What Is the ROX Score and How Is It Calculated?

The ROX index is a bedside scoring tool used in intensive care to predict whether a patient on high-flow nasal cannula oxygen therapy is improving or heading toward intubation. It is calculated by dividing the patient’s oxygen saturation (SpO2) by the fraction of inspired oxygen (FiO2), and then dividing that result by the respiratory rate.¹ The formula is simple enough to calculate at the bedside without lab draws, which is the main reason it has become so widely adopted. But the number it produces only becomes useful once you understand what the thresholds mean, when to check it, and where the tool falls short.

The Formula and What Each Piece Means

Written out, the ROX index equals (SpO2 / FiO2) ÷ respiratory rate. SpO2 is the oxygen saturation reading from a standard pulse oximeter, expressed as a percentage. FiO2 is the fraction of oxygen being delivered to the patient, expressed as a decimal (so 60% oxygen would be 0.60). Respiratory rate is simply how many breaths per minute the patient is taking.

Each component captures a different dimension of how the patient is doing. The SpO2/FiO2 ratio by itself tells you how well the lungs are absorbing oxygen relative to how much oxygen they are being given. A patient saturating at 97% on only 30% oxygen is in much better shape than one saturating at 92% on 80% oxygen. The respiratory rate adds a measure of how hard the patient is working to breathe. A patient who looks oxygenated on the monitor but is breathing 35 times a minute is compensating, and that effort is often unsustainable. By rolling all three variables into a single number, the ROX index captures both the oxygenation picture and the breathing effort simultaneously, giving clinicians a snapshot of ventilatory failure risk.²

A practical advantage is that none of these inputs require an arterial blood gas draw. SpO2 comes from a finger clip, FiO2 is set on the high-flow device, and respiratory rate can be counted or read from a monitor. That makes the ROX index something a nurse or respiratory therapist can calculate in seconds, repeatedly, without waking a sleeping patient or ordering labs. One validation study described this simplicity as the index’s main advantage over scores that depend on blood gas results or complex calculations.³

Key Thresholds and What They Tell You

The original landmark work on the ROX index proposed that a score of 4.88 or higher at 12 hours after starting high-flow nasal cannula (HFNC) therapy was associated with a lower risk of needing intubation. That 4.88 cutoff became widely cited in protocols, with sensitivity around 70% and specificity around 72% for predicting HFNC failure.³ In practical terms, a patient whose ROX index climbs above roughly 5 during the first half-day on HFNC is more likely to avoid the ventilator. A patient whose score stays flat or drops below that range is at higher risk of deterioration.

The thresholds shift depending on the clinical context and the time window. A systematic review and meta-analysis of COVID-19 patients found an optimal cutoff of about 5.23, with sensitivity of 73% and specificity of 69%. The same analysis proposed different working thresholds depending on how long the patient had been on HFNC. In the first two to six hours, a score below roughly 5.33 was useful for flagging likely failure. Between six and twelve hours, the concerning threshold dropped to below about 3.69. After twelve to twenty-four hours, a score above roughly 6.07 was useful for ruling failure out.⁴ These shifting thresholds reflect the fact that patients who are going to tolerate HFNC tend to show improving numbers over time, while those headed for intubation plateau.

Why the Trend Matters More Than a Single Number

A single ROX score at one moment in time gives some information, but serial measurements are far more revealing. In a study of COVID-19 patients in the ICU, the ROX index was checked at multiple time points during therapy. Patients who were successfully weaned from HFNC showed a pattern: their ROX stayed relatively flat during the first eight hours but then climbed and held steady after twelve hours. Patients who ultimately failed HFNC showed no such upward trend, with their scores essentially stagnating throughout.⁵

This trajectory effect is why many ICU protocols call for checking the ROX index repeatedly, often at 2, 6, 12, and 24 hours after starting high-flow oxygen. A patient whose score is borderline at 2 hours but climbing by 12 hours tells a very different story than one whose score was borderline at 2 hours and has not budged. The score at 12 hours appears to be particularly predictive. One study of 60 COVID-19 patients found that the ROX index at 12 hours was the strongest predictor of HFNC outcome, with a cutoff of 5.81 achieving sensitivity above 97% and specificity above 82%, though the authors cautioned these estimates came from a small, single-center cohort and need external validation.⁶

