The association between peri-haematomal oedema and functional outcome after intracerebral haemorrhage - Individual participant data meta-analysis

Dr N Samarasekera
Dr G Mair
Professor C Weir
Dr X Wang
Dr T Moullaali
Dr A Parry-Jones
Mr J Drever
Mrs Sharon Tuck
Professor R Al-Shahi Salman

Intracerebral haemorrhage (ICH) accounts for approximately 10% of strokes in high income countries and 20% of strokes in low/middle income countries.1 Approximately 40% of patients die within the first month, and 86% are dead or dependent within one year.2 Haemostatic drugs, blood pressure lowering and haematoma evacuation for supratentorial ICH have not clearly improved outcome, so there is still no effective acute treatment for ICH.
Peri-haematomal oedema (PHO) is visible around ICH on computed tomography (CT) and magnetic resonance imaging (MRI) in most patients. On CT, PHO appears as a hypoattenuated area around the ICH and on T2-weighted MRI it appears as hyperintense area. PHO results from red blood cell breakdown, thrombin accumulation and neuro-inflammation. It contributes to mass effect and is a promising potential therapeutic target after ICH.3 ‘Treating the global and secondary effects of ICH including oedema formation’ is a clinical research priority set by The Second Haemorrhagic Stroke Academia Industry (HEADS-2) consortium in 2020,4 and by The Stroke Association’s Haemorrhagic Stroke Review.5 However, the HEADS-2 consortium also acknowledged ‘the debatable relationship between peri-haematomal oedema and functional outcome.’ Knowledge of factors which might modify this association would help to stratify patients according to their risk of PHO, to help select participants for trials of treatments targeted at PHO.
In a systematic review of 6681 articles on 4 April 2021 we identified 44 studies (using CT [n=35], MRI [n=7], both CT and MRI [n=2]) of 8,374 participants, which assessed both PHO and functional outcome in adults after spontaneous ICH. Of 37 studies which examined the association between PHO and functional outcome, 30 studies found modest associations between PHO and worse outcomes, three found that PHO was associated with better outcomes and four found no association. Only eight studies adjusted for other variables known to affect outcome after ICH,6-13 and a meta-analysis of five9-13 of these studies that assessed outcome using the modified Rankin scale at 90 days did not find an association between PHO and outcome (odds ratio [OR] 1.05, 0.99-1.11). However, the existence, strength, direction, and modifiers of an association remain uncertain, partly because of heterogeneity between these five studies, including the time of PHO assessment in relation to ICH onset (admission [n=1]10 over first 24 hours [n=2],9,11 over first 72 hours [n=1]13 and within first 12 days [n=1]12).
Although PHO within 24 hours of ICH onset may be associated with poorer outcome,14 the strength of the association and its existence when PHO is measured beyond 24 hours are unclear. Little is known about other variables which might modify the association between PHO and outcome. Sex,15 blood pressure, hyperglycaemia and statin use have all been suggested as potential effect modifiers,16 but variables have rarely been studied more than once, findings have been inconsistent, and small sample sizes and heterogeneity both in PHO measurement and timing of brain imaging following ICH have precluded firm conclusions.
Because the differences between studies preclude standard meta-analysis based on aggregated data we intend to use an individual participant data meta-analysis (IPDMA) to (1) explore the strength of the association between PHO and outcome after ICH and (2) to identify variables which may modify the strength of the association of PHO with outcome.
Hypotheses

Our hypothesis is that PHO is likely to affect functional outcome after ICH and that the time point of measurement and clinical variables such as age and blood pressure on admission may modify this association.

Research design
We will conduct an individual participant data meta-analysis.

Study identification
On 4 April 2021 we completed a comprehensive search of electronic databases (Embase (1980-) and Medline (1950-); appendix 1), hand searched bibliographies of relevant studies, and performed forward citation searching for each included article by using Google Scholar to identify relevant studies (registered on PROSPERO CRD42021253263). We excluded conference abstracts.

