This will be a single-center, noninferiority, open-label randomized controlled trial conducted in the Emergency Medicine Department of National University Hospital, Singapore, a 1225-bed tertiary academic medical center with 24/7 specialist coverage and approximately 110,000 annual attendances. Adult patients aged 21 years and above requiring NIPPV for T2RF due to any cause will be randomized to HVNI or NIPPV. The trial has been registered in ClinicalTrials.gov (NCT07065656). The protocol was prepared in accordance with the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) 2025 statement [20]. The SPIRIT participant timeline is shown in Figure 1.

This study received ethics approval from the National Healthcare Group Domain Specific Review Board (reference 2024-4329) on February 24, 2025. Because patients with T2RF are deemed critically ill and hypoxia may affect their mental capacity, informed consent will be obtained from a legally acceptable representative (LAR) where feasible prior to randomization. If an LAR is unavailable and enrollment cannot be delayed, patients will be enrolled under emergency waiver, with delayed consent obtained from the LAR or the patient at the earliest opportunity prior to hospital discharge. This approach complies with Singapore’s regulations for clinical trials under emergency situations [21]. The waiver and delayed consent will be certified by a study investigator and an independent emergency medicine specialist. Individual participant data will be deidentified prior to sharing.

Adult patients aged 21 years and above with T2RF, defined as arterial blood gas (ABG) showing PaCO₂ above 45 mm Hg with pH below 7.35 due to any cause for whom ventilatory support is indicated in the ED will be included. Patients with do-not-resuscitate orders are eligible if NIPPV use is permitted. The clinical indications for initiating ventilatory support are identical for both HVNI and NIPPV in this trial as both modalities target the same pathophysiology of acute hypercapnic respiratory failure requiring augmented ventilation. The following patients will be excluded: patients with do-not-resuscitate orders that specifically do not allow for NIPPV, clinical suspicion or confirmed base of skull fracture or severe facial trauma precluding mask or cannula placement, vulnerable patient populations (eg, pregnant women or prisoners), refusal of consent (by the patient or LAR), hemodynamic instability requiring immediate resuscitation (eg, systolic blood pressure <90 mm Hg after initial fluids), cardiac or respiratory arrest, and immediate need for endotracheal intubation as judged by the attending physician. Additional exclusion criteria include contraindications specific to either modality: complete nasal obstruction or anatomical abnormalities precluding nasal cannula placement (contraindication to HVNI), recent upper-airway or esophageal surgery (contraindication to both NIPPV and HVNI), undrained pneumothorax (contraindication to positive pressure ventilation), active uncontrolled upper gastrointestinal bleeding with high aspiration risk, and inability to protect the airway (eg, excessive secretions with absent cough reflex).

Enrollment will be through convenience sampling of eligible ED patients triaged as “urgent” or “immediate.” All study investigators are full-time clinicians with an average of 16 to 20 shifts per month to cover as many shifts as possible for enrollment. Participants will be randomized 1:1 to HVNI or NIPPV using variable block sizes generated by an independent web-based platform [22]. The specific block sizes are not disclosed in this manuscript to prevent prediction of the allocation sequence. Allocation will be implemented via sealed, opaque, sequentially numbered envelopes, with allocation concealment maintained until randomization is completed.

Due to the nature of the interventions, participants, treating clinicians, and bedside nursing staff cannot be blinded to group allocation. ABG is analyzed at the bedside using i-STAT Alinity (Abbott Laboratories). Device allocation will not be recorded on ABG stickers or forms to minimize detection bias. The data analyst performing the primary and secondary outcome analyses will also be blinded to treatment allocation and will receive deidentified datasets with treatment arms coded (eg, group A and group B) without revealing the intervention assignment. Data entry staff who are not involved in clinical care will be trained to abstract outcomes using standardized case report forms (CRFs). Nevertheless, the primary outcome evaluated is objective and should not be confounded by lack of blinding [23].

In the HVNI arm, patients will receive oxygen therapy through the Vapotherm high-velocity therapy 2.0 via a single-prong nasal cannula (Unicorn). Initial settings are flow rate of 35 L per minute [14], temperature of 37 °C, and fraction of inspired oxygen (FiO₂) titrated up to 1.0 to achieve target oxygen saturation (SpO₂). The titration algorithm will be to maintain flow rate at 35 L per minute when tolerated to optimize CO₂ washout. If intolerance occurs (eg, agitation or discomfort), stepwise reduction in flow by 5 L per minute will be performed to the minimally tolerated level while maintaining target SpO₂. Adjustments in temperature and FiO₂ will be up to the attending physician’s discretion to achieve the target SpO₂ value. Flow rate may be uptitrated as tolerated upon assessment of respiratory rate, work of breathing, and comfort. Study investigators will verify that the single-prong cannula remains securely positioned in the nostril and has not been dislodged due to patient movement. Mouth state (open vs closed) will be recorded at each assessment time point as open-mouthed breathing may reduce the effectiveness of nasopharyngeal dead-space washout. If the cannula is found to be displaced, it will be repositioned immediately, and the event will be documented. Persistent inability to maintain cannula position will be recorded as a protocol deviation.

