Stroke patients have increased frequency and extent of cerebral white matter lesions (WMLs), as seen on MRI.1 WMLs also accumulate during aging and so are also labelled as age related white matter changes (ARWMCs). ARWMCs are regarded as a marker of cerebral small vessel disease, which is associated with arterial hypertension and other vascular risk factors.2

In the general population in persons aged 65–84 years, moderate to severe ARWMCs occur in up to one-third of cases.3 ARWMCs associate with functional decline leading to loss of independence mainly due to decline in cognitive and motor functioning.4 ARWMCs relate to risk of recurrent stroke.5^–9

Stroke alone is associated with an increased risk of death.10 11 In a study on community dwelling elderly patients aged over 75 years without evidence of prior stroke or neurological diseases, in a long term follow-up of 11.8 years, severe ARWMCs were associated with a twofold increased risk of death,12 and more than half of the deaths were due to vascular causes.12 Severe degrees of ARWMC predicted poor survival in a follow-up of 2 years.5 In a large non-institutionalised cohort of subjects over 65 years without a history of stroke, ARWMCs predicted total, cardiovascular and non-cardiovascular mortality in a follow-up study of more than 10 years.7

At the moment, however, there are no data on the long term predictive value of ARWMCs in stroke patients referred to hospital. In the present study, based on a large well defined and detailed post-stroke cohort with a follow-up of up to 12 years, we hypothesised that severe ARWMCs (ie, confluent WMLs) are linked to impaired long term survival and brain related causes of death.

METHODS

Patients

The Helsinki Stroke Aging Memory cohort comprised a consecutive series of all Finnish (caucasian) patients with suspected stroke admitted to Helsinki University Central Hospital (n = 1622) between 1 December 1993 and 30 March 1995, as described in detail previously.13 14 Patients without ischaemic stroke (n = 175), presenting with intracerebral (n = 229) or subarachnoid haemorrhage (n = 69), were excluded. Of the 1149 patients with ischaemic stroke, we further excluded those younger than 55 years (n = 258) or older than 85 years (n = 88), those not living in Helsinki (n = 158) and those not speaking the Finnish language (n = 3). A total of 642 patients fulfilled the inclusion criteria and were invited to a follow-up visit 3 months later. Of these, 71 died (11.1%) before the 3 month follow-up, 82 refused (12.8%) and three were lost (0.5%) as a result of undefined causes. Finally, 486 (85.1%; 246 men, 240 women) of the living patients were included in the final cohort.15 The 85 patients who refused or were not identified were compared with the 486 patients in the cohort: mean age of the former group was 79.2 (7.68) years and that of the final cohort was 71.2 (7.6) (NS).14 The percentage of women was 67.1% and 49.4% (p = 0.023) respectively, and that of hospitalised subjects at the time of examination 60.0% and 16.8% (p = 0.0001), respectively. A detailed medical and neurological history was taken and stroke subtypes defined.14

Hypertension was defined as systolic blood pressure ⩾160 mm Hg or diastolic blood pressure ⩾95 mm Hg. Atrial fibrillation (AF) was defined by clinical criteria, and cardiac arrhythmias other than AF were defined as other arrhythmias. Myocardial infarction (MI) and cardiac failure (CF) were based on clinical diagnosis. Diabetes was defined as previously documented diagnosis, current use of insulin or oral hypoglycaemic medication, or fasting blood glucose >7.0 mmol/l. Peripheral atherosclerosis (PAD) was considered if the patient had claudication, >2 peripheral pulses missing or a history of amputation or peripheral arterial surgery due to atherosclerosis. Smoking habits were scored at admission as non-smokers and smokers (current or former). Laboratory analyses included total and high density lipoprotein cholesterol, triglycerides and fasting blood glucose. Hypercholesterolaemia was defined as total cholesterol >5.0 mmol/l. Hypertriglyceridaemia was defined as serum triglycerides >2.0 mmol/l. Low HDL was defined as <1.2 mmol/l. Stroke severity was assessed using the modified Rankin score.16

The study was approved by the ethics committee of the Department of Clinical Neurosciences, Helsinki University Central Hospital, Finland. The study was explained to the patients and informed consent was obtained.

