Abstract
Background
Closed-loop oxygen control (CLOC) systems improve accuracy of oxygen delivery compared with standard manual oxygen adjustment. Determination of their efficacy in delivering oxygen noninvasively to adults, using different delivery methods and across different hospital locations, would enhance clinical practice.
Methods
We conducted a systematic review and meta-analysis, searching Medline, Embase, Scopus Basic and Advanced, and Cochrane databases. Randomised controlled trials comparing time spent within a target peripheral oxygen saturation (SpO2) range between CLOC and manual oxygen control in hospitalised adults (>18 years) were included. Outpatient studies and studies using mechanical ventilation were excluded. The primary outcome was proportion of time spent within a target SpO2 range. Meta-regression was performed for pre-determined study-level covariates, including oxygen delivery mode and locations of care.
Results
The search identified 3777 papers; of these, seven were included, with 635 participants. The mean proportion of time spent within the target SpO2 range was between 81.2 and 94.9% for CLOC and 51.3–84.0% for manual oxygen control. The mean difference was 21.2% (95% CI 18.5–23.8, p<0.0001), favouring CLOC. Evidence of heterogeneity (I2=87.9) was explained by three studies using low-flow oxygen therapy, mean difference 30.9% (95% CI 27.0–34.8, p<0.0001); compared with four studies using high-flow nasal oxygen therapy, mean difference 12.7% (95% CI 9.1–16.4, p<0.001).
Conclusions
CLOC improves adherence to a target SpO2 range compared to manual oxygen control and is an effective approach for achieving a target SpO2 range when delivering oxygen noninvasively in hospitalised adults.
Shareable abstract
Use of closed-loop oxygen control in adults receiving noninvasive oxygen improves the accuracy of oxygen delivery and represents an effective strategy for achieving a target oxygen saturation range in acute inpatient care. https://bit.ly/4znFoqr
Introduction
Supplemental oxygen is a widely administered therapy for managing respiratory failure. When used to correct hypoxaemia, supplemental oxygen can be life-preserving. Yet, when administered in excess, hyperoxaemia can be potentially harmful [1–3]. It is therefore recommended that, to balance the risks of both hypoxaemia and hyperoxaemia, oxygen be titrated to maintain peripheral oxygen saturation (SpO2) within a prespecified target range [1, 4].
The usual method for titrating oxygen is to manually adjust the inspiratory oxygen fraction (FIO2) to achieve a prescribed SpO2 target range. Manually controlling oxygen in this way is generally inaccurate. Patients on a standard hospital ward receiving manual adjustments are reported to spend approximately 45% of their time outside of a safe SpO2 range [5]. Efforts to reduce time spent outside the target range are likely to reduce risk and potentially minimise patient harm.
The recent development of closed-loop oxygen control (CLOC) systems represents a significant technological breakthrough in delivering oxygen to patients with respiratory failure. CLOC systems operate algorithms that automatically adjust the delivered FIO2 in response to continuously measured SpO2, to achieve stable SpO2 values within a prespecified target range (figure 1). Randomised trials assessing efficacy report that, compared with standard manual oxygen control, the proportion of time spent within the target SpO2 range using a CLOC system is significantly increased [6–13]. In turn, this increase in time spent within the target SpO2 range likely improves the safety of oxygen delivery by reducing the time spent in potentially harmful hypoxaemic and hyperoxaemic states. The use of CLOC may also reduce the time clinicians spend manually adjusting oxygen [11–13] and shorten hospital length of stay by enabling more expedient weaning from oxygen therapy [7].
FIGURE 1.
Diagrammatic representation of a closed-loop oxygen control system. Pulse oximetry continuously measures peripheral oxygen saturation (SpO2), with recordings analysed by a computer algorithm that dynamically controls the inspiratory oxygen fraction (FIO2) output. Available mask interfaces for closed-loop FIO2 delivery include low-flow oxygen therapy (nasal cannula or face mask), high-flow nasal oxygen therapy and continuous positive airway pressure or noninvasive ventilation masks.
Previously reported meta-analyses [14, 15] have broadly assessed CLOC systems across all standard modes of oxygen delivery, including invasive mechanical ventilation, in both adults and children, and were conducted before the publication of several important trials delivering CLOC via high-flow nasal oxygen therapy (HFNOT) [11–13]. The aim of this systematic review and meta-analysis is to establish the effect of CLOC on time spent within a target SpO2 range compared with manual oxygen control in hospitalised adults receiving noninvasive oxygen delivery.
Methods
Search strategy and selection criteria
For this systematic review and meta-analysis, we searched Medline, Embase, Scopus Basic, Scopus Advanced, Cochrane, and Web of Science databases without date limitations, for randomised controlled trials that compared the use of a CLOC system to manual oxygen control. An example of the search strategy is presented in the supplemental material (text S1). Studies were included if CLOC and manual oxygen control were delivered noninvasively, reported the proportion of time spent within the target SpO2 range, included adults (≥18 years of age) and were conducted in a hospital inpatient setting.
We excluded studies that used invasive mechanical ventilation and recruited outpatients or patients younger than 18 years old. Two reviewers (L. Kirton and R. Biggs), independently and in duplicate, screened titles and abstracts using a standardised data form (Covidence; Veritas Health Innovation, Melbourne, Australia). Disagreements about inclusion were resolved through consensus.
This study is reported in accordance with the Cochrane Handbook for Systematic Reviews of Interventions and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement [16]. This study is registered with PROSPERO as CRD42024500598.
Data analysis
Two reviewers (L. Kirton and R. Biggs) independently and in duplicate extracted data using a standardised data form (Covidence; Veritas Health Innovation, Melbourne, Australia).
