Interpretable clinical visualization model for prediction of prognosis in osteosarcoma: a large cohort data study

Area of Science:

• Oncology
• Biostatistics
• Clinical Informatics

Background:

• Existing osteosarcoma prediction models lack external validation and clinical utility due to reliance on single-center data.
• This limitation hinders the reliability and practical application of current predictive tools for osteosarcoma patients.

Purpose of the Study:

• To develop and validate a robust clinical prediction model for osteosarcoma with improved predictive ability and credibility.
• To enhance the clinical application value of osteosarcoma prediction models through external validation.

Main Methods:

• Data from SEER database and four Chinese medical centers (2010-2016) were used for osteosarcoma patient information.
• Factor selection involved full subset regression, univariate Cox, and LASSO regression, optimizing for minimum AIC and maximum AUC.
• Model visualization included nomograms, a web calculator, and decision trees, with external validation performed across diverse medical centers.

Main Results:

• Eight key predictors (age, grade, laterality, M stage, surgery, bone/lung metastases, tumor size) were identified.
• The model demonstrated good internal and external consistency, with Area Under the Curve (AUC) > 0.8 indicating strong predictive ability.
• Decision curve analysis confirmed excellent clinical utility for 3- and 5-year predictions in both North American and Chinese cohorts.

Conclusions:

• A validated clinical prediction model for osteosarcoma was successfully developed and visualized via nomogram and web calculator.
• The model exhibits strong consistency, predictive power, and significant clinical application value for osteosarcoma patients.