Impact of ventilation sufficiency in prognosis of high-flow treated hypoxemic respiratory failure: A retrospective study

Abstract

Using high-flow nasal cannula (HFNC) in patients with hypoxemic respiratory failure to avoid intubation raises concerns about its potential to increase mortality due to delayed intubation. Identifying at-risk patients is essential. While the literature predicts risk with oxygen-based indices (ROX, SpO₂/FiO₂, PaO₂/FiO₂), we aimed to detect ventilation insufficiency. To achieve this, we developed a simplified and modified “ventilatory ratio” (VRm), assessed its predictive value, and compared it with the ROX index. In this retrospective, observational study, patients in the intensive care unit who received high-flow nasal cannula therapy were divided into 2 groups: a failure group, consisting of those who underwent intubation, and a success group. After excluding metabolic and organic disorders that could affect ventilation (such as chronic obstructive pulmonary disease, obesity hypoventilation, renal failure, and sepsis), data from 15 patients (7 non-intubated, 8 intubated) were analyzed. There were no significant differences between the 2 groups in terms of age, Apache II score, length of stay in the intensive care unit, arterial blood gas values, SpO₂/FiO₂ or PaO₂/FiO₂. The ROX index and VRm values were significantly different in both groups. The diagnostic accuracy (area under the ROC curve – AUC) of the ROX index for predicting the need for intubation was significantly high, at 0.89 (95% CI: 0.63–0.99; P < .001). The optimal cutoff value for the ROX index was ≤ 3.92. The VRm index demonstrated a diagnostic accuracy of AUC = 0.93 (95% CI: 0.67–1) and was also significantly high (P < .001). The optimal cutoff value for VRm was determined to be >2.68. Comparison of ROC curves between the ROX and VRm indices was not statistically significant (z = 0.41; P = .685). The VRm index may serve as a useful additional bedside tool for predicting HFNC therapy failure in hypoxemic pneumonia patients, especially in the gray zone of the ROX index.

1. Introduction

Hypoxic respiratory failure, a type of acute respiratory failure, is one of the leading causes of admission to intensive care. Although the treatment approach varies based on the severity of hypoxic respiratory failure, some patients who are managed conservatively require invasive mechanical ventilation.^[1,2] However, due to the association of invasive mechanical ventilation with increased mortality in some patient groups, high-flow nasal cannulae (HFNC) have become widespread,^[3] and with the use of HFNC, there have been hesitations regarding the timing of the transition to invasive ventilation.^[3–5] This has led to the development of easy-to-use, point-of-care indices, the PaO₂/FiO₂ ratio, the SpO₂/FiO₂ ratio, and the ROX index, which are simple indices that can be calculated at the patient’s bedside.^[6–9] These studies aimed to reduce the morbidity and mortality that could be caused by delayed intubation in patient groups monitored with HFNC. Although their validity is not yet generally accepted, these developed indices may be determinants of the need for invasive mechanical ventilation, however their adequacy remains controversial,^[10–12] and research is still ongoing to predict HFNC failure.

