Table 1.
Classification accuracies (%) of different sets of features.
PCHIP and moving average features classify better than conventional parameters, and slightly better than all nonzero initial conditions.
Fig 1.
MDGini variation (in the classifier built with 200s-MA features) with time for the five selected chemical species.
Tf-fVIIa-fXa and Tf-fVIIa-fX are most significant during 1000–1600 seconds, and IIa during 1400–2500 seconds from the addition of the trigger. Significance of fIXa-fVIIIa-fX increases monotonically and remains most significant at 3600 seconds suggesting that it is a long-lived species. 200s-MA—200-second moving average; MDGini—Mean Decrease in Gini index; Tf-fVIIa-fXa—Tissue factor-factor VIIa-factor Xa; Tf-fVIIa-fX—Tissue factor-factor VIIa-factor X; fIXa-fVIIIa-fX—factor IXa-factor VIIIa-factor X; IIa—activated alpha-thrombin.
Table 2.
Classification accuracies (%), mean (SD), for 200s-MA values of selected species.
Classification using all 18 200s-MA features of Tf-fVIIa-fXa, Tf-fVIIa-fX, fIXa-fVIIIa-fX, and IIa result in similar accuracies. Classification accuracies of the best 3 and the best feature from each species indicate significance is most localized in Tf-fVIIa-fXa.
Fig 2.
Means and 90% quantiles for Tf-fVIIa-fXa and fXa simulation profiles in ACS and CAD populations.
A: Tf-fVIIa-fXa concentration profiles from the two groups split significantly from about 1000 to 1500 seconds. B: In fXa concentration profiles, there is a huge variation in both ACS and CAD populations. However, the profiles from the two groups overlap and make the features of this species weak for classification. Tf-fVIIa-fXa—Tissue factor-factor VIIa-factor Xa; fXa—factor Xa.
Fig 3.
Means and 90% quantiles for fIXa-fVIIIa-fX and IIa simulation profiles in ACS and CAD populations.
A: fIXa-fVIIIa-fX profiles show that this species is more long-lived in ACS than CAD cases. B: Though IIa concentration profiles appear to reach zero by 2000 seconds, MDGini suggests that the dynamics between the two groups is most significant during that time. fIXa-fVIIIa-fX—factor IXa-factor VIIIa-factor X; IIa—activated alpha-thrombin; MDGini—Mean Decrease in Gini index.
Table 3.
Classification accuracies (%), for classifiers built using combinations of best 200s-MA features.
An efficient way to assay the entire system is by measuring three species at three specific time intervals of 200 seconds. Tf-fVIIa-fXa, IIa and fIXa-fVIIIa-fX make the best combination.
Fig 4.
A single decision tree built with just two of the best 200s-MA features, one each from fIXa-fVIIIa-fX and Tf-VIIa-Xa, is shown. ACS and CAD populations separate well in just those two features. fIXa-fVIIIa-fX—factor IXa-factor VIIIa-factor X; Tf-VIIa-Xa—Tissue factor-factor VIIa-factor Xa; 200s-MA—200-second moving average.