The intensive care unit (ICU) is the primary facility for admitting critically ill patients, most of whom present with varying degrees of hypoxemia [1]. Some patients may also experience acute respiratory failure, acute respiratory distress syndrome, trauma, or respiratory dysfunction caused by other illnesses, which can be life-threatening [1]. Therefore, early implementation of supportive respiratory therapy and maintenance of oxygen saturation in critically ill patients is crucial. Currently, the main treatment approaches for respiratory support in the ICU include conventional oxygen therapy (COT), high-flow nasal cannula (HFNC) therapy, and mechanical ventilation (MV) [2].

Nasal cannula oxygen therapy is a simple and commonly used method of oxygen supplementation. However, its main limitation is the inability to deliver high oxygen concentration [2]. The maximum achievable oxygen concentration through a nasal cannula is approximately 44% at the highest flow rates, which may not be sufficient for patients with severe hypoxemia [3]. Furthermore, it remains challenging to accurately adjust the oxygen concentration delivered through a nasal cannula [3].

Mask oxygen therapy can deliver a slightly higher oxygen concentration than nasal cannula therapy [2]. However, it also presents drawbacks, such as limited precision in oxygen concentration control and interference with sputum clearance and oral intake [2]. The mask may obstruct eating and drinking, and can also impede coughing and clearance of secretions, thereby increasing the risk of respiratory infection [2].

Both nasal cannula and mask oxygen therapies have distinct advantages and disadvantages, and the choice of oxygen therapy depends largely on the condition of the patient and the severity of hypoxemia [4]. In some cases, HFNC or MV may be necessary to ensure adequate oxygenation and respiratory support [4].

HFNC delivers high oxygen concentrations and at large flow rates, allowing for rapid improvement in oxygenation among hypoxemic patients. It is generally well-tolerated by patients and may reduce the need for invasive MV [4]. Additionally, HFNC alleviates respiratory distress and decreases the work of breathing by generating a low level of positive end-expiratory pressure (PEEP) in the airways, promoting lung recruitment and improving oxygenation [5].

MV, in contrast, is a more invasive respiratory support modality that provides positive pressure ventilation to correct respiratory failure in patients with hypoxemia and hypercapnia [6]. It can also mitigate respiratory muscle fatigue and facilitate blood flow redistribution in patients with shock [6]. However, MV is associated with several potential complications, including the risk of lung injury or barotrauma from high airway pressures, which impede venous return and lead to decreased cardiac output and hypotension, as well as patient discomfort [7]. Moreover, prolonged MV increases the risk of ventilator-associated pneumonia [7].

Although HFNC therapy offers an alternative treatment option for hypoxemic patients with significant clinical benefits, it is not universally suitable for all patients. MV remains a frequently utilized life-saving intervention among hospitalized patients [8]. Therefore, clinicians often face a dilemma when choosing between these two treatment modalities. On one hand, the profound instability of many critically ill patients, combined with continuously evolving treatment regimens, means that overreliance on HFNC may delay initiation of essential interventions, including MV, thereby worsening patient outcomes [9]. On the other hand, premature initiation of MV may heighten patient resistance and reduce adherence [10].

Currently, there are no universally accepted criteria for determining the optimal timing for transitioning from HFNC to MV intervention in ICU patients, and most clinical decisions in this regard rely heavily on empirical judgment. Therefore, developing a predictive model for the transition from HFNC to MV could support clinicians in making more evidence-based treatment decisions. This study focused on ICU patients receiving HFNC therapy and aimed to predict the probability of transition to MV, providing clinicians with a simple and effective predictive model to guide optimal treatment timing. Such a prognostic model may facilitate early identification of patient deterioration, thereby enabling timely initiation of MV when necessary. On the other hand, it can help prevent unnecessary or premature intubation, thus avoiding associated discomfort and potential complications. In this manner, the prognostic model supports the delivery of more personalized and effective care for patients undergoing HFNC therapy.

In critically ill patients, hypoxemia and respiratory dysfunction remain major clinical challenges. Conventional oxygen therapy methods, such as nasal cannulas or face masks, are often limited in their effectiveness among critically ill patients. HFNC therapy is a relatively recent modality that delivers warmed and humidified oxygen to the nasal cavity through a nasal cannula, with a high gas flow rate generating a mild positive pressure in the upper respiratory tract [20]. This mechanism reduces physiological dead space and alleviates respiratory effort. The proportion of inspired oxygen (FiO₂%) can be adjusted by modifying the proportion of oxygen in the driving gas [20].

