Article Text


Research paper
Associations with health-related quality of life after intracerebral haemorrhage: pooled analysis of INTERACT studies
  1. Candice Delcourt1,2,
  2. Danni Zheng1,
  3. Xiaoying Chen1,
  4. Maree Hackett1,3,
  5. Hisatomi Arima1,4,
  6. Jun Hata5,
  7. Emma Heeley6,
  8. Rustam Al-Shahi Salman7,
  9. Mark Woodward1,8,
  10. Yining Huang9,
  11. Thompson Robinson10,
  12. Pablo M Lavados11,
  13. Richard I Lindley1,12,
  14. Christian Stapf13,
  15. Leo Davies1,2,
  16. John Chalmers1,2,
  17. Craig S Anderson1,2,14,
  18. Shoichiro Sato1,15
  19. for the INTERACT Investigators
  1. 1The George Institute for Global Health, The University of Sydney, Sydney, New South Wales, Australia
  2. 2Neurology Department, Royal Prince Alfred Hospital, Sydney, New South Wales, Australia
  3. 3The University of Central Lancashire, Lancashire, UK
  4. 4Department of Preventive Medicine and Public Health, Faculty of Medicine, Fukuoka University, Fukuoka, Japan
  5. 5Center for Cohort Studies, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan
  6. 6Centre for Health Record Linkage, NSW Ministry of Health, Sydney, New South Wales, Australia
  7. 7Centre for Clinical Brain Sciences, University of Edinburgh, Edinburgh, UK
  8. 8Nuffield Department of Population Health, The George Institute for Global Health, Oxford University, Oxford, UK
  9. 9Department of Neurology, Peking University First Hospital, Beijing, China
  10. 10Department of Cardiovascular Sciences and NIHR Biomedical Research Unit for Cardiovascular Diseases, University of Leicester, Leicester, UK
  11. 11Clínica Alemana de Santiago, Facultad de Medicina Clínica Alemana Universidad del Desarrollo, Universidad de Chile, Santiago, Chile
  12. 12Westmead Hospital Clinical School, Westmead, New South Wales, Australia
  13. 13Department of Neuroscience, CRCHUM, University of Montreal, Montreal, Quebec, Canada
  14. 14The George Institute for Global Health China at Peking University Health Sciences Center, Beijing, China
  15. 15Department of Cerebrovascular Medicine, National Cerebral and Cardiovascular Center, Osaka, Japan
  1. Correspondence to Professor Craig S Anderson The George Institute for Global Health, PO Box M201, Missenden Road, Camperdown, NSW 2050, Australia; canderson{at}


Background and purpose Limited data exist on health-related quality of life (HRQoL) after intracerebral haemorrhage (ICH). We aimed to determine baseline factors associated with HRQoL among participants of the pilot and main phases of the Intensive Blood Pressure Reduction in Acute Cerebral Haemorrhage Trials (INTERACT 1 and 2).

Methods The INTERACT studies were randomised controlled trials of early intensive blood pressure (BP) lowering in patients with ICH (<6 hours) and elevated systolic BP (150–220 mm Hg). HRQoL was determined using the European Quality of Life Scale (EQ-5D) at 90 days, completed by patients or proxy responders. Binary logistic regression analyses were performed to identify factors associated with poor overall HRQoL.

Results 2756 patients were included. Demographic, clinical and radiological factors associated with lower EQ-5D utility score were age, randomisation outside of China, antithrombotic use, high baseline National Institutes of Health Stroke Scale (NIHSS) score, larger ICH, presence of intraventricular extension and use of proxy responders. High (≥14) NIHSS score, larger ICH and proxy responders were associated with low scores in all five dimensions of the EQ-5D. The NIHSS score had a strong association with poor HRQoL (p<0.001). Female gender and antithrombotic use were associated with decreased scores in dimensions of pain/discomfort and usual activity, respectively.

Conclusions Poor HRQoL was associated with age, comorbidities, proxy source of assessment, clinical severity and ICH characteristics. The strongest association was with initial clinical severity defined by high NIHSS score.