In the NIPPV arm, patients will receive ventilation through the Respironics V60 respirator (Philips Healthcare) with an oronasal mask of appropriate size. Initial settings are IPAP of 12 to 20 cm H₂O, EPAP of 5 to 10 cm H₂O, and FiO₂ of up to 1.0 to achieve the target SpO₂. The titration algorithm will be to adjust IPAP to alleviate dyspnea and reduce respiratory rate and PaCO₂, adjust EPAP for oxygenation and to counter intrinsic positive end-expiratory pressure when indicated, and optimize mask fit to minimize leak.

Cointerventions such as bronchodilators, systemic steroids, antibiotics, diuretics, and other standard-of-care therapies are permitted as clinically indicated and should follow departmental protocols. To reduce confounding, all cointerventions will be documented with time stamps. Initiation of major therapy changes will be avoided within 10 minutes preceding scheduled ABG draws when clinically safe.

All patients will be put on continuous pulse oximetry. For presumed acute respiratory acidosis without known chronic hypercapnia, SpO₂ will be maintained above 95%, and for known or suspected chronic hypercapnia with acute decompensation, SpO₂ will be kept between 88% and 92%. Each patient’s target range will be declared at baseline and maintained throughout the initial 60-minute assessment window unless clinical changes require adjustment. FiO₂ will be titrated to maintain the declared target. Physiologic monitoring will include vital signs per ED standard, with respiratory rate recorded at baseline, 30 minutes, and 60 minutes.

The independent attending physician may change modality or escalate care if deterioration occurs. To standardize decisions, treatment failure is defined by any of the following: persistent or worsening acidosis (pH decrease ≥0.03 from baseline or pH <7.25 with respiratory rate of >30 breaths per minute after 30 minutes of therapy, PaCO₂ increase ≥5 mm Hg at 30 minutes compared with baseline, and SpO₂ persistently below target for >5 minutes despite FiO ₂≥0.8 [or the maximal tolerated settings]), severe intolerance to the device (eg, sustained agitation or inability to tolerate the interface) despite comfort measures, and clinical signs of impending respiratory failure (eg, decreasing consciousness, marked fatigue, and hemodynamic deterioration). Intubation criteria follow departmental standards for acute respiratory failure and will be applied consistently in both arms. All crossovers and intubations will be recorded with timing and reason.

The primary outcome is the percentage change in PaCO₂ from baseline to 60 minutes after initiation of therapy. Baseline is defined as the pretreatment ABG obtained before initiation of the assigned therapy (HVNI or NIPPV). This ensures that the primary outcome reflects the ventilatory effect of the intervention itself. The 30- and 60-minute assessment time points are measured from the time of therapy initiation. The ABG results and other variables (Table 1) to be collected will be documented in real time on paper-based CRFs by trained study investigators.

Percentage change (relative difference) as the primary outcome was selected to account for the wide range of baseline PaCO₂ values expected in a heterogeneous, all-cause T2RF population. An absolute change in PaCO₂ may disproportionately favor patients with higher baseline values as they have greater room for absolute reduction. The percentage change normalizes the treatment effect relative to each patient’s starting value, facilitating more equitable comparison across the spectrum of disease severity. To address the potential limitation of relative outcomes, a prespecified sensitivity analysis using absolute change in PaCO₂ at 60 minutes will be conducted to confirm the robustness of the primary finding.

Secondary outcomes include the following: percentage change in PaCO₂ from baseline to 30 minutes; change in respiratory rate at 30 and 60 minutes; modified Borg Dyspnea Scale (0-10) at 30 and 60 minutes [24]; proportion of treatment failure (as defined above), conversion to another modality of oxygenation, and need for intubation in the ED; adverse events due to therapy rendered (eg, facial pressure injury, vomiting, or aspiration); length of intensive care unit (ICU) and hospital stay; and inpatient mortality. Inpatient mortality will be assessed at hospital discharge. ICU length of stay (if applicable) and total hospital length of stay will be recorded in days from the date of ED presentation to the date of ICU discharge and hospital discharge, respectively. These data will be abstracted from the electronic health record (EHR) following patient discharge.

A total of 84 patients (42 per group) will be enrolled to provide 80% power and one-sided α of 2.5% to demonstrate noninferiority [25] assuming an SD of 6.65% for percentage change in PaCO₂ levels before and after treatment [26] and a noninferiority margin of 4.3%. The sample size includes an estimated 10% allowance for missing primary end point data. The noninferiority margin of a 4.3% absolute difference in percentage change was informed by clinician judgment (number of specialist accredited [board-certified] emergency physicians, n=14) of clinically acceptable differences and literature on PaCO₂ responses to NIPPV. For example, emergency physicians opined that a change from 70 mm Hg to 63 mm Hg using HVNI (−10.0%) was considered noninferior to a change in PaCO₂ from 70 mm Hg to 60 mm Hg after treatment with NIPPV (−14.3%).

