Predicting the ability to walk after traumatic spinal cord injury is of utmost importance in the clinical setting. Nevertheless, only a small fraction of predictive models are evaluated on their performance by other authors using external data. The Dutch Clinical Prediction Rule for long-term walking ability was developed and validated using neurological assessments performed within 15 days postinjury. However, in reality, this assessment is most often performed between 11 and 55 days. When considering a longer time from injury to neurological assessments, the Dutch Clinical Prediction Rule has only been externally validated for patients after non-traumatic spinal cord injury. We aimed to validate the Dutch Clinical Prediction Rule with neurological assessment performed within 3-90 days after traumatic spinal cord injury, using (a) the Dutch Clinical Prediction Rule logistic regression coefficients (Equation 1); (b) the Dutch Clinical Prediction Rule weighted coefficients (Equation 2); and (c) the reestimated (using a Spanish population) weighted coefficients (Equation 3). We conducted a retrospective (STROBE-compliant) study involving 298 adults with traumatic spinal cord injury admitted to a hospital between 2010 and 2019 in Spain. The Spinal Cord Independence Measure item-12 was used for walking assessment. Using Equation 1, the model yielded 86.2% overall classification accuracy, 94.5% sensitivity, and 83.4% specificity (area under the curve [AUC] = 0.939, 95% confidence interval [CI]: 0.915-0.965; p < .001).Using Equation 2 yielded 86.2% overall classification accuracy, 93.2% sensitivity, and 83.9% specificity (AUC = 0.9392, 95% CI: 0.914-0.964; p < .001).Using Equation 3 yielded 86.9% overall classification accuracy, 68.9% sensitivity, and 92.8% specificity (AUC = 0.939, 95% CI: 0.914-0.964; p < .001). This study validates the Dutch Clinical Prediction Rule in a Spanish traumatic spinal cord injury population with assessments performed up to 90 days postinjury with similar performance, using the original coefficients and including a reestimation of the coefficients.
Read full abstract