Scholarlink

Background <p>Prostate cancer (PCa) often metastasizes to bone or lung, leading to poor prognosis and higher mortality. Early identification of patients at high risk for metastasis is critical for improving treatment outcomes. However, current prediction models are limited in accuracy and generalizability. This study aimed to develop and validate machine learning models to predict bone and/or lung metastasis in PCa patients, using data from both the SEER database and an external cohort.</p> Methods <p>Data from 56,107 PCa patients in the SEER database (2018–2021) were analyzed, with feature selection performed using LASSO regression, identifying variables such as Gleason score, PSA levels, and N stage. Six machine learning models were constructed, including XGBoost, Random Forest, and Logistic Regression. The SMOTE technique was applied to address data imbalance, and the models were externally validated using a cohort of 201 patients from the First Affiliated Hospital of Nanchang University.</p> Results <p>In the SEER cohort, 2868 (5.1%) patients had bone and/or lung metastasis, while 53,239 (94.9%) showed no evidence of metastasis. The external validation cohort, consisting of 201 patients, had a significantly higher metastatic rate, with 51 (25.4%) patients presenting with bone and/or lung metastasis. XGBoost showed superior performance, achieving an AUC of 0.979 in the SEER training set and 0.924 in the external validation cohort. Key predictors were Gleason score, PSA, and N stage. The model correctly identified 2768 metastatic cases (true positives) and 52,712 non-metastatic cases (true negatives). Feature importance analysis confirmed Gleason score and PSA as the most significant predictors. In external validation, the model maintained high accuracy, sensitivity, and specificity, confirming its robustness and generalizability across different patient populations.</p> Conclusion <p>XGBoost proved to be a highly accurate tool for predicting bone and/or lung metastasis in PCa patients. Incorporating this model into clinical practice could improve early detection of high-risk patients, enabling more personalized and timely interventions, ultimately enhancing patient outcomes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of bone or lung metastasis in prostate cancer using machine learning: a SEER-based study with external validation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月19日

查看全文
  • Mingyan Zhong,
  • Ru Chen

摘要

Background

Prostate cancer (PCa) often metastasizes to bone or lung, leading to poor prognosis and higher mortality. Early identification of patients at high risk for metastasis is critical for improving treatment outcomes. However, current prediction models are limited in accuracy and generalizability. This study aimed to develop and validate machine learning models to predict bone and/or lung metastasis in PCa patients, using data from both the SEER database and an external cohort.

Methods

Data from 56,107 PCa patients in the SEER database (2018–2021) were analyzed, with feature selection performed using LASSO regression, identifying variables such as Gleason score, PSA levels, and N stage. Six machine learning models were constructed, including XGBoost, Random Forest, and Logistic Regression. The SMOTE technique was applied to address data imbalance, and the models were externally validated using a cohort of 201 patients from the First Affiliated Hospital of Nanchang University.

Results

In the SEER cohort, 2868 (5.1%) patients had bone and/or lung metastasis, while 53,239 (94.9%) showed no evidence of metastasis. The external validation cohort, consisting of 201 patients, had a significantly higher metastatic rate, with 51 (25.4%) patients presenting with bone and/or lung metastasis. XGBoost showed superior performance, achieving an AUC of 0.979 in the SEER training set and 0.924 in the external validation cohort. Key predictors were Gleason score, PSA, and N stage. The model correctly identified 2768 metastatic cases (true positives) and 52,712 non-metastatic cases (true negatives). Feature importance analysis confirmed Gleason score and PSA as the most significant predictors. In external validation, the model maintained high accuracy, sensitivity, and specificity, confirming its robustness and generalizability across different patient populations.

Conclusion

XGBoost proved to be a highly accurate tool for predicting bone and/or lung metastasis in PCa patients. Incorporating this model into clinical practice could improve early detection of high-risk patients, enabling more personalized and timely interventions, ultimately enhancing patient outcomes.