Novel g-computation algorithms for time-varying actions with recurrent and semi-competing events

Alena Sorensen D'Alessio; Chantel L. Martin; Jessie K Edwards; Lucas M. Neuroth; Paul N Zivich

arxiv: 2603.10169 · v2 · submitted 2026-03-10 · 📊 stat.ME · stat.AP

Novel g-computation algorithms for time-varying actions with recurrent and semi-competing events

Alena Sorensen D'Alessio , Lucas M. Neuroth , Jessie K Edwards , Chantel L. Martin , Paul N Zivich This is my paper

Pith reviewed 2026-05-15 12:59 UTC · model grok-4.3

classification 📊 stat.ME stat.AP

keywords g-computationsemi-competing eventstime-varying confoundingcausal inferencehypertensionsmokingepidemiologyMonte Carlo simulation

0 comments

The pith

Novel g-computation algorithms estimate causal effects with semi-competing events and time-varying actions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops two new g-computation algorithms to assess the effects of time-varying public health interventions when a terminal event such as death can prevent observation of non-terminal events like disease onset. This addresses a gap in epidemiology where long follow-up periods make semi-competing events common and time-varying confounding complicates causal inference. Simulations demonstrate that the new estimators have little bias, proper confidence interval coverage, and perform better than existing methods across different sample sizes. An application to data on smoking prevention and hypertension prevalence in midlife shows a small protective effect of never smoking.

Core claim

We propose two novel g-computation algorithms for causal effects with semi-competing events and time-varying actions. Our simulations show that the novel g-computation estimators had little bias and appropriate confidence interval coverage. They outperformed existing alternative estimators across sample sizes. In the illustrative application, the novel estimator identified a small reduction in prevalence of hypertension and risk of death in midlife had all cigarette smoking been prevented across follow-up compared to the observed smoking patterns.

What carries the argument

The novel g-computation algorithms that sequentially model the conditional distributions of the outcome and treatment processes while accounting for the semi-competing structure and time-varying confounding.

If this is right

Long-running cohort studies of aging can now estimate intervention effects without bias from death as a competing event.
The methods provide a way to study how preventing behaviors like smoking affects chronic disease prevalence despite mortality.
Existing alternative estimators can be replaced with these for better performance in similar settings.
Applications to life course analyses become feasible as death becomes more prevalent in samples.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

These algorithms might be adapted to handle recurrent non-terminal events more explicitly in future extensions.
Applying the method to other datasets with time-varying exposures could reveal intervention effects on other aging outcomes.
Relaxing the correct specification assumption through doubly robust versions could strengthen robustness.

Load-bearing premise

The models used in g-computation for the conditional distributions of outcomes and treatments are correctly specified and there is no unmeasured time-varying confounding.

What would settle it

Running the Monte Carlo simulations with the paper's data-generating mechanism and observing that the novel estimators have more bias or worse coverage than the existing alternatives would falsify the performance advantage.

read the original abstract

Background: A core aspect of epidemiology is determining the impacts of potential public health interventions over time. With long follow-up periods, epidemiologists may need to consider semi-competing events, in which a terminal event, like death, precludes a non-terminal event, like hypertension. Time-varying confounding poses an additional challenge when studying time-varying interventions or actions. Existing methods do not simultaneously address semi- competing events and time-varying confounding. Methods: We propose two novel g-computation algorithms for causal effects with semi- competing events and time-varying actions. To explore performance of our novel g-computation estimators, we conducted a Monte Carlo simulation study. We then applied our estimator to investigate how cigarette smoking prevention throughout young and middle adulthood might impact prevalent hypertension using data from Waves III (aged 18-26 years) - VI (aged 39-51 years) of the National Longitudinal Study of Adolescent to Adult Health. Results: Our simulations show that the novel g-computation estimators had little bias and appropriate confidence interval coverage. They outperformed existing alternative estimators across sample sizes. In the illustrative application, the novel estimator identified a small reduction in prevalence of hypertension and risk of death in midlife had all cigarette smoking been prevented across follow-up compared to the observed smoking patterns. Conclusion: As long-running cohorts progress in age, death within the study sample will become an increasing concern for studies of aging-related outcomes, life course analyses, and investigations into chronic disease development. Our novel g-computation estimators provide a simultaneous solution.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Two new g-computation algorithms for semi-competing events plus time-varying actions, backed by simulations that assume correct specification.

read the letter

The paper introduces two g-computation algorithms that estimate causal effects when interventions change over time and a terminal event like death can preclude observing non-terminal outcomes such as hypertension. Existing tools do not handle both features together, so this targets a practical gap in long-running cohort studies of aging and chronic disease. The Monte Carlo results show low bias, nominal coverage, and better performance than alternatives across sample sizes, and the Add Health application produces a small estimated reduction in midlife hypertension prevalence and death risk under a no-smoking intervention. That is the concrete advance. The simulations and application rest on the usual identifying assumptions plus correct specification of the conditional models for outcomes, treatments, and recurrent events. The stress-test note is on target here: the reported performance properties are shown under data-generating processes that match the estimator's parametric families, with no additional checks for misspecification or residual confounding that would appear in real data. The application effect is small and plausible, but without those robustness checks the transferability remains untested. This work is aimed at epidemiologists and methodologists who need tools for time-varying causal questions in settings with competing terminal events. Readers working on life-course or aging analyses will find usable algorithms and an illustrative example. It deserves a serious referee because it fills a documented methodological hole with new estimators and supporting simulation evidence, even though the robustness section would benefit from expansion.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes two novel g-computation algorithms for estimating causal effects of time-varying actions in the presence of recurrent and semi-competing events (e.g., death precluding non-terminal outcomes like hypertension), while handling time-varying confounding. Monte Carlo simulations are used to assess performance, claiming little bias, appropriate confidence interval coverage, and outperformance over alternatives across sample sizes. The method is then applied to Waves III-VI of the National Longitudinal Study of Adolescent to Adult Health to estimate the effect of preventing cigarette smoking throughout young and middle adulthood on midlife hypertension prevalence and mortality risk, yielding small reductions.

