Development of a Predictive Model for the Risk of Myelosuppression in Patients With Nasopharyngeal Carcinoma Undergoing Chemoradiotherapy.

Background: Myelosuppression is a frequent complication in patients with nasopharyngeal carcinoma (NPC) undergoing chemoradiotherapy. Current clinical practice relies predominantly on treatment-phase monitoring for myelosuppression risk assessment, while effective pretreatment prediction tools are lacking. This study developed a predictive model based on pretreatment clinical indicators to facilitate early identification of high-risk patients and support clinical decision-making.

Methods: We conducted a retrospective cohort study using electronic medical records of 210 patients with NPC who received chemoradiotherapy at the First Affiliated Hospital of Bengbu Medical University between May 2016 and December 2021. Using R software, patients were randomly allocated into a training set (n = 150) and an internal validation set (n = 60) at a 7:3 ratio. Variable selection was performed using Least Absolute Shrinkage and Selection Operator regression, followed by univariable and multivariable logistic regression analyses to identify potential predictors. Following categorization of these identified potential predictors, Firth penalized-likelihood regression was employed to correct for small-sample bias, while multicollinearity was rigorously assessed using variance inflation factors (VIFs). A predictive nomogram was subsequently constructed. Model performance was evaluated through multiple validation metrics, including the concordance index (C-index), receiver operating characteristic curve analysis, clinical decision curve analysis, and calibration curve.

Results: Multivariable logistic regression analysis identified 3 potential predictors of myelosuppression: pretreatment plateletcrit (PCT), direct bilirubin (DBIL), and sodium ions (Na⁺) (all P < .05). All these potential predictors met strict stability criteria after conversion to categorical variables (all VIF < 2.1, with a predefined threshold of VIF < 5). Model evaluation demonstrates that the developed nomogram exhibits favorable predictive performance.

Conclusion: Pretreatment PCT, DBIL, and Na⁺ may serve as potential predictors of myelosuppression in patients with NPC undergoing chemoradiotherapy. This nomogram could serve as a risk stratification tool to identify high-risk patients before treatment, enabling early interventions for myelosuppression prevention.