MRI analysis

A total of 396 patients (81.5%) from the original cohort underwent MRI at 3 months and constituted the final study population. There were 312 patients (78.8%) with first ever and 84 patients (21.2%) with recurrent stroke. The reasons for not performing MRI in 90 patients were as follows: contraindication (27 patients), refusal (33 patients), claustrophobia (two patients), severe illness (27 patients) and obesity (one patient). The excluded patients were older (72.9 (7.4) vs 70.8 (7.7); p = 0.03), the number of women was smaller (28.9% vs 51.8%; p = 0.03) and patients were more often hospitalised (34.4% vs 12.9%, p<0.01) compared with the final study population.

MRI was performed with 1.0 T imaging equipment (Siemens Magnetom),15 as detailed previously. The protocol included transaxial T2, PD and T1 weighted 5 mm thick slices (conventional spin echo technique) and a three dimensional gradient echo sequence yielding 643 mm thick coronal sections. ARWMCs were rated on PD weighted images in accordance with the LADIS (Leukoaraiosis and Disability in the Elderly) ARWMC rating as none to mild, moderate and severe. The rating atlas has been detailed previously.17 In the none to mild degree of ARWMC, periventricular lesions included no more than small cap or thin lining, and in the other WM areas, no more than large focal lesions. In the moderate degree of ARWMC, periventricular lesions included no more than large cap and smooth halo, and in the other WM areas, no more than focal confluent lesions. The severe degree of ARWMC included cases with extending caps or irregular halo in the periventricular area and diffusely confluent lesions or extensive WM changes in other WM areas. Intra- and interobserver reliabilities for rating basic WMLs in periventricular and other WM areas were tested previously and found to be good.15 18 19However, we did not re-test intra- and interobserver reliabilities using atlas based assessment of the present follow-up study.17

Survival and causes of death

Long term survival data and causes of death on 21 September 2006 were obtained from Statistics Finland. Mean (SD) follow-up time was 7.5 (4.0) years (range 0.3–12.8 years). Of the 396 patients with MRI data, 277 (69.9%) had died during the follow-up and data on survival were obtained in all cases. The causes of death based on ICD-9 and ICD-10 classifications were also obtained and divided further into cardiac, brain related (ischaemic stroke, bleeding, vascular dementia), cancer, infection, trauma and other categories. The cause of death could not be verified for eight cases (2.9%) with death certificates.

Statistical analysis

SPSS/WIN (V.12.0, SPSS Inc) software was used. The effect of risk factors (sex, age, MI, CF, AF, other arrhythmias, arterial hypertension, PAD, diabetes, hypercholesterolaemia, hypertriglyceridaemia, low HDL), poor modified Rankin score (3–5 vs 0–2) and ARWMC degree on survival were first analysed using the Kaplan–Meier log rank test. The cumulative hazard function was also used and, based on these analyses, the proportional hazards assumption was met for each parameter included in further models. When analysing survival using cardiac death as an endpoint, the proportional hazards assumption was not met due to convergence of survival and hazard curves. Because of the low number of cases at the end of follow-up (12 years), log rank analysis was also performed after 10 years of follow-up. In Cox regression proportional hazards survival analysis, in the forced entry model (model 1), the potential predictors were used as covariates (sex, age, MI, CF, AF, other arrhythmias, arterial hypertension, PAD, diabetes, total and high density lipoprotein cholesterol, triglycerides, history of smoking and poor modified Rankin score). The final model (model 2) was constructed using likelihood ratio for selection of significant variables. The probability for variable entry was set at p<0.05 and for removal at p>0.10. As there were no differences in survival analyses between moderate and mild ARWMC categories, these categories were combined in Cox regression analyses. All analyses were also performed including patients with first ever stroke only (78.8%). As the results were similar in both analyses, all stroke cases were included in the study. Statistical significance was set at p<0.05.

