当代的大型语言模型能提供因果推理所需的域知识吗? 通过ASCVD病例研究评估自动化因果图发现
Maryam Aziz1, M Alan Brookhart1
1Department of Population Health Sciences, Duke University School of Medicine, Durham, NC, USA.
Clinical epidemiology
|November 5, 2025
概括
大型语言模型 (LLM) 在为流行病学研究生成定向非循环图 (DAG) 方面表现有前途,但需要专家监督. 快速工程,特别是思维链,提高了因果推理的DAG完整性和一致性.
科学领域:
- 流行病学 流行病学
- 公共卫生 公共卫生
- 因果推理因果推理
- 医疗信息学 医疗信息学
背景情况:
- 定向非循环图 (DAG) 对于流行病学研究设计和偏差减少至关重要.
- 开发准确的DAG用于因果推理需要大量的领域专业知识.
- 由于广泛的培训数据,大型语言模型 (LLM) 提供了自动化DAG生成的潜力.
研究的目的:
- 评估LLM快速工程策略的有效性,以产生DAG用于人口健康研究.
- 评估OpenAI的GPT-4o和GPT-o1在创建描绘因果关系的DAG方面的表现.
主要方法:
- 测试了四种快速工程策略:零射击,一射击,基于指令和思维链 (CoT).
- 作为一个案例,我们使用了一个关于他类药物用于预防心血管疾病的假设研究.
- 生成的DAG被评估为一致性,循环性,来源准确性,完整性 (ASCVD标准) 和及时遵守.
主要成果:
- 所有生成的DAG都是非循环的,除了一个基于指令的提示实例外.
- 超过一半的DAG符合6/7的ASCVD标准,但种族一直被遗漏.
- 思想链提示产生了最完整的DAG; 一次性提示提供了最高的一致性和坚持.
- 零射击提示在GPT-o1上表现更好,提供理由和来源.
结论:
- LLM 证明了产生与基本流行病学标准保持一致的 DAG 的基本能力.
- 局限性包括缺乏理由,系统地忽略种族,以及频繁的源头幻觉,需要人类专家的审查.
- 当前的LLM作为DAG开发的头脑风暴或预分析工具是有价值的,而不是取代专家判断.
相关概念视频
Vector Algebra: Graphical Method
16.7K
Vectors can be multiplied by scalars, added to other vectors, or subtracted from other vectors. The vector sum of two (or more) vectors is called the resultant vector or, for short, the resultant.
We use the laws of geometry to construct resultant vectors, followed by trigonometry to find vector magnitudes and directions. For a geometric construction of the sum of two vectors in a plane, we follow the parallelogram rule. Suppose two vectors are at arbitrary positions. Translate either one of...
We use the laws of geometry to construct resultant vectors, followed by trigonometry to find vector magnitudes and directions. For a geometric construction of the sum of two vectors in a plane, we follow the parallelogram rule. Suppose two vectors are at arbitrary positions. Translate either one of...
16.7K
Causality in Epidemiology
1.5K
Causality or causation is a fundamental concept in epidemiology, vital for understanding the relationships between various factors and health outcomes. Despite its importance, there's no single, universally accepted definition of causality within the discipline. Drawing from a systematic review, causality in epidemiology encompasses several definitions, including production, necessary and sufficient, sufficient-component, counterfactual, and probabilistic models. Each has its strengths and...
1.5K
Deductive Reasoning
63.8K
Deductive reasoning, or deduction, is the type of logic used in hypothesis-based science. In deductive reasoning, the pattern of thinking moves in the opposite direction as compared to inductive reasoning, which means that it uses a general principle or law to predict specific results. From those general principles, a scientist can deduce and predict the specific results that would be valid as long as the general principles are valid.
For example, a researcher can deduce specific predictions...
For example, a researcher can deduce specific predictions...
63.8K
Correlation and Causation
41.1K
Statistical tests can calculate whether there is a relationship, or correlation, between independent and dependent variables. An indirect relationship of the variables signifies a correlation, while a direct relationship shows causation. If it is determined that no connection exists between the variables, then the correlation is a coincidence.
Correlation versus Causation
If the dependent variable increases or decreases when the independent variable increases, there is a positive or negative...
Correlation versus Causation
If the dependent variable increases or decreases when the independent variable increases, there is a positive or negative...
41.1K
Graphs of Equations in Two Variables
182
An equation with two variables, typically written in the form y = f(x) or Ax + By = C, describes a relationship between quantities represented by x and y. Each solution to such an equation is an ordered pair (x, y) that satisfies the equation when substituted. These pairs can be represented graphically to understand the variables' relationship visually.A common technique for constructing the graph of a two-variable equation is to create a value table. Begin by choosing several values for the...
182


