Eliminating Bias in Survival Estimation: Statistical Bias Mitigation Is the First Step Forward.
Eliminating Bias in Survival Estimation: Statistical Bias Mitigation Is the First Step Forward.
复制标题
消除生存估计中的偏差：减少统计偏差是前进的第一步。
DOI：
10.1097/ccm.0000000000006110
复制
发表时间：
2024
影响因子：
8.8
通讯作者：
Kohn,Rachel
中科院分区：
文献类型：
--
作者：
Kohn,Rachel
生存测量自古代历史以来一直在发展，反映了我们对健康，疾病以及后来患者特征和观点的不断变化的理解。大约公元前1600年，来自古埃及的埃德温·史密斯纸莎草纸是已知最古老的关于手术和创伤的文献。与其他早期的医学文献相比，埃德温·史密斯纸莎草纸接近医学，特别是改善健康和生存，科学而不是植根于魔法。16 - 18世纪标志着欧洲出现了一种结构化的方法来维护生命统计数据，这种方法植根于频繁的流行病和大流行病（2）。在19世纪世纪，人寿保险公司开发了基于大型数据集的精算表来评估死亡率风险，提供了对人口水平生存趋势的见解（3，4）。流行病学随后在世纪成为一门学科，专注于了解疾病的原因和分布，包括患者水平的特征，如吸烟，饮食，遗传学和生存分析中的医疗保健（4）。在现代，生存测量已经发展成为一个多方面的领域，采用各种统计方法，使用来自众多来源的数据，此外还结合患者的特征和观点。尽管有这些非凡的科学进步，我们才刚刚开始了解影响我们准确估计生存率的各种偏见来源，包括方法学偏见（例如，针对给定用例的生存估计的最适当建模策略;将其纳入给定模型的协变量）和结构性偏差（即“正常化和合法化的政策，做法和态度，经常对少数群体产生累积和长期的不利后果”[5]）。
Survival measurement has evolved since ancient history, reflecting our changing understanding of health, disease, and much later, patient characteristics and perspectives. The Edwin Smith Papyrus from Ancient Egypt in~ 1600 BCE is the oldest known document addressing surgery and trauma. Compared with other early medical texts, the Edwin Smith Papyrus approached medicine, specifically improving health and survival, scientifically rather than rooted in magic (1). The 16th–18th centuries marked the emergence of a structured approach to maintaining vital statistics in Europe, rooted in frequent epidemics and pandemics (2). In the 19th century, life insurance companies developed actuarial tables based on large datasets to assess mortality risk, providing insights into population-level survival trends (3, 4). Epidemiology subsequently emerged as a discipline in the 20th century, focused on understanding causes and distributions of diseases, including patient-level characteristics, such as smoking, diet, genetics, and access to healthcare in survival analyses (4). In the modern era, survival measurement has evolved into a multifaceted field that employs various statistical approaches and uses data from numerous sources, additionally incorporating patient characteristics and perspectives. Despite these extraordinary scientific advances, we are just beginning to understand the various sources of bias that impact our ability to accurately estimate survival, including methodologic biases (eg, the most appropriate modeling strategy for survival estimation for a given use case; which covariates to include in a given model) and structural biases (ie,“the normalized and legitimized range of policies, practices, and attitudes that routinely produce cumulative and chronic adverse outcomes for minority populations”[5]).