Inclusion and exclusion criteria
Study level inclusion criteria:
• Acute spontaneous (non-traumatic) supratentorial or infratentorial ICH confirmed by CT brain imaging.
• Repeat CT brain imaging performed at least once up to 14 days after the first brain imaging study.
• Observational cohort studies or control arms of randomised trials that measured PHO and functional outcome (using modified Rankin scale or similar measure) after ICH.
• Randomised trials which measured PHO and functional outcome (using modified Rankin scale or similar measure) after ICH, where the intervention used should not affect PHO; or the control arm of randomised trials only if the intervention may have affected PHO (for example, steroids, mannitol, hypertonic saline).

Exclusion criteria:
Study level:
• Studies reporting participants who are included in other publications that reported a larger cohort.
• Studies solely using MRI to assess PHO (since CT is the most frequently used imaging modality for participants with ICH and is reliable for the semiautomated assessment of PHO)17
• Studies involving <5 participants with spontaneous ICH

Participant level:
• Age less than 18 years.
• Secondary causes of ICH (including trauma, tumour, intracranial aneurysm, arteriovenous malformation, arteriovenous fistula, cavernous malformation, venous thrombosis, moyamoya syndrome, reversible cerebral vasoconstriction syndrome, or haemorrhagic transformation of ischaemic stroke).
• Exclusively intraventricular/subarachnoid/subdural/extradural haemorrhage.
• Participants treated with surgery, mannitol or steroids following their ICH (since these might also affect PHO)
• Participants where first (diagnostic) CT was performed more than 72 hours after ICH onset or where the time from ICH onset to first CT was unknown.
•Participants who might be eligible for inclusion in the IPDMA but whose data were not originally published in the specific study. Data collected

Study level variables
• First author of study and date
• Study period (years)
• Country
• Funding source
• Participants identified prospectively vs. retrospectively
• Inclusion and exclusion criteria (for example age range, admission blood pressure, GCS)
• Prespecified CT scanning protocol used
• Prospective outcome ascertainment used
• Time point of measuring functional outcome and method such as modified Rankin Scale
• Number of participants provided by each study to dataset and the reasons, if applicable for excluding any participant(s) from the dataset
• Method of PHO measurement including analysis software used
• Technique for measuring PHO – manual vs. semi-automated
o the PHO measuring technique used such as ABC/2, largest diameter (where a manual method has been used)
o Hounsfield unit threshold used for assessment of PHO (if applicable)

Participant level variables
Required
• Sex
• Age at presentation (years)
• Nature of symptom onset (awoke from sleep, last seen well, or awake at onset)
• Time from symptom onset to first scan (hours preferred or days, if hours not available)
• Glasgow Coma Scale (GCS) at presentation
• National Institutes of Health Stroke Scale (NIHSS) at presentation
• Single or multiple ICH (yes/no)
o If participants with multiple ICH were included, how PHO was assessed
• Location of ICH – as per CHARTS location where lobar = frontal, parietal, temporal, occipital, insular regions; deep = basal ganglia, thalamus, internal capsule, external capsule, corpus callosum, and deep and periventricular white matter defined as white matter adjacent to or within approximately 10 mm of the lateral ventricular margin); infratentorial = (brainstem or cerebellum), uncertain = (probably lobar, probably deep, holohemispheric)
• ICH volume (mm3 or ml) on first (diagnostic) scan
• PHO volume (mm3 or ml) on first (diagnostic) scan
• Intraventricular extension on first (diagnostic) scan (yes/no)
• Time from symptom onset to second scan (hours preferred or days if hours not available)
• ICH volume (mm3 or ml) on second scan
• PHO volume (mm3 or ml) on second scan
• Time from symptom onset to third scan (hours preferred or days if hours not available)
• ICH volume (mm3 or ml) on third scan
• PHO volume (mm3 or ml) on third scan
• Type of functional outcome (ideally modified Rankin Scale score)
• Date of assessment of functional outcome
• Score on functional outcome
• Occurrence and date of death