Trial registration numbers NCT00226096 and NCT00716079; Post-results.

View Full Text

Statistics from


As most stroke survivors have some degree of disability, assessment of their quality of life is important for healthcare providers, policymakers and researchers. Health-related quality of life (HRQoL) measures key physical and emotional aspects of disease on a person's quality of life. Although several studies of mixed ischaemic stroke and intracerebral haemorrhage (ICH) patient groups are consistent in reporting age and physical disability as predictors of poor HRQoL,1 ,2 few have focused specifically on ICH, the least treatable and the most disabling type of stroke. Analysis from the Factor Seven for Acute Haemorrhagic Stroke Treatment (FAST) trial showed that low overall HRQoL, as measured by the European Quality of Life Scale (EQ-5D)3 utility score, was associated with age, clinical factors (stroke severity, systolic blood pressure (BP) and neurological deterioration) and imaging features (larger and deep ICH).4 Yet, the number of participants was still relatively small (n=621) and individual HRQoL dimensions within EQ-5D were not reported. We aimed to determine the clinical and imaging factors associated with overall HRQoL and its constituents in the pooled data set of participants of the pilot and main phases of the Intensive Blood Pressure Reduction in Acute Cerebral Haemorrhage Trial (INTERACT 1 and 2).5 ,6



The INTERACT studies5 ,6 were multicentre, randomised, controlled clinical trials that included 3233 participants from 22 countries during 2005–2013. We pooled the pilot and main phase INTERACT studies to increase the sample size and the precision of the estimates. Patients with imaging-confirmed ICH were randomly assigned to receive either early intensive BP lowering treatment (<140 mm Hg systolic goal) or contemporaneous guideline-recommended standard BP lowering (<180 mm Hg systolic goal) within 6 hours of onset. The study protocol was approved by appropriate ethics committees at each site and registered with (NCT00226096 and NCT00716079). Written informed consent was obtained from patients or their legal surrogates when they were unable to do so.


Demographics and clinical characteristics were recorded at patient enrolment, with stroke severity measured on the Glasgow Coma Scale (GCS) and the National Institutes of Health Stroke Scale (NIHSS). CT scans were performed according to standardised techniques at baseline and centrally analysed for the volume and location of ICH and the presence of intraventricular extension of ICH (IVH).


The outcome for this analysis was HRQoL, as assessed directly by a patient or by a proxy responder, using the EQ-5D3 questionnaire at 90 days. This descriptive system defines the state of general health across five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) with three levels (no problems, some/moderate problems and severe problems). The EQ-5D utility score integrates the ratings of the five dimensions into a single score, calculated using population-based preference weights for each subscale. In the present analysis, we used the weights obtained from the UK population. Utility scores express HRQoL quantitatively as a fraction of perfect health, with a score of 1 representing perfect health, a score of 0 representing death and negative scores (minimum score −0.109) representing health states considered worse than death. The average utility score in disease-free populations range between 0.8 and 0.9.7–9 When patients were not able to answer the questionnaire themselves, proxy responders, such as their caregiver or doctor, were asked to rate the patient's HRQoL. The EQ-5D has previously been validated for use in proxy responders.10 ,11 The protocol did not stipulate the process of proxy responder selection; the decision was opportunistic and arose during a telephone or face-to-face interview between the responsible neurologically competent person (blinded to treatment arm) and the patient or caregiver at the scheduled time of follow-up.

Statistical analysis

We evaluated baseline characteristics of the patients by utility score groups (≤0.7 (median utility score) and >0.7) summarised by means and SDs, median and IQR for continuous variables and numbers (%) for categorical variables. The baseline differences in patient characteristics between utility score groups were also assessed by the χ2 test and the Wilcoxon test. Similarly, patient characteristics between the included (alive at 90 days and with complete EQ-5D data) and excluded (dead at 90 days or incomplete EQ-5D data) study participants were compared.

