall injuries added one point; cervical level injury added seven points; clinical suspicion of central cord syndrome (CCS) added one point, whereas the presence of associated injuries deducted three points. At the seven-point cutoff, sensitivity was 97.3%, specificity was 21.9%, PPV was 35.8%, and NPV was 94.7%. The cervical calculator assigned one point for ages 45–60, fall injuries, and CCS, but deducted two points if there were associated injuries. At the zero-point threshold, sensitivity was 96.6%, specificity was 8.5%, PPV was 36.4%, and NPV was 82.4%. Conclusions The calculators demonstrated high sensitivity and may be invaluable adjuncts for assessing suspected CT-negative TSCI. External validation is necessary to determine their generalizability.