Desirable
• Ethnicity
• First-ever or recurrent ICH
• Premorbid level of function; for example as measured by modified Rankin scale
• History of hypertension before ICH which led to inclusion in the study (yes/no/unknown)
• History of diabetes mellitus (yes/no/unknown)
• History of ischaemic stroke (yes/no/unknown)
• Whether on any oral or parenteral anticoagulant at ICH symptom onset (yes/no/unknown)
• Whether on antiplatelet therapy at ICH symptom onset (yes/no/unknown)
• Whether taking beta-blockers at ICH symptom onset (yes/no/unknown)
• Whether taking a statin at ICH symptom onset (yes/no/unknown)
• Whether taking immunosuppressive agents at ICH symptom onset -systemically administered steroids, steroid sparing agents (including Azathioprine, methotrexate, cyclosporin, tacrolimus, mycofenolate mofetil), disease modifying therapies for coexisting neurological disorders such as multiple sclerosis (yes/no/unknown)
• Pyrexia (defined as temperature >37.5 degrees Celsius) at presentation (yes/no)
• Systolic and diastolic blood pressure on admission (mmHg) or mean arterial blood pressure (mmHg) if systolic and diastolic blood pressure are unavailable
• Blood glucose (mmol/l or mg/dL), fibrinogen, haematocrit, platelet count (platelet number per litre) and plasma sodium (mmol/l) at presentation
• Whether an acute blood pressure lowering intervention (including – beta blocker, ACE-inhibitor, angiotensin 2 receptor blocker, calcium channel blocker, nitric oxide donor, alpha blocker, diuretic, centrally acting agent, other) was used (yes/no) following ICH
o If an acute blood pressure lowering intervention was used, the timing of the intervention used in relation to symptom onset (48 hours after onset of ICH)
• Use of mannitol in first week after ICH onset (yes/no)
• Use of hypertonic saline in first week after ICH onset (yes/no)
• Presence of subarachnoid haemorrhage on first CT scan (yes/no)
• Any finger like projections of ICH on diagnostic scan
• Presence of CTA spot sign when CTA concurrently acquired with first CT scan (yes/no)
• Do-not-attempt resuscitation order instigated following ICH (yes/no)
o If ‘Yes’, interval (days or hours) between ICH onset and do-not-attempt resuscitation order

Data management
1. Where data from an identified study are available within the Virtual International Stroke Trials Archive-ICH (VISTA-ICH http://www.virtualtrialsarchives.org/vista-ich; n=725), we will apply to obtain these data. We will also invite corresponding authors of eligible studies identified by the systematic review and not already included in VISTA-ICH (including both observational studies and the control arms of randomised trials) to participate in the IPDMA and request anonymised datasets containing the study level variables and participant variables listed above.
2. We will use individual participant data from all collaborating studies, and consider, where possible using aggregate data from other eligible studies which do not contribute individual participant data.
3. We will require collaborating studies to supply a data dictionary with their data explaining the variables provided.
4. For each dataset, we will check the completeness, ranges, and values of the data provided. We will request the same dataset that has already been published. We will standardize the format and coding of the variables across the collaborating studies. Datasets obtained from collaborating studies will be combined to form a new master dataset, which will include a variable to indicate the original study. Data provided by VISTA will be identified as one dataset, because VISTA does not permit identification of the contributing studies.
5. We will apply a consistent measure of PHO on the first, and any subsequent scans. Since previous studies have used different measures of PHO, we will request ICH volume and PHO volume on each participant, which will then permit calculation of the different measurements that have been used in previous studies, including absolute PHO volume (total lesion volume-ICH volume), relative PHO volume ([total lesion volume-ICH volume]/ICH volume), PHO expansion rate and oedema extension distance (which is the difference between the radius of a sphere equal to the total lesion volume and the radius of a sphere equal to ICH volume alone).18 We will use oedema extension distance as our primary measure of PHO since it is less dependent on ICH volume and is thought to increase the power to detect effects of PHO on outcome.18 In a sensitivity analysis we will use absolute volume as a second measure of PHO, since it is likely to be closely related to mass effect generated by an ICH, which is a predictor of outcome.

