Individual survival prediction model for patients with leptomeningeal metastasis.

Background: Survival prediction for patients with leptomeningeal metastasis (LM) is crucial for making proper management plans and counseling patients. Prognostic models in this patient domain have been limited, and existing models often include predictors that are not available in resource-limited settings. Our aim was to develop a practical, individualized survival prediction model for patients diagnosed with LM.

Methods: We collected a retrospective cohort of patients diagnosed with LM from cerebrospinal fluid at Chiang Mai University Hospital from January 2015 to July 2021. Nine candidate predictors included male gender, age > 60 years, presence of extracranial involvement, types of primary cancer, the time between primary cancer and LM diagnosis, presence of cerebral symptoms, cranial symptoms, spinal symptoms, and abnormal CSF profiles. Flexible parametric survival analysis was used to develop the survival prognostic model for predicting survival at 3, 6, and 12 months after diagnosis. The model was evaluated for discrimination and calibration.

Results: 161 patients with 133 events were included. The derived individual survival prediction model for patients with LM, or the LMsurv model, consists of three final predictors: types of primary cancer, presence of cerebral symptoms, and presence of spinal symptoms. The model showed acceptable discrimination (Harrell's C-statistics: 0.72; 95% confidence interval 0.68-0.76) and was well calibrated at 3, 6, and 12 months.

Conclusions: The LMsurv model, incorporating three practical predictors, demonstrated acceptable discrimination and calibration for predicting survival in LM patients. This model could serve as an assisting tool during clinical decision-making. External validation is suggested to confirm the generalizability of the model.