合并痛风患者住院期间痛风发作预测模型的开发
基于新西兰惠灵顿2017年625例合并痛风住院患者的回顾性队列,开发住院痛风发作预测模型。临床知识驱动模型(A)含9个预测因子,C统计量0.82、校准斜率0.93,优于统计驱动模型(B,15个因子)和决策树模型(C,5个因子)。Bootstrap验证C统计量0.80、校准斜率0.78。
要点
- 合并痛风住院患者中,临床知识驱动模型含9个简单实用的预测因子。
- 模型A的C统计量0.82、校准斜率0.93,Bootstrap验证分别为0.80和0.78。
- 预测因子包括入院前尿酸>0.36 mmol/L、痛风石、未用降尿酸治疗等。
核心结论
- 提出一套含9个预测因子的合并痛风患者住院痛风发作预测模型,简单实用且有临床知识支持。
研究设计与样本
- 回顾性队列研究,纳入新西兰惠灵顿2017日历年625例合并痛风的住院患者,其中87例发生住院痛风发作。
- 采用三种建模方法:临床知识驱动模型(A)、统计驱动模型(B)和决策树模型(C),最终模型经Bootstrap验证。
主要发现
- 模型A产生9个预测因子,模型B产生15个,模型C产生5个。
- 模型A因简单且C统计量(0.82)和校准斜率(0.93)更优而被选为最终模型。
- 最终模型Bootstrap验证的C统计量为0.80,校准斜率为0.78。
关键数值与推荐
- 最终9个预测因子为:入院前尿酸>0.36 mmol/L、痛风石、入院前未用降尿酸治疗(ULT)、入院前未用痛风预防治疗、急性肾损伤、手术、开始或增加痛风预防治疗、调整ULT、发作前使用利尿剂。
临床怎么用
- 可用于识别合并痛风住院患者中痛风发作高风险者,预测因子均为临床常规可获取信息。
局限性
- 原文未报告模型的外部验证结果。
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: Hospitalisation is a risk factor for flares in people with gout. However, the predictors of inpatient gout flare are not well understood. The aim of this study was to develop a prediction model for inpatient gout flare among people with comorbid gout. METHODS: We used data from a retrospective cohort of hospitalised patients with comorbid gout from Wellington, Aotearoa/New Zealand, in 2017 calendar year. For the development of a prediction model, we took three approaches: (A) a clinical knowledge-driven model, (B) a statistics-driven model and (C) a decision tree model. The final model was chosen based on practicality and performance, then validated using bootstrap procedure. RESULTS: The cohort consisted of 625 hospitalised patients with comorbid gout, 87 of whom experienced inpatient gout flare. Model A yielded 9 predictors of inpatient gout flare, while model B and C produced 15 and 5, respectively. Model A was chosen for its simplicity and superior C-statistics (0.82) and calibration slope (0.93). The final nine-item set of predictors were pre-admission urate >0.36 mmol/L, tophus, no pre-admission urate-lowering therapy (ULT), no pre-admission gout prophylaxis, acute kidney injury, surgery, initiation or increase of gout prophylaxis, adjustment of ULT and diuretics prior to flare. Bootstrap validation of the final model showed adequate C-statistics and calibration slope (0.80 and 0.78, respectively). CONCLUSION: We propose a set of nine predictors of inpatient flare for people with comorbid gout. The predictors are simple, practical and are supported by existing clinical knowledge.
引用
- Jatuworapruk K, Grainger R, Dalbeth N, et al. Development of a prediction model for inpatient gout flares in people with comorbid gout. Ann Rheum Dis. 2019. PMID: 31811060. doi:10.1136/annrheumdis-2019-216277
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。