Author + information
- Tom Stuckey1,
- Narendra Singh2,
- Robi Goswami3,
- Jeremiah Depta4,
- Roger Gammon5,
- Michael Roberts6,
- John Steuter7,
- Sunny Gupta8 and
- William Sanders JR.9
- 1Lebauer Research, Greensboro, North Carolina, United States
- 2Atlanta Heart Specialists, Atlanta, Georgia, United States
- 3Piedmont Heart Institute, Atlanta, Georgia, United States
- 4Rochester General Hospital, Rochester, Georgia, United States
- 5Austin Heart, Austin, Texas, United States
- 6Lexington Cardiology, West Columbia, South Carolina, United States
- 7Bryan Heart Institute, Lincoln, Nebraska, United States
- 8Analytics 4 Life, Toronto, Ontario, Canada
- 9Analytics 4 Life, Morrisville, North Carolina, United States
Machine-learned solutions are rapidly being implemented for the analysis of health care system based big data. Cardiac phase space tomography analysis (cPSTA) analyzes thoracic physiological signals, without the use of radiation, exercise or pharmacological stress to identify the presence of flow-limiting coronary artery disease (CAD). The ongoing Coronary Artery Disease Learning & Algorithm Development (CADLAD) study is determining the diagnostic performance of cPSTA in assessing CAD among patients with chest pain, referred for coronary angiography (ANGIO). It is known that conventional diagnostic care pathways for detecting CAD are less accurate in women than men.
This prospective, multicenter, non-significant risk study was designed to develop and test machine-learned algorithms to assess the presence of CAD (defined as one or more ≥ 70% stenosis, or fractional flow reserve < 0.80) The overall interim results of CADLAD have been submitted elsewhere for presentation. This analysis focuses on CADLAD by gender. cPSTA signals were collected from subjects at rest, just prior to ANGIO. Features (mathematical and tomographic) were extracted from the signals, used for machine learning, and then blindly, prospectively tested in a validation cohort.
ANGIO results were paired with resting cPSTA data from 513 subjects were used to generate a machine-learned algorithm to assess significant CAD. A separate validation cohort of 94 subjects was used to prospectively test the accuracy of cPSTA in assessing significant CAD. Results are shown in the table.
|Table: cPSTA Performance by Gender||Men||Women|
|Sensitivity (95% CI)||83% (56%, 95%)||100% (100%,100%)|
|Specificity (95% CI)||64% (49%, 76%)||73% (42%, 92%)|
|AUC (95% CI) (Area Under the receiver-operator characteristic Curve)||0.76 (0.62, 0.86)||0.82 (0.60, 0.96)|
|NPV (95% CI) (Negative Predictive Value)||91% (76%, 97%)||100% (100%,100%)|
|TP (True Positive)||15||7|
|TN (True Negative)||30||16|
|FP (False Positive)||17||6|
|FN (False Negative)||3||0|
These initial data, while of limited power, suggest that resting cPSTA imaging performs well overall, and that the results in women are equivalent or better as compared with men. The ongoing CADLAD trial is designed to test the utility of cPSTA in a larger population that includes a large number of women.
OTHER: Womens Health Issues