In recent years, HFNC has been increasingly applied as an effective clinical intervention to improve oxygenation in critically ill patients, with significant effects. Evidence indicates that HFNC is more comfortable and effective than face mask or nasal cannula oxygen therapy for individuals with respiratory failure of various etiologies [20]. A meta-analysis further demonstrated that, compared with conventional oxygen therapy or non-invasive mechanical ventilation (NIMV), HFNC significantly reduces intubation rates, MV requirements, and the need for respiratory support device replacement [21]. In particular, HFNC therapy, which incorporates humidification, offers significant benefits for critically ill patients.

However, as the disease progresses, MV may become necessary to optimize oxygenation during resuscitation, and HFNC has inherent limitations as a first-line resuscitative measure [9]. Therefore, early identification of patient deterioration and timely preparation for MV are crucial. A well-constructed predictive model can assist clinicians in recognizing patients at increased risk of requiring MV at an early stage.

The present study analyzed data from 4069 patients who met the inclusion criteria and received HFNC therapy, as recorded in the MIMIC-IV database. The primary endpoint of the study was the initiation of MV among HFNC-treated patients. The analysis incorporated the demographic characteristics, clinical parameters, and laboratory indicators of the patients. Eventually, body weight, Sepsis-3 score, pulmonary score in the LODS, urine output score in the OASIS score, PaO₂/FiO₂ ratio, GCS, and creatinine score in the SOFA score were identified as significant independent factors influencing the likelihood of patients receiving HFNC requiring MV. These factors were subsequently integrated into the decision curve analysis to construct the predictive model.

In this study, body weight was identified as a significant independent risk factor for patients receiving HFNC therapy who eventually required MV. These findings are consistent with previous studies demonstrating that obesity is associated with an increased likelihood of requiring MV among critically ill patients. For example, a prospective study by Tafelski et al. (2016) [22] reported that obesity was significantly associated with an increased need for MV. Similarly, a retrospective cohort study conducted by Shao et al. (2020) [23] found that obese patients were more likely to require prolonged MV.

The mechanism underlying this association may be that, following weight gain, the diaphragm is subjected to increased mechanical load from the abdominal and thoracic walls, leading to reduced lung compliance, alveolar collapse, and hypoxemia [24]. To maintain sufficient oxygen saturation (SpO₂), higher pleural pressure (Ppl) and airway pressure are required, which may subsequently reduce cardiac output, offsetting the benefit of improved SpO₂ and leading to no change or even a decrease in oxygen delivery (DO₂), a critical determinant of tissue oxygenation [25]. This study also observed that higher body weight corresponded to higher individual scores and increased risk ratios for MV, indicating that body weight is an important factor to consider when predicting the need for MV in critically ill patients receiving HFNC therapy.

The SOFA score, LODS, and OASIS are widely used tools for estimating ICU mortality, with the SOFA score being the most common measure for assessing the severity of organ dysfunction [26]. With the SOFA scoring system, creatinine is a key indicator of renal function [26]. Additionally, Bhattacharya et al. (2011) [26] demonstrated that calculating the SOFA score aids in predicting the prognosis of patients requiring MV.

In this study, variations in creatinine levels within the SOFA score indicated a direct correlation with increased risk of MV. The PaO₂/FiO₂ ratio is a critical indicator of respiratory function, reflecting the efficiency of pulmonary gas exchange under oxygen therapy [27]. Lee et al. (2021) [27] further validated that the PaO₂/FiO₂ ratio can predict the success of HFNC therapy in patients with pneumonia-induced acute respiratory failure.

The GCS is an established indicator for evaluating neurological function [28], and its value was not directly proportional to the risk of MV in this study. Overall, the findings of our study suggest that critically ill patients with severe physiological compromise may not be suitable for HFNC oxygen therapy and may instead require MV as a first-line treatment.

The nomogram established in this study revealed that patient weight, pulmonary function score, and urine output scores carry the highest predictive weights and rank among the top three indicators for evaluating the risk of MV in critically ill patients receiving HFNC therapy. This finding suggests that optimizing these three indicators may potentially reduce the likelihood of requiring MV. Consequently, focusing on these indicators during patient management may offer a promising approach to preventing the need for MV in critically ill patients receiving HFNC oxygen therapy.

Targeting interventions aimed at weight loss, improvement of pulmonary function, and optimization of urine output through interventions such as diuresis, fluid balance management, and pulmonary rehabilitation may lower the risk of MV in critically ill patients. Therefore, early identification of patients at high risk for MV using the developed nomogram can help clinicians prioritize these key indicators, potentially preventing the need for MV and improving overall patient outcomes.

