Derivation and validation of a prognostic prediction rule from clinical and stress CMR data characterizes cardiac prognostication in patients with suspected myocardial ischemia
Journal of Cardiovascular Magnetic Resonance volume 13, Article number: O4 (2011)
We sought to derive and validate a parsimonious predictive rule incorporating clinical and stress CMR data to characterize major adverse cardiac events (MACE) in patients with suspected ischemia.
Out of 736 patients referred for assessment of suspected ischemia by vasodilator stress perfusion CMR, 25 (3%) were excluded due to images of insufficient quality. The remaining 711 patients (297 females, mean age 56±15 years) were followed for cardiac events (MACE) that occurred within the first 3 years after CMR (100% complete follow-up; median of 21.4 months, range 2.5 months to 8.2 years). We randomized patients in this clinical cohort 1:1 into a training set (SETTraining, n=356) and a testing set (SETTesting, n=355). Forty-six (6.5%) patients experienced MACE: 24 (7%) in the SETTraining and 22 (6%) in the SETTesting. A parsimonious prediction rule was built by stepwise regression in the SETTraining, considering all clinical, ECG, and CMR parameters using P<0.05 for both entry and stay. History of PCI, wall motion abnormality at rest, non-sinus rhythm, and CMR myocardial ischemia score (ISCH-SCORE) formed the prediction rule (Table-1). This prediction rule was validated in the SETTesting divided into 3 progressive categories based on their ascending predicted probabilities of MACE (low: 1.5 to 10% and high:>10%). The utility of the model was first determined by comparing the predicted and observed numbers of MACE in each risk category in the SETTesting (Figure-1).
We then performed ROC analyses, in either patient set and both sets combined, to determine how well the range of probabilities by the prediction rule can determine MACE occurence. MACE prediction was excellent in SETTraining, SETTesting, and both sets combined (AUC 83%, 86%, and 79%, P=NS, figure-2).
In addition, we performed an internal validation of the same prediction rule using 400 bootstrap simulations. Each bootstrap sample was the same size as the original derivation sample, but patients were drawn randomly with replacement from the sample. By bootstrap simulation the prediction rule’s estimated optimism was very low (8%±3) and mean AUC of the prediction rule was 82.4% (95%CI: 70%-93%) reinforcing its validity and utility (table-2).
We derived and validated a prediction rule for MACE using clinical and ISCH-SCORE for patients with suspected myocardial ischemia. ISCH-SCORE provided strong and major prognostic utility in this prediction rule.
About this article
Cite this article
Coelho-Filho, O.R., Mongeon, FP., Jerosch-Herold, M. et al. Derivation and validation of a prognostic prediction rule from clinical and stress CMR data characterizes cardiac prognostication in patients with suspected myocardial ischemia. J Cardiovasc Magn Reson 13 (Suppl 1), O4 (2011). https://doi.org/10.1186/1532-429X-13-S1-O4