利用电子健康记录信息估算澳大利亚养老院痴呆患者死亡风险
该研究基于IMPETUS-D集群随机试验的二次分析,纳入澳大利亚24家养老院922名中度至重度认知障碍的痴呆患者,利用电子健康记录数据构建Cox比例风险模型。最终模型保留年龄、男性、经口进食所需协助程度、基线体重和6个月内体重变化5个预测因子,模型区分度中等(表观c统计量0.64,乐观校正后0.61),校准可接受(乐观校正校准斜率0.82)。
要点
- 年龄、男性、经口进食协助程度、基线体重及6个月体重变化与痴呆患者死亡风险相关。
- 基于常规电子健康记录数据的死亡风险模型区分度中等,校准可接受。
- 模型经进一步验证后,或可借助养老院电子健康记录系统改善预见性照护规划。
核心结论
- 利用常规收集的电子健康记录数据构建的模型可识别澳大利亚养老院痴呆患者死亡的关键预测因子。
- 该模型经进一步验证后,或可利用养老院电子健康记录系统改善对恶化风险增加患者的预见性照护规划。
研究设计与样本
- 研究为IMPETUS-D集群随机试验的二次分析,该试验针对养老院员工进行姑息治疗教育干预。
- 研究对象为来自澳大利亚24家养老院的中度至重度认知障碍痴呆患者(老年护理资助工具认知技能C或D级)。
- 基线数据收集于2018年10月至2019年9月,死亡随访数据截至2020年9月30日。
- 候选预测变量从电子健康记录信息中识别,基于文献综述选择与死亡风险增加相关的因素,并限制为10个变量以确保每个候选预测变量至少20个事件。
- 采用Cox比例风险模型分析变量与死亡的关系,通过后向 elimination 法在P值阈值>0.10下构建模型。
主要发现
- 共纳入922名痴呆患者,随访期间发生296例死亡。
- 最终模型保留5个候选预测变量:年龄、男性、经口进食所需协助程度、基线体重和6个月内体重变化。
- 模型显示中等死亡区分度(表观c统计量0.64;乐观校正后c统计量0.61)。
- 模型校准可接受(乐观校正校准斜率0.82)。
关键数值与推荐
- 最终模型纳入5个预测变量:年龄、男性、经口进食所需协助程度、基线体重、6个月内体重变化。
- 模型区分度:表观c统计量0.64,乐观校正后c统计量0.61。
- 模型校准:乐观校正校准斜率0.82。
临床怎么用
- 可利用养老院电子健康记录中常规收集的年龄、性别、经口进食协助程度、基线体重及6个月体重变化等信息识别死亡高风险痴呆患者。
- 识别高风险患者后,可考虑提前进行预见性照护规划,以改善姑息治疗的可及性。
- 该模型需进一步验证后方可用于临床实践。
局限性
- 原文未明确报告研究局限性。
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: End-of-life care for people with dementia is mainly delivered in nursing homes, but prognostic uncertainty often hinders timely access to specialist palliative care and anticipatory planning. Existing electronic health record (EHR) systems in Australian nursing homes do not adequately support the identification of residents at risk of deterioration. This study explored nursing home EHR data in Australia to identify factors associated with mortality risk among residents with dementia. DESIGN: Secondary analysis of the IMPETUS-D study, a cluster randomized trial of a palliative care education intervention for nursing home staff. SETTING AND PARTICIPANTS: People with dementia from 24 nursing homes in Australia with moderate to severe cognitive impairment (aged care funding instrument cognitive skills C or D). METHODS: Baseline data were collected between October 2018 and September 2019, with follow-up mortality data until September 30, 2020. Candidate predictor variables (CPVs) were identified from EHR information, selected based on a literature review of factors associated with increased mortality, and limited to 10 variables to ensure a minimum of 20 events per CPV. A Cox proportional hazards model for the relationship between variables and mortality was developed using backward elimination at a P value threshold of >0.10. RESULTS: Of 922 people with dementia, 296 deaths occurred during follow-up. Five CPVs were retained in the final model: age, male sex, level of assistance required for oral intake, baseline weight, and change in weight over 6 months. The model showed moderate mortality discrimination (apparent c-statistic, 0.64; optimism-corrected c-statistic, 0.61) and acceptable calibration (optimism-corrected calibration slope, 0.82). CONCLUSIONS AND IMPLICATIONS: A model using routinely collected EHR data identified key predictors of mortality among nursing home residents with dementia in Australia. With further validation, these predictors could leverage nursing home EHR systems to improve anticipatory care planning for people with dementia at increased risk of deterioration.
引用
- Bicknell R, McLean A, Lim WK, et al. Using EHR Information to Estimate Mortality Risk for People With Dementia in an Australian Nursing Home Setting. J Am Med Dir Assoc. 2026. PMID: 42391767. doi:10.1016/j.jamda.2026.106298
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。