Written by Clay Smith
Among nine syncope rules, the Canadian Syncope Risk Score was the best predictor of which patients should be admitted and which were safe for discharge.
Why does this matter?
We have covered syncope in the past. The San Francisco Syncope Rule (SFSR) has been externally validated but seems to have a higher than desirable miss rate, with a pooled sensitivity of 86%. The Canadian Syncope Risk Score (CSRS) has performed well and has also been externally validated. What does the pooled analysis show when we look at nine different syncope rules: SFSR, CSRS, OESIL, ROSE, FAINT, Boston syncope, Syncope Risk Score, Canadian Syncope Arrhythmia Score, and the Ottawa ECG rule?
This was a systematic review of 17 studies, 24,234 patients, evaluated for syncope. They pooled the studies for each of the scores listed above and calculated positive and negative likelihood ratios (LR+; LR-) for each syncope score. For context, if a LR+ is ≥4, it meaningfully increases the likelihood of a positive outcome; a LR+ of ≥10 is a very powerful positive discriminator. A LR- of 0.25 or lower meaningfully decreases the likelihood of an outcome; ≤0.1 is a very powerful negative discriminator. A LR of 1 means a test is useless. For all the syncope rules that had been externally validated, except the CSRS, the LR+ ranged from a lackluster 1 to 2.5; LR- were better, with several <0.1. However, by far the best performer was the CSRS. If simplified and dichotomized, any CSRS ≥4 had a LR+ of 11, and any CSRS score <1 (scores can go from 0 to -3) had a LR- of 0.1. This is clinically meaningful for decision making. I have started using the CSRS (and this handy calculator from MDCalc) in practice.
Multivariable risk scores for predicting short term outcomes for emergency department patients with unexplained syncope: a systematic review. Acad Emerg Med. 2020 Dec 31. doi: 10.1111/acem.14203. Online ahead of print.
2 thoughts on “Canadians Win Again – Syncope Score Systematic Review”
The abstract of this paper is ambiguous. Is it the case that scores were not calculated by the authors in this work? If so, the number of patients included – 24000 – is simply the sum of the populations in each study included in the review. This number has no statistical relevance. The findings would simply be a comparison of the accuracies reported in each included study and of limited interest.
(By the way, I love journalfeed reviews – they definitely improve my practice! Thanks.)