Exploring the effects of mechanical ventilator settings with modified vector-valued treatment policies

Ziren Jiang, Philip S. Crooke, John J. Marini Jared D. Huling, [8pt] Division of Biostatistics, Health Data ScienceUniversity of Minnesota, [4pt] Department of Mathematics, Vanderbilt University, [4pt] Department of Pulmonary, Critical Care MedicineUniversity of Minnesota[8pt] · 2025

Read Exploring the effects of mechanical ventilator settings with modified vector-valued treatment policies by Ziren Jiang, Philip S. Crooke, John J. Marini Jared D. Huling, [8pt] Division of Biostatistics, Health Data ScienceUniversity of Minnesota, [4pt] Department of Mathematics, Vanderbilt University, [4pt] Department of Pulmonary, Critical Care MedicineUniversity of Minnesota[8pt] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-09809.txt.

Introduction

Breathing machines save lives but can injure the lungs they rest, and a 2025 paper by Ziren Jiang and co-authors builds a cleaner way to judge their settings. Several settings act at once, and hospital rules tie them together, which makes cause and effect hard to untangle.

The team proposes a policy framework that shifts many settings together, paired with weights that balance the patient mix. Their design dodges common traps, such as impossible setting mixes and shaky assumptions, and adds checks for hidden confounding. In a large hospital database, equal cuts in total breathing power through different settings led to different expected outcomes. Lowering airway pressure looked better for survival than matching cuts made other ways.

Catalog record

CREATORZiren Jiang, Philip S. Crooke, John J. Marini Jared D. Huling, [8pt] Division of Biostatistics, Health Data ScienceUniversity of Minnesota, [4pt] Department of Mathematics, Vanderbilt University, [4pt] Department of Pulmonary, Critical Care MedicineUniversity of Minnesota[8pt]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-09809

