Association between early hyperoxaemia exposure and intensive care unit mortality in intracerebral haemorrhage - an observational cohort analysis

doi:10.21203/rs.3.rs-3626842/v1

Background

Supraphysiologic levels of oxygen could have potential adverse effects on the brain. These effects may be dose and time-dependent in brain injured patients. We therefore aimed to assess whether exposure to excess supplemental oxygen, measured as time-weighted mean exposure to hyperoxaemia, was associated with Intensive Care Unit (ICU) mortality in Intracerebral Haemorrhage (ICH).

Methods

In this single-centre retrospective cohort study we included all patients admitted to our ICU with a diagnosis of primary spontaneous ICH and an ICU length of stay over 24 hours. To provide a longitudinal measure of hyperoxaemia exposure we calculated the hyperoxaemia dose, defined as the area under the PaO₂ time curve above the threshold PaO₂ value of 13.3 kPa (100 mmHg) divided by the number of hours of potential exposure. To provide consistent potential exposure windows, nested subsets were created with progressively longer exposure periods (0–1 day, 0–2 days, 0–3 days, 0–4 days, 0–5 days, 0–6 days, 0–7 days). We used Cox regression to model the time-dependent association between hyperoxaemia dose and ICU mortality.

Results

Between September 2019 and July 2022, 275 patients met the inclusion criteria, with 24,588 ABG results available for analysis. Almost all patients (97.45%) were exposed to hyperoxaemia during their ICU admission. Cox regression modelling showed an association between hyperoxaemia dose and ICU mortality (Hazard Ratio (HR) 1.15, CI 1.05–1.25, p = 0.003). This association was observed in the 0–1 day exposure window in a step-function extended Cox model (HR 1.19, CI 1.06–1.35, p = 0.005) but not in any of the subsequent exposure periods.

Conclusion

In ICH patients admitted to ICU, early exposure to hyperoxaemia was associated with increased ICU mortality. Further prospective study is required to inform guidance on early systemic oxygen targets in ICH.

Intracerebral Hemorrhage Stroke

Hyperoxia

Oxygen Inhalational Therapy

Blood Oxygen Levels

Mechanical Ventilation

Critical Care.

Supraphysiologic levels of oxygen in general intensive care unit (ICU) patients have been associated with increased mortality¹. Such concerns have led to the development of clinical practice guidelines which recommend conservative oxygen administration targets^2–3. It has, however, become clear that subgroups of patients with acute brain pathologies taken from these large general ICU populations⁴ need to be considered independently⁵.

Neuronal cells limited tolerance of low oxygen partial pressures mean that avoidance of hypoxaemia and maintenance of adequate cerebral oxygen delivery remain a crucial component of neurocritical care⁶. However, liberal oxygen therapy is controversial^7–8 and could have potential adverse effects on the brain. The mechanisms remain poorly understood, although may include oxidative stress⁹, neuroinflammation¹⁰, excitotoxicity¹¹, and vasoconstriction¹².

Early exposure to hyperoxaemia has been associated with increased mortality in Hypoxic-Ischaemic Encephalopathy (HIE)¹³, aneurysmal Subarachnoid Haemorrhage (aSAH)¹⁴, Traumatic Brain Injury (TBI)¹⁵, and Acute Ischaemic Stroke (AIS)¹⁶, and has also been associated with poor neurological outcomes in these conditions^17,5. However, equivalent retrospective studies in these models of brain injury have yielded conflicting conclusions¹⁸.

Intracerebral haemorrhage (ICH) is a form of haemorrhagic stroke with persistently high mortality¹⁹. Initial neurocritical care is mainly focused on limiting damage to surrounding brain tissue by prevention of haematoma expansion through meticulous blood pressure control and anticoagulant reversal²⁰. Pathophysiological processes triggered by the haematoma that influence secondary brain injury include oxidative stress, neuroinflammation, excitotoxicity, iron toxicity, and thrombin formation¹⁰. The role that oxygen may have in limiting or potentiating these processes is poorly understood, and guidance on systemic oxygen targets in ICH is lacking.

Early exposure to hyperoxaemia has been associated with increased mortality in ICH¹⁶.

However, many of the observational studies examining the effects of hyperoxaemia in brain injury are limited by defining excess oxygen exposure by a single PaO₂ reading. Time weighted hyperoxaemia exposure across an ICU admission might be a more comprehensive measure²¹. Similar to intracranial pressure dose²², calculating hyperoxaemia dose considers a wider range of oxygen exposure, including the potentially harmful effects of low levels of hyperoxaemia for prolonged periods, and allows the potentially time-dependent effects of oxygen exposure to be investigated²³.

We therefore aimed to assess whether exposure to hyperoxaemia, measured as time weighted mean exposure, was associated with ICU mortality in ICH.

Study Design

This study is a single-centre retrospective cohort analysis. The study design was guided by the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement on observational cohort studies²⁴. Ethical approval was sought from the local Health Research Authority National Health Service Research Ethics Committee (REC ref. 23/LO/0038).

Patients

The study was conducted at the Department of Critical Care Medicine at King’s College Hospital NHS Foundation Trust, a large university-affiliated teaching hospital and tertiary critical care unit serving a diverse ethnic and socio-economic population.

