Resident Physician University Of Toronto Toronto, Ontario, Canada
Background: Clinical prediction models play a central role in guiding diagnostic testing for patients with stable chest pain. However, current models may underperform in women due to lack of insight into sex-related differences in the clinical presentation and pathophysiology of obstructive coronary artery disease (OCAD). This systematic review aims to evaluate the overall and sex-specific performance of externally validated clinical prediction models for OCAD.
METHODS AND RESULTS: We systematically searched MEDLINE, EMBASE, and Cochrane databases for studies published from 2011 onward, to align with the pivotal publication of the updated Diamond-Forrester model. Eligible studies evaluated clinical prediction tools for OCAD which used ≥3 clinical variables not including imaging or dedicated cardiac testing, and were applicable at the bedside to individuals with stable chest pain and no prior CAD. Primary analysis included external validation studies of predictive models defining OCAD as ≥ 50% stenosis on coronary CT angiography and/or invasive coronary angiography. Data was presented using summary statistics, and a one-sample t-test was used to compare model performance between men and women.
In total, 49 analyses evaluating 14 models in 18 unique external validation cohorts comprising 78,863 patients were included. The prevalence of OCAD varied between cohorts, ranging from 8.1-72.1% (mean 33.0%). The CAD Consortium Clinical model was the most frequently studied (n=5) and the most common predictive variables were sex (n=14), age (n=13), chest pain typicality (n=12), diabetes mellitus (n=8), and hypertension (n=6).
Area under the receiver operating characteristic curve (AUC) values for all included models ranged from 0.62 to 0.79 (mean 0.72), reflecting moderate discrimination. At a clinically relevant pre-test probability (PTP) threshold of 15% (n=12), sensitivity ranged from 58.5% to 97.4% with a mean of 78.0%, and mean NPV was 89.9% (range 79.7-98.6%). Calibration was inconsistently reported, limiting conclusions.
The proportion of female participants ranged from 24.4% to 66.3% (mean 44.9%). Despite relatively balanced representation, only 7/49 analyses (14.3%), representing five models, validated sex-specific discriminatory performance. There was a numerical trend towards reduced AUC in women (p=0.124). Of these, the CAD consortium clinical model performed best (AUC 0.722) in women.
Conclusion: Most clinical prediction tools for OCAD demonstrate only moderate discriminatory performance, and sex-specific validation remains critically lacking, highlighting the need for sex-specific validation of predictive models to support equitable, evidence-based OCAD care.
