Causal Forest Model Identifies Immunocompromised Patients Benefiting From High Flow Oxygen

A machine-learning analysis derived from the HIGH trial demonstrates that although high flow nasal oxygen does not produce a statistically significant average reduction in day-28 invasive mechanical ventilation across all immunocompromised patients presenting with acute hypoxemic respiratory failure, treatment responses vary substantially depending on baseline individual characteristics. The study, published in Intensive Care Medicine by Alexis Maillard, Elie Azoulay, and colleagues, evaluated whether specific clinical profiles influence the effectiveness of high flow nasal oxygen compared to conventional oxygen therapy.

Acute hypoxemic respiratory failure serves as the leading cause of intensive care unit admission among immunocompromised patients. While high flow nasal oxygen has emerged as a common first-line clinical practice for managing acute hypoxemic respiratory failure, a previous randomized trial reported no overall benefit from high flow nasal oxygen within this vulnerable patient group. In response to this lack of overall efficacy, the investigators sought to determine whether underlying baseline patient characteristics could identify individuals who might still experience clinical benefit in preventing the need for invasive mechanical ventilation.

To investigate these heterogeneous treatment responses, the researchers applied a causal forest machine-learning algorithm to data from the randomized HIGH trial, which originally enrolled 776 patients allocated to receive either high flow nasal oxygen or conventional oxygen. Patients with an established do-not-intubate order, numbering 72 individuals, were excluded from the analysis, leaving an active study population of 704 participants. The predictive model was trained to estimate the individualized effects of high flow nasal oxygen on the incidence of day-28 invasive mechanical ventilation, and its performance was internally validated using 500 bootstrap iterations before participants were categorized into terciles according to their predicted level of treatment benefit.

Across the 704 analyzed participants, the causal forest algorithm identified substantial treatment heterogeneity, revealing predicted outcomes that ranged from a 20.2 percent absolute risk reduction in day-28 invasive mechanical ventilation with high flow nasal oxygen (interquartile range, 15.5 percent to 23.9 percent) to a 7.4 percent risk reduction favoring conventional oxygen (interquartile range, 2.9 percent to 11.2 percent). Patients categorized in the subgroup with the highest predicted benefit from high flow nasal oxygen were characteristically older, more frequently solid organ transplant recipients, less neutropenic, and exhibited more severe hypoxemia and tachycardia, while simultaneously presenting lower non-respiratory sequential organ failure assessment scores. Within this specific high-benefit subgroup, high flow nasal oxygen was associated with a statistically significant lower incidence of invasive mechanical ventilation.

The investigators also assessed the external transportability of their causal forest model within an independent observational cohort consisting of 818 patients. In this external cohort, the model preserved its ability to rank patient benefit, although the absolute effect estimates were less calibrated than in the original trial cohort. The researchers concluded that high flow nasal oxygen might provide meaningful clinical benefits for select immunocompromised patients presenting with acute hypoxemic respiratory failure, while emphasizing that further prospective studies remain necessary to validate these findings before broader clinical implementation.

Subject of Research: Individualized treatment effects of high flow nasal oxygen in immunocompromised patients

Article Title: Individualized treatment effects of high flow nasal oxygen in immunocompromised patients

Article References: Maillard, A., Métais, M., Lemiale, V., Mokart, D., Moreau, A.-S., Canet, E., Kouatchet, A., Buell, K. G., Jaber, S., Argaud, L., Wallet, F., Pène, F., Demoule, A., & Azoulay, E. (2026). Individualized treatment effects of high flow nasal oxygen in immunocompromised patients. Intensive Care Medicine. https://doi.org/10.1007/s00134-026-08635-8

Image Credits: AI Generated

DOI: 10.1007/s00134-026-08635-8

Keywords: high flow nasal oxygen, acute hypoxemic respiratory failure, immunocompromised patients, invasive mechanical ventilation, causal forest, intensive care medicine, HIGH trial, machine learning, Individualized, treatment, effects, high