Outcomes of interest for assessing the efficacy of CLOC were the proportion of time spent within a pre-specified SpO2 target range, below and ≥2% below the target SpO2 range, and above and ≥2% above the target SpO2 range. These “proportion of time” measurements account for the differing target SpO2 ranges across studies. Other outcomes were the number of FIO2 adjustments over the study period and hospital length of stay. As CLOC systems are not fully autonomous, they still require some manual setting adjustments. The frequency of FIO2 adjustment outcome was therefore relevant to both CLOC and manual oxygen control groups. Hospital length of stay in this study was defined as the reported duration (in days) between randomisation and hospital discharge. Outcomes were treated on an intention-to-treat basis, including all patients who were randomly assigned to any treatment arm.
Because the difference between CLOC and manual control may be different dependent on the type of hospital care and mode of oxygen delivery pre-specified study level covariates included in meta-regression models were study locations including emergency department (ED) versus not ED, standard ward versus not standard ward, intensive care unit (ICU) versus not ICU, as well as modes of oxygen delivery including low-flow oxygen therapy (LFOT) versus HFNOT. We considered a standard hospital ward to include all medical and surgical wards and coronary care units, but excluded high-dependency units. LFOT comprises oxygen delivered through either a low-flow nasal cannula or a standard face mask.
Meta-analysis of scale variables used inverse variance weighted meta-analysis on the scale of measurement using reported means, standard deviations and counts. Heterogeneity was assessed using the Chi2 test and the I2 statistic. A correlation coefficient assessed publication bias. In the forest plots, the studies are ordered by effect size to provide a visual assessment of potential publication bias, rather than presenting separate funnel plots. The horizontal lines are the 95% confidence, the squares are proportional to the study variance and the diamonds are the pooled estimates. The weights are the proportions of the pooled variance (weights) for each study in a fixed-effects model. Both fixed- and random-effects estimates are shown. Heterogeneity was evaluated by the method of partitioning the Chi2 statistic for those variables with sufficient heterogeneity and/or where there were three or more studies contributing data. For interpretation, if the “between groups” partition of the Chi2 statistic is statistically significant, this means there is evidence that the subgroups explain variability and have different point estimates for the effect size. If the “Within groups” partition of the Chi2 statistic is statistically significant, this suggests there is evidence that at least one of the groups is still heterogeneous. If the “between groups” is statistically significant and the “within groups” is not, this means the subgroup effects explain the variation.
Analyses were performed in SAS 9.4 using the published methods outlined above.
Results
A total of 3777 studies were screened, and seven were included in the analysis, comprising 635 randomised participants, including 316 in the CLOC group and 319 in the manual oxygen control group (figure 2). The data extracted from the seven included studies are presented in table 1. The mean proportion of time spent within the target SpO2 range was between 81.2 and 94.9% for CLOC and 51.3–84.0% for manual oxygen control. For the fixed-effect model of the primary outcome, CLOC is estimated to increase the mean time spent within the target SpO2 range by 21.2% (95% CI 18.5–23.8, p<0.0001) (I2=87.9, 95% CI 77.4–93.5), compared with manual oxygen control (table 2, figure 3).
FIGURE 2.
PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) diagram outlining study selection. IQR: interquartile range; RCT: randomised controlled trial.
TABLE 1.
Trials of closed-loop oxygen delivery meeting the inclusion criteria
| First author [ref.], year | Population | Study type | Recruitment dates | Noninvasive oxygen delivery mode | Study location | Duration of receiving study therapy | Target SpO2 range options | Proportion of time (%) within target SpO2 range, mean±sd | |
|---|---|---|---|---|---|---|---|---|---|
| Intervention group: closed-loop oxygen control | Control group: manual oxygen control | ||||||||
| Harper [9] 2021 | Adult medical or surgical inpatients requiring HFNOT | RCT | January–October 2020 | HFNOT | Standard ward | Up to 24 h | 88–92% or 92–96% | 93.8±7.1 n=11 |
71.3±15.6 n=11 |
| Kirton [11] 2024 | Adult inpatients requiring oxygen following cardiac surgery | RCT | September 2021–December 2022 | HFNOT | Intensive care unit | Up to 24 h | 92–96% | 94±7.5 n=28 |
84±18.9 n=28 |
| Kirton [13] 2026 | Adult patients in an emergency department requiring oxygen | RCT | October 2022–December 2023 | HFNOT | Emergency department | Up to 3 h | 88–92% or 92–96% | 94.9±6.2 n=28 | 84±14.5 n=24 |
| Kirton [12] 2025 | Adult medical inpatients requiring oxygen | RCT | April 2022–April 2023 | HFNOT | Standard ward and intensive care unit | Up to 8 h | 88–92% or 92–96% | 87.2±12.5 n=36 |
74.2±17.5 n=34 |
| Lellouche [6] 2016 | Adults hospitalised for acute COPD exacerbation requiring oxygen | RCT | August 2011–February 2015 | LFOT | Standard ward | Up to 8 days | 88–92% | 81.2±15.9 n=25 |
51.3±19.7 n=25 |
| L’Her [7] 2017 | Adults in an emergency department with acute respiratory failure requiring ≥3 L·min−1 oxygen | RCT | August 2011–October 2014 | LFOT | Emergency department | Up to 3 h | 88–92% or 92–96% | 81.3±20.7 n=93 |
51.8±30 n=94 |
| L’Her [8] 2021 | Extubated post-surgical adults with intermediate to high risk of post-operative pulmonary complications requiring oxygen | RCT | Jan 2016 – Sept 2018 | LFOT | Standard ward | Up to 3 days | 88–92% or 92–96% | 94±11.3 n=95 |
62.1±23.3 n=103 |
HFNOT: high-flow nasal oxygen therapy; LFOT: low-flow oxygen therapy, RCT: randomised controlled trial; SpO2: peripheral oxygen saturation.
TABLE 2.