Data was collected retrospectively between the introduction of the electronic patient healthcare record in September 2019 and July 2022. We screened all patients over 18 years of age admitted to our ICU with a diagnosis of primary spontaneous ICH confirmed on computed tomography (CT) and computed tomography angiogram (CTA). Patients were included if their ICU length of stay was over 24 hours. Those admitted for less than 24 hours were typically admitted for early withdrawal of life-sustaining treatment, in addition to not providing a sufficient exposure window for the effects of hyperoxaemia to be assessed. For patients that were readmitted to ICU, only the index admission was included to limit confounding by unmeasured oxygen exposure outside the ICU.

Arterial blood gas (ABG) results and data concerning demographics, interventions, and outcomes were collected retrospectively from the electronic patient healthcare record and radiology imaging system. Initial disease severity was quantified using the ICH score and Acute Physiology And Chronic Health Evaluation (APACHE) II score. General and disease specific management conformed to international guidelines²⁰.

Hyperoxaemia defintion

In the absence of an agreed definition, and to avoid categorisation of PaO₂ data, we defined hyperoxaemia as PaO₂ > 13.3kPa (100 mmHg), as values exceeding this can only be achieved with supplementary oxygen regardless of treatment indication. In order to provide a time-weighted, longitudinal, measure of hyperoxaemia exposure we calculated the hyperoxaemia dose which was defined as the area under the PaO₂ time curve above the threshold PaO₂ value of 13.3 kPa (100 mmHg) divided by the number of hours of potential exposure²¹. Linear changes between PaO₂ values obtained from ABG were assumed (Fig. 1).

To provide consistent potential hyperoxaemia exposure windows, nested subsets were created with progressively longer exposure periods (0–1 day, 0–2 days, 0–3 days, 0–4 days, 0–5 days, 0–6 days, 0–7 days). Patients were only included in these subsets if they had complete data for the full exposure window, therefore limiting potential bias from non-random censoring (death or ICU discharge) within the assessed time period. Hyperoxaemia dose was calculated for each nested subset.

Patients were followed up until ICU discharge and missing variables were under 1%. The primary outcome of ICU mortality was chosen as being unaffected by unmeasured oxygen exposure after ICU discharge.

Statistics

Continuous variables were assessed for normality with the Shapiro-Wilk test, and reported as mean ± standard deviation (SD) for parametric data, or median and interquartile range (IQR) for non-parametric data. Categorical variables were reported as counts and proportions in each group.

Differences in baseline characteristics between the nested subsets were assessed with one-way analysis of variance (ANOVA) for parametric variables, the Kruskal-Wallis test for non-parametric variables, and the Pearson Chi-squared test for dichotomous categorical variables.

To assess the influence of variations in hyperoxaemia exposure over time on ICU mortality we fitted a multivariate Cox regression model with hyperoxaemia dose as a time-dependent coefficient²⁵. This survival analysis was chosen as a statistical approach that limited any potential survival bias, and allowed non-proportional hazards to be expressed. A standard multivariate Cox regression model was initially fitted. Advanced cardiac support (yes/no), age, APACHE II score, craniotomy (yes/no), ICH score, sex, and weight were selected a priori as predictor covariables based on the likelihood of influence on the research question. The proportional-hazards assumption of the Cox model was tested by calculating Schoenfeld residuals and visualised by a line fitted to the residual plot²⁶. A multivariate Cox regression model was then fitted, with hyperoxaemia dose as a time-dependent covariate. Finally, a step-function extended Cox model was fitted with hyperoxaemia dose calculated at 1 day intervals. Overall model fit was assessed by calculating Cox–Snell residuals and the Nelson-Aalen cumulative hazard estimator. Deviance and DFbeta residuals were calculated to examine model accuracy and identify outliers.

In view of the lack of an agreed definition of hyperoxaemia, and to allow comparison to other studies, we performed three sensitivity analyses to determine how different models and hyperoxaemia definitions may influence outcomes. In the first, we applied the methodology used in a previous investigation of hyperoxia and mortality in ICH¹⁶ to our dataset. Patients were divided into three exposure groups: hyperoxaemia (> 39.99kPa), normoxaemia (7.99-39.99kPa), and hypoxaemia (< 7.99kPa), based on their initial ICU ABG PaO₂ reading. ICU mortality was compared between groups and the influence of hyperoxaemia exposure assessed with multivariable logistic regression. In the second, we adjusted the PaO₂ threshold value from our original methodology from 13.3kPa to 20kPa (150mmHg), a threshold value adopted from previous studies¹⁴. We then calculated hyperoxaemia dose above this threshold for each nested subset and modelled ICU mortality as a function of hyperoxaemia dose with Cox regression as described previously. In the third, we modelled ICU mortality as a function of hyperoxaemia dose using multivariable logistic regression using methodology described in more detail elsewhere²¹. The co-variates used for this logistic regression model were the same as those described above for the Cox regression. Individual models were fitted for each exposure subset, and model validation was performed using the area under the receiver operating characteristic curve (AUROC). Precision-recall, brier scores and akaike information criterion (AIC) were also calculated to assess model performance. The average treatment effect of exposure to hyperoxaemia was calculated by fitting each model with each individuals hyperoxaemia exposure set to 0 and comparing it with the true hyperoxaemia exposure.

All analyses were performed with R Version 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria). A p value of < 0.05 was considered significant.

337 consecutive patients with ICH were identified within the 3-year study period. From these, 275 patients were admitted to our critical care unit for over 24 hours and met the inclusion criteria, with 24,588 ABG results available for analysis (Figure 2).

A summary of the patients baseline characteristics for the 0-1 day exposure cohort, along with selected other hyperoxaemia exposure window subsets, is shown in Table 1. Almost all patients (97.45%) were exposed to hyperoxaemia during their critical care admission.

