Construction And Evaluation Of A Prediction Model For Primary Central Nervous System Lymphoma

Objective:To construct a prognostic prediction model applied to the overall survival and diseasespecific survival of patients with primary central nervous system lymphoma(PCNSL).Methods:Patients with PCNSL diagnosed by pathology from January 1,2009,to December 31,2018,in the SEER database were extracted by SEER*Stat software,and the training and validation sets were distinguished according to 2:1 stratification numbers.Data from the training set were used to identify predictors affecting the prognosis of patients with PCNSL by combining the survival analysis results with univariate and multifactorial Cox regression analysis.The prediction models were visualized using R version 4.1.2 to plot nomogram models predicting overall and disease-specific survival in PCNSL.The models were internally validated by consistency index(C-index)and calibration curves and calibrated using the Bootstrap self-sampling method.The models were externally validated by validation set data.Results:3052 patients in the training set and 1056 patients in the validation set were included,and there was no statistically significant difference in the population distribution of the two general conditions.Age,gender,type of pathology,primary site,degree of disease progression,surgery,and chemotherapy were included as independent predictors by univariate and multifactorial Cox regression analysis combined with survival analysis results(P < 0.001),and radiotherapy was added as a model influencing factor according to clinical practice.The prognostic prediction model was constructed from the above eight factors,and the results were presented as column line plots with a C-index(95% CI)of 0.718(0.731-0.705)for the OS model and 0.711(0.725-0.697)for the DSS model.The temporal C-index curves suggest that the OS model has good predictive discrimination performance over five years and the DSS model over 1.5 years.The calibration curves suggest high accuracy for<80% survival probability for the OS model and 20%-80% for the DSS model,with the high accuracy site fitting the 45° reference line.The external validation data trended close to the training set time C-index and calibration curves.Conclusion:The nomogram model prognostic prediction model constructed in this study has good discrimination and accuracy and can provide a reference for clinical prognostic assessment and treatment strategy for PCNSL patients.There was no significant difference between the nomogram model and the all-factor model,which proved the simplicity of this model.The C-index of the model was >0.7,with good discrimination;most areas of the calibration curve fit the reference line,with reasonable accuracy of the model;the trend of the external validation data was close to that of the training set data,and the model can be tried to be extended to other populations.