Hypoxic patients are usually characterized by dyspnea and tachypnea, indicating increased respiratory drive. The follow-up indices are derived from measurements of oxygen levels. However, the “Happy hypoxia” phenomenon, where there is no dyspnea–tachypnea response to hypoxia, has gained attention during the COVID-19 outbreak and can also occur in other cases of respiratory failure.^[13–15] The most important factor influencing respiratory drive is the partial pressure of carbon dioxide, rather than hypoxia. In cases of hypercarbia with normal lung mechanics, tidal volume increases before respiratory rate does.^[16] Ventilation-perfusion (V/P) imbalance directly affects respiratory drive and is fundamental to ARDS, the leading cause of hypoxemia in the ICU. It has been shown that the dead space ratio (VD/VT), a marker of V/P mismatch, rises in the early stages of ARDS.^[17,18] As the disease progresses, lung mechanics change.^[14,19] The V/P imbalance worsens, VD/VT ratio increases, and carbon dioxide removal becomes less efficient, resulting in heightened respiratory drive. Simultaneously, increased impairment of lung mechanics makes raising tidal volume difficult. Consequently, partial pressure of carbon dioxide rises further with increased respiratory workload, which is compensated by an increase in respiratory rate rather than tidal volume. Although it seems logical to assess disease progression using oxygen-related indices when monitoring patients with hypoxic respiratory failure, this approach appears insufficient based on the underlying pathophysiology. Studies have shown that indices measuring oxygen adequacy do not reliably predict mortality in intubated ARDS patients, whereas measurements like the dead space fraction, ventilatory ratio, and corrected minute ventilation are more accurate predictors.^[20–24] Therefore, we aimed to predict progression to intubation by assessing ventilation adequacy in patients with hypoxic respiratory failure secondary to pneumonia. To do this, we modified the “Ventilatory ratio” for spontaneously breathing patients and called this new version the simplified-modified ventilatory ratio (VRm). Minute ventilation (VE) depends on 2 factors: respiratory rate and tidal volume. Due to the reasons mentioned earlier, patients tend to increase their respiratory rate rather than tidal volume. Since measuring minute ventilation or tidal volume is impractical in spontaneously breathing patients, we used respiratory frequency instead of minute ventilation in the formula.^[20] The VRm formula is: Fr_tot × PaCO₂/ Fr_ideal × 37.5.

This study aimed to evaluate the VRm index as a predictor of HFNC failure in hypoxic patients secondary to pneumonia. Additionally, we aimed to reveal the relationships between VRm and the SpO₂/FiO₂, PaO₂/FiO_2, and ROX indices.

2. Materials and methods

2.1. Study design

This is a retrospective study, and data were collected from files of patients hospitalized in the Bursa City Hospital A Block, Chest Diseases Intensive Care Unit, between August 1, 2024 and February 1, 2025, after receiving approval from the Non-Interventional Ethics Committee of the Bursa City Hospital (approval number: 2025 3/11, date: February 5, 2025).

2.2. Patients

Patients whose admission to the intensive care unit from the emergency department and diagnosis were hypoxic respiratory failure secondary to pneumonia, and who were given HFNC support as respiratory support treatment, were included in the study. The following patients were excluded: patients hospitalized due to hypercapnic respiratory failure, such as chronic obstructive pulmonary disease (COPD), asthma attacks and severe emphysematous disease, those diagnosed with muscle diseases such as ALS and Guillain–Barré, those admitted to intensive care due to respiratory failure secondary to primary central nervous system pathology, intoxications, those with emergency intubation or do-not-intubate situations, patients with severe hemodynamic instability despite adequate fluid support and vasopressors, patients receiving HFNC support for diagnostic or therapeutic procedures, those with acute or chronic renal failure, those with decompensated cardiac failure, patients diagnosed with lung cancer, patients with massive pleural effusion, those who underwent pneumonectomy, those diagnosed with obesity-hypoventilation, those who received noninvasive mechanical ventilation (NIMV) with HFNC and finally, patients whose arterial blood gases were not measured.

2.3. Data collection

All respiratory and hemodynamic data of patients and treatments applied in this intensive care unit were transferred to a digital database system (intensive care information management system validated with HIMSS level 6 – CEIBA Health version 5.6.4). The CEIBA data was scanned and the oxygen fraction (FiO₂) given to the patient, patient’s minute respiratory rate (Fr_tot) and pulse oximetry (SpO₂) values obtained during the same period as the ABG, were recorded. From these values, the modified ventilation ratio [Fr_tot × PaCO₂/Fr_ideal × 37.5 (Fr_tot: patient’s respiratory rate per minute, Fr_ideal: normal respiratory frequency, taken as 12/min)] and ROX indices (SpO₂/FiO₂/Fr_tot), were calculated. Our hospital’s information management system (FONET v4.24.11.1) was utilized to gather patients’ hospitalization diagnoses, demographic data (including age and gender), comorbid conditions, length of stay in the intensive care unit, 28-day mortality rate, and duration of HFNC treatment. ABG values taken at least 1 hour after HFNC therapy in the first 24 hours of hospitalization of patients were obtained.

