文献速递类风湿关节炎

平衡随机森林模型识别类风湿关节炎高危一级亲属

Identification of individuals at high risk of developing rheumatoid arthritis: a balanced random forest model in a cohort of 1544 first-degree relatives.
PubMed 文献 · RMD Open · 2025-11-27

研究基于SCREEN-RA前瞻性队列1540例类风湿关节炎患者一级亲属,采用平衡随机森林模型结合流行病学危险因素、血清学标志物、遗传学及临床体征预测类风湿关节炎发生。平均随访7.1年,27例发展为类风湿关节炎,126例发展为血清阳性炎症性关节炎;模型预测性能中等,敏感度高(≥0.85)但特异度有限,类风湿因子在预测中重要性最高。

要点

  • 在1540例类风湿关节炎一级亲属中,平衡随机森林模型预测类风湿关节炎的敏感度≥0.85,但特异度中等。
  • 类风湿因子是预测类风湿关节炎最重要的变量,其次为临床疑似关节痛(CSA)量表症状。
  • 预测血清阳性炎症性关节炎的主要变量为年龄、性别和抗RA33自身抗体。

核心结论

  • 结合遗传学、血清学标志物、流行病学危险因素和临床体征预测类风湿关节炎具有前景,但模型泛化仍具挑战性。
  • 队列中类风湿关节炎患病率低,使开发高准确度预测模型复杂化。
  • 未来应关注外部验证并可能纳入额外生物标志物以提高预测测试的敏感度和整体性能。

研究设计与样本

  • 研究使用SCREEN-RA前瞻性队列中1540例类风湿关节炎患者一级亲属(RA-FDRs)的数据。
  • 主要结局为类风湿关节炎的发生,次要结局为血清阳性炎症性关节炎(IA),用于探索性分析。
  • 拟合平衡随机森林(BRF)模型,并通过五折交叉验证进行评估以避免过拟合。
  • 研究选择以高敏感度为目标的分类阈值。

主要发现

  • 平均随访7.1年后,27例参与者发展为类风湿关节炎,126例发展为血清阳性炎症性关节炎。
  • BRF模型预测性能中等,特征为高敏感度(≥0.85)但特异度有限。
  • 类风湿因子(RFs)在类风湿关节炎预测中重要性最高,其次为「临床疑似关节痛」(CSA)量表症状。
  • 年龄、性别和抗RA33自身抗体是预测血清阳性炎症性关节炎的主要变量。

关键数值与推荐

  • BRF模型预测类风湿关节炎的敏感度≥0.85。
  • 平均随访7.1年期间,1540例类风湿关节炎一级亲属中27例发展为类风湿关节炎,126例发展为血清阳性炎症性关节炎。
  • 类风湿关节炎预测中重要性最高的变量为类风湿因子,其次为CSA量表症状。
  • 血清阳性炎症性关节炎预测的主要变量为年龄、性别和抗RA33自身抗体。

临床怎么用

  • 在类风湿关节炎一级亲属中,可考虑结合类风湿因子、CSA量表症状、年龄、性别、抗RA33自身抗体及遗传和流行病学因素进行风险分层。
  • 采用高敏感度分类阈值有助于识别更多高风险个体,但需注意特异度有限。

局限性

  • 原文述模型泛化仍具挑战性。
  • 原文述队列中类风湿关节炎患病率低,使开发高准确度预测模型复杂化。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

OBJECTIVES: To identify in a genetically susceptible population individuals at higher risk of developing rheumatoid arthritis (RA) using a classification approach combining known epidemiological risk factors, serological biomarkers, genetics, clinical signs and symptoms. METHODS: We used data from the prospective SCREEN-RA (Evaluation of a SCREENing strategy for Rheumatoid Arthritis) cohort of 1540 first-degree relatives of RA patients (RA-FDRs). The primary outcome was the development of RA. Additionally, we used seropositive inflammatory arthritis (IA) as a secondary outcome for exploratory analyses. Balanced random forest (BRF) models were fit and evaluated through fivefold cross-validation to avoid overfitting. We chose a classification threshold that targeted high sensitivity. RESULTS: After a mean follow-up of 7.1 years, 27 participants developed RA and 126 developed seropositive IA. The BRF demonstrated moderate predictive performance, characterised by high sensitivity (≥0.85) but modest specificity. Rheumatoid factors (RFs) had the highest importance in RA prediction, followed by symptoms of 'clinically suspected arthralgia' (CSA) scale. Age, gender and anti-RA33 autoantibodies were the main variables for the prediction of seropositive IA. CONCLUSIONS: Overall, the results demonstrate that predicting RA by combining genetics, serological biomarkers, epidemiological risk factors and clinical signs is promising, although model generalisation remains challenging. The low prevalence of RA in the cohort complicates the development of highly accurate prediction models. Future efforts should focus on including external validation and potentially incorporating additional biomarkers to enhance the sensitivity and overall performance of the predictive tests.

来源:RMD Open,查看原文。版权归原作者所有。

引用

  1. Aymon R, Lamacchia C, Gilbert BTP, et al. Identification of individuals at high risk of developing rheumatoid arthritis: a balanced random forest model in a cohort of 1544 first-degree relatives. RMD Open. 2025. PMID: 41314668. doi:10.1136/rmdopen-2025-005773

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。