A Nomogram-Based Prognostic Model for Lymphoma Patients Initially Presenting with Fever of Unknown Origin.

BACKGROUND

Patients with lymphoma who present with fever of unknown origin (FUO) as an initial symptom lack specific clinical feature analysis, prognostic factor analysis, and existing prognostic models. We aim to create a prognostic model for these patients to improve prognosis and risk assessment.

METHODS

A total of 555 lymphoma patients with FUO as initial symptom studied at Huadong Hospital affiliated with Fudan University. Univariable Cox regression identified outcome predictors, analyzed by LASSO Cox. Multifactorial Cox on screened coefficients determined independent prognostic factors and nomogram model. The validity of the nomogram was evaluated through bootstrap sampling, calibration curves for model calibration, time-dependent ROC curve analysis for discrimination assessment, and decision curve analysis for evaluating clinical usefulness. Further validation involved utilizing Kaplan-Meier curves and Log rank tests. Lastly, X-tile software determined the optimal cutoff point for the nomogram score by comparing it with the traditional International Prognostic Index (IPI) scoring system.

RESULTS

The entire cohort was divided into a training cohort (n=388) and a validation cohort (n=167). These risk factors (cell pathologic type, performance status score, Ann Arbor staging, thrombocytopenia, and raised direct bilirubin) were used to construct a web-based dynamic survival rate calculator for lymphoma patients initially presenting with FUO. The lymphoma-specific nomogram demonstrated good consistency and efficacy in predicting the model's risk stratification. Compared to the IPI scoring system, the nomogram model had higher AUC values for different clinical endpoints. The new nomogram prognostic model showed better differentiation of risk groups compared to traditional IPI scoring.

CONCLUSION

Our study developed and validated a prognostic nomogram for lymphoma patients initially presenting with FUO, demonstrating robust predictive efficacy and risk stratification ability. Furthermore, we have successfully implemented this model into a web-based dynamic survival rate calculator.