RESULTS

In the Kaplan–Meier log rank analysis, patients with severe ARWMCs (n = 214, 54.0%) had shorter median survival after stroke (6.1, 95% CI 4.6 to 7.5 years) than patients with moderate (n = 71, 17.9%; 8.7, 95% CI 6.8 to 10.5 years; p = 0.040) and mild ARWMCs (n = 111, 28.0%; 9.7, 95% CI 7.5–11.8 years, p<0.0001). Accordingly, severe ARWMCs predicted poor survival (n = 214, 54.0%; 6.1, 95% CI 4.6 to 7.5 years) compared with mild to moderate ARWMCs (n = 182, 46.0%; 9.3, 95% CI 8.0 to 10.6 years; p<0.0001) (fig 1). There were no differences between mild and moderate ARWMC categories (p = 0.278). Predictors of poor survival in addition to ARWMCs were CF (4.5, 95% CI 3.5 to 5.5 vs 8.6, 95% CI 7.7 to 9.5 years; p<0.0001), AF (4.9, 95% CI 3.9 to 5.9 vs 8.6, 95% CI 7.7 to 9.4 years; p<0.0001), other arrhythmias (5.4, 95% CI 3.9 to 6.8 vs 8.3, 95% CI 7.4 to 9.3 years; p = 0.003), PAD (4.6, 95% CI 3.1 to 6.1 vs 8.2, 95% CI 7.4 to 8.9 years; p = 0.004) and poor Rankin score (3–5 vs 0–2) (4.9, 95% CI 4.2 to 5.7 vs 10.0, 95% CI 8.7 to 11.4 years; p<0.0001).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Effect of age related white matter changes (ARWMC) detected on MRI on overall post stroke survival (endpoint: all cause death) in Stroke Aging Memory cohort. Kaplan–Meier (K-M) log rank analysis.

To account for potential confounders, a multivariate Cox regression analysis utilising two models (models 1 and 2) was used (table 1). In addition to severe ARWMCs (HR 1.34, 95% CI 1.03 to 1.73; p = 0.029), advanced age (HR 1.07, 95% CI 1.05 to 1.09; p<0.0001), CF (HR 1.59, 95% CI 1.17 to 2.15; p = 0.003), AF (HR 1.68, 95% CI 1.24 to 2.27; p = 0.001), PAD (HR 1.59, 95% CI 1.11 to 2.26; p = 0.011), smoking (HR 1.60, 95% CI 1.23 to 2.08; p<0.0001), diabetes (HR 1.44, 95% CI 1.08 to 1.92; p = 0.013) and poor modified Rankin score (HR 1.65, 95% CI 1.26 to 2.14; p<0.0001) remained as independent predictors (model 2) and were associated with poor long term survival (table 1).

In univariate analysis, brain related causes of death (p = 0.001) and specifically due to infarction (p<0.0001) were associated with severe ARWMCs (table 2).

In Kaplan–Meier analysis with brain associated cause of death as the endpoint, severe ARWMCs predicted poor survival (8.3, 95% CI 6.9 to 9.7 years) compared with the mild–moderate category (11.1, 95% CI 9.3 to 12.8 years; p = 0.001) (fig 2A). No associations with other causes of death were found and ARWMC category was not associated with cardiac death (p = 0.513) (fig 2B).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Effect of age related white matter changes (ARWMC) on MRI on (A) brain related post stroke survival (endpoint: brain related cause of death) and (B) cardiac related post stroke survival (endpoint: cardiac related cause of death) in Stroke Aging Memory cohort. Kaplan–Meier (K-M) log rank analysis.

Following on from these findings, in multivariate Cox regression analysis using brain associated causes of death as the endpoint (models 1 and 2), severe ARWMCs remained as a predictor of poor survival (HR 1.80, 95% CI 1.10 to 2.96; p = 0.019) (table 3). Other predictors were advanced age (HR 1.05, 95% CI 1.01 to 1.09; p = 0.009), AF (HR 1.82, 95% CI 1.08 to 3.07; p = 0.026), PAD (HR 2.17, 95% CI 1.19 to 3.95; p = 0.012) and poor Rankin score (HR 2.75, 95% CI 1.67 to 4.54; p<0.0001) (table 3).

DISCUSSION

In the present study in post stroke patients, we have shown that patients with severe ARWMC (ie, confluent WMLs) had impaired survival compared with those with mild to moderate WMLs in a follow-up study of up to 12 years. In addition to ARWMC, independent predictors of poor post stroke survival included increased age and severity of stroke, as well as vascular factors, including CF, AF, PAD, diabetes and smoking. Severe ARWMCs were associated with death due to stroke and brain related causes of death. Our results on the effect of ARWMC on survival are in line with previous studies in non-institutionalised elderly subjects with5 and without a history of a previous stroke7 12 with mid-term follow-up. The major difference in our cohort is the lower proportion of patients with hypertension and double the amount of patients with AF or a history of smoking.5

As stroke alone seems to be associated with more than 70% risk of death over 10 years,10 11 in several studies the interest has been to investigate subjects without a history of stroke. Twenty per cent of patients in our cohort had evidence of a previous stroke which may have an effect on the predictive value of ARWMCs. Therefore, all analyses were also confirmed including patients with first-ever stroke only, and the results were similar to those in the present paper.