The associations of baseline characteristics and EQ-5D utility score at 90 days were evaluated in a binary regression model that included clinically important variables, as described elsewhere12 ,13 and/or variables found significant in the crude analysis (age, sex, country of enrolment, medical history, antithrombotic use, onset to randomisation time, systolic BP, NIHSS score (≥14 vs<14), CT findings, intensive BP lowering treatment and proxy responders). All the analysis included ‘study’ as a covariate to account for the differential effect between the pilot and main phases of the INTERACT study. Each dimension of EQ-5D was treated as a categorical variable with two levels (answer of 2 or 3, meaning some/moderate or severe problems in the corresponding dimension, vs answer 1, meaning no problem in the corresponding dimension). The associations of low HRQoL in each dimension of EQ-5D were also assessed in binary logistic regression models (answer of 2 or 3 vs 1), including the same covariates as the overall utility score model.

To test the robustness of results, we first conducted stratified analyses for associations of utility score as a binary outcome variable in patients who completed the EQ-5D questionnaire themselves and proxy responders. Second, we applied Chinese norms to the Chinese study population14 ,15 and UK preference weights to the other participants and repeated the logistic regression analysis. Third, we forced all baseline variables into our analysis model. Finally, we conducted linear regression analysis with overall utility score as a continuous outcome to estimate changes in the utility score associated with variations in the independent covariates.

A standard level of significance (p<0.05) was used and the data are reported with ORs and 95% CIs. All analyses were performed using the SAS software (V.9.3, SAS Institute, Cary, North Carolina, USA).


Of 3233 patients in the INTERACT pooled cohort, 2756 with information on HRQoL at 90 days were included in the present analyses (see online supplementary material figure S1). The distribution of EQ-5D utility scores was left skewed (see online supplementary material figure S2) with 1251 patients having a utility score equal or lower than the median (≤0.7), and 1505 patients with a utility score higher than the median (>0.7). The variables associated with lower utility scores were age, female gender, randomisation outside of China, previous acute coronary syndrome, being on an antithrombotic, having a high baseline systolic BP, receiving non-intensive BP lowering (as per INTERACT protocol), clinical severity (lower GCS and higher NIHSS), larger and deep ICH, intraventricular extension of the ICH, completion of the questionnaire by a proxy responder and being an INTERACT2 patient (table 1).

Table 1

Patient characteristics

supplementary data

After excluding 477 patients either due to death within 90 days (n=382) or because of missing HRQoL data (n=95), there remained 2756 patients who were included in these analyses (supplementary material figure S1). Online supplementary table S1 outlines the clinical characteristics of the included and excluded patients.

The EQ-5D questionnaire was completed either by face-to-face or telephone interview directly with patients (n=1710, 1294 randomised in China and 416 outside of China) or by proxy responders (n=1046, 749 randomised in China and 297 outside of China).

Table 2 shows the multivariable binary logistic regression analysis for associations between decreased utility score (≤0.7) at 90 days. Variables significantly associated with decreased EQ-5D utility score were older age, randomisation outside of China, antithrombotic use, shorter onset to randomisation time, higher baseline NIHSS score (≥14), larger and deep ICH, presence of IVH and proxy responders.

Table 2

Multivariable binary logistic regression analysis for associations with decreased EQ-5D utility score* at 90 days

The binary logistic regression model explained a very large proportion of the variation in the utility score (C statistics=0.77). Supplementary analysis of direct responses from patients-only (without proxy responders) (n=1710) showed similar results. In the subgroup of proxy-responders-only (n=1046), age, antithrombotics, high NIHSS, large baseline ICH volume and IVH remained significant but country of randomisation, onset to treatment time and deep location became non-significant (see online supplementary material tables S2 and S3). Supplementary analysis with Chinese utility weights for the Chinese participants and UK weights for others showed similar results with the exception of ‘randomisation in China’, which was no longer associated with increased quality of life (see online supplementary material table S4). Chinese weights applied to the Chinese population produced less variation in the utility score with higher minimum scores than when UK norms were used (−0.149 to 1 vs −0.594 to 1, respectively). Supplementary analysis forcing all baseline variables into the model with linear regression showed broadly similar results (see online supplementary material tables S5 and S6).

