Scholarlink

Background <p>Gout is the most common inflammatory arthritis. Recurrent flares are common among hospitalised patients and contribute to substantial clinical and economic burden. However, the accurate prediction of inpatient recurrence remains challenging, particularly in individuals with comorbid gout. This study aims to evaluate the predictive value of variables and develop a multidimensional, interpretable artificial intelligence (AI) model to predict gout recurrence (GoutRe).</p> Methods <p>We conducted a real-world, retrospective, prospective, multicentre cohort study across five tertiary hospitals in China (2010-2024), enrolling 6,526 hospitalised patients with comorbid gout. A total of 82 multidimensional features encompassing laboratory findings, comorbidities, medications, and clinical indicators were collected. We developed 3,744 AI models and the optimal model was chosen based on practicality and performance. SHapley Additive exPlanation (SHAP) was used for feature importance analysis and interpret the model. Subgroup and stratified analyses were performed by age, tophus, comorbidities, and key biomarkers such as serum urate (SU), neutrophil count (NEUT), and estimated glomerular filtration rate (eGFR).</p> Results <p>The final GoutRe model achieved strong discrimination across all cohorts, with AUCs of 0.832 (training, 95%CI 0.820-0.845), 0.785 (internal validation, 95%CI 0.763-0.808), 0.742 (external validation, 95%CI 0.690-0.793), and 0.744 (prospective validation, 95%CI 0.673-0.814). Calibration plots and low Brier scores indicated good agreement between predicted and observed outcomes. Decision curve analysis demonstrated favorable net clinical benefit. SHAP analysis identified 20 key predictors, including length of stay, glucocorticoid use, SU, neutrophil count, and tophus. The model showed enhanced performance in patients aged ≥60 years, those with tophus, and in subgroups with neoplasms, genitourinary, neurological, and cardiovascular diseases. Stratified analyses revealed significant positive correlations between recurrence risk and elevated SU, NEUT, and reduced eGFR. A secure web-based application was developed for clinical implementation.</p> Conclusions <p>The multidimensional and interpretable GoutRe model demonstrated robust predictive performance across diverse clinical settings and validation cohorts. Its web-based tool may assist clinicians in early identification of patients at high risk of GoutRe, thereby enabling individualized treatment planning.</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

Development and validation of a multidimensional and interpretable artificial intelligence model to predict gout recurrence in hospitalised patients: a real-world, ambispective multicentre cohort study in China

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月4日

查看全文
  • Meng Li,
  • Hui Zhang,
  • Shixian Chen,
  • Fei Zhong,
  • Jiani Liu,
  • Juan Wu,
  • Ruifeng Lin,
  • Ruichang Li,
  • Yu Wu,
  • Danning Xie,
  • Kangyu Zhang,
  • Bowen Zheng,
  • Xiaoling Chen,
  • Zhipeng Cheng,
  • Yinxiu Jiang,
  • Haixin Ye,
  • Li Cai,
  • Ruixia Xie,
  • Dongsheng Li,
  • Junqing Zhu,
  • Juan Li

摘要

Background

Gout is the most common inflammatory arthritis. Recurrent flares are common among hospitalised patients and contribute to substantial clinical and economic burden. However, the accurate prediction of inpatient recurrence remains challenging, particularly in individuals with comorbid gout. This study aims to evaluate the predictive value of variables and develop a multidimensional, interpretable artificial intelligence (AI) model to predict gout recurrence (GoutRe).

Methods

We conducted a real-world, retrospective, prospective, multicentre cohort study across five tertiary hospitals in China (2010-2024), enrolling 6,526 hospitalised patients with comorbid gout. A total of 82 multidimensional features encompassing laboratory findings, comorbidities, medications, and clinical indicators were collected. We developed 3,744 AI models and the optimal model was chosen based on practicality and performance. SHapley Additive exPlanation (SHAP) was used for feature importance analysis and interpret the model. Subgroup and stratified analyses were performed by age, tophus, comorbidities, and key biomarkers such as serum urate (SU), neutrophil count (NEUT), and estimated glomerular filtration rate (eGFR).

Results

The final GoutRe model achieved strong discrimination across all cohorts, with AUCs of 0.832 (training, 95%CI 0.820-0.845), 0.785 (internal validation, 95%CI 0.763-0.808), 0.742 (external validation, 95%CI 0.690-0.793), and 0.744 (prospective validation, 95%CI 0.673-0.814). Calibration plots and low Brier scores indicated good agreement between predicted and observed outcomes. Decision curve analysis demonstrated favorable net clinical benefit. SHAP analysis identified 20 key predictors, including length of stay, glucocorticoid use, SU, neutrophil count, and tophus. The model showed enhanced performance in patients aged ≥60 years, those with tophus, and in subgroups with neoplasms, genitourinary, neurological, and cardiovascular diseases. Stratified analyses revealed significant positive correlations between recurrence risk and elevated SU, NEUT, and reduced eGFR. A secure web-based application was developed for clinical implementation.

Conclusions

The multidimensional and interpretable GoutRe model demonstrated robust predictive performance across diverse clinical settings and validation cohorts. Its web-based tool may assist clinicians in early identification of patients at high risk of GoutRe, thereby enabling individualized treatment planning.