Primary outcome data summary, mean differences, pooled estimates of effect, heterogeneity, publication bias and sources of variation
| Proportion of time (%) within SpO2 target range | ||||
|---|---|---|---|---|
| First author [ref.], year | CLOC, mean±sd | Manual, mean±sd | Mean difference (95% CI) | p-value |
| Harper [9] 2021 | 93.8±7.1 n=11 | 71.3±15.6 n=11 | 22.5 (12.4–32.6) | 0.0003 |
| Kirton [11] 2024 | 94±7.5 n=28 | 84±18.9 n=28 | 10.0 (2.5–17.5) | 0.012 |
| Kirton [13] 2026 | 94.9±6.2 n=28 | 84±14.5 n=24 | 10.9 (5–16.8) | 0.0007 |
| Kirton [12] 2025 | 87.2±12.5 n=36 | 74.2±17.5 n=34 | 13.0 (5.9–20.1) | 0.0006 |
| Lellouche [6] 2016 | 81.2±15.9 n=25 | 51.3±19.7 n=25 | 29.9 (20–39.8) | <0.0001 |
| L’Her [7] 2017 | 81.3±20.7 n=93 | 51.8±30 n=94 | 29.5 (22.1–36.9) | <0.0001 |
| L’Her [8] 2021 | 94±11.3 n=95 | 62.1±23.3 n=103 | 31.9 (26.7–37.1) | <0.0001 |
| Pooled fixed effect | 21.2 (18.5–23.8) | <0.0001 | ||
| Pooled random effect | 21 (13.7–28.3) | <0.0001 | ||
| Publication bias | ||||
|---|---|---|---|---|
| Method | Test statistic | |||
| Correlation coefficient | −0.05 p=0.88 |
|||
| Heterogeneity | Chi2 (DF) | p-value | I2 (95% CI) | |
|---|---|---|---|---|
| 49.6 (6) | <0.0001 | 87.9 (77.4–93.5) | ||
| Meta-regression | Chi2 (DF) | p-value | CLOC minus manual Mean difference (95% CI) |
|
|---|---|---|---|---|
| HFNOT versus LFOT | ||||
| Source of variation | ||||
| Between groups | 44.84 (1) | <0.0001 | HFNOT | 12.7 (9.1–16.4) |
| Within groups | 4.77 (5) | 0.44 | LFOT | 30.9 (27–34.8) |
| HFNOT | 4.45 (3) | 0.22 | Overall | 21.2 (18.5–23.8) |
| LFOT | 0.32 (2) | 0.85 | ||
| Total | 49.61 (6) | <0.0001 | ||
| ED versus not ED | ||||
| Source of variation | ||||
| Between groups | 2.48 (1) | 0.11 | Not ED | 22.7 (19.4–25.9) |
| Within groups | 47.13 (5) | <0.0001 | ED | 18.1 (13.5–22.8) |
| Not ED | 32.3 (4) | <0.0001 | Overall | 21.2 (18.5–23.8) |
| ED | 14.83 (1) | 0.0001 | ||
| Total | 49.61 (6) | <0.0001 | ||
| Ward versus not ward | ||||
| Source of variation | ||||
| Between groups | 28.39 (1) | <0.001 | Not Ward | 15.2 (11.8–18.7) |
| Within groups | 21.22 (5) | 0.0007 | Ward | 29.9 (25.8–34.1) |
| Not ward | 18.6 (3) | 0.0003 | Overall | 21.2 (18.5–23.8) |
| Ward | 2.63 (2) | 0.27 | ||
| Total | 49.61 (6) | <0.0001 | ||
| ICU versus non-ICU | ||||
| Source of variation | ||||
| Between groups | 9.68 (1) | 0.00019 | Not ICU | 22.8 (19.9–25.6) |
| Within groups | 39.94 (5) | <0.0001 | ICU | 10 (2.5–17.5) |
| Not ICU | 39.94 (5) | <0.0001 | Overall | 21.2 (18.5–23.8) |
| ICU | NA | NA | ||
| Total | 49.61 (6) | <0.0001 | ||
CLOC: closed loop oxygen control; DF: degrees of freedom; ED: emergency department; HFNOT: high-flow nasal oxygen therapy; ICU: intensive care unit; LFOT: low-flow oxygen therapy; Manual: manual oxygen control; NA: not applicable/available; SpO2: peripheral oxygen saturation; Ward: standard hospital ward.
FIGURE 3.
Forest plot of the difference in proportion of time within the target peripheral oxygen saturation range between closed-loop oxygen control (CLOC) and manual oxygen control. The horizontal lines are the 95% confidence intervals, the squares are proportional to the study variances and the diamonds are the fixed-pooled estimates, both overall and by subgroups of high-flow nasal oxygen therapy (HFNOT) and low-flow oxygen therapy (LFOT).
For all secondary outcomes except the proportion of time greater than or equal to 2% below range and hospital length of stay, there was evidence favouring CLOC over manual oxygen control (table 3 and supplemental figures S1–S4). In the random-effects model, CLOC significantly reduced the mean time spent below the target range by 6.1% (95% CI 2.7–9.5, p=0.0005) compared with manual oxygen control. For the mean time spent above and ≥2% above the target SpO2 range, the reductions with CLOC were 4.4% (95% CI 1.3–7.6, p=0.006) and 12.5% (95% CI −20.7–−4.4), respectively.
TABLE 3.
