Scholarlink

Background <p>Cancer imposes a growing burden on health systems, particularly in middle-income countries where fragmented services limit timely diagnosis and treatment. Effective patient stratification is essential to improve care delivery and optimize the use of scarce hospital resources.</p> Objective <p>This study presents an explainable machine learning framework for segmenting hospitalized oncology patients and supporting resource allocation in oncology services.</p> Methods <p>A dataset of 5,296 patients from the Michoacán State Oncology Center (Mexico) was analyzed. Patient profiles were identified using K-Prototypes clustering, which integrates numerical and categorical clinical variables. Cluster assignments were validated through a CatBoost classifier, achieving near-perfect accuracy (≈ 99%). SHAP (Shapley Additive Explanations) was applied to interpret classification outcomes and quantify the contribution of individual clinical features.</p> Results <p>Three distinct and clinically coherent clusters emerged: (i) older patients undergoing minimal interventions with the highest in-hospital mortality; (ii) middle-aged women with surgically treatable cancers and the lowest mortality; and (iii) younger patients undergoing complex gynecological surgeries with intermediate risk. Length of stay, number of procedures, and urinary catheter use were identified as the dominant predictors of cluster membership.</p> Conclusion <p>The integration of mixed-data clustering, robust classification, and explainable AI provides a transparent and accurate approach to oncology patient segmentation. These insights support data-driven surgical scheduling and hospital bed management, enhancing efficiency in oncology care. The methodology is transferable to other oncology services and contributes to advancing precision oncology within healthcare systems under resource constraints.</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

Explainable machine learning for cancer patient segmentation and hospital resource optimization in oncology care

  • 期刊论文

  • 发表日期：

    2026年3月7日

查看全文
  • Francisco Javier Lopéz-Flores,
  • Alma Yunuen Raya-Tapia,
  • Nagib Alexey Zetina-Alonso,
  • José María Ponce-Ortega

摘要

Background

Cancer imposes a growing burden on health systems, particularly in middle-income countries where fragmented services limit timely diagnosis and treatment. Effective patient stratification is essential to improve care delivery and optimize the use of scarce hospital resources.

Objective

This study presents an explainable machine learning framework for segmenting hospitalized oncology patients and supporting resource allocation in oncology services.

Methods

A dataset of 5,296 patients from the Michoacán State Oncology Center (Mexico) was analyzed. Patient profiles were identified using K-Prototypes clustering, which integrates numerical and categorical clinical variables. Cluster assignments were validated through a CatBoost classifier, achieving near-perfect accuracy (≈ 99%). SHAP (Shapley Additive Explanations) was applied to interpret classification outcomes and quantify the contribution of individual clinical features.

Results

Three distinct and clinically coherent clusters emerged: (i) older patients undergoing minimal interventions with the highest in-hospital mortality; (ii) middle-aged women with surgically treatable cancers and the lowest mortality; and (iii) younger patients undergoing complex gynecological surgeries with intermediate risk. Length of stay, number of procedures, and urinary catheter use were identified as the dominant predictors of cluster membership.

Conclusion

The integration of mixed-data clustering, robust classification, and explainable AI provides a transparent and accurate approach to oncology patient segmentation. These insights support data-driven surgical scheduling and hospital bed management, enhancing efficiency in oncology care. The methodology is transferable to other oncology services and contributes to advancing precision oncology within healthcare systems under resource constraints.