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Exploring the effects of mechanical ventilator settings with modified vector-valued treatment policies Ziren Jiang, Philip S. Crooke, John J. Marini Jared D. Huling, [8pt] Division of Biostatistics, Health Data ScienceUniversity of Minnesota, [4pt] Department of Mathematics, Vanderbilt University, [4pt] Department of Pulmonary, Critical Care MedicineUniversity of Minnesota[8pt] Abstract Mechanical ventilation is critical for managing respiratory failure, but inappropriate ventilator settings can lead to ventilator-induced lung injury (VILI), increasing patient morbidity and mortality. Evaluating the causal impact of ventilator settings is challenging due to the complex interplay of multiple treatment variables and strong confounding due to ventilator guidelines. In this paper, we propose a modified vector-valued treatment policy (MVTP) framework coupled with energy balancing weights to estimate causal effects involving multiple continuous ventilator parameters simultaneously in addition to sensitivity analysis to unmeasured confounding. Our approach mitigates common challenges in causal inference for vector-valued treatments, such as infeasible treatment combinations, stringent positivity assumptions, and interpretability concerns. Using the MIMIC-III database, our analyses suggest that equal reductions in the total power of ventilation (i.e., the mechanical power) through different ventilator parameters result in different expected patient outcomes. Specifically, lowering airway pressures may yield greater reductions in patient mortality compared to proportional adjustments of tidal volume alone. Moreover, controlling for respiratory-system compliance and minute ventilation, we found a significant benefit of reducing driving pressure in patients with acute respiratory distress syndrome (ARDS). Our analyses help shed light on the contributors to VILI. Keywords— Causal inference, observational studies, mechanical ventilation, covariate balancing weights, sensitivity analysis 1 Introduction Mechanical ventilation is a lifesaving intervention for patients with respiratory failure, providing critical support to maintain adequate gas exchange. However, the application of mechanical ventilation is not without risks. With each breath delivered by a mechanical ventilator, a specific amount of energy is transferred to the respiratory system, primarily to overcome airway resistance and expand lung tissue. Only a portion of this energy is recovered through the elastic recoil of the lung, while the remainder is dissipated in the lung parenchyma ( Busana et al. 2022 ) . If ventilator settings are not carefully optimized, this process may result in ventilator-induced lung injury (VILI), a phenomenon believed linked to poor outcomes, including increased morbidity and mortality ( Huhle et al. 2018 ) . Understanding how the various settings impact mortality is thus of great interest. The process of one ventilation cycle in terms of the airway pressure is illustrated in the left panel of Figure 3 . Each cycle can be characterized by the following ventilator settings: (1) tidal volume ( {V_{T}} ), which represents the amount of air delivered per ventilation cycle; (2) respiratory rate ( {{RR}} ), which is the frequency at which ventilation cycles occur per minute; (3) peak inspiratory pressure ( {P_{\textnormal{peak}}} ), which is the maximum pressure applied to the lungs during inhalation (4) plateau pressure ( {P_{\textnormal{plateau}}} ), which is the static airway pressure when airflow is paused at the end of inhalation; and (5) positive end-expiratory pressure ( PEEP ), which is the baseline pressure maintained at the end of exhalation to prevent alveolar collapse. The possibility of VILI has long been recognized in clinical practice, prompting extensive research aimed at understanding its underlying mechanisms and developing lung-protective ventilation strategies to mitigate this complication. Pivotal randomized controlled trials (RCTs) showed that mechanical ventilation with low tidal volumes ( {V_{T}} ) improves patient survival ( Amato et al. 1998 ; Acute Respiratory Distress Syndrome Network 2000 ) . Various variables alone have been demonstrated experimentally to associate with VILI, including respiratory rate ( RR ) ( Rich et al. 2003 ) , and driving pressure (defined as \textnormal{DP}=P_{\textnormal{plateau}}-\textnormal{PEEP} ) ( Bellani et al. 2011 ) . However, each setting contributes to a different aspect of ventilation and their combination is what results in a particular ventilation cycle. Under passive conditions, the product of energy per cycle and RR per minute has become known in medical practice as ‘power’. Mechanical power ( MP ) is defined as: \textnormal{MP}=0.098\times V_{T}\times\textnormal{RR}\times[P_{\textnormal{peak}}-\frac{1}{2}\times(P_{\textnormal{plateau}}-\textnormal{PEEP})], (1) which is a putative summary metric that aggregates the contribution of the ventilator settings to VILI ( Neto et al. 2018 ; Vasques et al. 2018 ) . Reasoning that energy is required to accomplish alveolar stretch, the rate of mechanical energy delivery per minute has been viewed as a unifying metric to quantify VILI risk—whatever its components. Literature has shown that the larger values of MP are associated with higher mortality ( Neto et al. 2018 ) , indicating that indeed ventilator settings may cause harm to patients. However, as a summary measurement, MP is not directly manipulable, making a single causal relationship between MP and mortality ill-posed. Further, Marini and Rocco 2020 hypothesized that it ‘ seems unlikely that all combinations of frequency, tidal volume, and pressure (flow resistive pressure, driving pressure, and PEEP) that sum to the same power value are equally dangerous to a given lung .’ Providing sharper insights into the different contributors of VILI are thus essential for providing better ventilation recommendations. However, existing literature with observational studies has primarily focused on a single ventilation variable at each time, which may lead to biased conclusions because settings are not set in isolation from each other in practice. Only targeting one of the settings at a time, manipulating one setting among the many that exist may intrinsically change the value of other settings, thus entangling the interpretation of the causal effect. For example, {P_{\textnormal{plateau}}} and {P_{\textnormal{peak}}} both measure respiratory system pressures, though at different ventilation stages, and exhibit a strong empirical correlation of 0.86 in our motivating dataset. Thus, assessment of the causal effect of reducing {P_{\textnormal{peak}}} alone implicitly involves comparing patients who have lower {P_{\textnormal{plateau}}} . Methods that explicitly allow for joint manipulation of treatments are thus needed to provide clearer insights into the causes of VILI. In this work, we analyze the MIMIC-III (Medical Information Mart for Intensive Care) dataset Johnson et al. 2016 with simultaneous consideration of multiple variables for mechanical ventilation in exploring their relationship to ICU patient mortality. We use MIMIC-III in particular, as the time-frame during which data were collected was prior to a shift in ventilator use due to guideline changes in part as a result of the analyses of Amato et al. 2015 of data from several randomized trials. Post 2015, variation in ventilation use is likely less ( Del Sorbo et al. 2017 ) , making positivity violations even more severe and thus causal analyses more challenging. 1.1 Causal inference for vector-valued treatments Consider data \{\boldsymbol{X}_{i},{\boldsymbol{A}}_{i},Y_{i}\}_{i=1}^{n} collected from an observational study, where for subject i=1,...,n , {\boldsymbol{X}}_{i}\in\mathcal{X}\subseteq\mathbb{R}^{p} denotes a p -dimensional vector of pre-treatment covariates, {\boldsymbol{A}}_{i}\in\mathcal{A}\subseteq\mathbb{R}^{k} denotes the vector of k real-valued intervention variables, and Y_{i}\in\mathbb{R} denotes the outcome. We are interested in assessing the causal effects of manipulating multiple treatment variables (exposures) simultaneously. While this work focuses on mechanical ventilation, our methods have utility in far broader areas of research, including the impact of health policy on childhood obesity ( Williams and Crespi 2020 ) , environmental health considering the joint effects of different pollutants ( Zhu et al. 2020 ) and many others ( Wages et al. 2011 ) . Several approaches have been proposed for estimating

Read the complete text