Baseline factors associated with early and late death in intracerebral haemorrhage survivors

Background and purpose The aim of this study was to determine whether early and late death are associated with different baseline factors in intracerebral haemorrhage (ICH) survivors. Methods This was a secondary analysis of the multicentre prospective observational CROMIS‐2 ICH study. Death was defined as ‘early’ if occurring within 6 months of study entry and ‘late’ if occurring after this time point. Results In our cohort (n = 1094), there were 306 deaths (per 100 patient‐years: absolute event rate, 11.7; 95% confidence intervals, 10.5–13.1); 156 were ‘early’ and 150 ‘late’. In multivariable analyses, early death was independently associated with age [per year increase; hazard ratio (HR), 1.05, P = 0.003], history of hypertension (HR, 1.89, P = 0.038), pre‐event modified Rankin scale score (per point increase; HR, 1.41, P < 0.0001), admission National Institutes of Health Stroke Scale score (per point increase; HR, 1.11, P < 0.0001) and haemorrhage volume >60 mL (HR, 4.08, P < 0.0001). Late death showed independent associations with age (per year increase; HR, 1.04, P = 0.003), pre‐event modified Rankin scale score (per point increase; HR, 1.42, P = 0.001), prior anticoagulant use (HR, 2.13, P = 0.028) and the presence of intraventricular extension (HR, 1.73, P = 0.033) in multivariable analyses. In further analyses where time was treated as continuous (rather than dichotomized), the HR of previous cerebral ischaemic events increased with time, whereas HRs for Glasgow Coma Scale score, National Institutes of Health Stroke Scale score and ICH volume decreased over time. Conclusions We provide new evidence that not all baseline factors associated with early mortality after ICH are associated with mortality after 6 months and that the effects of baseline variables change over time. Our findings could help design better prognostic scores for later death after ICH.


Introduction
Most research on outcomes following intracerebral haemorrhage (ICH) has focussed on short-term Correspondence: G. Banerjee, Stroke Research Centre, Department of Brain Repair and Rehabilitation, UCL Queen Square Institute of Neurology, Russell Square House, 10-12 Russell Square, London WC1B 5EH, UK (tel.: + 44 (0)20 3448 4072; fax: +44 (0)20 3108 7537; e-mail: g.banerjee@ucl.ac.uk). *CROMIS-2 collaborators are given in Appendix 1. mortality (within 6 months), reflecting the high rates of early death associated with this stroke subtype [1,2]. Many factors associated with early mortality relate to ICH severity and this is reflected in prognostic scores that aim to predict outcome in the short term [3][4][5][6][7][8][9][10][11]. Policies of active acute management, including blood pressure lowering, prompt reversal of anticoagulation and neurosurgical referral, aim to improve prognosis in patients with ICH [12]. A better understanding of the factors that influence 'late' death following ICH might identify potentially modifiable risk factors that could improve long-term outcomes [1].
Our aim was to evaluate whether early and late death are associated with different baseline factors in ICH survivors using data from the prospective multicentre CROMIS-2 ICH study. We hypothesized that factors relating to the severity of the acute ICH would not be associated with death at later (beyond 6 months) time points.

Data availability statement
Analyses for the CROMIS-2 study are ongoing; once all of these analyses are completed, the CROMIS-2 Steering Committee will consider applications from other researchers for access to anonymized source data.

Participants
We included adults from the CROMIS-2 (Clinical Relevance of Microbleeds in Stroke) ICH study; full details of the study protocol have been published previously [13]. Further details are provided in the Supporting Information. The study was approved by the National Research Ethics Service (IRAS reference 10/ H0716/61). Informed written consent was obtained for all participants.

Outcomes
The outcome of interest for this project was time to death. Mortality notifications were received from NHS Digital (previously the Health and Social Care Information Centre) as detailed in the previously published study protocol [13]. NHS Digital is a national centralized body that collects data on health and social care in the UK; mortality data are derived from 'hospital episode statistics' (records of all NHS patient admissions) and information on registered deaths from the Office of National Statistics (death registration is a legal requirement in the UK).
Patients were censored at either 3 years following the ICH that resulted in study entry or at last available follow-up for vital status (the time of the study's last notification of deaths from NHS Digital, i.e. 31 October 2017), depending on which was earlier.

Imaging
Brain computed tomography imaging was acquired acutely at the time of the index event as part of the patient's routine clinical care. Further details are provided in the Supporting Information.

Statistics
Statistical analysis was performed using Stata (version 15.1, StataCorp LLC, College Station, TX, USA). We dichotomized time following ICH into 'early' (before 6 months) and 'late' (after 6 months) periods. We used univariable Cox regression to calculate hazard ratios (HRs) for all baseline variables collected to review for associations during these two time periods. Variables where the 95% confidence intervals (CI) did not cross 1 were considered as statistically significant. To explore this further, the effect of each baseline variable was then allowed to vary linearly with time; further details are provided in the Supporting Information.