Table 3 shows the multivariable binary logistic regression analysis of HRQoL by EQ-5D dimensions. The associations with mobility problems (1762 scoring 2 or 3 [some/moderate or severe problems] vs 994 scoring 1 [no problem]) were age, past history of an acute coronary syndrome and hypertension, onset to randomisation time, NIHSS score, volume and deep location of ICH, IVH and proxy responders. Factors associated with problems in self-care (1320 scoring 2 or 3 vs 1436 scoring 1) were age, country of randomisation, diabetes mellitus, onset to randomisation time, NIHSS score, volume and deep location of ICH, IVH and proxy responders. Usual activity (1694 scoring 2 or 3 vs 1062 scoring 1) was independently associated with age, country of randomisation, antithrombotic use, onset to randomisation time, NIHSS score, volume and deep location of ICH, IVH and proxy responders. Pain/discomfort (1154 scoring 2 or 3 vs 1602 scoring 1) were associated with gender, history of an acute coronary syndrome, NIHSS score, ICH volume and proxy responders. Finally, anxiety/depression (982 scoring 2 or 3 vs 1774 scoring 1) was associated with age, NIHSS score, volume of ICH and proxy responders.

Table 3

Multivariable binary logistic regression analyses for associations with poor HRQoL by dimension (some/moderate or severe problems vs no problem)


In this study involving a large multiethnic group of ICH patients, poor HRQoL at 90 days was associated with older age, randomisation outside of China, antithrombotic use, greater severity (higher baseline NIHSS score, larger ICH, and presence of IVH) and proxy assessment. In particular, a highly significant association was found between stroke severity and poor HRQoL, across all five dimensions of the EQ-5D. The relationship of other variables across specific dimensions varied, with ICH volume and having a proxy responder being associated with at least some problems across all dimensions, while being female but not age was associated with poor scores for pain/discomfort. Antithrombotic use, diabetes mellitus and coronary disease were associated with low scores in usual activity, self-care and mobility, and pain/discomfort, respectively; hypertension was associated with reduced mobility; and Chinese patients reported better scores in usual activity and self-care.

As in the FAST trial,4 our study showed that ICH imaging features of large ICH with associated IVH was related to poor HRQoL. In studies of HRQoL after aneurysmal subarachnoid haemorrhage, IVH predicted worse HRQoL on the Aachen Life Quality Inventory (ALQI), and particularly for the psychosocial dimension which is a summary score of free-time activities, family relations, social contact, communication and cognition.16 Since the EQ-5D rating cover social interactions within the usual activities dimension (family and leisure activities), our results show a similar relationship in which IVH was found to affect mobility, self-care and daily activities dimensions. This is probably due to the direct influence of IVH, producing a high risk of reduced consciousness, disability and death17–19 in relation to a given volume of IVH.

In the present analysis, prior antithrombotic use (which may be regarded as a marker of cardiovascular disease risk) was associated with poor overall utility score and usual activity. Comorbidities of diabetes mellitus, hypertension and coronary artery disease were each associated with the poor utility score within these domains, which is consistent with prior studies. Comorbidities may negatively affect physical function, memory and thinking, but also social participation, as measured by the Stroke Impact Scale (SIS).12 ,20 ,21 Our study results corroborate prior findings and suggest that the added burden of pre-existing comorbidities affects not only functional outcome but also HRQoL among ICH patients.22

In our analysis, female gender was not associated with poor overall HRQoL, except with respect to pain/discomfort. Conversely, several prior studies have indicated that this variable is a common predictor of poor HQoL after stroke.1 ,2 In these studies, though, low education and different work and household living among females compared to males were postulated as possible reasons. Intriguingly, a recent American multicentre, longitudinal registry study has shown a similar tendency for a higher proportion of women than men to report problems in the EQ-5D dimension of pain/discomfort after ischaemic stroke or transient ischaemic attack.23 Although we speculated that poststroke headache, spasticity or joint contracture could be possible underlying reasons, data pertaining to these variables were not available in our study.