Performance During the COVID-19 Pandemic

The ROX index saw massive uptake during COVID-19, when hospitals worldwide were managing enormous numbers of patients on high-flow oxygen and needed a fast way to triage who was deteriorating. An early systematic review and meta-analysis pooling over 1,300 patients across eight studies found that the ROX index had good discriminatory power for predicting HFNC failure in COVID-19 patients, with a pooled sensitivity of about 70% and specificity of about 79%.⁷ A later, larger meta-analysis that included more COVID-19 studies found somewhat lower pooled accuracy, with an overall area under the curve of about 0.77, suggesting the index is useful but not perfect.⁴

The takeaway from the COVID-era literature is that the ROX index is a genuinely helpful early warning signal, but it is not a crystal ball. About a third of the time, patients who fell below the danger threshold still managed to avoid intubation, and some patients above the “safe” threshold still ended up needing a ventilator. It works best as one input among several in a clinical decision, not as an automatic intubation trigger.

When ROX-Guided Decisions Affect Survival

One of the more consequential questions around the ROX index is whether using it to guide the timing of intubation actually improves patient outcomes. There is growing evidence that it does. A study comparing early versus delayed intubation in COVID-19 patients who were deteriorating on HFNC found that patients who were intubated early, guided by ROX-based criteria, had a hospital survival rate of about 92%, compared with significantly worse survival in the delayed group. The median survival time was roughly 15 days in the early intubation group versus about 10 days in the delayed group.⁸

A larger study using target trial emulation methods found similar results. When researchers modeled what would happen if intubation were triggered at a ROX threshold below 4.88, estimated 30-day mortality dropped by about 7 percentage points compared with usual care. The usual care mortality was roughly 27%, and the ROX-guided strategy brought it to about 20%.⁹ The mechanism is straightforward: patients who linger too long on high-flow oxygen while continuing to deteriorate arrive at intubation in worse physiological shape. Using the ROX index to catch that deterioration earlier gives them a better chance.

Modifications That Fold In Heart Rate

The original ROX formula deliberately keeps things simple, but some researchers have argued that adding heart rate makes the score more sensitive. A rapidly climbing heart rate can be an early sign of respiratory distress that the standard three variables miss. The ROX-HR index divides the standard ROX score by heart rate, so that a rising pulse pushes the score down, flagging the patient as higher risk. In one study, the ROX-HR index showed slightly better predictive accuracy than the standard ROX index during the first 12 hours of HFNC therapy, with the difference most visible at around the 10-hour mark.¹⁰

Another variant, sometimes called the modified ROX index score, builds heart rate into the formula differently. A prospective study compared this modified version against the standard ROX index at one hour after starting high-flow oxygen. The modified version achieved an area under the curve of 0.79 compared with 0.70 for the original, suggesting better discrimination at that early time point. Its optimal cutoff was 4.63, with sensitivity of about 74% and specificity of about 70%.¹¹ A separate study also explored an HROX index that multiplied the ROX by a heart rate improvement factor calculated from baseline to the day of weaning.¹² None of these modified versions have displaced the original in widespread clinical use, but they illustrate the ongoing effort to squeeze more predictive power out of bedside numbers.

How It Stacks Up Against Other Scoring Systems

The ROX index is not the only bedside tool for gauging whether non-invasive respiratory support is working. The HACOR scale, which incorporates heart rate, acidosis, consciousness, oxygenation, and respiratory rate, is another commonly used tool. A prospective study in emergency department patients found that the ROX index generally outperformed the HACOR scale for predicting HFNC success, with areas under the curve of roughly 0.82 to 0.85 for ROX at different time points, compared with 0.69 to 0.76 for HACOR.¹³

That said, a multicenter Japanese study that externally validated both tools in COVID-19 patients found much more modest performance from both, with their discrimination converging to similar, middling levels by 24 and 48 hours.¹⁴ The lesson is that single-site studies tend to produce flattering accuracy numbers, while validation in new populations brings everyone back to earth. Both ROX and HACOR are useful, but neither is infallible.

Machine learning models trained on electronic health record data have also been pitted against the ROX index. One model using a broader set of variables achieved an area under the curve of 0.76 at 24 hours for predicting HFNC failure, compared with 0.70 for the ROX index alone. The machine learning approach was statistically better, though the margin was not dramatic. For many hospitals without the infrastructure to deploy real-time predictive algorithms, the ROX index remains the more practical option.

