arXiv subjects

Mats J. Stensrud

Publications and source records attributed to Mats J. Stensrud.

At least 19 recordsLinked to original sources

arXiv2026

Optimal sequential decision-making with initiation regimes

Consider an optimal dynamic treatment regime, $g^{\textbf{opt}}$ correctly identified from a large, perfectly executed sequentially randomized experiment. Even when the experimental results are generalizable to a future target population, there is no guarantee that $g^{\textbf{opt}}$ outperforms human decision-makers; human experts can do better than $g^{\textbf{opt}}$ whenever they have access to relevant information beyond the covariates recorded in the experiment. Motivated by this observation, we derive results on a new class of regimes called initiation regimes, which generalize existing results on superoptimal regimes. These regimes follow human decision-makers up to the point where it becomes more beneficial to initiate a sequential optimal regime, and are guaranteed to outperform both purely human and purely algorithmic decision rules, e.g., based on reinforcement learning algorithms. Furthermore, we present modified experimental designs that identify the best initiation regimes, show how the best initiation regime can be identified from classical observational data under explicit assumptions, and give estimation and statistical inference methodology for these regimes. To illustrate the practical utility of the methods, we consider initiation regimes in a case study on treatment of lower back pain.

stat.ME↗

arXiv2026

Causal inference with staggered entries and effects that change over calendar time

Studies with staggered entry, in which individuals enroll at different calendar times, are ubiquitous in medicine and related disciplines. Because these studies usually have a fixed administrative end of follow-up, identification of the estimand of interest relies on assumptions about the right-censoring mechanism. The assumptions are often considered plausible only conditional on covariates, including the time an individual entered the study (E). Yet, censoring assumptions formulated conditional on E are ill-posed due to positivity violations. Here, we study the consequences of this issue for common procedures in causal survival analysis, such as those based on marginal structural models. We further give conditions for valid identification and introduce sensitivity analyses to assess practical implications. We illustrate our methodology through two case studies. The first builds on the seminal article by Hernán et al. (2000) on the effect of zidovudine treatment. The second reanalyzes a recent study on the effect of the mRNA vaccine in patients receiving immune checkpoint inhibitors.

stat.ME↗

arXiv2026

Causal inference for N-of-1 trials

The aim of personalized medicine is to tailor treatment decisions to individuals' characteristics. N-of-1 trials are within-person crossover trials that hold the promise of targeting individual-specific effects. While the idea behind N-of-1 trials might seem simple, analyzing and interpreting N-of-1 trials is not straightforward. Here we ground N-of-1 trials in a formal causal inference framework and formalize intuitive claims from the N-of-1 trials literature. We focus on causal inference from a single N-of-1 trial and define a conditional average treatment effect (CATE) that represents a target in this setting, which we call the U-CATE. We discuss assumptions sufficient for identification and estimation of the U-CATE under different causal models where the treatment schedule is assigned at baseline. A simple mean difference is an unbiased, asymptotically normal estimator of the U-CATE in simple settings. We also consider settings where carryover effects, trends over time, time-varying common causes of the outcome, and outcome-outcome effects are present. In these more complex settings, we show that a time-varying g-formula identifies the U-CATE under explicit assumptions. Finally, we analyze data from N-of-1 trials about acne symptoms and show how different assumptions about the data generating process can lead to different analytical strategies.

stat.ME↗

arXiv2026

Effects conditional on post-treatment events generated by independent mechanisms

In both observational studies and randomized trials, post-treatment events such as dropout, nonadherence, and truncation by death occur frequently. In some studies, conditioning on post-treatment variables is a deliberate strategy to isolate particular treatment effects on the outcome. However, naive comparisons of outcomes conditional on post-treatment events generally lack a causal interpretation, even when treatment is randomly assigned. There exist causal estimands that account for post-treatment events, including survivor average causal effects and conditional separable effects, but identification usually requires measurement of common causes of the post-treatment event and the outcome. In this article, we show that such measurements are not always necessary. Conceptually, what we require is that the treatment and other unmeasured causes of the outcome generate the post-treatment event through "independent mechanisms". Then, conditional separable effects and survivor average causal effects can be identified without adjustment for common causes of the post-treatment event and the outcome. We illustrate the results in studies with truncating events, differential nonadherence, and the birth weight paradox.