Severe ARWMCs on CT predicted morbidity and mortality, independent of neurological deficits in the mid-term.8 Similarly, in geriatric patients, ARWMCs predicted death, especially vascular deaths, after long term follow-up (relative risk 2.81).9 In the Dutch TIA trial, the risk of stroke was twice that in patients with ARWMCs compared with those without ARWMCs.20 In the NASCET trial, in patients with carotid endarterectomy, the 3 year risk of stroke was double in patients with severe ARWMCs compared with patients with no ARWMCs.21

We consider ARWMCs as important predictors of survival. According to our statistical models, the classical risk factors for stroke have a significant predictive value. Brain related death and ischaemic stroke were the dominant causes of death in patients with mild to moderate or severe ARWMCs. This was also confirmed in both the Kaplan–Meier analysis and Cox regression analysis adjusted for potential confounders in which ARWMCs predicted brain related cause of death and not other causes of death. Therefore, it is unlikely that ARWMCs led to deaths caused by falls or pneumonia related to hospitalisation. The rate of bleeding events and dementia was not statistically different in different ARWMC categories. ARWMCs can be regarded as surrogates for small vessel disease.4 It would be expected that the rate of death due to cerebral bleeding events and dementia would be elevated in the severe ARWMC category because there is an intriguing association between cerebral amyloid angiopathy, cerebral bleeding and white matter lesions.22^–26 Cerebral amyloid angiopathy can lead to intracerebral haemorrhage as a result of vessel fragility. Hypertension has also been demonstrated to be associated with this complex interaction.27 We believe our study is limited in size to allow for detailed analysis of these interactions. Our modified Fazekas rating scale17 has been used previously in the large LADIS study4 and was demonstrated to correlate well with the more complex Scheltens rating scale and semiautomated volumetric methods.28 In the present study, the modified rating scale allowed large enough subgroups allowing sufficient statistical power.

In Finland, determination of cause of death has been based on autopsy in approximately 30% of all deaths in the past two decades (www.tilastokeskus.fi) which is high compared with other European countries. In addition, the death certificates of all deceased, whether or not they underwent autopsy, are reviewed by the district forensic physician. The official cause of death has been demonstrated to be an accurate means of evaluating disease specific mortality in Finland.29 This adds to the reliability of the present study. In a previous study, more than half of the deaths in patients with severe ARWMCs were caused by vascular causes while only 33% and 6% were attributable to vascular causes in moderate and mild WMHs, respectively.12 These findings support our observation that ARWMCs are an independent predictor of death, especially brain related causes.

A potential weakness of our study is the possibility of selection bias, as the cohort was formed 3 months after the index stroke. This may limit generalisation of the results. Therefore, we retrospectively obtained additional data on stroke related deaths in Helsinki University Hospital district during the collection of the cohort from an independent organisation (Statistics Finland). In these retrospective data, up to 64% of stroke related deaths occurred in women. While the proportions of both sexes in the present study were equal, this suggests that some women may have died before hospital assessment at 3 months. Because of exclusion of patients, the true survival rate may be underestimated. One of the strengths of our study is that the cohort was a consecutive one and that all patients with suspected cerebrovascular ischaemic event were reviewed by senior neurologists and that the neuropsychological and clinical characteristics of the patients were strictly evaluated and also the severity of stroke was quantitated according to the modified Rankin score. A potential strength of our study is that our unit is responsible for primary stroke management of all inhabitants living in the Helsinki area. Also, the survival data are comprehensive with a negligible amount of unresolved deaths.

According to our results, patients with severe ARWMCs are at risk of poor post stroke survival. As treatment of hypertension seems to slow down the advancement of ARWMCs,30 future studies elucidating the effect of antihypertensive strategies on post stroke survival in different ARWMC categories is needed.