Using the ROX Index Beyond High-Flow Nasal Cannula

The ROX index was designed for patients on HFNC, but clinicians have naturally wondered whether it works for other forms of respiratory support. For noninvasive ventilation, the results are mixed. A large study of patients with new-onset acute respiratory failure receiving NIV found that the ROX index had moderate predictive power for NIV failure, with areas under the curve climbing from 0.64 before NIV started to 0.77 after 24 hours of therapy. Failure rates dropped in a clear gradient: patients with a pre-NIV ROX index below 2 had a 92% failure rate, while those with a score above 10 had a failure rate of about 30%.¹⁵ That gradient is clinically useful even though the discrimination is not as sharp as for HFNC.

For COPD exacerbations treated with HFNC or NIV, the ROX index at treatment initiation and at one and six hours predicted avoidance of invasive mechanical ventilation or death with areas under the curve around 0.72 to 0.73, and an optimal cutoff of about 6.88.¹⁶ However, when one study specifically looked at NIV patients and tested the ROX index alongside its variants, the predictive accuracy was consistently low to moderate across all time windows, with areas under the curve never exceeding about 0.63.¹⁷ The bottom line is that the ROX index retains some value outside its original HFNC niche, but the further you get from that original setting, the weaker the predictions become.

Post-Extubation Monitoring

Another application gaining attention is using the ROX index after a patient has been extubated and placed on HFNC as a bridge. Extubation failure, where a patient needs to be reintubated, carries its own risks. In a retrospective cohort study, the ROX index was significantly lower in patients who failed post-extubation HFNC compared with those who tolerated it successfully. After adjusting for severity and comorbidities, a ROX index above 4.88 was associated with modestly lower odds of reintubation. However, the overall predictive capacity for extubation failure was limited, with an area under the curve of only 0.64.¹⁸ The tool provides a small signal in this context, but post-extubation decisions clearly depend on additional clinical judgment.

Adapting the ROX Index for Children

Pediatric patients present a challenge because normal respiratory rates in infants and toddlers are much higher than in adults. A raw ROX score of 5 means something very different in a newborn breathing 50 times per minute than in an adult breathing 25 times per minute. For infants with bronchiolitis on HFNC, one study found that a ROX index measured at 12 hours could distinguish between success and failure groups, with the failure group averaging about 5.8 and the success group about 7.7. The best cutoff range at 12 hours fell between 6.50 and 7.18.¹⁹ These cutoffs are noticeably higher than the adult threshold of around 4.88 to 5.23, reflecting the age-specific respiratory physiology.

Some pediatric researchers have developed a separate “pediatric ROX” (pROX) index that accounts for age-adjusted respiratory norms. In one study of children with acute respiratory distress on HFNC, the pROX index at 12 hours had an area under the curve of 0.77 for predicting failure, with a cutoff below 132 achieving sensitivity of about 69% and specificity of about 85%.²⁰ The adult version is not directly transferable to children, and pediatric units applying ROX-based protocols need to use age-appropriate thresholds.

The Pulse Oximetry Problem

Because the ROX index depends entirely on a pulse oximeter reading for its oxygenation input, any limitation of pulse oximetry becomes a limitation of the ROX index. The most clinically significant issue is that pulse oximeters can overestimate oxygen saturation in patients with darker skin. Two large cohort analyses demonstrated that occult hypoxemia, meaning an actual arterial oxygen saturation below 88% despite the pulse oximeter reading 92% to 96%, occurred roughly three times more often in Black patients compared with white patients.²¹

If the SpO2 number feeding into the ROX formula is falsely elevated, the resulting ROX index will look artificially reassuring. A patient who appears to have a comfortable ROX of 6 might actually have a true score closer to 4 if their real oxygen saturation is several points lower than the monitor shows. This means that in patients where pulse oximetry is known to be less reliable, including those with darker skin, poor peripheral circulation, hypothermia, or nail polish, the ROX index should be interpreted with extra caution. Some clinicians advocate pairing the ROX with periodic arterial blood gases in high-risk situations, which somewhat undermines the tool’s appeal as a fully non-invasive measure, but may be necessary to avoid false reassurance.

Other limitations are more mundane but still relevant. The FiO2 delivered by a high-flow nasal cannula is not as precisely controlled as it would be on a ventilator, because some room air gets entrained around the nasal prongs, especially if the patient is breathing with their mouth open. And respiratory rate counted manually has its own measurement variability. The ROX index treats all three of its inputs as precise numbers, but in reality each one carries a margin of error that gets compounded in the final score. Clinicians who rely heavily on narrow threshold cutoffs should keep this measurement noise in mind.