stat.ME↗

arXiv2026

A simple and powerful test of vaccine waning

Determining whether vaccine efficacy wanes is important for individual and public decision making. Yet, quantification of waning is a subtle task. The classical approaches cannot be interpreted as measures of declining efficacy unless we impose unreasonable assumptions. Recently, formal causal estimands designed to quantify vaccine waning have been proposed. These estimands can be bounded under weaker assumptions, but the bounds are often too wide to make claims about the presence of waning. We propose a different approach: a formal test to assess whether a treatment effect is constant over time at the individual level. This test provides a considerable power gain over existing approaches and is valid under interpretable assumptions in vaccine trials. We illustrate the increase in power through real and simulated examples, using three different approaches to compute the test statistics. Two of these approaches are based solely on summary data, accessible from existing clinical trials. Beyond our test, we also give new results that bound the waning effect. We use our methods to reanalyze data from a randomized controlled trial of the BNT162b2 COVID-19 vaccine. While prior analysis did not establish waning, our test rejects the null hypothesis of no waning.

stat.ME↗

arXiv2026

Nonparametric tests of treatment effect homogeneity for policy-makers

Recent work has focused on nonparametric estimation of conditional treatment effects, but inference has remained relatively unexplored. We propose a class of nonparametric tests for both quantitative and qualitative treatment effect heterogeneity. The tests can incorporate a variety of structured assumptions on the conditional average treatment effect, allow for both continuous and discrete covariates, and do not require sample splitting to obtain a tractable asymptotic null distribution. Furthermore, we show how the tests are tailored to detect alternatives where the population impact of adopting a personalized decision rule differs from using a rule that discards covariates. The proposal is thus relevant for guiding treatment policies. The utility of the proposal is borne out in simulation studies and a re-analysis of an AIDS clinical trial.

stat.ME↗

arXiv2025

Counterfactual Harm: A Counter-argument

As AI systems are increasingly used to guide decisions, it is essential that they follow ethical principles. A core principle in medicine is non-maleficence, often equated with ``do no harm''. A formal definition of harm based on counterfactual reasoning has been proposed and popularized. This notion of harm has been promoted in simple settings with binary treatments and outcomes. Here, we highlight a problem with this definition in settings involving multiple treatment options. Illustrated by an example with three tuberculosis treatments (say, A, B, and C), we demonstrate that the counterfactual definition of harm can produce intransitive results: B is less harmful than A, C is less harmful than B, yet C is more harmful than A when compared pairwise. This intransitivity poses a challenge as it may lead to practical (clinical) decisions that are difficult to justify or defend. In contrast, an interventionist definition of harm based on expected utility forgoes counterfactual comparisons and ensures transitive treatment rankings.

stat.AP↗

arXiv2025

Quantification of vaccine waning as a challenge effect

Knowing whether vaccine protection wanes over time is important for health policy and drug development. However, quantifying waning effects is difficult. A simple contrast of vaccine efficacy at two different times compares different populations of individuals: those who were uninfected at the first time versus those who remain uninfected until the second time. Thus, the contrast of vaccine efficacy at early and late times can not be interpreted as a causal effect. We propose to quantify vaccine waning using the challenge effect, which is a contrast of outcomes under controlled exposures to the infectious agent following vaccination. We identify sharp bounds on the challenge effect under non-parametric assumptions that are broadly applicable in vaccine trials using routinely collected data. We demonstrate that the challenge effect can differ substantially from the conventional vaccine efficacy due to depletion of susceptible individuals from the risk set over time. Finally, we apply the methods to derive bounds on the waning of the BNT162b2 COVID-19 vaccine using data from a placebo-controlled randomized trial. Our estimates of the challenge effect suggest waning protection after 2 months beyond administration of the second vaccine dose.

stat.ME↗

arXiv2025

Refining the Notion of No Anticipation in Difference-in-Differences Studies

We address an ambiguity in identification strategies using difference-in-differences, which are widely applied in empirical research, particularly in economics. The assumption commonly referred to as the "no-anticipation assumption" states that treatment has no effect on outcomes before its implementation. However, because standard causal models rely on a temporal structure in which causes precede effects, such an assumption seems to be inherently satisfied. This raises the question of whether the assumption is repeatedly stated out of redundancy or because the formal statements fail to capture the intended subject-matter interpretation. We argue that confusion surrounding the no-anticipation assumption arises from ambiguity in the intervention considered and that current formulations of the assumption are ambiguous. Therefore, new definitions and identification results are proposed.

