Skip navigation
Link to Back issues listing | Back Issue Listing with content Index | Subject Index

Predicting early stroke survival

Study
Results
Comment

Early prediction of recovery with good quality of life after stroke can be done using a scoring system that combines magnetic resonance imaging, stroke scores and time from admission [1], and has been summarised on the Bandolier Internet site . What about the corollary, that of predicting likelihood of death in the first month, and when magnetic resonance imaging is not available? A study gives a way of doing just that using a simple clinical scoring system.

Study


The study involved a retrospective cohort of consecutive 440 patients admitted with a diagnosis of acute ischaemic stroke. Exclusions were intracerebral or subarachnoid haemorrhage, patients with unclear diagnosis, or where the diagnosis was recorded as acute cerebrovascular accident.

Patients were randomly allocated to a derivation and a validation group. The derivation group was used to develop the prediction model, and the validation sample was used to test the prediction model. The primary outcome was death within 30 days of admission, ascertained using linkage of patient records to mortality data. A number of baseline measures were investigated for the model. Included were level of consciousness (including impaired consciousness, defined as drowsy but responsive to verbal stimuli), dysphagia with moderate or severe swallowing difficulty.

Results


The 440 patients came from a larger sample of 544 before exclusions. Of the 440, 45 (10%) died within 30 days. The mean age was 70 years. About a quarter had impaired consciousness or were unconscious, 20% had dysphagia, and 10% had faecal or urinary incontinence.

Death rates were higher with impaired consciousness (Figure 1) and in those with dysphagia (40%), had urinary incontinence (61%) had a body temperature above 36.5ºC (15%) and had hyperglycaemia without a history of diabetes (11%). These were all factors that remained significant in a regression model (combining unconscious and impaired consciousness), and they were given "points" according to the hazard ratios from the statistics - greater levels of significance were given more points.

Figure 1: Level of consciousness and mortality after stroke


Table 1: Scoring prognostic factors

Prognostic index for 30-day stroke mortality
Factor Points
Impaired consciousness 5
Urinary incontinence 4
Dysphagia 3
Admission temp >36º5C 2
Hyperglycaemia with no history of diabetes 2
Maximum 16

The final model is given in Table 1. Mortality clearly increased with prognostic index (Figure 2). The risk of 30-day mortality with patients with scores of less than 11 was 3%, and for those with scores of 11 or more was 75%. The likelihood ratio for a score of 11 or more was 34, and for one of less than 11 was 0.3. The validation group gave almost identical results to the derivation group.

Figure 2: Mortality and prognostic score



Comment


This is a good exemplar of how to construct a useful clinical scoring system. It tells us something we want to know, and allows us to perform a simple calculation from readily available clinical data, without resort to high technology. Ideally we would like to see a prospective evaluation, and the authors tell us they are doing this.

References:

  1. EA Baird et al. A three-item scale for the early prediction of stroke recovery. Lancet 2001 357: 2095-2099.
  2. Y Wang et al. A prognostic index for 30-day mortality after stroke. Journal of Clinical Epidemiology 2001 54: 766-773.
previous or next story in this issue