Statistical analysis
1. In adults who have a spontaneous ICH:
a. Is the volume of PHO at a certain time point in the first two weeks after ICH associated with longer term functional outcome (for example, three months after ICH onset)?
b. Is the change in PHO volume between two selected time points in the first two weeks after ICH associated with longer term functional outcome (for example, three months after ICH onset)?
We will perform a single-stage individual participant data meta-analysis of the association between PHO and outcome using logistic regression modelling. A random intercept and random coefficient for the study will be included in each model. We will explore each potential effect modifier of the association between PHO and outcome in turn, by examining the effect of PHO (measured either at a certain time point (research question 1a) or the change over time (research question 1b); for example in the first 24 hours after ICH onset), on the likelihood of a poor outcome (e.g. modified Rankin scale 3-6) in a model. We will also consider examining the effect of PHO using an ordinal analysis of functional outcome through a proportional odds logistic regression. In all models polynomial terms will be considered to account for any associations which are non-linear on the logit scale. We will use these findings to inform development of a multivariable model to determine the association between PHO and outcome after accounting for clinical and imaging prognostic variables (such as GCS on admission, participant age, ICH volume, ICH location (supratentorial vs. infratentorial). Such a model will quantify the relationship between PHO and outcome, and will identify those with ICH most likely to be adversely affected by PHO.

2. How does the association between PHO and functional outcome vary
a. according to the time at which PHO volume is measured in the first two weeks after ICH onset?
b. according to the time points between which change in PHO volume is measured in the first two weeks after ICH onset?
We will assess whether the association between PHO and functional outcome varies according to PHO growth since ICH onset, and determine the time window of growth from ICH onset which is most closely associated with outcome. This will be established through graphical representation of the log-odds ratio for the association and its 95% confidence interval. We will look at the development of PHO over time by comparing PHO measurements at certain time points in relation to ICH onset – such as at 24 hours and 72 hours after ICH; but the specific time points explored will depend on data availability at each time point.
3. Is the association between PHO volume and functional outcome, as outlined in 1a affected by:
c. clinical variables such as participant age, blood pressure on admission?
d. radiological variables such as ICH volume or ICH location?
4. Is the association between change in PHO volume and functional outcome, as outlined in question 1b affected by:
e. clinical variables such as participant age, blood pressure on admission?
f. radiological variables such as ICH volume or ICH location?
Where adequate data are available, heterogeneity in the association between PHO and functional outcome will be assessed in the subgroups specified above to determine whether these factors modify any association between PHO and outcome. Significant effect modifiers identified will be used to inform potential extensions to the multivariable models developed for research questions 1a and 1b.
We will handle potential effect modifiers either singly or in groups based upon our knowledge of potential interactions between them; for example prior anticoagulant use increasing ICH volume size and PHO volume likely to be in turn affected by ICH volume. These analyses may lack sufficient statistical power so may be hypothesis generating only.

Risk of bias
We will assess bias risk (using established tools such as the Newcastle Ottawa scale for observational studies19 and the Cochrane collaboration tool for trials http://methods.cochrane.org/bias/assessing-risk-bias- included-studies), for all included studies. We will compare studies which contribute data to this IPDMA with studies which do not and assess publication bias by visual inspection for funnel plot asymmetry (with and without studies where IPD is obtained). The Grades of Recommendation, Assessment, Development and Evaluation20 will be used to evaluate the quality of the synthesised evidence.

Transfer of data and confidentiality
The preferred formats for data transfer are the following: Excel spreadsheet or a delimited text file, but we will accept other formats (Access database, SAS transport file, SPSS portable file, fixed-format text file). We will not collect participant-identifiable data. Authors are asked to de-identify data before transfer by removing all participant-identifiable data, and replacing them with a unique study ID number for each participant. All data sent will be held securely in the strictest confidence on password protected servers, and the final dataset will be held in a secure repository (DataShare https://datashare.is.ed.ac.uk/) which has achieved the ‘Data Seal of Approval,’ a peer-reviewed status as a digital trusted repository.