stat.ME↗

arXiv2025

Interpretational errors with instrumental variables

Instrumental variables (IV) are often used to identify causal effects in observational settings and experiments subject to non-compliance. Under canonical assumptions, IVs allow us to identify a so-called local average treatment effect (LATE). The use of IVs is often accompanied by a pragmatic decision to abandon the identification of the causal parameter that corresponds to the original research question and target the LATE instead. This pragmatic decision presents a potential source of error: an investigator mistakenly interprets findings as if they had made inference on their original causal parameter of interest. We conducted a systematic review and meta-analysis of patterns of pragmatism and interpretational errors in the applied IV literature published in leading journals of economics, political science, epidemiology, and clinical medicine (n = 309 unique studies). We found that a large fraction of studies targeted the LATE, although specific interest in this parameter was rare. Of these studies, 61% contained claims that mistakenly suggested that another parameter was targeted -- one whose value likely differs, and could even have the opposite sign, from the parameter actually estimated. Our findings suggest that the validity of conclusions drawn from IV applications is often compromised by interpretational errors.

econ.EM↗

arXiv2024

Causal inference with recurrent and competing events

Many research questions concern treatment effects on outcomes that can recur several times in the same individual. For example, medical researchers are interested in treatment effects on hospitalizations in heart failure patients and sports injuries in athletes. Competing events, such as death, complicate causal inference in studies of recurrent events because once a competing event occurs, an individual cannot have more recurrent events. Several statistical estimands have been studied in recurrent event settings, with and without competing events. However, the causal interpretations of these estimands, and the conditions that are required to identify these estimands from observed data, have yet to be formalized. Here we use a formal framework for causal inference to formulate several causal estimands in recurrent event settings, with and without competing events. We clarify when commonly used classical statistical estimands can be interpreted as causal quantities from the causal mediation literature, such as (controlled) direct effects and total effects. Furthermore, we show that recent results on interventionist mediation estimands allow us to define new causal estimands with recurrent and competing events that may be of particular clinical relevance in many subject matter settings. We use causal directed acyclic graphs and single world intervention graphs to illustrate how to reason about identification conditions for the various causal estimands based on subject matter knowledge. Furthermore, using results on counting processes, we show that our causal estimands and their identification conditions, which are articulated in discrete time, converge to classical continuous time counterparts in the limit of fine discretizations of time.

stat.ME↗

arXiv2024

Variant Specific Treatment Effects with Applications in Vaccine Studies

Pathogens usually exist in heterogeneous variants, like subtypes and strains. Quantifying treatment effects on the different variants is important for guiding prevention policies and treatment development. Here we ground analyses of variant-specific effects on a formal framework for causal inference. This allows us to clarify the interpretation of existing methods and define new estimands. Unlike most of the existing literature, we explicitly consider the (realistic) setting with interference in the target population: even if individuals can be sensibly perceived as iid in randomized trial data, there will often be interference in the target population where treatments, like vaccines, are rolled out. Thus, one of our contributions is to derive explicit conditions guaranteeing that commonly reported vaccine efficacy parameters quantify well-defined causal effects, also in the presence of interference. Furthermore, our results give alternative justifications for reporting estimands on the relative, rather than absolute, scale. We illustrate the findings with an analysis of a large HIV1 vaccine trial, where there is interest in distinguishing vaccine effects on viruses with different genome sequences.

stat.AP↗

arXiv2024

Improved bounds and inference on optimal regimes

Point identification of causal effects requires strong assumptions that are unreasonable in many practical settings. However, informative bounds on these effects can often be derived under plausible assumptions. Even when these bounds are wide or cover null effects, they can guide practical decisions based on formal decision theoretic criteria. Here we derive new results on optimal treatment regimes in settings where the effect of interest is bounded. These results are driven by consideration of superoptimal regimes; we define regimes that leverage an individual's natural treatment value, which is typically ignored in the existing literature. We obtain (sharp) bounds for the value function of superoptimal regimes, and provide performance guarantees relative to conventional optimal regimes. As a case study, we consider a commonly studied Marginal Sensitivity Model and illustrate that the superoptimal regime can be identified when conventional optimal regimes are not. We similarly illustrate this property in an instrumental variable setting. Finally, we derive efficient estimators for upper and lower bounds on the superoptimal value in instrumental variable settings, building on recent results on covariate adjusted Balke-Pearl bounds. These estimators are applied to study the effect of prompt ICU admission on survival.