Secondary outcome data summaries, mean differences and pooled estimates of effect, comparing closed-loop and manual oxygen delivery
| First author [ref.], year | CLOC, mean±sd | Manual, mean±sd | Mean difference (95% CI) | p-value |
|---|---|---|---|---|
| Proportion of time (%) below target range | ||||
| Kirton [11] 2024 | 3.9±6.7 n=28 | 9.6±16.9 n=28 | −5.7 (−12.4–1) | 0.10 |
| Kirton [13] 2026 | 3.2±5.2 n=28 | 11.7±13.3 n=24 | −8.5 (−13.8– −3.2) | 0.003 |
| Kirton [12] 2025 | 8.3±11.8 n=36 | 11.8±12.9 n=34 | −3.5 (−9.3–2.3) | 0.24 |
| Pooled fixed effect | −6.1 (−9.5– −2.7) | 0.0005 | ||
| Pooled random effect | −6.1 (−9.5– −2.7) | 0.0005 | ||
| Proportion of time (%) greater than or equal to 2% below the target range | ||||
| Harper [9] 2021 | 0.8±2.2 n=11 | 5.6±9 n=11 | −4.8 (−10.3–0.7) | 0.10 |
| L’Her [7] 2017 | 3.2±8.9 n=93 | 5.1±11.5 n=94 | −1.9 (−4.8–1) | 0.21 |
| Pooled fixed effect | −2.6 (−5.1–0.04) | 0.054 | ||
| Pooled random effect | −2.6 (−5.1–0.04) | 0.054 | ||
| Proportion of time (%) above target range | ||||
| Kirton [11] 2024 | 2.2±3.7 n=28 | 6.4±10.8 n=28 | −4.2 (−8.4–0) | 0.056 |
| Kirton [13] 2026 | 1.9±2.8 n=28 | 4.3±6.2 n=24 | −2.4 (−5–0.2) | 0.071 |
| Kirton [12] 2025 | 4.5±7.3 n=36 | 14.1±16.4 n=34 | −9.6 (−15.5– −3.7) | 0.002 |
| Pooled fixed effect | −3.7 (−5.7– −1.6) | 0.004 | ||
| Pooled random effect | −4.4 (−7.6– −1.3) | 0.006 | ||
| Proportion of time (%) greater than or equal to 2% above the target range | ||||
| Harper [9] 2021 | 2.0±2.5 n=11 | 8.8±12.7 n=11 | −6.8 (−14.4–0.8) | 0.10 |
| L’Her [7] 2017 | 4.1±8.6 n=93 | 21.6±30.2 n=94 | −17.5 (−23.9– −11.1) | <0.0001 |
| Pooled fixed effect | −13.1 (−18– −8.2) | <0.0001 | ||
| Pooled random effect | −12.5 (−20.7– −4.4) | 0.0026 | ||
| Number of FIO2 adjustments | ||||
| Harper [9] 2021 | 4.0±3.9 n=11 | 4.8±5.4 n=11 | −0.8 (−4.7–3.1) | 0.69 |
| Kirton [11] 2024 | 2.1±2.5 n=28 | 11.8±11.4 n=28 | −9.7 (−14–−5.4) | <0.0001 |
| Kirton [13] 2026 | 0.2±0.5 n=28 | 0.8±1.4 n=24 | −0.6 (−1.2–0) | 0.039 |
| Kirton [12] 2025 | 2.7±4.5 n=36 | 5.7±5.2 n=34 | −3.0 (−5.3–−0.7) | 0.012 |
| Pooled fixed effect | −0.9 (−1.4–−0.3) | 0.0013 | ||
| Pooled random effect | −2.7 (−4.9– −0.5) | 0.014 | ||
| Hospital length of stay (days) | ||||
| Harper [9] 2021 | 8.1±6 n=11 | 4.8±2.6 n=11 | 3.3 (−0.6–7.2) | 0.11 |
| L’Her [8] 2021 | 12.5±12.4 n=95 | 13.3±11.3 n=103 | −0.8 (−4.1–2.5) | 0.64 |
| L’Her [7] 2017 | 9.2±6.9 n=93 | 11.1±7 n=94 | −1.9 (−3.9–0.1) | 0.069 |
| Lellouche [6] 2016 | 5.8±4.4 n=25 | 8.4±6 n=25 | −2.6 (−5.5–0.3) | 0.087 |
| Pooled fixed effect | −1.2 (−2.6–0.2) | 0.086 | ||
| Pooled random effect | −0.9 (−2.9–1.1) | 0.37 | ||
CLOC: closed loop oxygen control; FIO2: inspiratory oxygen fraction.
Across included studies, the effect of CLOC compared to manual oxygen control on the proportion of time spent within the target SpO2 range was heterogeneous; Chi2 49.6 (degrees of freedom (DF) 6), I2=87.9 (95% CI 77.4–93.5) (table 2). Most of this variation is explained by subgrouping based on whether studies used an HFNOT or LFOT device to deliver CLOC, with homogeneity within these categories. The mean increase in time spent within the target SpO2 range when using CLOC compared to manual oxygen control was 12.7% (95% CI 9.1–16.4) with HFNOT and 30.9% (95% CI 27.0–34.8) with LFOT (table 2, figure 3).
In the HFNOT studies, the mean proportion of time spent within the target SpO2 range was between 87.2% and 94.9% for CLOC and 71.3% and 84% for manual oxygen control. For LFOT studies, the mean proportion of time spent within the target SpO2 range was 81.2–94% for CLOC and 51.3 to 62.1% for manual oxygen control.
For other study-level covariates, there was some evidence that ED versus non-ED studies explained variability in the primary outcome, but residual heterogeneity remained within these subgroups. There was a smaller effect size in ED-based studies compared to non-ED-based studies. There was some evidence that standard ward-based studies versus nonstandard ward-based studies explained variability in the primary outcome. Still, the estimates were homogeneous within the standard ward group and heterogeneous within the nonstandard ward group. There was a larger effect size in the standard ward group than in the nonstandard ward group. Only one study had ICU data; ICU-versus-not-ICU-based studies explained variability. However, they were still heterogeneous within the non-ICU group. The effect size was smaller in the ICU studies than in the non-ICU studies. For secondary end-points, there was either insufficient heterogeneity or too few studies to support meaningful subgroup analyses.