2.4. Management

In this intensive care unit, the primary intensive care specialist selects the respiratory support system (noninvasive ventilation, nasal high-flow cannula, or invasive mechanical ventilator) for patients based on clinical and laboratory findings. A protocol is used for all patients with hypoxemic respiratory failure. If the SpO₂ is <90% despite using a reservoir mask at 14 L/min, HFNC therapy is initiated, with initial settings of 30 L/min and FiO₂: 100%. If the patient remains tachypneic and dyspneic after the initial settings, the flow is increased by 5 minutes until it reaches 60 L/min. With the relevance of clinical signs, a sample of arterial blood gas analysis (ABG) is drawn after 1 hour. If the initial ABG shows partial oxygenation (PaO₂) improvement, FiO₂ is adjusted to reduce it, and if oxygenation fails to improve, then the flow is increased gradually. With maximal FiO₂ and flow, if the patient remains tachypneic and dyspneic, the primary physician considers intubation.

2.5. Statistical analysis

The data was evaluated using Jamovi 2.5.5.0 statistical program. Descriptive statistics are presented as the unit number (n), percentage (%), mean ± standard deviation (mean ± ss) for normally distributed quantitative variables and median (min–max) values, for non-normally distributed quantitative variables. The normal distribution of data for quantitative variables was evaluated using the Shapiro–Wilk normality test. Comparisons between groups were made via Student t test for variables that were normally distributed and via the Mann–Whitney U test for variables that were not normally distributed. The ROC analysis of the ROX index and VRm was evaluated. The cutoff value and area under the curve (AUC) were calculated for VRm and P < .05 was considered statistically significant

3. Results

A total of 214 patient files were reviewed, and 15 patients who met the criteria were included in the study. A total of 8 patients were women (53%) and 7 (47%) were men. Two patients had no comorbid conditions, and the comorbid conditions in the other patients were lymphoma (2 patients), previous cerebrovascular accident (CVA), compensated cirrhosis, HT (4 patients, 1 patient additionally had diabetes), Alzheimer (2 patients), vasculitis, and ischemic heart disease (Fig. 1).

Figure 1.

Flow diagram.

There were 8 intubated patients and 7 patients who did not require intubation. There was no significant difference between the 2 groups in terms of age, Apache II score, length of stay in the intensive care unit, or arterial blood gas values, such as pH, PO₂, PCO₂, HCO₃, FiO₂, SpO₂, SpO₂/FiO₂, or PaO₂/FiO₂. At the initial ABG analysis, 1 patient in the intubated group had a PaO₂ of 112 mm Hg, and all the other PaO₂ values were under 100 mm Hg. There was a significant difference in respiratory rates between the 2 groups (P = .006). The ROX index and VRm values were significantly different between the 2 groups (Table 1).

Table 1.

Comparisons between groups.

	HFNC failure (n:8)	HFNC success (n:7)	P-value
Female (36%)	4 (27%)	3 (20%)	.782
Male (64%)	4 (27%)	4 (27%)
Age (mean ± SD)	71 ± 13	77 ± 9	.371
APACHE II (mean ± SD)	22 ± 3	22 ± 10	.981
ICU-LOS (mean ± SD)	18 ± 11	11 ± 5	.120
Day of HFNC Median (min–max)	2.5 (1–18)	7 (3–11)	.131
pH (mean ± SD)	7.36 ± 0.11	7.45 ± 0.04	.076
PCO2 (mean ± SD)	46 ± 12	39 ± 9	.227
PaO2 (mean ± SD)	66 ± 23	75 ± 11	.259
SpO2 Median (min–max)	91 (63–95)	94 (91–95)	.218
FiO2 Median (min–max)	100 (80–100)	100 (70–100)	>.999
RR	33 ± 7	22 ± 7	.006
SpO2/fiO2 (mean ± SD)	93.6 ± (15.6)	102 ± (16.8)	.314
PaO2/FiO2 (mean ± SD)	68 ± 26.6	83.1 ± 20.6	.246
ROX	2.97 ± 0.97	4.91 ± 1.13	.003
VRm	3.34 ± 0.88	1.87 ± 0.54	.002