Kaplan-Meier estimated survival plot is shown in Figure 3. Initial survival analysis with multivariate Cox regression showed an association between hyperoxaemia dose and ICU mortality (Hazard Ratio (HR) 1.38, CI 1.24 - 1.54, p=0.000000002) (Table 2). The time-dependent variation in hyperoxaemia exposure was tested by calculating Schoenfeld residuals (Table 3). Figure 4 illustrates the non-proportional hazards observed in initial multivariate Cox regression analysis.

Multivariate Cox regression modelling using hyperoxaemia dose as a time-dependent coefficient showed an association between hyperoxaemia dose and ICU mortality (HR 1.15, CI 1.05 - 1.25, p=0.003) (Table 4). The step-function extension of the Cox model revealed 0 - 1 day as the only exposure window where a significant association between hyperoxaemia dose and ICU mortality was observed (HR 1.19, CI 1.06 - 1.35, p=0.005), Table 5. Cox-Snell, DFbeta and deviance residuals confirmed acceptable model performance (Figure 5 and Figure 6).

In the first sensitivity analysis, by dividing patients into 3 hyperoxaemia exposure groups we did not observe a significantly higher mortality in the hyperoxaemia group compared to the normoxaemia group (crude OR 2.16 (0.21-107.83), p=0.6608), or in the hyperoxaemia group compared to the hypoxia group (crude OR 0.99 (0.35 - 2.59), p=1) (Table 6). There was also no association between hyperoxaemia exposure, by this definition, and ICU mortality in multivariable logistic regression analysis (adjusted OR 0.64 (0.03-6.17), p=0.7151) (Table 7).

In the second sensitivity analysis, adjusting the threshold PaO₂ value when calculating hyperoxaemia dose to 20kPa yielded equivalent results in Cox regression analysis to the threshold value of 13.3kPa (Table 8 and Table 9).

In the third sensitivity analysis, we did not observe an association between hyperoxaemia dose and ICU mortality in multivariable logistic regression modelling for any of the exposure subsets (Figure 7 and Table 10). Overall model fit was good until the 5 day exposure window (Figure 8 and Figure 9). The average treatment effect of exposure to hyperoxaemia, calculated by setting each individuals hyperoxaemia exposure to 0 and comparing it with the observed hyperoxaemia exposure, for the 0-1 day, 0-2 days, 0-3 days, 0-4 days, and 0-5 days exposure subsets were 1.02%, 0.57%, 1.42%, 0.17%, and 1.21%, favouring no exposure to hyperoxaemia (Figure 10).

Early exposure to hyperoxaemia, within the first 24 hours of ICU admission, was observed to be associated with increased ICU mortality. This builds significantly on previous analysis of hyperoxaemia exposure in ICH¹⁶ by demonstrating a dose-dependent and time-varying relationship with ICU mortality through calculation of a longitudinal measure of oxygen exposure.

As this dataset demonstrates, brain injured patients managed on ICU often require ventilatory support with associated oxygen therapy and the subsequent risk of exposure to excess supplemental oxygen. Observational studies of various models of brain injury^{12–15,27-28} have shown an association between hyperoxaemia exposure and increased mortality. Comparison of these studies^17–18 can be complicated by the broad range of definitions of hyperoxaemia exposure used.

Although high FiO₂ may be the causative mechanism for lung parenchymal damage, from a neurocritical care perspective (in the absence of brain tissue oxygen (PbtO₂) monitoring) measuring PaO₂ seems the most appropriate way to assess the potential pathophysiological effects of oxygen on the brain. However, many previous retrospective studies have modelled outcomes using a single PaO₂ measurement, the first or highest PaO₂ result, to represent oxygen exposure for the entire ICU admission. Modelling oxygen exposure as a longitudinal measure may be a more comprehensive approach, considering a wider range of oxygen exposure across the admission, with less vulnerability to the influence of treatment effects compared to a single measure²¹.

The level at which hyperoxaemia may become harmful is not clear, but defining hyperoxaemia as any measurement over 13.3kPa excludes hypoxia exposure from this oxygen parameter, which itself may affect outcomes²⁸. This approach also avoids categorisation into different levels of PaO₂ which could increase false-positive findings. Almost all patients in this study were exposed to hyperoxaemia making it difficult to test any hyperoxaemia exposure against no exposure. Instead we tested dose-response relationships which have previously been difficult to demonstrate clearly^14,21,29 .

In an attempt to address the time varying effects of oxygen exposure over an ICU admission we used a Cox regression model with a time-dependent coefficient extension. Survival analyses such as this have the potential to be exposed to bias from informative censoring, however, we attempted to minimise this by using nested subsets, where hyperoxaemia dose was calculated for patients who had complete data until the end of that time exposure window, without non-random censoring from discharge or death.

The association between hyperoxaemia dose and ICU mortality was not observed to be constant over time, and appears to be most significant in the first 24 hours. This is important as a significant amount of hyperoxaemia exposure is likely to occur within the first 24 hours in ICH. Patients are commonly placed on high FiO₂ levels in the Emergency Department for transfer to and from initial CT scans, in addition to inter-hospital transfers to receiving neurosurgical centres. This may be an argument to support a more closely titrated oxygen therapy approach during this period.

