Path specific analysis reveals how treatments cause harm
Path-specific harm decomposition: A partial identification framework
Machine Learning
Summary
When treatments help most people but hurt some, it’s important to know exactly how the harm happens. The authors separate the harm caused directly by the treatment from harm caused indirectly through other factors influenced by the treatment. Because it’s hard to measure these effects exactly, they create a new way to estimate and bound the amount of harm along each pathway. They also provide methods to measure these harm bounds accurately and reliably. This approach gives clearer insights into the sources of harm in treatment effects.
What this means in practice
- •For clinical trial teams: Estimate and separate direct and indirect harm pathways from treatments to improve patient safety assessments in trials.
- •For policy evaluation analysts: Distinguish between direct and mediator-driven harm in social programs to better target interventions that avoid unintended negative effects.
Authors
Ruizi Yan, Dennis Frauen, Maresa Schröder, Stefan Feuerriegel
Abstract
A central goal when designing treatment policies is often to "do no harm", that is, to avoid interventions that improve average outcomes while worsening outcomes for some individuals. A widely used notion for harm is the fraction of negatively affected (FNA), defined as the probability that an intervention decreases an individual's outcome. However, in many applications, treatments operate through mediators, and a single "total" FNA can obscure whether harm arises primarily through direct pathways or indirect (mediator-induced) pathways. In this work, we introduce a path-specific analogue of the FNA. For this, we disentangle total harm into direct and indirect harm in causal mediation settings. However, these quantities depend on joint distributions of potential outcomes that are not point-identified even in randomised controlled trials. As a remedy, we develop a novel partial identification framework for direct and indirect FNA. In our framework, we (i) derive sharp Makarov bounds for the FNA, and (ii) propose a semiparametrically efficient estimator with valid confidence intervals for these bounds under mild margin conditions. We demonstrate our framework across various numerical experiments. To the best of our knowledge, we are the first to study path-specific decomposition of causal harm and to develop an orthogonal inference framework for its analysis.