APACHE II = Acute Physiologic and Chronic Health Evaluation, HFNC = high-flow nasal cannula, ICU-LOS = intensive care unit length of stay, ROX = respiratory rate oxygenation index, RR = respiratory rate, VRm = simplified-modified ventilatory ratio index.

The bold values indicate statistically significant *P < .05.

The diagnostic accuracy (area under the ROC curve [AUC]) of the ROX index for predicting the need for intubation was significantly high at 0.89 (95% CI: 0.63–0.99; P < .001). The optimal cutoff value for the ROX index was ≤ 3.92, yielding a sensitivity of 87.50%, specificity of 85.71% and a Youden index of 0.73. The intubation rate was 87.50% in patients with ROX ≤ 3.92, whereas it was14.30% in patients with ROX > 3.92, indicating a significantly greater risk of intubation below this threshold (P = .005; OR: 42; 95% CI: 2.14–825.72).

The VRm index demonstrated a diagnostic accuracy of 0.93 (95% CI: 0.67–1.00), which was also significantly high (P < .001). The optimal cutoff value for VRm was determined to be >2.68, with a sensitivity of 75.00%, a specificity of 100%, and a Youden index of 0.75. The intubation rate was 85.70% in patients with VRm > 2.68, whereas it was 25% in patients with VRm ≤ 2.68, indicating a significantly greater risk of intubation above this threshold (P = .019; OR: 18; 95% CI: 1.27–255.74). A comparison of the ROC curves between the ROX and VRm indices revealed a difference in the AUC of 0.04 (95% CI: −0.14 to 0.21), which was not statistically significant (z = 0.41; P = .685).

4. Discussion

Our study shows that VRm, a simplified index, effectively predicts HFNC failure in hypoxic pneumonia patients. The ROC analysis yielded an AUC of 0.93, suggesting it is a reliable predictor of HFNC failure (Fig. 2).

Figure 2.

ROC curves for the ROX index and VRm. ROC = receiver operating characteristic curve, ROX = ratio of oxygen saturation, VRm = simplified-modified ventilatory ratio index.

In our study, the cutoff value of the ROX index was 3.9. Roca et al^[25] reported that the cutoff value for the ROX index was 4.8. Additionally, they mentioned a gray zone between 3.8 and 4.8 and suggested that patients in these ranges should undergo frequent reevaluations. The differences in cutoff values can be attributed to the variations in the homogeneity of the patient population between our study and other studies. Both patient groups in our study were from the severe hypoxic population, and this may explain why we obtained lower values of the ROX index.^[6,9,25]

In our study, the cutoff value of VRm for intubation risk was 2.6, and the AUC value was found to be quite high (AUC = 0.93 [95% CI: 0.67–1]). These findings suggest that the VRm may be a successful index for predicting the need for intubation in severe hypoxic patients, followed by HFNC therapy. VR is essentially related to the VD/VT ratio and it has been shown that the VD/VT ratio increases, even in the early stages of ARDS pathophysiology.^[17,18,26] Studies have shown that the VD/VT ratio is correlated with VR; therefore, VR may be a good predictor of mortality.^[21–23] One of the main determinants of VE (minute ventilation) in the formulation of VR is respiratory frequency, suggesting that this factor may play a greater role than tidal volume in determining VE in patients with severe ARDS. Indeed, in the study by Spinelli et al,^[14] the respiratory rate was increased by the combination of both hypoxia and hypercarbia. They also reported that the respiratory rate increased with decreasing respiratory volume due to the occurrence of lung injury and vagal stimulation, which is the pathophysiological basis supporting our hypothesis. Since the strict exclusion criterion in our study excluded metabolic and organic disorders that could affect respiratory rate, the respiratory rate may have a greater role than the respiratory volume in determining VE in the hypoxic patients in our study.