Our finding of an association between early ICU hyperoxaemia exposure and in-hospital mortality is in line with results from the only previous observational study of hyperoxaemia exposure in ICH patients¹⁶. However, in a sensitivity analysis we were not able to reproduce the results from this previous studies methodology. This may represent a change in practice over time, as the deleterious effects of hyperoxaemia have become more widely recognised, as very few patients in our dataset were exposed to such high individual PaO₂ readings (> 40kPa).

In a further sensitivity analysis, using multivariable logistic regression modelling, we were unable to show a significant dose-response relationship between hyperoxaemia exposure and ICU mortality. This is consistent with results from a previous retrospective cohort study in general ICU patients employing this methodology²¹ where a dose-dependent effect was not observed, but a consistent association between any exposure to hyperoxemia and ICU mortality was found, suggesting harm from exposure to supraphysiologic levels of oxygen. Calculating hyperoxaemia dose compresses serial PaO₂ readings into a single measure. Including the time to ICU mortality in Cox survival analysis adds back statistical information which is lost in the binary outcome of logistic regression, and may explain the detection of a dose-dependent relationship with this method.

Limitations of this study include the single-centre design which mean the results may not be generalisable to other patient populations. The association between hyperoxaemia exposure and increased ICU mortality was not significant across models, limiting interpretation of hyperoxaemia exposure as a causative mechanism. It remains possible that the observed association is influenced by patients who would be expected to have high levels of oxygen exposure and high mortality, such as patients requiring multiple transfers for CT imaging. Although, as outlined, there are multiple pathophysiological processes that could theoretically be potentiated by hyperoxia in ICH.

Pre-ICU oxygen exposure was not measured in this study, but has previously been shown to negatively effect outcomes³⁰. Although mean pre-ICU length of stay was very short in this dataset, under 1 day, the oxygen exposure in the Emergency Department is likely to be highly variable. We did not include assessment of neurological outcomes in this study. Unfavourable outcomes have previously been reported to be associated with hyperoxaemia exposure^5,17, however, there is likely to be an influence from oxygen exposure after ICU discharge in addition to other unknown factors which are difficult to adequately control for.

In ICH patients admitted to ICU, early exposure to hyperoxaemia was associated with increased ICU mortality. Further prospective study is required to inform guidance on early systemic oxygen targets in ICH.

ICU Intensive Care Unit, HIE Hypoxic-Ischaemic Encephalopathy, aSAH aneurysmal Subarachnoid Haemorrhage, TBI Traumatic Brain Injury, AIS Acute Ischaemic Stroke, ICH Intracerebral Haemorrhage, PaO₂ Partial pressure of oxygen in arterial blood, ICP Intracranial pressure, CT Computed tomography, CTA Computed tomography angiogram, STROBE Strengthening the Reporting of Observational Studies in Epidemiology, APACHE Acute Physiology And Chronic Health Evaluation, SD Standard deviation, IQR Interquartile range, AUROC Area Under the Receiver Operating characteristic Curve, AIC Akaike Information Criterion, OR Odds ratio, CI Confidence interval, HR Hazard ratio, FiO₂ Fraction of inspired oxygen.

I confirm that this manuscript complies with all instructions to authors.

I confirm that authorship requirements have been met and the final manuscript was approved by all authors.

I confirm that this manuscript has not been published elsewhere and is not under consideration by another journal.

I confirm adherence to ethical guidelines and indicate ethical approvals and use of informed consent, as appropriate.

I disclose no Conflicts of Interest for all authors .

I confirm the use of STROBE reporting checklist.

There were no sources of funding for this study.

Acknowledgements

Author Contributions

ME and RM were involved in the study idea and design, data acquisition and analysis, writing and drafting of the manuscript. All authors read and approved this version of the final manuscript.

