Scholarlink

Background <p>Although machine learning is often used in medical diagnosis, its effectiveness in cancer diagnosis remains uncertain.</p> Objective <p>To explore the ability of machine learning to predict cancer postoperative complications and early recurrence.</p> Methods <p>From the creation of the database until October 4, 2024, we conducted a comprehensive search of PubMed, Web of Science (WoS), Embase, Scopus, Cochrane Library, Wanfang, and the China National Knowledge Infrastructure (CNKI). The pooled sensitivity, specificity, Fagan plot analysis, and area under the curve (AUC) were used to assess the overall test performance of machine learning. In addition, meta-regression analysis was used to explore the sources of heterogeneity further. Furthermore, Deeks’ funnel plot asymmetry test was used to assess publication bias.</p> Results <p>Ultimately, 31 publications were identified and incorporated into this meta-analysis. In the subgroup of postoperative complications, the combined sensitivity, specificity, and AUC values of all studies were 0.75 (95% CI, 0.65–0.83), 0.78 (95% CI, 0.65–0.87), and 0.83 (95% CI, 0.79–0.86), respectively. Moreover, the combined sensitivity, specificity, and AUC values of proposed studies (studies that proposed the best predictive model) were 0.85 (95% CI, 0.71–0.93), 0.76 (95% CI, 0.39–0.94), and 0.88 (95% CI, 0.85–0.91), respectively. In the subgroup of early recurrence, the combined sensitivity, specificity, and AUC values of all studies were 0.74 (95% CI, 0.68–0.80), 0.73 (95% CI, 0.67–0.77), and 0.80 (95% CI, 0.76–0.83), respectively. Furthermore, the combined sensitivity, specificity, and AUC values of proposed studies were 0.78 (95% CI, 0.70–0.85), 0.76 (95% CI, 0.70–0.82), and 0.84 (95% CI, 0.80–0.87), respectively. In addition, Deeks’ Funnel Plot, <i>p</i>-value &gt; 0.05, indicating no publication bias. Furthermore, meta-regression analysis showed that sample size and machine learning may be the main influencing factors.</p> Conclusion <p>Machine learning can accurately predict cancer postoperative complications and early recurrence. However, its accuracy is influenced by multiple factors, including the type of machine learning model, tumor type, sample size, year of publication, and country of publication. Therefore, more studies with larger sample sizes and more standardized methodology are needed to improve the reliability of its prediction.</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

Machine learning for postoperative complication prediction and early recurrence risk assessment across cancer types: a systematic review and meta-analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月28日

查看全文
  • Wen Chen,
  • Xinliang Liu,
  • Zhenheng Wu,
  • Haifen Tan,
  • Fuqian Yu,
  • Dongmei Wang,
  • Hengyi Gao,
  • Zhigang Chen

摘要

Background

Although machine learning is often used in medical diagnosis, its effectiveness in cancer diagnosis remains uncertain.

Objective

To explore the ability of machine learning to predict cancer postoperative complications and early recurrence.

Methods

From the creation of the database until October 4, 2024, we conducted a comprehensive search of PubMed, Web of Science (WoS), Embase, Scopus, Cochrane Library, Wanfang, and the China National Knowledge Infrastructure (CNKI). The pooled sensitivity, specificity, Fagan plot analysis, and area under the curve (AUC) were used to assess the overall test performance of machine learning. In addition, meta-regression analysis was used to explore the sources of heterogeneity further. Furthermore, Deeks’ funnel plot asymmetry test was used to assess publication bias.

Results

Ultimately, 31 publications were identified and incorporated into this meta-analysis. In the subgroup of postoperative complications, the combined sensitivity, specificity, and AUC values of all studies were 0.75 (95% CI, 0.65–0.83), 0.78 (95% CI, 0.65–0.87), and 0.83 (95% CI, 0.79–0.86), respectively. Moreover, the combined sensitivity, specificity, and AUC values of proposed studies (studies that proposed the best predictive model) were 0.85 (95% CI, 0.71–0.93), 0.76 (95% CI, 0.39–0.94), and 0.88 (95% CI, 0.85–0.91), respectively. In the subgroup of early recurrence, the combined sensitivity, specificity, and AUC values of all studies were 0.74 (95% CI, 0.68–0.80), 0.73 (95% CI, 0.67–0.77), and 0.80 (95% CI, 0.76–0.83), respectively. Furthermore, the combined sensitivity, specificity, and AUC values of proposed studies were 0.78 (95% CI, 0.70–0.85), 0.76 (95% CI, 0.70–0.82), and 0.84 (95% CI, 0.80–0.87), respectively. In addition, Deeks’ Funnel Plot, p-value > 0.05, indicating no publication bias. Furthermore, meta-regression analysis showed that sample size and machine learning may be the main influencing factors.

Conclusion

Machine learning can accurately predict cancer postoperative complications and early recurrence. However, its accuracy is influenced by multiple factors, including the type of machine learning model, tumor type, sample size, year of publication, and country of publication. Therefore, more studies with larger sample sizes and more standardized methodology are needed to improve the reliability of its prediction.