The fact that SpO₂/FiO₂ did not differ between the groups contradicts the literature.^[27,28] This may be explained by the small number of cases in our study, but it can also be explained by the decrease in the diagnostic value of SpO₂/FiO₂ at values < 92%, as mentioned in the study by Bonaventura et al.^[10] Since the average SpO₂ value of our patients was approximately 92%, we may not have been able to detect a difference between the 2 groups in our study. Both groups were almost severe hypoxic groups, and both SpO₂/FiO₂ and PaO₂/FiO₂ were found to be quite low. Similarly, we could not detect a difference between the groups for PaO₂/FiO₂, which contradicts the literature. In the study by Calle-Peña et al,^[29] the PaO₂/FiO₂ values in the successful HFNC group were >100 and were significantly different from those in the unsuccessful group. In our study, the PaO₂/FiO₂ values of the HFNC success group, similar to those of the unsuccessful group, were <100. Therefore, we can say that the same situation is valid for SPO₂/FiO₂ here. In other words, very low values of PaO₂/FiO₂ may be less reliable in predicting HFNC failure.

The most significant limitations of our study were the small sample size and the single-center setting, as well as its retrospective nature. Additionally, we evaluated the patients within a short observation period, and they were severely hypoxic. These values may not be applicable to cases with milder hypoxia who require more extended follow-up periods. Therefore, we cannot generalize our results to a broader population. The small number of cases resulted from the strict exclusion criteria used in our methodology. The principal reason for exercising such strict exclusions was to obtain exclusively the respiratory rate and PCO₂ that would result from ventilation insufficiency. The respiratory rate and PCO₂ are affected by various metabolic or organic pathologies that can impact the respiratory drive, including a low Glasgow Coma Scale score, septic shock, chronic obstructive pulmonary disease and emphysematous diseases, acute or chronic renal failure, and the obesity-hypoventilation syndrome.^[16] These factors were purposefully excluded from our study. Another condition that may affect respiratory frequency and PCO₂ is increased CO₂ production (VCO₂), and our study could not eliminate this effect. Because there was a difference in respiratory rates between our groups, we cannot say that respiratory efforts and therefore increased VCO₂, were similar. It is also possible that a high-calorie diet can increase the volume of carbon dioxide (VCO₂) in the blood. Although the caloric intake of the patients in our study was unobserved, which may be an additional limitation, we can infer that since our cases were patients with respiratory distress, we remained below the target level in caloric intake on the first day due to decreased oral intake. However, Ravenscraft et al^[30] reported that the effect of CO₂ production on increased minute ventilation in the early stages of ARDS is minimal. In our cases, the effects of increased CO₂ production on minute ventilation and, indirectly, on respiratory frequency were also minimal, as observed between the groups.

5. Conclusion

Although the ROX index is noninvasive, there is a patient population in the indeterminate gray area and a more sensitive index is called for. Since these types of patients present with a serious condition, intensive care follow-up is invasive. We propose the VRm index as an additional bedside tool for predicting HFNC therapy failure in hypoxemic patients with pneumonia who are determined to be in the gray area of the ROX index.

Author contributions

Conceptualization: Melike Seyda Dagdelen.

Data curation: Melike Seyda Dagdelen.

Formal analysis: Melike Seyda Dagdelen.

Investigation: Melike Seyda Dagdelen.

Methodology: Melike Seyda Dagdelen, Ibrahim Koc.

Supervision: Ibrahim Koc.

Visualization: Melike Seyda Dagdelen.

Writing – original draft: Melike Seyda Dagdelen, Ibrahim Koc.

Writing – review & editing: Melike Seyda Dagdelen, Ibrahim Koc.