Amalgamating Evidence About Causes: Medicine, the Medical Sciences, and Beyond

在许多科学领域，在推断因果关系时，可以依赖来自不同方法、专家和学科的各种证据。合并证据以产生因果知识对于科学家和那些旨在在决策中依赖因果主张的人来说是一个普遍的挑战。这在生物医学科学和医疗实践中非常重要。在医学中，至少有四个领域需要从业者合并因果知识：在常规临床实践中治疗一系列患者，从多个医学试验中测量效应量并将其汇总为总体效应量，根据不同证据推断干预效果，以及合并一组专家的判断。在每个领域中，与推定因果关系有关的证据都有不同的形式和性质，可靠性也各不相同，而且不同领域之间融合不同证据的方式也各不相同。这个项目的主要目的是利用科学哲学的工具来评估医学中因果证据的融合。证据的融合最近在科学哲学中受到了一些关注（参见弗莱彻，Landes & Poellinger 2019的概述）。证据融合问题的一种有影响力的哲学方法建立在Bovens & Hartmann（2003）开发的贝叶斯网络框架基础上（Menon & Stegenga 2017; Landes，Osimani & Poellinger 2018）。另一种方法以著名的阿罗不可能性定理为起点，询问证据的合并是否面临与偏好合并类似的约束（Stegenga 2013; Cresto & Tajer 2020）。布拉德利、迪特里希和李斯特（2014）使用了关于判断聚合的研究结果，阐明了因果判断合并的限制。科学哲学中证据融合的另一个途径是阐明科学实践中证据融合的方法论问题。

In many areas of science, a variety of evidence from different methods, experts, and disciplines can be relied on when inferring causal claims. The amalgamation of evidence to produce causal knowledge is a widespread challenge for scientists and those aiming to rely on causal claims in decision-making. This is acutely important in the biomedical sciences and in medical practice. In medicine there are at least four domains in which practitioners are required to amalgamate causal knowledge: treating a sequence of patients in routine clinical practice, measuring effect sizes from multiple medical trials and aggregating them into an overall effect size, making inferences about intervention effects based on diverse evidence, and amalgamating a group of experts' judgements. In each domain the evidence pertaining to the putative causal relations has distinct forms and properties and varying reliability, and the ways in which that disparate evidence can be amalgamated itself varies between the domains. The broad aim of this project is to evaluate the amalgamation of causal evidence in medicine using tools from philosophy of science. Amalgamation of evidence has received some recent attention in philosophy of science (see Fletcher, Landes & Poellinger 2019 for a general overview). One influential philosophical approach to the question of evidence amalgamation builds off the Bayesian network framework developed in Bovens & Hartmann (2003) (Menon & Stegenga 2017; Landes, Osimani & Poellinger 2018). Another approach takes as its starting point the famous Arrow impossibility theorem, asking if the amalgamation of evidence faces similar constraints as the amalgamation of preferences (Stegenga 2013; Cresto & Tajer 2020). Bradley, Dietrich, & List (2014) use results from work on judgement aggregation to articulate constraints on the amalgamation of causal judgements. Still another approach to evidence amalgamation in philosophy of science is to articulate methodological problems of evidence amalgamation in scientific practice.