Chu DK, Kim LH, 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(10131):1693-1705. doi:10.1016/S0140-6736(18)30479-3
Siemieniuk RAC, Chu DK, Kim LH, et al. Oxygen therapy for acutely ill medical patients: a clinical practice guideline. BMJ. 2018;363:k4169. Published 2018 Oct 24. doi:10.1136/bmj.k4169
O'Driscoll BR, Howard LS, Earis J, Mak V. British Thoracic Society Guideline for oxygen use in adults in healthcare and emergency settings. BMJ Open Respir Res. 2017;4(1):e000170. Published 2017 May 15. doi:10.1136/bmjresp-2016-000170
ICU-ROX Investigators and the Australian and New Zealand Intensive Care Society Clinical Trials Group, Mackle D, Bellomo R, et al. Conservative Oxygen Therapy during Mechanical Ventilation in the ICU. N Engl J Med. 2020;382(11):989-998. doi:10.1056/NEJMoa1903297
Young PJ, Mackle D, Hodgson C, et al. Conservative or liberal oxygen therapy for mechanically ventilated adults with acute brain pathologies: A post-hoc subgroup analysis. J Crit Care. 2022;71:154079. doi:10.1016/j.jcrc.2022.154079
Okonkwo DO, Shutter LA, Moore C, et al. Brain Oxygen Optimization in Severe Traumatic Brain Injury Phase-II: A Phase II Randomized Trial. Crit Care Med. 2017;45(11):1907-1914. doi:10.1097/CCM.0000000000002619
Singer M, Young PJ, Laffey JG, et al. Dangers of hyperoxia. Crit Care. 2021;25(1):440. Published 2021 Dec 19. doi:10.1186/s13054-021-03815-y
Diringer MN, Aiyagari V, Zazulia AR, Videen TO, Powers WJ. Effect of hyperoxia on cerebral metabolic rate for oxygen measured using positron emission tomography in patients with acute severe head injury. J Neurosurg. 2007;106(4):526–529
Helmerhorst, H.J.F., Schultz, M.J., van der Voort, P.H.J. et al. Bench-to-bedside review: the effects of hyperoxia during critical illness. Crit Care 19, 284 (2015). https://doi.org/10.1186/s13054-015-0996-4
Aronowski J, Zhao X. Molecular pathophysiology of cerebral hemorrhage: secondary brain injury. Stroke. 2011;42(6):1781-1786. doi:10.1161/STROKEAHA.110.596718
Quintard, H., Patet, C., Suys, T. et al. Normobaric Hyperoxia is Associated with Increased Cerebral Excitotoxicity After Severe Traumatic Brain Injury. Neurocrit Care 22, 243–250 (2015). https://doi.org/10.1007/s12028-014-0062-0
Fukuda S, Koga Y, Fujita M, Suehiro E, Kaneda K, Oda Y, Ishihara H, Suzuki M, Tsuruta R. Hyperoxemia during the hyperacute phase of aneurysmal subarachnoid hemorrhage is associated with delayed cerebral ischemia and poor outcome: a retrospective observational study. J Neurosurg. 2021;134(1):25–32
Kilgannon JH, Jones AE, Shapiro NI, et al. Association between arterial hyperoxia following resuscitation from cardiac arrest and in-hospital mortality. JAMA. 2010;303(21):2165-2171. doi:10.1001/jama.2010.707
Grensemann J, Mader MM, Westphal M, Kluge S, Czorlich P. Hyperoxia is Dose-Dependently Associated with an Increase of Unfavorable Outcomes in Ventilated Patients with Aneurysmal Subarachnoid Hemorrhage: A Retrospective Cohort Study. Neurocrit Care. 2022;37(2):523-530. doi:10.1007/s12028-022-01534-y
Rezoagli E, Petrosino M, Rebora P, et al. High arterial oxygen levels and supplemental oxygen administration in traumatic brain injury: insights from CENTER-TBI and OzENTER-TBI [published correction appears in Intensive Care Med. 2023 Feb;49(2):269-272]. Intensive Care Med. 2022;48(12):1709-1725. doi:10.1007/s00134-022-06884-x
Rincon F, Kang J, Maltenfort M, et al. Association between hyperoxia and mortality after stroke: a multicenter cohort study. Crit Care Med. 2014;42(2):387-396. doi:10.1097/CCM.0b013e3182a27732
Hirunpattarasilp, C., Shiina, H., Na-Ek, N. et al. The Effect of Hyperoxemia on Neurological Outcomes of Adult Patients: A Systematic Review and Meta-Analysis. Neurocrit Care 36, 1027–1043 (2022). https://doi.org/10.1007/s12028-021-01423-w
Stolmeijer R, Bouma HR, Zijlstra JG, Drost-de Klerck AM, Ter Maaten JC, Ligtenberg JJM. A Systematic Review of the Effects of Hyperoxia in Acutely Ill Patients: Should We Aim for Less?. Biomed Res Int. 2018;2018:7841295. Published 2018 May 14. doi:10.1155/2018/7841295
Jolink WM, Klijn CJ, Brouwers PJ, Kappelle LJ, Vaartjes I. Time trends in incidence, case fatality, and mortality of intracerebral hemorrhage. Neurology. 2015; 85:1318–1324. doi: 10.1212/WNL.0000000000002015
Greenberg SM, Ziai WC, Cordonnier C, et al. 2022 Guideline for the Management of Patients With Spontaneous Intracerebral Hemorrhage: A Guideline From the American Heart Association/American Stroke Association. Stroke. 2022;53(7):e282-e361. doi:10.1161/STR.0000000000000407
Palmer E, Post B, Klapaukh R, et al. The Association between Supraphysiologic Arterial Oxygen Levels and Mortality in Critically Ill Patients. A Multicenter Observational Cohort Study. Am J Respir Crit Care Med. 2019;200(11):1373-1380. doi:10.1164/rccm.201904-0849OC
Åkerlund CA, Donnelly J, Zeiler FA, et al. Impact of duration and magnitude of raised intracranial pressure on outcome after severe traumatic brain injury: A CENTER-TBI high-resolution group study. PLoS One. 2020;15(12):e0243427. Published 2020 Dec 14. doi:10.1371/journal.pone.0243427
Zhu, Z., Zhou, M., Wei, Y. et al. Time-varying intensity of oxygen exposure is associated with mortality in critically ill patients with mechanical ventilation. Crit Care 26, 239 (2022). https://doi.org/10.1186/s13054-022-04114-w
von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007;370:1453–1457
Zhang Z, Reinikainen J, Adeleke KA, Pieterse ME, Groothuis-Oudshoorn CGM. Time-varying covariates and coefficients in Cox regression models. Ann Transl Med. 2018;6(7):121. doi:10.21037/atm.2018.02.12
Grambsch, P., Therneau, T. Proportional hazards tests and diagnostics based on weighted residuals, Biometrika, 1994;81(3): 515–526. https://doi.org/10.1093/biomet/81.3.515
Yokoyama S, Hifumi T, Kawakita K, Tamiya T, Minamino T, Kuroda Y. Early Hyperoxia in The Intensive Care Unit is Significantly Associated With Unfavorable Neurological Outcomes in Patients With Mild-to-Moderate Aneurysmal Subarachnoid Hemorrhage. Shock. 2019;51(5):593-598. doi:10.1097/SHK.0000000000001221
Davis DP, Meade W, Sise MJ, Kennedy F, Simon F, Tominaga G, et al. Both hypoxemia and extreme hyperoxemia may be detrimental in patients with severe traumatic brain injury. J Neurotrauma 2009;26:2217–2223
Helmerhorst HJ, Arts DL, Schultz MJ, et al. Metrics of Arterial Hyperoxia and Associated Outcomes in Critical Care. Crit Care Med. 2017;45(2):187-195. doi:10.1097/CCM.0000000000002084
Page, D, Ablordeppey, E, Wessman, BT, et al. Emergency department hyperoxia is associated with increased mortality in mechanically ventilated patients: a cohort study. Crit Care 22, 9 (2018). https://doi.org/10.1186/s13054-017-1926-4

