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

Risk of stroke with AF

Risk classification scheme

Bandolier was once made aware of the GP's crie de coeur: don't tell us what treatment to use, but tell us what patients to treat! The trouble is that there's not an awful lot of good evidence around. Some rejoicing, then, when a paper helps by giving a simple clinical way of estimating the risk of stroke in patients with non-rheumatic atrial fibrillation [1].


The patients in this study were 1733 people aged 65 to 95 years discharged from hospital with a diagnosis of non-rheumatic atrial fibrillation, without any treatment apart from aspirin in some. A national registry was generated using anonymous information from five quality improvement or peer review organisations serving seven US states. Medicare records could be reviewed for proper assessment and diagnosis of atrial fibrillation, documented risk factors, any therapy or comorbid conditions. Standardised abstraction forms were used, with excellent agreement between abstractors.

The study outcome was hospital admission for first ischaemic stroke, as determined by Medicare claims. The minimum follow up was 365 days, and the maximum was 1000 days.

Risk classification scheme

Two risk classification schemes used in clinical trials to classify patients with atrial fibrillation at low, medium and high risk of stroke were combined. The new scheme involved independent risk factors identified in the original two schemes: one point was given for the presence of congestive heart failure, hypertension (systolic >160 mmHg), age greater than 75 years, and diabetes, and two points given for prior cerebral ischaemia (Table 1). The name of the new classification scheme, CHADS2, is also an acronym.

Table 1: Components of CHADS2

CHADS2 item Points
Congestive heart failure 1
Hypertension (systolic >160 mmHg) 1
Age greater than 75 years 1
Diabetes 1
Prior cerebral ischaemia 2


There were 94 strokes in the 1733 patients over an average 1.2 year follow up, a crude average rate of 4.5% a year. Within the cohort, the crude rate for patients with no risk factors (120 people) was 1.2%. With increasing risk scores, the crude annual risk rose also. Figure 1 shows the 95% confidence interval of the rate after smoothing through an exponential survival model. Table 2 gives the numbers..

Figure 1: CHADS score and risk of stroke

Table 2: CHADS score and risk of stroke

  Number of:  
CHADS2 score Patients Strokes Adjusted annual stroke rate (95%CI)
0 120 2 1.9 (1.2 to 3.0)
1 463 17 2.8 (2.0 to 3.8)
2 523 23 4.0 (3.1 to 5.1)
3 337 25 5.9 (4.6 to 7.3)
4 220 19 8.5 (6.3 to 11.1)
5 65 6 12.5 (8.2 to 17.5)
6 5 2 18.2 (10.5 to 27.4)


This is another example of a clinical scoring system being tested and showing how useful they can be. Here the very system, being an acronym, is helpful in bringing to mind important risk factors for stroke with non-rheumatic atrial fibrillation. The annual risk of stroke rose from under 2% a year with no risk factors to over 10% a year for five or six. At some point the balance tips to the use of anticoagulants.

It would be terrific to see this scoring system tested on a separate data set, but for now we have a useful way of targeting warfarin treatment at patients who will benefit most, those with a high baseline risk. Patients with a lower risk can be offered aspirin.


  1. BF Gage et al. Validation of clinical classification schemes for predicting stroke. Results from the national registry of atrial fibrillation. JAMA 2001 285: 2864-2870.
previous or next story in this issue