A Novel Index to Predict the Failure of High-Flow Nasal Cannula in Patients with Acute Hypoxemic Respiratory Failure: A Pilot Study

When to intubate in acute hypoxaemic respiratory failure? Options and opportunities for evidence-informed decision making in the intensive care unit

The optimal timing of intubation in acute hypoxaemic respiratory failure is uncertain and became a point of controversy during the COVID-19 pandemic. Invasive mechanical ventilation is a potentially life-saving intervention but carries substantial risks, including injury to the lungs and diaphragm, pneumonia, intensive care unit-acquired muscle weakness, and haemodynamic impairment. In deciding when to intubate, clinicians must balance premature exposure to the risks of ventilation with the potential harms of unassisted breathing, including disease progression and worsening multiorgan failure. Currently, the optimal timing of intubation is unclear. In this Personal View, we examine a range of parameters that could serve as triggers to initiate invasive mechanical ventilation. The utility of a parameter (eg, the ratio of arterial oxygen tension to fraction of inspired oxygen) to predict the likelihood of a patient undergoing intubation does not necessarily mean that basing the timing of intubation on that parameter will improve therapeutic outcomes. We examine options for clinical investigation to make progress on establishing the optimal timing of intubation.

Diaphragm excursions as proxy for tidal volume during spontaneous breathing in invasively ventilated ICU patients

There is a need to monitor tidal volume in critically ill patients with acute respiratory failure, given its relation with adverse clinical outcome. However, quantification of tidal volume in non-intubated patients is challenging. In this proof-of-concept study, we evaluated whether ultrasound measurements of diaphragm excursion could be a valid surrogate for tidal volume in patients with respiratory failure. Diaphragm excursions and tidal volumes were simultaneously measured in invasively ventilated patients (N = 21) and healthy volunteers (N = 20). Linear mixed models were used to estimate the ratio between tidal volume and diaphragm excursion. The tidal volume-diaphragm excursion ratio was 201 mL/cm in ICU patients [95% confidence interval (CI) 161-240 mL/cm], and 361 (294-428) mL/cm in healthy volunteers. An excellent association was shown within participants (R2 = 0.96 in ICU patients, R2 = 0.90 in healthy volunteers). However, the differences between observed tidal volume and tidal volume as predicted by the linear mixed models were considerable: the 95% limits of agreement in Bland-Altman plots were ± 91 mL in ICU patients and ± 396 mL in healthy volunteers. Likewise, the variability in tidal volume estimation between participants was large. This study shows that diaphragm excursions measured with ultrasound correlate with tidal volume, yet quantification of absolute tidal volume from diaphragm excursion is unreliable.

Optimal machine learning methods for prediction of high-flow nasal cannula outcomes using image features from electrical impedance tomography

BACKGROUND

High-flow nasal cannula (HNFC) is able to provide ventilation support for patients with hypoxic respiratory failure. Early prediction of HFNC outcome is warranted, since failure of HFNC might delay intubation and increase mortality rate. Existing methods require a relatively long period to identify the failure (approximately 12 h) and electrical impedance tomography (EIT) may help identify the patient's respiratory drive during HFNC.

OBJECTIVES

This study aimed to investigate a proper machine-learning model to predict HFNC outcomes promptly by EIT image features.

METHODS

The Z-score standardization method was adopted to normalize the samples from 43 patients who underwent HFNC and six EIT features were selected as model input variables through the random forest feature selection method. Machine-learning methods including discriminant, ensembles, k-nearest neighbour (KNN), artificial neural network (ANN), support vector machine (SVM), AdaBoost, xgboost, logistic, random forest, bernoulli bayes, gaussian bayes and gradient-boosted decision trees (GBDT) were used to build prediction models with the original data and balanced data proceeded by the synthetic minority oversampling technique.

RESULTS

Prior to data balancing, an extremely low specificity (less than 33.33%) as well as a high accuracy in the validation data set were observed in all the methods. After data balancing, the specificity of KNN, xgboost, random forest, GBDT, bernoulli bayes and AdaBoost significantly reduced (p<0.05) while the area under curve did not improve considerably (p>0.05); and the accuracy and recall decreased significantly (p<0.05).

CONCLUSIONS

The xgboost method showed better overall performance for balanced EIT image features, which may be considered as the ideal machine learning method for early prediction of HFNC outcomes.

Analysis of risk factors for the failure of respiratory support with high-flow nasal cannula oxygen therapy in children with acute respiratory dysfunction: A case-control study

BACKGROUND

Evidence-based clinical practice guidelines regarding high-flow nasal cannula (HFNC) use for respiratory support in critically ill children are lacking. Therefore, we aimed to determine the risk factors for early HFNC failure to reduce the failure rate and prevent adverse consequences of HFNC failure in children with acute respiratory dysfunction.

METHODS

Demographic and laboratory data were compared among patients, admitted to the pediatric intensive care unit between January 2017 and December 2018, who were included in a retrospective cohort study. Univariate and multivariate analyses were performed to determine risk factors for eventual entry into the predictive model for early HFNC failure and to perform an external validation study in a prospective observational cohort study from January to February 2019. Further, the association of clinical indices and trends pre- and post-treatment with HFNC treatment success or failure in these patients was dynamically observed.

RESULTS

In total, 348 pediatric patients were included, of these 282 (81.0%) were included in the retrospective cohort study; HFNC success was observed in 182 patients (64.5%), HFNC 0-24 h failure in 74 patients (26.2%), and HFNC 24-48 h failure in 26 patients (9.2%). HFNC 24 h failure was significantly associated with the pediatric risk of mortality (PRISM) III score [odds ratio, 1.391; 95% confidence interval (CI): 1.249-1.550], arterial partial pressure of carbon dioxide-to-arterial partial pressure of oxygen (PaCO₂/PaO₂) ratio (odds ratio, 38.397; 95% CI: 6.410-230.013), and respiratory rate-oxygenation (ROX) index (odds ratio, 0.751; 95% CI: 0.616-0.915). The discriminating cutoff point for the new scoring system based on the three risk factors for HFNC 24 h failure was ≥ 2.0 points, with an area under the receiver operating characteristic curve of 0.794 (95% CI, 0.729-0.859, P < 0.001), sensitivity of 68%, and specificity of 79%; similar values were noted on applying the model to the prospective observational cohort comprising 66 patients (AUC = 0.717, 95% CI, 0.675-0.758, sensitivity 83%, specificity 44%, P = 0.009). In this prospective cohort, 11 patients with HFNC failure had an upward trend in PaCO₂/PaO₂ ratio and downward trends in respiratory failure index (P/F ratio) and ROX index; however, opposite directions of change were observed in 55 patients with HFNC success. Furthermore, the fractional changes (FCs) in PaCO₂/PaO₂ ratio, P/F ratio, percutaneous oxygen saturation-to-fraction of inspired oxygen (S/F) ratio, and ROX index at 2 h post-HFNC therapy onset were statistically significant between the two groups (all, P < 0.05).

CONCLUSION

In the pediatric patients with acute respiratory insufficiency, pre-treatment PRISM III score, PaCO₂/PaO₂ ratio, and ROX index were risk factors for HFNC 24 h failure, and the direction and magnitude of changes in the PaCO₂/PaO₂ ratio, P/F ratio, and ROX index before and 2 h after HFNC treatment were warning indicators for HFNC 24 h failure. Further close monitoring should be considered for patients with these conditions.