Table 1. Patient demographics, interventions and outcomes.

	0-1d exposure (n=275)	0-3d exposure (n=197)	0-5d exposure (n=161)	0-7d exposure (n=132)	p-value
Demographics
Age (years), mean (SD)	57.19 (13.99)	55.53 (12.77)	55.41 (12.76)	55.95 (12.68)	0.447
Male:Female	59:41	60:40	61:39	56:44	0.9013
BMI (kg/m²), median (IQR)	25.11 (23.15 - 28.34)	24.93 (23.03 - 28.03)	25.11 (23.15 - 28.34)	25.73 (23.38 - 28.40)	0.785
Ethnicity
White or white British, % (n)	52.00 (143)	47.72 (94)	47.20 (76)	46.97 (62)	0.9135
Black or black British, % (n)	24.00 (66)	30.45 (60)	26.71 (43)	28.79 (38)	0.6753
Asian or Asian British, % (n)	6.55 (18)	6.59 (13)	6.83 (11)	6.82 (9)	0.9994
Admission APACHE II, median (IQR)	14 (10 - 20)	12 (9-17)	12 (9-16)	11 (8-16)	0.0001751
Admission ICH score, median (IQR)	3 (2-3)	2 (2-3)	2 (2-3)	2 (2-3)	0.0234
Admission GCS, median (IQR)	8 (5-10)	8 (6-11)	8 (6-11)	8 (6-11)	0.1051
Pre-ICU length of hospital stay (d), median (IQR)	1 (1-1)	1 (1-1)	1 (1-1)	1 (1-1)	0.9638
Interventions
EVD % (n)	30.90 (85)	38.07 (75)	45.34 (73)	48.48 (64)	0.07997
Decompressive hemi-craniectomy, % (n)	9.45 (26)	12.18 (24)	14.29 (23)	16.67 (22)	0.281
Craniotomy, % (n)	24.00 (66)	28.93 (57)	29.81 (48)	29.55 (39)	0.6824
Mechanical Ventilation, % (n)	93.09 (256)	94.92 (187)	97.52 (157)	99.24 (131)	0.974
Pneumonia, % (n)	14.91 (41)	19.79 (39)	23.60 (38)	26.52 (35)	0.1039
Tracheostomy, % (n)	20.72 (57)	28.43 (56)	34.78 (56)	41.67 (55)	0.007928
ICP monitoring, % (n)	6.91 (19)	9.14 (18)	9.94 (16)	9.09 (12)	0.739
Number of CT head scans, median (IQR)	2 (1-4)	3 (2-5)	3 (2-5)	4 (3-5)	0.0000000000005263
Outcomes
Number of ventilated days (days), median (IQR)	5 (2-13)	9 (4-16)	11(7-18)	14 (9-21)	0.00000000000000022
ICU LOS (days), median (IQR)	7 (4-18)	13 (7-22)	16 (10-25)	19 (13-26)	0.00000000000000022
ICU Mortality, % (n)	35.64 (98)	24.87 (49)	18.01 (29)	12.88 (17)	0.000331
Oxygenation
Hyperoxaemia dose (kPa), median (IQR)	0.64 (0.19-1.33)	0.46 (0.21-0.93)	0.41 (0.24-0.76)	0.43 (0.26-0.82)	0.004932
Any hyperoxia exposure (yes), % (n)	92.36 (254)	96.95 (191)	98.76 (159)	100 (132)	0.9446
Cumulative hyperoxaemia exposure (kPa.h), median (IQR)	15.47 (4.50 - 31.99)	33.24 (15.37 - 67.27)	48.64 (28.41 - 91.19)	61.47 (37.13 - 96.24)	0.00000000000000022

SD standard deviation, BMI body mass index, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage, GCS Glasgow coma score, EVD external ventricular drain, ICP intracranial pressure, CT Computed tomography, ICU intensive care unit, LOS length of stay

Table 2. Multivariate Cox regression

Variable	HR	2.5%	97.5%	p-value
Hyperoxaemia dose	1.38	1.24	1.54	0.00000000231
Advanced cardiac support	1.45	0.92	2.27	0.1080
Age	1.00	0.98	1.02	0.9775
APACHE II	1.09	1.05	1.12	0.000000479
Craniotomy	0.45	0.23	0.89	0.0218
ICH Score	2.51	1.90	3.31	0.0000000000706
Sex	0.93	0.59	1.47	0.7574
Weight	1.01	1.00	1.02	0.1788

HR Hazard Ratio, CI Confidence Interval, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage

Table 3. Multivariate Cox regression Schoenfeld residuals

Variable	Chi-sq	DoF	Schoenfeld test
Hyperoxaemia dose	7.170	1	0.0074
Advanced cardiac support	1.964	1	0.1610
Age	0.014	1	0.9059
APACHE II	2.068	1	0.1504
Craniotomy	2.165	1	0.1412
ICH Score	4.119	1	0.0424
Sex	3.932	1	0.0474
Weight	2.100	1	0.1473
Global	20.634	8	0.0082

Chi-sq Chi-squared test, DoF Degrees of freedom, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage

Table 4. Multivariate Cox regression with time-dependent co-efficient

Variable	HR	2.5%	97.5%	p-value
Hyperoxaemia dose	1.15	1.05	1.25	0.00319
Advanced cardiac support	1.34	0.85	2.11	0.21450
Age	1.00	0.98	1.02	0.97119
APACHE II	1.10	1.06	1.13	0.0000000261
Craniotomy	0.23	0.09	0.56	0.00120
ICH Score	2.79	2.08	3.75	0.0000000000104
Sex	0.86	0.54	1.37	0.52609
Weight	1.00	0.99	1.02	0.81448

HR Hazard Ratio, CI Confidence Interval, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage

Table 5. Multivariate Cox regression with time-dependent coefficient step function

Variable	HR	2.5%	97.5%	p-value
Hyperoxaemia dose strata
Hyperoxaemia dose 1d	1.19	1.06	1.35	0.00482
Hyperoxaemia dose 2d	1.07	0.92	1.25	0.39237
Hyperoxaemia dose 3d	1.24	0.99	1.55	0.06582
Hyperoxaemia dose 4d	1.28	0.86	1.91	0.22031
Hyperoxaemia dose 5d	1.44	0.95	2.19	0.08883
Hyperoxaemia dose 6d	1.15	0.37	3.54	0.81167
Hyperoxaemia dose 7d	1.06	0.30	3.73	0.92321
Advanced cardiac support	1.37	0.87	2.18	0.17781
Age	1.00	0.98	1.02	0.94741
APACHE II	1.10	1.06	1.13	0.0000000282
Craniotomy	0.20	0.08	0.53	0.00104
ICH Score	2.81	2.08	3.80	0.0000000000158
Sex	0.82	0.52	1.31	0.41110
Weight	1.00	0.99	1.02	0.83135

HR Hazard Ratio, CI Confidence Interval, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage

Table 6. Patient numbers in each group for sensitivity analysis following Rincon et al¹⁶ methodology

Group	Alive	Dead
Hyperoxaemia (pO2>40 kPa) group (n)	4	1
Normoxaemia (pO2 8-39.99 kPa) group (n)	157	85
Hypoxaemia (pO2<7.99 kPa) group (n)	15	8

Table 7. Logistic regression model output for sensitivity analysis following Rincon et al¹⁶ methodology.

Variable	OR	2.5%	97.5%	p-value	Chi-sq	DoF	AIC	Brier Score
Age	0.9872	0.9619	1.0127	0.32238	0.28375	1	242.37	0.1466954
Weight	1.0104	0.9903	1.0306	0.30104	0.70600	1
Sex	0.5381	0.2500	1.1389	0.10747	0.33703	1
Advanced Cardiac Support	2.6321	1.2876	5.4920	0.00863	0.00743	1
APACHE II	1.1367	1.0746	1.2092	0.000019	0.00000000003975	1
Craniotomy	0.3997	0.1601	0.9317	0.03973	0.00382	1
ICH Score	2.7502	1.9009	4.1295	0.000000284	0.000000009753	1
Admission PaO₂>40 kPa (yes/no)	0.6350	0.0285	6.1720	0.71508	0.70836	1

OR odds ratio, Chi-sq Chi-squared test, DoF degrees of freedom, AUROC area under the receiver operator curve, AIC akaike information criterion, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage, PaO₂ Partial pressure of oxygen in arterial blood

Table 8. Multivariate Cox regression with time-dependent co-variate for the sensitivity analysis with hyperoxamemia definition set to threshold PaO2> 20kPa.

Variable	HR	2.5%	97.5%	p-value
Hyperoxaemia dose	1.16	1.03	1.31	0.01495
Advanced cardiac support	1.35	0.86	2.14	0.19461
Age	1.00	0.98	1.02	0.95311
APACHE II	1.10	1.06	1.13	0.0000000176
Craniotomy	0.24	0.10	0.58	0.00154
ICH Score	2.78	2.07	3.74	0.0000000000111
Sex	0.86	0.54	1.37	0.49692
Weight	1.00	0.99	1.02	0.94301

HR Hazard Ratio, CI Confidence Interval, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage, PaO₂ Partial pressure of oxygen in arterial blood

Table 9. Multivariate Cox regression with time-dependent co-efficient step function for the sensitivity analysis with hyperoxamemia definition set to threshold PaO2> 20kPa.