There was a mean of 2.7 (95% CI 0.5–5.9) fewer FIO2 adjustments over the study period with CLOC compared to manual oxygen control (p=0.014). Hospital length of stay did not significantly differ between groups (table 3). All seven included studies were rated as low risk of bias, with the risk-of-bias assessment tables presented in the supplemental material (table S1).
Discussion
This systematic review and meta-analysis shows that the proportion of time spent within a target SpO2 range is significantly higher with automated CLOC than with standard manual oxygen control in hospitalised patients receiving noninvasive oxygen therapy. CLOC likely reduced the risk of both hypoxaemia and hyperoxaemia, as its use was associated with less time spent both below and above the target SpO2 range.
The use of CLOC narrows the distribution of observed SpO2 values in patients receiving oxygen by bringing them closer to the target SpO2 range. Compared with manual oxygen control, CLOC resulted in a disproportionately greater reduction in time spent above or ≥2% above the target SpO2 range than in time spent below or ≥2% below the target range. This aligns with the observation that when delivering oxygen manually, clinicians are more inclined to over-oxygenate than under-oxygenate [17]. In this regard, CLOC can be seen as a tool to minimise risks associated with both hypoxaemia and hyperoxaemia, but with a more prominent role in lowering iatrogenic hyperoxaemia.
The detrimental physiological effects of hyperoxaemia are well elucidated. These include worsening of ventilation–perfusion matching through reversal of hypoxaemic pulmonary vasoconstriction in poorly ventilated lung areas [18–20], coronary and cerebral vasoconstriction [21], and accumulation of reactive oxygen species that cause oxidative cellular damage. The potential clinical relevance of these effects is suggested by evidence that even modest hyperoxaemia is associated with increased risk of mortality [2, 22, 23]. Conversely, hypoxaemia may result in myocardial and other critical organ ischaemia, leading to critical organ dysfunction, which, if severe, increases the risk of death [1, 24–26]. These competing harms have led to widely adopted international guidelines recommending that oxygen be delivered to maintain SpO2 within a target range to avoid both hyperoxaemia and hypoxaemia [1, 4, 27].
The mode of noninvasive respiratory support used to deliver CLOC showed a significant subgroup effect. In studies using LFOT devices (either a low-flow nasal cannula or a standard face mask), the mean proportion of time spent within the target SpO2 range for manual oxygen control was substantially lower (51.3–62.1%) than in studies using HFNOT (71.3–84.0%). This observation may be explained by differences in the performance of manual oxygen control between the LFOT and HFNOT groups. Physiological advantages of HFNOT over LFOT include flow-dependent increases in clearance of dead-space gases, increased end-expiratory lung volumes, inspiratory flow matching, reduced entrainment of ambient air and humidification, all of which are likely to contribute to more stable oxygenation and improved accuracy of manual oxygen control with HFNOT [28, 29]. As a result, the relative benefit of adding CLOC appears greater for LFOT than for HFNOT. However, with CLOC, the time spent within the target SpO2 range for both HFNOT and LFOT is similar, at 87.2–94.9% and 81.2–94%, respectively.
As HFNOT is now a widely available and increasingly recommended therapy, with some evidence of reduced intubation rates in respiratory failure compared with LFOT [30–33], it may be that, for the noninvasive management of acute hypoxaemic respiratory failure, the combination of HFNOT with CLOC represents an approach to more effectively maintain SpO2 within a target range. Whether an estimated 12.7 percentage-point increase in time spent within the target SpO2 range, compared with manually controlled HFNOT, improves clinical outcomes, including mortality risk, remains to be determined in future large clinical trials.
We found evidence that the magnitude of CLOC's effect on the proportion of time spent within the target SpO2 range, compared with manual oxygen control, varied by patient care location. The effect size was smaller in the single ICU study than in those conducted outside the ICU. This likely reflects markedly higher ICU staffing levels and more intensive monitoring than in a standard hospital ward, an interpretation supported by the observation that manual oxygen control performed better in the ICU than in non-ICU settings.
This analysis included studies comparing CLOC with manual oxygen control across populations with diverse clinical characteristics, which could not be separated into subgroups for meta-analysis. It is plausible that subgroups respond differently to CLOC systems, as well as the risks associated with marked hypoxaemia and hyperoxaemia. Only one study in our analysis specifically examined CLOC in COPD patients [6], highlighting a particularly high-risk subgroup that warrants further research. Additionally, patients with conditions such as obesity, obstructive sleep apnoea syndrome, bronchiectasis and neuromuscular or chest wall diseases may also experience disproportionate harm from hyperoxaemia, with potential to derive greater benefit from CLOC use [34, 35].
Our analysis also found that the number of FIO2 setting adjustments required of clinicians was significantly lower with CLOC than with manual oxygen control. This finding suggests that compared to manual oxygen control, the use of CLOC is likely to reduce the time clinicians spend operating the oxygen device, potentially affecting clinician workload. Outpatient studies have employed more direct methods to measure the time clinicians spend titrating oxygen settings between CLOC and manual oxygen control [36], which warrant consideration in future protocol development.
The length of hospital stays between the CLOC and manual oxygen control groups did not differ significantly; however, our analysis included only a small number of studies with relatively short treatment durations. Further dedicated studies are required to assess the impact of early initiation and longer use of CLOC on hospital length of stay and liberation from oxygen.
Best-practice guidelines clearly recommend delivering oxygen to a target SpO2 range. In contrast, relatively little guidance is offered on how best to achieve this. Our systematic review and meta-analysis provide a comprehensive and up-to-date summary of CLOC's efficacy in acute inpatient care, highlighting it as an effective strategy for maintaining SpO2 within a target range.