Overall, this model is designed to assist frontline ICU clinicians in the early identification of deterioration among patients receiving HFNC therapy, thereby supporting timely escalation of care when appropriate, while avoiding unnecessary mechanical ventilation in patients who may still respond to non-invasive oxygen support.

This study has several limitations. First, the dataset used consisted solely of cases derived from the MIMIC-IV database, without incorporating external datasets from other reputable databases to evaluate the applicability and reliability of the model. This limitation may restrict the generalizability of the findings. Second, the study only examined the occurrence of MV during hospitalization as the primary outcome and did not conduct long-term follow-up to assess patient survival or post-discharge outcomes. Third, the analysis did not include variables related to medication usage or adjunctive therapies, which could influence respiratory status and thereby affect the accuracy of the model. Although the constructed nomogram demonstrated moderate discriminative capacity, its sensitivity and specificity were modest, underscoring the limited ability of the model to accurately classify patients at risk of HFNC failure. This limitation likely reflects the heterogeneity of critically ill patients and the constrained range of clinical variables extracted from the MIMIC-IV database. Moreover, the model demonstrated clinical usefulness primarily within low-to-moderate risk thresholds, as demonstrated by the DCA. The absence of net benefit beyond a probability threshold of 0.4 suggests limited clinical value at higher predicted risks, a common challenge in models developed from single-center ICU datasets, where high-risk populations may be underrepresented. Therefore, further extensive, well-designed prospective studies and robust external validations are necessary to enhance model performance and develop a more reliable, precise, dependable, and user-friendly predictive tool for clinical applications. Future research could integrate additional dynamic physiological measurements, laboratory biomarkers, and multi-center datasets to improve the discriminative power and clinical relevance of the model.

Furthermore, this study did not examine the potential effects of combining HFNC oxygen therapy with mask oxygen therapy, despite the clinical dilemma often encountered in choosing between these two approaches. This limitation was primarily due to the lack of relevant combination-therapy cases within the MIMIC-IV database. Future prospective investigations are warranted to evaluate whether sequential or combined use of these two modalities could improve clinical outcomes in patients with hypoxemia. By addressing these limitations, future studies can enhance the clinical value of the proposed nomogram and contribute to improved prognostic accuracy and patient outcomes among critically ill individuals receiving HFNC oxygen therapy.

We developed and validated a predictive nomogram for patients receiving HFNC therapy who ultimately required MV. The nomogram demonstrates good predictive performance and may serve as a valuable adjunct tool for clinicians considering whether to continue HFNC therapy or transition to MV in critically ill patients.

• Although high-flow nasal cannula (HFNC) therapy is widely applied in the management of respiratory failure, clinical outcomes remain heterogeneous.

• This study developed a predictive nomogram to estimate treatment outcomes in patients undergoing HFNC therapy.

• The predictive model incorporated clinically relevant variables, including body weight, Sepsis-3 status, pulmonary score, urine output score, PaO₂/FiO₂ ratio, GCS score, and creatinine score, and demonstrated strong discrimination and calibration performance.

• Internal validation using bootstrap resampling confirmed the robustness and stability of the model’s predictive performance.

• The proposed nomogram may support clinicians in making individualized, evidence-based decisions regarding the initiation and continuation of HFNC therapy.

The clinical data used to support the findings of this study were supplied by the Medical Information Mart for Intensive Care IV version 2.0 (MIMIC-IV v.2.0; https://www.physionet.org/content/mimiciv/2.0/). Although the database is publicly and freely available, researchers must complete the National Institutes of Health’s web-based course known as Protecting Human Research Participants to apply for permission to access the database.

Conceptualization: JJL. Data curation: JJL, YY. Methodology: YL. Software: YY. Writing—original draft: JJL. Writing—review and editing: YL. All authors contributed to the interpretation of data and to the final version of the manuscript. All authors contributed to revising the manuscript critically for important intellectual content. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

The MIMIC-IV v.2.0 (https://www.physionet.org/content/mimiciv/2.0/) database has received ethical approval from the institutional review boards (IRBs) at Beth Israel Deaconess Medical Center and Massachusetts Institute of Technology. Because the database does not contain protected health information, a waiver of the requirements for informed consent was included in the IRB approval.

Not applicable.

This research received no external funding.

The authors declare no conflict of interest.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/BJHM50719.