The primary analysis will follow an intention-to-treat approach, with supportive per-protocol analysis excluding major deviations (eg, early crossovers or missing postbaseline ABG values) and a safety set analyzed as treated. The main end point, percentage change in PaCO₂ at 60 minutes, will be assessed using analysis of covariance (ANCOVA) adjusting for baseline PaCO₂ and prespecified covariates (eg, etiology, baseline pH, and age). Noninferiority will be concluded if the upper bound of the 2-sided 95% CI for the HVNI–NIPPV adjusted difference is below the 4.3% margin. A prespecified sensitivity analysis will repeat the primary ANCOVA model using absolute change in PaCO₂ at 60 minutes as the dependent variable to assess whether the conclusions are robust to the choice of relative vs absolute outcome metric. Secondary endpoints (percentage PaCO₂ changes at 30 minutes, respiratory rate, and modified Borg Dyspnea Scale) will use similar adjusted models, whereas binary outcomes (treatment failure, intubation, and adverse events) will be compared using chi-square or Fisher exact tests and logistic regression, and length of stay and mortality will be assessed via nonparametric or time-to-event and logistic methods. Missing data handling will focus primarily on the primary outcome (percentage change in PaCO₂ at 60 minutes). Missing 60-minute ABG results may occur due to early treatment crossover, intubation before the 60-minute time point, patient refusal of repeat ABG, or technical failure of the ABG analyzer. Multiple imputation by chained equations will be performed, with the imputation model including baseline PaCO₂, baseline pH, treatment arm, age, etiology, respiratory rate, and 30-minute PaCO₂ (if available). A minimum of 20 imputed datasets will be generated, and results will be pooled using the Rubin rules.

Sensitivity analyses will include (1) complete-case analysis (excluding participants with missing primary outcome data) and (2) worst-case imputation for the HVNI arm (assigning the worst observed percentage change to missing values in the HVNI arm and the best observed percentage change to missing values in the NIPPV arm), which is a conservative approach for noninferiority trials. Baseline covariates with missing data (expected to be minimal) will be handled through multiple imputation within the same multiple imputation by chained equations framework. Missing secondary outcomes will be analyzed using available data, with the extent of missingness reported for each outcome. Secondary endpoints will be treated as exploratory without formal multiplicity adjustment.

Prespecified exploratory subgroup analyses will be conducted to assess potential heterogeneity of treatment effect. Subgroups will include (1) COPD vs non-COPD etiology, (2) baseline PaCO₂ strata (eg, 45‐60 mm Hg vs >60 mm Hg), and (3) baseline pH strata (eg, 7.25‐7.35 vs <7.25). Subgroup effects will be evaluated by including interaction terms (subgroup × treatment) in the primary ANCOVA model. These analyses are exploratory and hypothesis generating; the trial is not powered for subgroup comparisons, and results will be interpreted with caution.

Data will be recorded on paper-based standardized CRFs using trial numbers only and will be stored under lock and key with restricted access. Electronic data will be stored on secure, access-controlled systems with password protection and audit trails. The ABG samples will be analyzed at the bedside using the i-STAT Alinity analyzer. The analyzer generates a printed result slip that includes PaCO₂, PaO₂, pH, bicarbonate, and other parameters with a time stamp. The printed result slip will be attached to the paper-based CRF for each participant. Should the printed result slip be unavailable, illegible, or degraded over time, the ABG results will be retrieved electronically from the analyzer and from the patient’s EHR. Study investigators will transcribe key ABG values onto the CRF in real time. Device settings (flow rate, FiO₂, and temperature for HVNI and IPAP, EPAP, and FiO₂ for NIPPV) will also be recorded on the CRF at baseline, 30 minutes, and 60 minutes. Accordingly, the study CRF serves as the primary source document to ensure complete and accurate capture of primary outcome data.

Apart from ABG readings, other variables and outcomes as stated will be collected and followed up on through electronic medical record review. To ensure data accuracy, the following quality checks are in place: (1) double data entry by 2 independent data entry staff members, with discrepancy resolution; (2) programmed range and logic checks during electronic data entry; (3) periodic source data verification by the study coordinator comparing CRF entries against ABG printouts and EHR records; and (4) audit trails for all electronic data modifications.

Study investigators will undergo training on device application, titration protocols, mask fitting and leak management, and CRF completion. An independent data monitoring and safety team within the institution will conduct periodic reviews to assess protocol adherence, data completeness, and safety. Any serious adverse events related to the study interventions will be reported to the National Healthcare Group Domain Specific Review Board per institutional requirements. Quarterly adherence checks will include verification of device settings, oxygen targets, ABG timing windows, and documentation of cointerventions.