These findings should be considered alongside the limitations of CLOC systems. Stabilising SpO2 within a target range with CLOC does not necessarily confer greater safety or effectiveness of respiratory support. Other critical respiratory parameters, such as the work of breathing, arterial blood gases, respiratory rate and fatigue, must also be evaluated to ensure comprehensive care. Although CLOC is an effective tool for oxygen titration, users should remain cognisant that it is not a standalone respiratory monitoring strategy and should not replace clinical judgement, especially when assessing the risk of treatment failure or the need to escalate care.
Our analyses are limited by the small number of studies available for inclusion, especially for the time below and ≥2% below the target SpO2 range, and above and ≥2% above the target SpO2 range end-points, with only two contributing studies, for which the risk that some findings might instead be explained by chance is amplified. Of the seven included studies, the four that used HFNOT were all conducted in the same hospital using the same brand of HFNOT device [9, 11–13], potentially confounding the HFNOT versus LFOT subgroup effect observed in this study, as well as reducing the generalisability of our findings across different geographic populations and devices. We also excluded an important study by Roca et al. [10], which, due to its crossover design, could not be included in our meta-analysis. This study, which was conducted in ICUs using HFNOT in patients with severe respiratory failure, reported that the median time spent within the target SpO2 range was 96.5% (interquartile range (IQR) 93.5–98.9) with CLOC and 89.0% (IQR 77.4–95.9) with manual oxygen control, similar to the results in our meta-analysis. Similarly, another recent study in a coronary care unit in patients with cardiovascular disease [37], which could not be included because the outcome variables were reported only as medians (IQRs) rather than means±sd, reported that the proportion of time spent within the target SpO2 range was 87.0% (IQR 81.3–93.8) with CLOC and 60.6% (IQR 32.4–71.8) with manual oxygen control.
Studies reported differing outcomes for time spent outside the target SpO2 range. Meta-analysis endpoints were restricted to the outcomes most consistently reported across studies, including the proportion of time spent above or below the specified target range, and the proportion of time spent ≥2% above or ≥2% below the target range. These measures cannot distinguish the extent and duration of SpO2 deviations from the target range, so the respective risks of severe hypoxaemia and hyperoxaemia could not be compared between the CLOC and manual oxygen control groups. However, individual study data provide some assurance that extreme SpO2 values with CLOC are very uncommon and likely reduced compared with manual oxygen control. In the study by L’Her et al. [8] of 199 post-operative participants using LFOT with CLOC versus manual oxygen control, the mean±sd time with SpO2 <85% was 0.1±0.3 min and 0.3±1.0 min, respectively, and mean time spent with SpO2 >98% under oxygen was 0.9±3.2 min and 45.2±57.8 min, respectively. Nonetheless, standardised measures of the distribution of SpO2 values outside the target range are required.
Further research is needed before CLOC can be broadly recommended as a standard therapy for the management of respiratory failure. Research priorities include assessing clinical end-points in specific populations most likely to benefit from its use. Additionally, little is known about the cost-effectiveness of CLOC devices, which is an important consideration for adoption into mainstream care. Similarly, the impact of CLOC on clinician workload, and the potential hidden costs of CLOC use, such as clinician de-skilling and complacency that may arise with the use of automated technology. Accordingly, future trials should assess differences between CLOC and manual oxygen control in alarm frequency and silencing.
As CLOC systems are primarily dependent on pulse oximetry for oxygen titration, future trials would be enhanced by measuring oximetry sensor displacement, motion artefacts, measurement bias, tissue hypoperfusion and other factors that can produce unreliable plethysmographic waveforms. The trials included in this review excluded common conditions that can impair pulse oximetry signals, so the reported accuracy of both CLOC and manual oxygen control may differ under real-world conditions [5]. Future studies that report the real-world efficacy of CLOC and manual oxygen control, correlated with arterial blood gas measurements, are likely to help identify the limitations of oximetry-dependent oxygen delivery systems and guide future improvements.
Overall, the use of CLOC systems in the management of respiratory failure has now been tested across a range of hospital locations, in patients with different underlying conditions, using both LFOT and HFNOT. Our analysis shows that CLOC use significantly improves the proportion of time spent within a prespecified target SpO2 level compared with usual care with manual adjustments and represents an effective approach for maintaining hospitalised adults within a target SpO2 range when using noninvasive oxygen therapies.
Points for clinical practice
• Use of CLOC improves the accuracy of noninvasive oxygen delivery in achieving a target SpO2 range in hospitalised adults.
• The magnitude of the improvement is greater for CLOC use in standard hospital wards and when delivered via low-flow oxygen devices.
• CLOC represents an effective noninvasive oxygen delivery strategy for maintaining SpO2 within a target range in acute inpatient care.
Questions for future research
• Further research is required to establish the effect of CLOC on liberation from supplemental oxygen, hospital length of stay, morbidity and mortality.
Acknowledgements
Thank you to Te Herenga Waka – Victoria University Wellington Librarians who assisted with generating the search strategy for this study.
Footnotes
Provenance: Submitted article, peer reviewed.
The systematic review protocol was registered with PROSPERO with identifier: CRD42024500598.
Author contributions: L. Kirton had full access to all the data in this study, and takes complete responsibility for the integrity and accuracy of the data analysis. Conceptualisation: L. Kirton, M. Weatherall and R. Beasley. Data curation: L. Kirton, A. Eathorne and M. Weatherall. Formal analysis: A. Eathorne and M. Weatherall. Investigation: L. Kirton, R. Biggs, N. Kivi, F. Lynch, T. Kingi, R. Sayers and J. Sparks. Methodology: L. Kirton, M. Weatherall and R. Beasley. Project administration: L. Kirton, J. Sparks and R. Biggs. Software: A. Eathorne and M. Weatherall. Supervision: L. Kirton and R. Beasley. Validation: L. Kirton and R. Biggs. Visualisation: L. Kirton, M. Weatherall and R. Beasley. Writing – original draft: L. Kirton. Writing – review and editing: all authors.
