Causal interpretation of sedation depth and mortality during venovenous extracorporeal membrane oxygenation

Dear Editor

We read with great interest the multicenter cohort study by Li et al. [1], which examined early metabolic trajectories and sedation depth in 1,484 adults receiving veno-venous extracorporeal membrane oxygenation (V-V ECMO) for severe pneumonia or acute respiratory distress syndrome (ARDS). The finding that 24-h lactate was strongly associated with in-hospital mortality, together with the lower mortality observed among patients classified as receiving light sedation, provides clinically relevant hypotheses for early ECMO management. We would, however, like to raise several considerations regarding the causal interpretation of the sedation analysis.

The principal concern is the temporal ordering of covariates used to estimate the association between sedation depth and mortality. Sedation exposure was defined from the actual RASS observed during the first 24 h after ECMO initiation. Yet the multivariable model adjusted for severe acute kidney injury requiring continuous renal replacement therapy (CRRT), and the authors defined this variable as a new requirement for CRRT during the ECMO run. The propensity-score analysis also described the matched sample as balanced for severe AKI requiring CRRT, while the overlap-weighting model additionally adjusted for 24-h lactate. These variables are therefore not clearly pre-exposure confounders. Indeed, Table 2 shows substantially less CRRT use with light sedation (29.4% vs 42.9%), and the Discussion explicitly proposes a pathway whereby lighter sedation reduces vasopressor exposure, improves organ perfusion, and subsequently decreases AKI and metabolic derangement. Under that proposed causal structure, CRRT and possibly 24-h lactate may lie downstream of sedation. Conditioning on an ­intermediate can attenuate or otherwise distort the total exposure effect [2]. A reanalysis restricted to covariates established before the sedation assessment would therefore be informative. If repeated RASS and physiological measurements are available, marginal structural models or related inverse-probability weighting approaches could more appropriately address time-varying confounding affected by prior treatment [3].

A related issue concerns what is being treated as the exposure. The study categorized “sedation strategy” according to an actual documented RASS score during the first 24 h, rather than a prespecified sedation target. RASS is a validated measure for monitoring a patient’s observed level of arousal and sedation over time [4], but an observed RASS is not equivalent to a clinician-assigned strategy. The manuscript also does not specify how patients with multiple RASS assessments during the 24-h window were classified, for example by the first, lowest, median, modal, or time-weighted score. This distinction matters in V-V ECMO because sedation depth and respiratory drive evolve together with hemodynamic and pulmonary recovery. Moreover, a RASS of −2 to +1 does not by itself identify a physiologically favorable spontaneous-breathing phenotype. In severe ARDS, spontaneous effort may preserve diaphragm activity, but excessive respiratory drive can increase regional lung stress and contribute to patient self-inflicted lung injury [5]. Thus, the proposed cardiopulmonary mechanism would be strengthened by reporting the timing and aggregation rule for RASS and, where available, respiratory effort or patient-ventilator interaction measures.

Finally, clarification of the propensity-score analysis appears necessary for reproducibility. The sedation subgroup contained 932 patients with deep sedation and 351 with light sedation, yet the Results state that “396 matched pairs” were generated using 1:1 nearest-neighbor matching. Under 1:1 matching without replacement, 396 distinct pairs cannot be formed when the smaller exposure group contains only 351 patients. If matching with replacement was used, this should be stated explicitly together with the matched sample size in each group and the corresponding variance estimation; if 396 refers to total matched patients rather than pairs, the wording should be corrected. Transparent reporting of matched samples and post-match balance is central to evaluating propensity-score analyses [6]. A smaller numerical inconsistency also deserves correction: Table 1 reports light sedation in 27.9% of survivors and 16.9% of non-survivors, whereas the accompanying text states 33.5% and 22.1%, respectively.

These issues do not diminish the value of the registry or the importance of dynamic metabolic assessment during V-V ECMO. They do, however, affect whether light sedation can be interpreted as a potentially modifiable intervention rather than a marker of early physiological recovery. Clarifying the exposure definition, preserving causal time order in covariate adjustment, and resolving the matching discrepancies would substantially strengthen the clinical interpretation of this important study.

Huijuan Cheng and Yanyan Ye
Respiratory Medicine, The Second Affiliated Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, Fujian Province, China
First Clinical Medical College, Fujian University of Traditional Chinese Medicine, Fuzhou, Fujian Province, China
Feng Lu
Respiratory Medicine, The Second Affiliated Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, Fujian Province, China
lufenglud@163.com

Funding Statement

This work received no specific funding.

Ethics statement

Ethics approval was not required because this correspondence was based solely on a previously published study and involved no new human participants or identifiable individual-level data.

Consent

Not applicable.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data availability is not applicable because no new data were generated or analyzed for this correspondence.