Scholarlink

Background <p>In multiple myeloma, progression within 24 months (POD24) is a strong adverse prognostic factor. However, its impact on overall survival (OS) remains underexplored through machine learning.</p> Methods <p>We retrospectively collected clinical data from 155 patients and divided them into POD24 and non-POD24 groups. Survival analysis was performed using Kaplan-Meier (KM) curves, univariate Cox regression, and multivariate Cox regression. We further evaluated the effect of these variables on overall survival using ten machine-learning algorithms.</p> Results <p>KM and Cox regression analyses revealed significantly poorer OS in the POD24 group (<i>p</i> &lt; 0.001). Among machine learning models, ANN achieved the best performance in the test set. PCA-based visualization showed clear class separation and prediction consistency with original feature distributions. SHAP analysis identified POD24 as the strongest model-derived predictor in this cohort of mortality. Force plots further demonstrated that non-POD24 substantially contributed to lower predicted death risk.</p> Conclusion <p>This study suggests that POD24 may be associated to the survival outcomes in multiple myeloma using both traditional statistical and machine learning approaches. Our findings highlight the potential value of POD24 in mortality risk prediction and demonstrate the utilization of ANN-based SHAP interpretation in enhancing model transparency.</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

The prognostic value of POD24 for multiple myeloma: a comprehensive analysis based on traditional statistics and machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月27日

查看全文
  • Quane Zhang,
  • Yifan Wang,
  • Qiuni Chen,
  • Jingjing Ma,
  • Chuanyang Lu,
  • Yue Chen,
  • Yuye Shi,
  • Wenting Shi,
  • Zhengmei He,
  • Liang Yu,
  • Chunling Wang

摘要

Background

In multiple myeloma, progression within 24 months (POD24) is a strong adverse prognostic factor. However, its impact on overall survival (OS) remains underexplored through machine learning.

Methods

We retrospectively collected clinical data from 155 patients and divided them into POD24 and non-POD24 groups. Survival analysis was performed using Kaplan-Meier (KM) curves, univariate Cox regression, and multivariate Cox regression. We further evaluated the effect of these variables on overall survival using ten machine-learning algorithms.

Results

KM and Cox regression analyses revealed significantly poorer OS in the POD24 group (p < 0.001). Among machine learning models, ANN achieved the best performance in the test set. PCA-based visualization showed clear class separation and prediction consistency with original feature distributions. SHAP analysis identified POD24 as the strongest model-derived predictor in this cohort of mortality. Force plots further demonstrated that non-POD24 substantially contributed to lower predicted death risk.

Conclusion

This study suggests that POD24 may be associated to the survival outcomes in multiple myeloma using both traditional statistical and machine learning approaches. Our findings highlight the potential value of POD24 in mortality risk prediction and demonstrate the utilization of ANN-based SHAP interpretation in enhancing model transparency.