Conflict of interest: L. Kirton reports support for attending meetings from Fisher and Paykel Healthcare. R. Biggs, N. Kivi, F. Lynch, T. Kingi, R. Sayers, J. Sparks, A. Eathorne and M. Weatherall have nothing to disclose. R. Beasley reports grants from Fisher and Paykel Healthcare.
Support statement: No funding declared.
Supplementary material
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Supplementary material
ERR-0167-2026.SUPPLEMENT
Data availability
Data included in this systematic review and meta-analysis can be made available upon request starting 1 year after publication and for a minimum of 5 years thereafter. Researchers must provide a methodologically sound proposal for consideration. Requests should be addressed to louis.kirton@mrinz.ac.nz
References
- 1.O'Driscoll BR, Howard LS, Earis J, et al. BTS guideline for oxygen use in adults in healthcare and emergency settings. Thorax 2017; 72: ii1–ii90. doi: 10.1136/thoraxjnl-2016-209729 [DOI] [PubMed] [Google Scholar]
- 2.Chu DK, Kim LH-Y, Young PJ, et al. Mortality and morbidity in acutely ill adults treated with liberal versus conservative oxygen therapy (IOTA): a systematic review and meta-analysis. Lancet 2018; 391: 1693–1705. doi: 10.1016/S0140-6736(18)30479-3 [DOI] [PubMed] [Google Scholar]
- 3.Hafner S, Beloncle F, Koch A, et al. Hyperoxia in intensive care, emergency, and peri-operative medicine: Dr. Jekyll or Mr. Hyde? A 2015 update. Ann Intensive Care 2015; 5: 42. doi: 10.1186/s13613-015-0084-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Beasley R, Chien J, Douglas J, et al. Thoracic Society of Australia and New Zealand oxygen guidelines for acute oxygen use in adults: ‘swimming between the flags’. Respirology 2015; 20: 1182–1191. doi: 10.1111/resp.12620 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Harper JCP, Semprini R, Kearns NA, et al. Determination of oxygen saturation compared to a prescribed target range using continuous pulse oximetry in acutely unwell medical patients. BMC Pulm Med 2021; 21: 332. doi: 10.1186/s12890-021-01700-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Lellouche F, Bouchard PA, Roberge M, et al. Automated oxygen titration and weaning with FreeO2 in patients with acute exacerbation of COPD: a pilot randomized trial. Int J Chron Obstruct Pulmon Dis 2016; 11: 1983–1990. doi: 10.2147/COPD.S112820 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.L'Her E, Dias P, Gouillou M, et al. Automatic versus manual oxygen administration in the emergency department. Eur Respir J 2017; 50: 1602552. doi: 10.1183/13993003.02552-2016 [DOI] [PubMed] [Google Scholar]
- 8.L'Her E, Jaber S, Verzilli D, et al. Automated closed-loop versus standard manual oxygen administration after major abdominal or thoracic surgery: an international multicentre randomised controlled study. Eur Respir J 2021; 57: 2000182. doi: 10.1183/13993003.00182-2020 [DOI] [PubMed] [Google Scholar]
- 9.Harper J, Kearns N, Bird G, et al. Automatic versus manual oxygen titration using a novel nasal high-flow device in medical inpatients with an acute illness: a randomised controlled trial. BMJ Open Respir Res 2021; 8: e000843. doi: 10.1136/bmjresp-2020-000843 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Roca O, Caritg O, Santafé M, et al. Closed-loop oxygen control improves oxygen therapy in acute hypoxemic respiratory failure patients under high flow nasal oxygen: a randomized cross-over study (the HILOOP study). Crit Care 2022; 26: 108. doi: 10.1186/s13054-022-03970-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Kirton LW, Cruz RS, Navarra L, et al. Effect of automated titration of oxygen on time spent in a prescribed oxygen saturation range in adults in the ICU after cardiac surgery. Crit Care Resusc 2024; 26: 64–70. doi: 10.1016/j.ccrj.2024.01.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Kirton LW, Eathorne AL, Cook JK, et al. Automated oxygen titration with noninvasive respiratory support in acute in-patient care. Respir Care 2025; 17: 1405–1414. doi: 10.1089/respcare.12745 [DOI] [PubMed] [Google Scholar]
- 13.Kirton L, Cook J, Hamill R, et al. Automated titration of nasal high flow oxygen in the emergency department: a randomised controlled trial. Emerg Med J 2026; 43: 215–222. doi: 10.1136/emermed-2024-214023 [DOI] [PubMed] [Google Scholar]
- 14.Denault M-H, Péloquin F, Lajoie A-C, et al. Automatic versus manual oxygen titration in patients requiring supplemental oxygen in the hospital: a systematic review and meta-analysis. Respiration 2019; 98: 178–188. doi: 10.1159/000499119 [DOI] [PubMed] [Google Scholar]
- 15.Mól CG, Vieira AGDS, Garcia BMSP, et al. Closed-loop oxygen control for critically ill patients–a systematic review and meta-analysis. PLoS One 2024; 19: e0304745. doi: 10.1371/journal.pone.0304745 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Higgins JPT, Thomas J, Chandler J, et al. Cochrane Handbook for Systematic Reviews of Interventions. Chichester, John Wiley & Sons, 2019. [Google Scholar]
