This study aimed to develop and validate a nomogram and present it on a website to be used to predict the overall survival at 16, 32, and 48 months in patients with prolonged disorder of consciousness (pDOC).
We retrospectively analyzed the data of 381 patients with pDOC at two centers. The data were randomly divided into training and validation sets using a ratio of 6:4. On the training set, Cox proportional hazard analyses were used to identify the predictive variables. In the training set, two models were screened by COX regression analysis, and based on clinical evidence, model 2 was eventually selected in the nomogram after comparing the receiver operating characteristic (ROC) of the two models. In the training and validation sets, ROC curves, calibration curves, and decision curve analysis (DCA) curves were utilized to measure discrimination, calibration, and clinical efficacy, respectively.
The final model included age, Glasgow coma scale (GCS) score, serum albumin level, and computed tomography (CT) midline shift, all of which had a significant effect on survival after DOCs. For the 16-, 32-, and 48-month survival on the training set, the model had good discriminative power, with areas under the curve (AUCs) of 0.791, 0.760, and 0.886, respectively. For the validation set, the AUCs for the 16-, 32-, and 48-month survival predictions were 0.806, 0.789, and 0.867, respectively. Model performance was good for both the training and validation sets according to calibration plots and DCA.
We developed an accurate, efficient nomogram, and a corresponding website based on four correlated factors to help clinicians improve their assessment of patient outcomes and help personalize the treatment process and clinical decisions.