Significance. If the estimators are valid and the simulation properties hold beyond idealized conditions, this would fill a methodological gap in causal inference for longitudinal epidemiologic studies involving semi-competing events and time-varying interventions, which is increasingly relevant as cohorts age. The simulation results and real-data illustration demonstrate potential utility for life-course analyses of chronic disease, though the small application effects highlight the need for further validation.

major comments (2)

[Monte Carlo simulation study] Monte Carlo simulation study: The data-generating mechanisms match the parametric families assumed by the novel g-computation estimators (correct specification of conditional outcome, treatment, and recurrent-event distributions plus no unmeasured time-varying confounding). No additional scenarios probe misspecification or residual confounding, so the reported low bias, nominal coverage, and outperformance do not necessarily transfer to the Add Health application where these conditions may not hold exactly.
[Application to Add Health data] Application section: The identifying assumptions required for g-computation with semi-competing events (no unmeasured time-varying confounding, correct handling of the terminal event) are invoked but not subjected to sensitivity analyses or explicit verification in the Add Health data; given the small reported effects on hypertension prevalence and death risk, it is unclear whether the conclusions are robust to plausible violations.

minor comments (2)

[Abstract] The abstract conclusion could more explicitly note the dependence on correct model specification and the identifying assumptions.
[Methods] Notation for the recurrent-event and semi-competing processes could be clarified with an explicit directed acyclic graph or timeline diagram to aid readers unfamiliar with the setting.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments on our manuscript. We address each major comment point by point below, providing the strongest honest defense of our work while acknowledging limitations where they exist. We have revised the manuscript to incorporate additional analyses as described.

read point-by-point responses

Referee: Monte Carlo simulation study: The data-generating mechanisms match the parametric families assumed by the novel g-computation estimators (correct specification of conditional outcome, treatment, and recurrent-event distributions plus no unmeasured time-varying confounding). No additional scenarios probe misspecification or residual confounding, so the reported low bias, nominal coverage, and outperformance do not necessarily transfer to the Add Health application where these conditions may not hold exactly.

Authors: We agree that the primary simulation scenarios assume correct model specification, which is a standard and necessary first step to verify that the estimators are unbiased and have proper coverage when the modeling assumptions hold. This baseline evaluation is common in methodological papers before exploring robustness. To directly address the concern, the revised manuscript will include additional Monte Carlo scenarios that introduce model misspecification (e.g., omitted time-varying covariates or incorrect link functions) and residual unmeasured time-varying confounding. These results will provide evidence on performance under more realistic conditions closer to the Add Health application. revision: yes
Referee: Application section: The identifying assumptions required for g-computation with semi-competing events (no unmeasured time-varying confounding, correct handling of the terminal event) are invoked but not subjected to sensitivity analyses or explicit verification in the Add Health data; given the small reported effects on hypertension prevalence and death risk, it is unclear whether the conclusions are robust to plausible violations.

Authors: We acknowledge that the small effect sizes make it particularly important to assess robustness to violations of the identifying assumptions. The manuscript already discusses the plausibility of no unmeasured time-varying confounding and proper handling of the terminal event in the Add Health context, drawing on the rich covariate data available. In the revision, we will add explicit sensitivity analyses, including quantitative bias analysis for unmeasured confounding and alternative specifications for the death process, to evaluate how the conclusions might change under plausible violations. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation extends standard g-computation independently

full rationale

The paper proposes two novel g-computation algorithms to handle semi-competing events and time-varying actions, extending the standard framework with new handling for recurrent events and terminal events that preclude non-terminal ones. The Monte Carlo simulations evaluate finite-sample performance under data-generating processes that match the parametric families used by the estimators, which is standard validation practice for causal estimators and does not reduce any claim to a tautology or fitted input renamed as prediction. The illustrative application to Add Health data is separate from the derivation. No self-definitional steps, load-bearing self-citations, uniqueness theorems imported from the authors' prior work, or ansatzes smuggled via citation appear in the derivation chain. The central claims rest on the explicit algorithmic construction and identifying assumptions (correct model specification, no unmeasured time-varying confounding) that are stated as prerequisites rather than derived from the results themselves.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard causal identification assumptions for g-computation in longitudinal data with competing risks; no free parameters or invented entities are introduced beyond the algorithmic extensions.

axioms (2)

domain assumption No unmeasured time-varying confounding
Required for g-computation to identify causal effects of time-varying actions.
domain assumption Correct specification of models for outcome and treatment mechanisms
G-computation relies on these models to simulate counterfactual outcomes accurately.

pith-pipeline@v0.9.0 · 5596 in / 1352 out tokens · 50845 ms · 2026-05-15T12:59:50.377640+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We propose two novel g-computation algorithms for causal effects with semi-competing events and time-varying actions... Monte Carlo simulation study... multistate illness-death-recovery model
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Our simulations show that the novel g-computation estimators had little bias and appropriate confidence interval coverage

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.