A Lancet-published study highlighted the limitations of the GRACE 2.0 heart attack mortality risk score – namely, the systematic underestimation of mortality risk in female patients – and offered a solution.
The internationally used Global Registry of Acute Coronary Events (GRACE) 2.0 system was developed and validated using a predominantly male patient population. And yet it “is used in both sexes alike,” without accounting for sex-specific disease characteristics.
In the new Lancet study, researchers from the UK and Switzerland reviewed GRACE 2.0’s performance in 421k patients with non-ST-elevation acute coronary syndromes, and found it to be less accurate in women:
- GRACE 2.0 accurately predicted in-hospital death for male patients but was notably less accurate with female patients (AUCs: 0.86 vs. 0.82).
- GRACE 2.0 was prone to incorrectly classify women as low-to-intermediate risk, keeping at-risk women from receiving early invasive treatment.
The authors didn’t stop there. They developed their own AI-powered algorithm that accounted for sex-specific disease differences. Their risk prediction model, dubbed GRACE 3.0, was trained on 310k patients, validated in 78k patients, and externally validated in 21k patients. GRACE 3.0 performed better for both women and men:
- Grace 3.0’s prediction accuracy exceeded GRACE 2.0 in both validation cohorts, irrespective of sex.
- On the external dataset, the model showed superior discrimination for both male and female patients (AUCs 0.91 & 0.87).
- GRACE 3.0 decreased the proportion of women (-5.3%) and increased the proportion of men (+2.2%) considered low-to-intermediate risk.
This study and the new GRACE 3.0 system take on an issue that has long plagued cardiology: male-biased data. The authors revealed clinically relevant limitations of the GRACE 2.0 score, which favors the under-treatment of female patients, and demonstrated GRACE 3.0’s advantages for males and females alike.