Aims: Indiscriminate coronary computed tomography angiography (CCTA) referrals for suspected coronary artery disease could result in a higher rate of equivocal and non-diagnostic studies, leading to inappropriate downstream resource utilization or delayed time to diagnosis. We sought to develop a simple clinical tool for predicting the likelihood of a non-diagnostic CCTA to help identify patients who might be better served with a different test.
Methods and results: We developed a clinical scoring system from a cohort of 21 492 consecutive patients who underwent CCTA between February 2006 and May 2021. Coronary computed tomography angiography study results were categorized as normal, abnormal, or non-diagnostic. Multivariable logistic regression analysis was conducted to produce a model that predicted the likelihood of a non-diagnostic test. Machine learning (ML) models were utilized to validate the predictor selection and prediction performance. Both logistic regression and ML models achieved fair discriminate ability with an area under the curve of 0.630 [95% confidence interval (CI) 0.618-0.641] and 0.634 (95% CI 0.612-0.656), respectively. The presence of a cardiac implant and weight >100 kg were among the most influential predictors of a non-diagnostic study.
Conclusion: We developed a model that could be implemented at the 'point-of-scheduling' to identify patients who would be best served by another non-invasive diagnostic test.
Keywords: cardiac imaging techniques; coronary computed tomographic angiography; image interpretation; non-diagnostic tests; prediction model.
© The Author(s) 2023. Published by Oxford University Press on behalf of the European Society of Cardiology.