Associations of 'early' versus 'late' death
Of the 306 deaths, 156 occurred within 6 months of the index haemorrhage event ('early') and 150 deaths occurred after 6 months and within 3 years of the index ICH ('late'). Baseline characteristics for both groups are shown in Table 1.
We then investigated which baseline characteristics showed a significant change in HR between the early and late periods ( Table 2). We found that HRs for the presence of APOE e2, GCS score, NIHSS score and ICH volume >60 mL showed evidence of significant change between the early and late periods ( Table 2). Percentage values were calculated using the total number of patients for whom data were available as the denominator. Data are given as mean AE SD, n (%) and median (interquartile range). GCS, Glasgow Coma Scale; ICH, intracerebral haemorrhage; IQR, interquartile range; mRS, modified Rankin scale; NIHSS, National Institutes of Health Stroke Scale; WMC, white matter changes.

Further exploratory analysis of time-varying effects
We then performed exploratory analyses where time was considered as a continuous measure (Table S1). In these analyses, variables that showed significant time-varying effects were history of a previous cerebral ischaemic event (P = 0.0214), admission GCS score (P = 0.0108), NIHSS score (P < 0.00001) and ICH volume (P = 0.0439). The HRs of previous cerebral ischaemic events increased with time, whereas those for NIHSS score and ICH volume decreased with time. The protective (negative association) of GCS score also decreased with time.

Discussion
We provide new evidence that not all baseline factors associated with early mortality after ICH are associated with mortality after 6 months. In analyses where the time-varying effects of baseline variables were allowed to vary continuously with time, we found that the influence of measures of acute ICH severity decreased over time, whereas those associated with established cerebrovascular disease (previous cerebral ischaemic events) increased over time. These results support the argument that definitions of 'early' or 'late' death are necessarily arbitrary, as the impact of some characteristics present at study entry vary continuously with time. In our study, the factors that we found to be independently associated with early death are in keeping with other studies and reflected in pre-existing prognostic scores that include these and other variables [4][5][6][7][8][9][10][11]. Differences between our results for associations with late death and those previously reported [14,15] are likely to reflect our method of considering early and late death independently; when considering all death events together, we observed effects that were similar to those previously reported. Additionally, we observed that four variables (APOE e2, GCS score, NIHSS score and ICH volume >60 mL) showed significant differences in the magnitude of their effect before and after 6 months (although the HRs for APOE e2 were not statistically significant in themselves). This result confirms that, whereas GCS score, NIHSS score and ICH volume are important predictors of early mortality, their effect changes significantly between the early and late periods, and thus they are less useful for predicting mortality in the longer term, as we hypothesized. Our analyses of linear time-varying effects on long-term mortality following ICH are novel and demonstrate the potentially complex interactions that can occur over time. These analyses highlight the difficulties in defining what is 'early' death or a 'short-term' outcome; further work that considers time-varying effects on mortality across longer time scales is needed to guide this.
Our finding of an association between intraventricular extension and late death seems counterintuitive, but illustrates the importance of our work and the complicated manner in which baseline variables might interact over time. Given that our cohort included patients with milder strokes, we hypothesize that the effect of intraventricular extension on early death was lost in the adjusted analyses because of larger magnitude effects associated with other factors associated with ICH severity (NIHSS score and ICH volume). We speculate that, when considering late death, the effects of acute factors such as NIHSS score and ICH volume were of smaller magnitude, and thus the impact of intraventricular extension as a measure of stroke severity became more apparent. We did observe, in unadjusted analyses, that intraventricular extension was associated with both early and late death, but the magnitude of the association was smaller for late death ( Table 2).
Our study has a number of strengths, including the number of patients, its multicentre design (which increases generalizability), robust ascertainment of follow-up events and the detailed clinical and radiological data available for each participant. However, there are limitations of our work. Our cohort is comprised of survivors with mild strokes, as reflected by the median NIHSS and GCS scores, low ICH volumes and low early death rates. This cohort is therefore unlikely to be representative of all patients with ICH, particularly those with more severe haemorrhages. We were unable to adjust for acute complications of ICH or details relating to immediate care, either active or care-limiting (i.e. do not resuscitate orders or palliative pathways), all of which would impact mortality. Additionally, we were unable to comment on cause of death in our patients. Finally, although we considered the time-varying effects of variables recorded at study entry, the status of these may have changed after this time point (e.g. antiplatelet or anticoagulant use) and this could have influenced our results. Univariable hazard ratios (HRs) for each characteristic obtained by fitting Cox regression models with time-varying effects (before/after 6 months). Data are given as HR (95% confidence intervals). The time-varying coefficient P-value compares the difference between the early and late HRs. GCS, Glasgow Coma Scale; mRS, modified Rankin scale; NIHSS, National Institutes of Health Stroke Scale.
We provide new evidence that not all baseline factors associated with early mortality after ICH are associated with mortality after 6 months. Our findings could help design better prognostic scores for later death after ICH.