Publication policy
We intend to include and order the authors according to (a) their contribution to the design and execution of this study and (b) the number of eligible participants they contribute with a complete dataset. We will apply the ICMJE criteria for authorship. The VISTA-ICH collaboration will be listed as a group author name. We will share the results of our analyses with participating groups. All authors will have the opportunity to review the manuscript and approve the final version before submission to a journal.

1. Feigin VL et al. Worldwide stroke incidence and early case fatality reported in 56 population-based studies: a systematic review. Lancet Neurol 2009;8:355–69
2. Samarasekera N et al. Influence of intracerebral haemorrhage location on incidence, characteristics and outcome: population-based study. Stroke 2015;46:361-8
3. Ironside N et al. Perihematomal edema after spontaneous intracerebral haemorrhage. Stroke 2019;50:1626-1633. DOI: 10.1161/STROKEAHA.119.024965
4. The Hemorrhagic Stroke Academia Industry (HEADS) Roundtable participants Recommendations for Clinical Trials in ICH. Stroke 2020;51:1333-1338. DOI: 10.1161/STROKEAHA.119.027882
5. The Stroke Association State of play review – Haemorrhagic Stroke Research Priority Document 2014 https://www.stroke.org.uk/news/haemorrhagic-stroke-workshop-priority-setting
6. Levine JM et al. Early edema in warfarin-related intracerebral hemorrhage. Neurocrit Care 2007;7:58–63
7. Appelboom G et al. Volume-dependent effect of perihaematomal oedema on outcome for spontaneous intracerebral haemorrhages. J Neurol Neurosurg Psychiatry 2013;84:488–93
8. Ozdinc S et al. Prognostic value of perihematomal edema area at the initial ED presentation in patients with intracranial hematoma. Am J Emerg Med 2016;34:1241–6
9. Yang J et al. Prognostic Significance of Perihematomal Edema in Acute Intracerebral Hemorrhage. Stroke 2015;46:1009-13
10. Murthy SB et al. Rate of perihaematomal oedema expansion is associated with poor clinical outcomes in intracerebral haemorrhage. J Neurol Neurosurg Psychiatry 2016;87:1169–73.
11. Urday S et al. Rate of Perihematomal Edema Expansion Predicts Outcome After Intracerebral Hemorrhage. Crit Care Med 2016;44:790-797
12. Volbers B et al. Peak perihemorrhagic edema correlates with functional outcome in intracerebral hemorrhage. Neurology 2018;90:e1005–12
13. Grunwald Z et al. Perihematomal Edema Expansion Rates and Patient Outcomes in Deep and Lobar Intracerebral Hemorrhage. Neurocrit Care 2017;26:205–12
14. Selim M et al. Perihematoma edema: a potential translational target in intracerebral hemorrhage? Transl Stroke Res 2015;6:104-6
15. Wagner M et al. Sex differences in perihemorrhagic edema evolution after spontaneous intracerebral haemorrhage. Eur J Neurol 2012;19:1477-1481
16. Urday S et al. Targeting secondary injury in intracerebral haemorrhage – perihaematomal oedema. Nat Rev Neurol 2015;11:111-22
17. Volbers B et al. Semi-automatic volumetric assessment of perihemorrhagic edema with computed tomography. European J Neurol 2011;18:1323–8
18. Parry-Jones AR et al. Edema extension distance: outcome measure for phase II clinical trials targeting edema after intracerebral haemorrhage. Stroke 2015;46:e137-40
19. Wells GA, Shea B, O’Connell D, et al. The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomised studies in meta-analyses. The Ottawa Hospital Research Institute. http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp
20. Iorio A, Spencer FA, Falavigna M, et al. Use of GRADE for assessment of evidence about prognosis: rating confidence in estimates of event rates in broad categories of patients. BMJ 2015;350:h870.

TBC