- 17.Helmerhorst HJ, Schultz MJ, Van Der Voort PH, et al. Self-reported attitudes versus actual practice of oxygen therapy by ICU physicians and nurses. Ann Intensive Care 2014; 4: 23. doi: 10.1186/s13613-014-0023-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Robinson TD, Freiberg DB, Regnis JA, et al. The role of hypoventilation and ventilation-perfusion redistribution in oxygen-induced hypercapnia during acute exacerbations of chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2000; 161: 1524–1529. doi: 10.1164/ajrccm.161.5.9904119 [DOI] [PubMed] [Google Scholar]
- 19.Aubier M, Murciano D, Milic-Emili J, et al. Effects of the administration of O2 on ventilation and blood gases in patients with chronic obstructive pulmonary disease during acute respiratory failure. Am Rev Respir Dis 1980; 122: 747–754. doi: 10.1164/arrd.1980.122.5.747 [DOI] [PubMed] [Google Scholar]
- 20.Cutaia M, Rounds S. Hypoxic pulmonary vasoconstriction. Chest 1990; 97: 706–718. doi: 10.1378/chest.97.3.706 [DOI] [PubMed] [Google Scholar]
- 21.Thomson AJ, Webb DJ, Maxwell SRJ, et al. Oxygen therapy in acute medical care. BMJ 2002; 324: 1406–1407. doi: 10.1136/bmj.324.7351.1406 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Helmerhorst HJF, Roos-Blom M-J, Van Westerloo DJ, et al. Association between arterial hyperoxia and outcome in subsets of critical illness: a systematic review, meta-analysis, and meta-regression of cohort studies. Crit Care Med 2015; 43: 1508–1519. doi: 10.1097/CCM.0000000000000998 [DOI] [PubMed] [Google Scholar]
- 23.Austin MA, Wills KE, Blizzard L, et al. Effect of high flow oxygen on mortality in chronic obstructive pulmonary disease patients in prehospital setting: randomised controlled trial. BMJ 2010; 341: c5462–c5462. doi: 10.1136/bmj.c5462 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Beasley R, Mackle D, Young P. Oxygen: a new look at an old therapy. J R Soc N Z 2019; 49: 126–142. doi: 10.1080/03036758.2019.1566154 [DOI] [Google Scholar]
- 25.Buist M, Bernard S, Nguyen TV, et al. Association between clinically abnormal observations and subsequent in-hospital mortality: a prospective study. Resuscitation 2004; 62: 137–141. doi: 10.1016/j.resuscitation.2004.03.005 [DOI] [PubMed] [Google Scholar]
- 26.Goodacre S. Prediction of mortality among emergency medical admissions. Emerg Med J 2006; 23: 372–375. doi: 10.1136/emj.2005.028522 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Piraino T, Madden M, Roberts KJ, et al. AARC clinical practice guideline: management of adult patients with oxygen in the acute care setting. Respir Care 2022; 67: 115–128. doi: 10.4187/respcare.09294 [DOI] [PubMed] [Google Scholar]
- 28.Roca O, Li J, Mauri T. High-flow nasal cannula: evolving practices and novel clinical and physiological insights. Intensive Care Med 2024; 50: 758–761. doi: 10.1007/s00134-024-07386-8 [DOI] [PubMed] [Google Scholar]
- 29.Mauri T, Turrini C, Eronia N, et al. Physiologic effects of high-flow nasal cannula in acute hypoxemic respiratory failure. Am J Respir Crit Care Med 2017; 195: 1207–1215. doi: 10.1164/rccm.201605-0916OC [DOI] [PubMed] [Google Scholar]
- 30.Oczkowski S, Ergan B, Bos L, et al. ERS clinical practice guidelines: high-flow nasal cannula in acute respiratory failure. Eur Respir J 2022; 59: 2101574. doi: 10.1183/13993003.01574-2021 [DOI] [PubMed] [Google Scholar]
- 31.Rochwerg B, Einav S, Chaudhuri D, et al. The role for high flow nasal cannula as a respiratory support strategy in adults: a clinical practice guideline. Intensive Care Med 2020; 46: 2226–2237. doi: 10.1007/s00134-020-06312-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Frat J-P, Thille AW, Mercat A, et al. High-flow oxygen through nasal cannula in acute hypoxemic respiratory failure. N Engl J Med 2015; 372: 2185–2196. doi: 10.1056/NEJMoa1503326 [DOI] [PubMed] [Google Scholar]
- 33.Frat J-P, Quenot J-P, Guitton C, et al. High-flow or standard oxygen in acute hypoxemic respiratory failure. N Engl J Med 2026; 394: 2095–2106. doi: 10.1056/NEJMoa2516087 [DOI] [PubMed] [Google Scholar]
- 34.Wijesinghe M, Williams M, Perrin K, et al. The effect of supplemental oxygen on hypercapnia in subjects with obesity-associated hypoventilation. Chest 2011; 139: 1018–1024. doi: 10.1378/chest.10-1280 [DOI] [PubMed] [Google Scholar]
- 35.Pilcher J, Thayabaran D, Ebmeier S, et al. The effect of 50% oxygen on PtCO 2 in patients with stable COPD, bronchiectasis, and neuromuscular disease or kyphoscoliosis: randomised cross-over trials. BMC Pulm Med 2020; 20: 125. doi: 10.1186/s12890-020-1132-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Cirio S, Nava S. Pilot study of a new device to titrate oxygen flow in hypoxic patients on long-term oxygen therapy. Respir Care 2011; 56: 429–434. doi: 10.4187/respcare.00983 [DOI] [PubMed] [Google Scholar]
- 37.Taraldsen IA, Grand J, Lukoschewitz JD, et al. Automated oxygen administration versus manual control in acute cardiovascular care: a randomised controlled trial. Heart 2024; 111: 27–34. doi: 10.1136/heartjnl-2024-324488 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Supplementary material
ERR-0167-2026.SUPPLEMENT
Data Availability Statement
Data included in this systematic review and meta-analysis can be made available upon request starting 1 year after publication and for a minimum of 5 years thereafter. Researchers must provide a methodologically sound proposal for consideration. Requests should be addressed to louis.kirton@mrinz.ac.nz