stat.ME↗

arXiv2024

Optimal regimes with limited resources

Policy-makers are often faced with the task of distributing a limited supply of resources. To support decision-making in these settings, statisticians are confronted with two challenges: estimands are defined by allocation strategies that are functions of features of all individuals in a cluster; and relatedly the observed data are neither independent nor identically distributed when individuals compete for resources. Existing statistical approaches are inadequate because they ignore at least one of these core features. As a solution, we develop theory for a general policy class of dynamic regimes for clustered data, covering existing results in classical and interference settings as special cases. We cover policy-relevant estimands and articulate realistic conditions compatible with resource-limited observed data. We derive identification and inference results for settings with a finite number of individuals in a cluster, where the observed dataset is viewed as a single draw from a super-population of clusters. We also consider asymptotic estimands when the number of individuals in a cluster is allowed to grow; under explicit conditions, we recover previous results, thereby clarifying when the use of existing methods is permitted. Our general results lay the foundation for future research on dynamic regimes for clustered data, including the longitudinal cluster setting.

stat.ME↗

arXiv2024

Rejoinder to "Perspectives on `harm' in personalized medicine -- an alternative perspective"

In our original article (Sarvet & Stensrud, 2024), we examine twin definitions of "harm" in personalized medicine: one based on predictions of individuals' unmeasurable response types (counterfactual harm), and another based solely on the observations of experiments (interventionist harm). In their commentary, Mueller & Pearl (2024) (MP) read our review as an argument that "counterfactual logic should [...] be purged from consideration of harm and benefit" and "strongly object [...] that a rational decision maker may well apply the interventional perspective to the exclusion of counterfactual considerations." Here we show that this objection is misguided. We analyze MP's examples and derive a general result, showing that determinations of harm through interventionist and counterfactual analyses will always concur. Therefore, individuals who embrace counterfactual formulations and those who object to their use will make equivalent decisions in uncontroversial settings.

stat.AP↗

arXiv2024

Optimal regimes for algorithm-assisted human decision-making

We consider optimal regimes for algorithm-assisted human decision-making. Such regimes are decision functions of measured pre-treatment variables and, by leveraging natural treatment values, enjoy a "superoptimality" property whereby they are guaranteed to outperform conventional optimal regimes. When there is unmeasured confounding, the benefit of using superoptimal regimes can be considerable. When there is no unmeasured confounding, superoptimal regimes are identical to conventional optimal regimes. Furthermore, identification of the expected outcome under superoptimal regimes in non-experimental studies requires the same assumptions as identification of value functions under conventional optimal regimes when the treatment is binary. To illustrate the utility of superoptimal regimes, we derive new identification and estimation results in a common instrumental variable setting. We use these derivations to analyze examples from the optimal regimes literature, including a case study of the effect of prompt intensive care treatment on survival.

stat.ME↗

arXiv2024

Perspectives on harm in personalized medicine

Avoiding harm is an uncontroversial aim of personalized medicine and other epidemiologic initiatives. However, the precise mathematical translation of "harm" is disputable. Here we use a formal causal language to study common, but distinct, definitions of "harm". We clarify that commitment to a definition of harm has important practical and philosophical implications for decision making. We relate our practical and philosophical considerations to ideas from medical ethics and legal practice.

stat.AP↗

arXiv2023

Causal effects of intervening variables in settings with unmeasured confounding

We present new results on average causal effects in settings with unmeasured exposure-outcome confounding. Our results are motivated by a class of estimands, e.g., frequently of interest in medicine and public health, that are currently not targeted by standard approaches for average causal effects. We recognize these estimands as queries about the average causal effect of an intervening variable. We anchor our introduction of these estimands in an investigation of the role of chronic pain and opioid prescription patterns in the opioid epidemic, and illustrate how conventional approaches will lead unreplicable estimates with ambiguous policy implications. We argue that our altenative effects are replicable and have clear policy implications, and furthermore are non-parametrically identified by the classical frontdoor formula. As an independent contribution, we derive a new semiparametric efficient estimator of the frontdoor formula with a uniform sample boundedness guarantee. This property is unique among previously-described estimators in its class, and we demonstrate superior performance in finite-sample settings. Theoretical results are applied with data from the National Health and Nutrition Examination Survey.

stat.ME↗