Variable	HR	2.5%	97.5%	p-value
Hyperoxaemia dose strata
Hyperoxaemia dose 1d	1.21	1.02	1.44	0.02961
Hyperoxaemia dose 2d	1.09	0.88	1.35	0.44961
Hyperoxaemia dose 3d	1.25	0.92	1.70	0.14938
Hyperoxaemia dose 4d	1.36	0.87	2.13	0.17262
Hyperoxaemia dose 5d	1.46	0.82	2.58	0.19693
Hyperoxaemia dose 6d	1.14	0.37	3.51	0.82509
Hyperoxaemia dose 7d	1.02	0.14	7.35	0.98193
Advanced cardiac support	1.36	0.86	2.16	0.18857
Age	1.00	0.98	1.02	0.92640
APACHE II	1.10	1.06	1.13	0.0000000178
Craniotomy	0.22	0.09	0.57	0.00193
ICH Score	2.81	2.08	3.79	0.0000000000124
Sex	0.83	0.52	1.32	0.43570
Weight	1.00	0.99	1.01	0.96338

HR Hazard Ratio, CI Confidence Interval, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage, PaO₂ Partial pressure of oxygen in arterial blood

Table 10. Logistic regression model output for sensitivity analysis following Palmer et al²¹ methodology. Model performance metrics include AUROC, Brier score, and AIC. AUROC > 0.7 associated with acceptable model performance. Brier score closer to 0 is preferable. Average treatment effects calculated as difference in mortality between observed exposure to hyperoxemia and the counterfactual scenario setting this exposure to 0.

Exposure	Variable	OR	2.5%	97.5%	p-value	Chi-sq	DoF	AUROC/c-index	AIC	Brier Score	Average treatment effect
1d exposure	Hyperoxaemia dose	1.0576	0.8943	1.2349	0.4543	0.603078	1	0.836677	246.61	0.1466707	1.02%
	Age	0.9884	0.9635	1.0137	0.3671	0.261282	1
	Weight	1.0105	0.9903	1.0309	0.2981	0.717552	1
	Sex	0.5593	0.2603	1.1820	0.1303	0.386533	1
	Advanced Cardiac Support	2.5291	1.2383	5.2637	0.0116	0.006234	1
	APACHE II	1.1361	1.0745	1.2078	0.0000171	0.00000000002145	1
	Craniotomy	0.3779	0.1495	0.8860	0.0307	0.002446	1
	ICH Score	2.8185	1.9444	4.2417	0.00000017	0.000000005002	1
2d exposure	Hyperoxaemia dose	1.0362	0.8639	1.2192	0.647657	0.493404	1	0.8057747	212.71	0.1473009	0.57%
	Age	0.9852	0.9581	1.0127	0.289709	0.861639	1
	Weight	1.0099	0.9892	1.0306	0.338763	0.725786	1
	Sex	0.5608	0.2489	1.2423	0.156267	0.254906	1
	Advanced Cardiac Support	2.7622	1.2960	6.0179	0.009145	0.005017	1
	APACHE II	1.1200	1.0502	1.2018	0.000916	0.0000004778	1
	Craniotomy	0.4210	0.1602	1.0262	0.065760	0.015243	1
	ICH Score	2.7163	1.8045	4.2749	0.00000497	0.0000003988	1
3d exposure	Hyperoxaemia dose	1.1159	0.8763	1.4537	0.344408	0.3838210	1	0.7636514	170.91	0.1435911	1.42%
	Age	0.9920	0.9593	1.0254	0.636034	0.7993441	1
	Weight	1.0036	0.9798	1.0261	0.754675	0.8233593	1
	Sex	0.6099	0.2431	1.5119	0.285755	0.4183934	1
	Advanced Cardiac Support	3.6483	1.5556	8.9127	0.003431	0.0041121	1
	APACHE II	1.1159	1.0297	1.2166	0.009398	0.0014126	1
	Craniotomy	0.4583	0.1579	1.2151	0.130439	0.0720259	1
	ICH Score	2.4766	1.5135	4.2919	0.000598	0.0001836	1
4d exposure	Hyperoxaemia dose	1.0167	0.6369	1.4106	0.92632	0.947170	1	0.7300725	147.73	0.1366861	0.17%
	Age	0.9988	0.9634	1.0352	0.94584	0.869110	1
	Weight	0.9865	0.9545	1.0146	0.37625	0.380116	1
	Sex	0.8649	0.3126	2.4091	0.77888	0.976341	1
	Advanced Cardiac Support	4.2682	1.6565	11.5665	0.00316	0.010345	1
	APACHE II	1.1097	1.0147	1.2218	0.02671	0.003565	1
	Craniotomy	0.4204	0.1263	1.2281	0.13050	0.077657	1
	ICH Score	1.7506	1.0196	3.1745	0.05151	0.042050	1
5d exposure	Hyperoxaemia dose	1.1316	0.7313	1.6830	0.5194	0.56113	1	0.6708464	139.84	0.1371839	1.21%
	Age	1.0009	0.9640	1.0392	0.9604	0.97060	1
	Weight	0.9938	0.9616	1.0218	0.6828	0.84500	1
	Sex	0.9198	0.3230	2.6655	0.8755	0.91825	1
	Advanced Cardiac Support	3.1935	1.1988	8.7118	0.0205	0.03642	1
	APACHE II	1.0711	0.9733	1.1824	0.1618	0.12623	1
	Craniotomy	0.6052	0.1900	1.7312	0.3673	0.34739	1
	ICH Score	1.4549	0.8112	2.7288	0.2218	0.21241	1

OR odds ratio, DoF degrees of freedom, AUROC area under the receiver operator curve, AIC akaike information criterion, APACHE acute physiology and chronic health evaluation, ICH Intracerebral haemorrhage

STROBEStatementChecklistofitemsthatshouldbeincludedinreportsofcohortstudies.docx

Association between early hyperoxaemia exposure and intensive care unit mortality in intracerebral haemorrhage - an observational cohort analysis

Status:

Version 1

Abstract

Figures

Introduction

Methods

Results

Discussion

Conclusion

Abbreviations

Declarations

References

Tables

Supplementary Files

Status:

Version 1