Kevin S. Attenhofer, MD
You can’t discuss stroke prevention without talking about antiplatelet drugs. Drugs like Aspirin and Clopidigrel are frequently used by stroke neurologists as the secondary prevention treatment of choice for patients with TIA and non-cardioembolic ischemic strokes. However, we must remember that bleeding is a clinically important and potentially life-threatening side effect of these agents. Reliably predicting who is most likely to bleed would dramatically inform the clinician’s decision-making process and likely result in improved patient outcomes.
To that end, the S2TOP-BLEED score was derived from patient data from six randomized trials (CAPRIE, ESPS-2, MATCH, CHARISMA, ESPRIT, and PRoFESS). It is a 28-point score that incorporates readily available patient characteristics: Sex, Smoking history, Type of antiplatelet, Outcome (mRS), Prior stroke, Blood pressure, Low BMI, Elderly, Ethnicity, Diabetes. It was validated in the PERFORM trial (Prevention of Cerebrovascular and Cardiovascular Events of Ischemic Origin Terutroban in Patients with a History of Ischemic Stroke or Transient Ischemic Attack Study). In this paper, Hilkens et al. externally validate the S2TOP-BLEED score in observational data from a real-world setting.
The setting the authors used is the Oxford Vascular Study (an ongoing population-based study on the incidence and outcome of acute vascular events in the UK). They looked at patients with TIA and ischemic stroke between 2002 and 2012 who were on antiplatelet after the event. Included patients had a median follow-up of 3.5 years.
This paper uses the c-statistic to compare predictive models. A brief refresher course on statistics: The c-statistic gives the probability that a randomly selected patient who experiences an event (in this case, bleed) had a higher risk score than a patient who did not experience the event. Values range from 0.5 to 1, where values less than 0.5 are unacceptably poor, values over 0.7 indicate a good model, values over 0.8 indicate a strong model, and a value of 1 means that the model perfectly predicts outcome. Notably, the c-statistic does not provide as much information as the receiver operating characteristic. For S2TOP-BLEED, the authors report a c-statistic of 0.69 for major bleeding at 3 years. Their score was best at predicting intracranial or upper GI bleeding (likely to be the most clinically important bleeds).
Compared to the REACH score (c-statistic 0.63 for major bleeding at 2 years) and the Intracranial-B2LEED3S score (c-statistic 0.60 for intracranial bleed at 2 years), the S2TOP-BLEED score showed better performance for prediction of both intracranial and major bleeds. Interestingly, the fact that the observed c-statistic is maintained/improved compared to the development cohort speaks to the generalizability of this prediction model to a wide range of stroke patients in clinical practice.
The ratio of ischemic events versus bleeds decreased from 7.5:1 in the low-risk group (scores 0-10) to 2.9:1 in the intermediate risk group (score 11-15) and 1.8:1 in the high-risk group (score 16+). Although the risk of recurrent ischemia will outweigh the risk of bleeding in most patients, this prediction score may help identify particularly high-risk patients in whom preventive measures should be taken.
Sitting at just below 0.70, this is a validated and generalizable tool that can useful for the clinician prescribing antiplatelet agents to stroke patients daily. It has a comparable c-statistic to prediction scores in anticoagulation such as HAS-BLED. The authors cite possible room for improvement by including factors such as renal dysfunction in a future version of this score. I think it’s also possible that this score (or one like it) may help identify a possible population in whom the benefit of dual antiplatelet therapy may outweigh the risk.
