Causal Counterfactual visualisation for human causal decision making - A case study in healthcare

因果关系的概念是我们理解世界的核心，也是人类决策的关键。因果关系从根本上不同于关联，因为它表现出干预的后果。有证据表明，不同的因果信念往往会导致不同的健康结果。目前，心理学研究仍然没有提供许多基本问题的答案，这些问题涉及人们如何在现实世界的决策中做出因果判断。根据最近的因果关系心理学理论，人们的因果判断是基于对假定原因是否(以及如何)对结果产生影响的评估。至关重要的是，这些判断取决于反事实的对比，即如果推定的原因没有发生，会发生什么？这包括设想在心理模拟中替代现实世界的可能性。为此，可视化具有支持人们反事实思维的巨大潜力，特别是在面对复杂情景时。然而，到目前为止，在帮助因果决策的背景下研究反事实可视化的工作很少。本研究将研究新的因果反事实可视化，与真实数据的直接可视化相比，它具有一种新的功能来呈现现实中没有发生的因果反事实。反事实将由用真实数据训练的反事实模拟模型生成。这通过将假设样本可视化到真实数据之外来扩展标准数据可视化。它将支持“举例说明”，使决策者能够互动地创建合成数据并检查“封闭的可能世界”(例如，一个微小的因果变化产生的不同结果)。可视化具体的样本将允许人们查看关键证据，并与反事实进行辩论，以获得可操作的见解。因果反事实可视化将得到该团队成员在心理学和人工智能领域的最新进展的支持。通过汇集多学科团队的专业知识，新的因果反事实可视化技术将提供一个有用的渠道，通过视觉呈现来评估和加深我们对人类在因果决策中的行为和表现的理解，特别是关于可视化的作用。我们将在临床案例研究的背景下进行一系列心理学实验，以探索医疗保健中的因果决策，这将涉及参与和与医生共同设计。医生将使用可视化工具来做出和测试因果假设，以便在涉及两个临床病例的心理实验中做出决策：(1)基于患肺癌的因果风险因素的临床判断，目标是个体患者的单位水平的因果关系；(2)通过衡量一种药物对糖尿病治疗的因果影响进行临床试验，目标是在人群层面上的因果关系。通过定制的模拟，可视化将揭示替代治疗的潜在结果及其在疗效和副作用方面的权衡，从而能够通过可视化具体的反事实(例如医学测试、筛查、图像)进行比较。此外，研究人员还可以在模拟临床试验中评估治疗组和对照组之间的差异，并查看违反他们假设的例子。通过这两个临床用例，研究结果将在医疗保健环境中直接衡量，在医疗保健环境中，临床医生面临着复杂的医学图景，来自多个来源的大量数据输入和相互作用，以做出重要的临床决策。这项工作即将展示新的可视化如何寻求解决这一问题，从而减少可变性，并通过临床医生可以解释的可操作的见解来支持稳健的决策。

The concept of causation is central to our understanding of the world and key to human decision making. Causation is fundamentally different from association as it exhibits consequences of interventions. Evidence shows that different causal beliefs tend to result in different health outcomes. At present, psychology research still does not provide answers to many fundamental questions about how people make causal judgements in real-world decisions. According to recent psychological theories of causation, people's causal judgments are based on assessing whether (and how) an presumed cause makes a difference to an outcome. Crucially, these judgements hinge on counterfactual contrasts, namely what would have happened if the presumed cause had not occurred? This involves envisaging alternative possibilities to the actual world in mental simulations. To this end, visualisation has great potential to support people's counterfactual thinking, especially in the face of complex scenarios. However, so far very little work has addressed counterfactual visualisation in the context of aiding causal decision making.This research will investigate novel causal counterfactual visualisation, which will, in contrast to the direct visualisation of real data, have a new functionality to render causal counterfactuals that did not occur in reality. The counterfactuals will be generated by a counterfactual simulation model that is trained with real data. This extends standard data visualisation by visualising hypothetical exemplars beyond real data. It will support "explanation-with-examples" by enabling decision makers to interactively create synthetic data and examine "close possible worlds" (e.g. different outcomes from a small causal change). Visualising concrete exemplars will allow people to view key evidence and contest their decisions against the counterfactuals to gain actionable insights.Causal counterfactual visualisation will be underpinned by the latest advance in both psychology and AI domains made by the members of this team. By bringing together expertise from a multidisciplinary team, the new causal counterfactual visualisation techniques will offer an useful channel to assess and further our understanding of human behaviour and performance in causal decision making with the aid of visual presentation, especially with respect to the role of visualisation.We will conduct a series of psychological experiments in the context of clinical case studies to probe causal decision making in healthcare, which will involve participations and co-design with doctors. Doctors will use the visualisation tool to make and test causal hypotheses for decision making in the psychological experiments involving two clinical cases: (1) Clinical judgement based on causal risk factors of developing lung cancer, which targets unit-level causality about individual patients; (2) Clinical trial of a drug by measuring its causal effect on diabetes treatment, which targets causation at the population-level. With customised simulations, the visualisation will reveal potential outcomes of alternative treatments and their trade-off in efficacy & side-effects to enable comparisons by visualising concrete counterfactuals (e.g. medical tests, screening, images). Also, researchers can assess difference between treatment and control groups in a simulated clinical trial, and view examples that violate their hypotheses. Through the two clinical use cases, the research outcomes will be directly measured in the healthcare setting, where clinicians face complex landscape of medicine with overwhelming number of input and interplay of data from multiple sources to make important clinical decisions. This work is about to demonstrate how the new visualisation can seek to resolve this leading to reduction variability and support robust decision making with actionable insights that clinicians can interpret.