Proxy-responder assessment was associated with decreased EQ-5D utility score and poor HRQoL across all five individual dimensions. The involvement of proxy responders may be a reflection of poststroke cognitive and communication disabilities preventing patients from answering the questionnaire themselves. Proxy responders provide their assessment of patients' HRQoL, which is in itself inherently subjective and can be related to various factors such as patient and caregiver education levels, family relationships and caregiver psychological distress and burden.24 It is also recognised that proxy responders have a tendency to report more problems and have a more pessimistic appreciation of patients' HRQoL than patients themselves, which introduces bias related to stroke severity.1 ,11 ,25 ,26

Interestingly, in China, although participants self-reported a better quality of life than patients randomised outside of China, proxy responders responded in the same way as non-Chinese proxy responders. Patients randomised in China reported better scores in the self-care and usual activity dimensions and this was not be explained by younger age, milder clinical severity or smaller ICH in the models. Chinese patients may have been more likely to receive informal caregiving and have higher perceived levels of family support, which may have led to a more positive life outlook as compared to participants in others countries,27 where the elderly and infirm are often cared for by health agency services or institutions. In addition, Chinese patients may be more engaged spiritually, which may also affect psychosocial aspects of the patient's life.28 Finally, the association between randomisation in China and increased total utility score that are present when adjusted UK norms are used but which disappear when Chinese norms are applied to patients randomised in China suggests that Chinese patients have a more optimistic view of their quality of life than residents of the UK. How much ethnic differences in HRQoL outcomes can be attributed to psychosocial characteristics29 as well as stroke management and rehabilitation practices in China requires further study.

Strengths of our analyses include the access to a large data set derived from an international multicentre study with a rigorous protocol and collection of outcome data by assessors blinded to the study groups. The robustness of our findings was validated through a range of sensitivity analyses involving ordinal logistic regression model, binary logistic regression model forcing all the baseline characteristics, stratification by patients versus proxy responders and by applying Chinese norms to patients randomised in China. However, we recognise that there are limitations, in particular, of selection bias as the findings were based on a clinical trial population where cases of large volume ICH were under-represented, and patients with a very high likelihood of early death and planned surgical evacuation of ICH were excluded. This and the proportion of surviving patients without complete assessment of quality of life at 90 days probably lead as to underestimating the total reduction in quality of life associated with ICH. The influence of socioeconomic factors could not be assessed in the INTERACT studies and about 40% of responses were given by proxies, who are known to give a more pessimistic view of HRQoL. Finally, the use of the UK weighted HRQoL equation for all patients in this study may not accurately reflect the true utility score in certain non-UK participants. Nevertheless, this reference has been used in prior investigations and we have confirmed our findings through sensitivity analyses using Chinese HRQoL norms for patients enrolled in China.14 ,15 ,30 ,31

In summary, significant associations were shown between clinical severity and poor HRQoL at 90 days in a large population of patients with acute ICH. Poor HRQoL was also associated with older age, comorbidities, ICH imaging characteristics (large and deep ICH and the presence of IVH) and proxy assessment.


View Abstract


  • Twitter Follow Rustam Al-Shahi Salman at @BleedingStroke

  • Collaborators INTERACT Investigators (see online only reference ‘Supplemental data 2_INTERACT Investigators.pdf’).

  • Contributors CD, DZ, CSA and SS contributed to the concept and rationale for the study. DZ and SS contributed to statistical analyses. CD, DZ, and Sato were responsible for the first draft of the manuscript. CD, DZ, XC, CSA and SS contributed to the interpretation of the results. All authors participated in the drafting and took responsibility for the content and integrity of this article.

  • Funding The National Health and Medical Research Council of Australia provided funding for this research.

  • Competing interests CSA received speaker fees from Takeda China and Advisory Committee reimbursement from Medtronic and Astra Zeneca. The other authors report no conflicts of interest.

  • Ethics approval Relevant ethics committee in countries involved.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.