Example 2

Analysis Populations

The “Discovery-Full Analysis Set” (“Discovery FAS”) consisted of pilot study patients with clinical data and a CT-based designation of either Revascularization CAD case, Native CAD case, or Control (N=748 for the Discovery-FAS group).

The “Discovery-Native CAD Set” was the subset of the Discovery-FAS with Native CAD as verified by CT, who had analyte (metabolomic) data (N=366 for the Discovery-Native CAD Set). These were subjects without previous revascularization procedures, such as percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG).

The “Discovery-Revasc CAD Set” was the subset of the Discovery-FAS who had undergone previous revascularization, such as percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG), and who had analyte data (N=44).

The “Discovery-All CAD Set” was the union of the Discovery-Native CAD Set and the Discovery-Revasc CAD Set (N=410).

The “Discovery-Control Set” was the subset of Discovery-FAS who had a calcium score of zero and were designated a Control after inspection of CT data, and who had analyte data. (N=338 for the Discovery-Control Set.)

The “Validation-Full Analysis Set” (“Validation-FAS”) consisted of pilot study patients with clinical data and a CT-based designation of either Revascularization CAD case, Native CAD case, or Control (N=348 for the Validation-FAS group).

The “Validation-Native CAD Set” was the subset of the Validation-FAS with Native CAD as verified by CT, who had analyte (metabolomic) data (N=207 for the Validation-Native CAD Set). These were subjects without previous revascularization procedures, such as percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG).

The “Validation-Revasc CAD Set” was the subset of the Validation-FAS who had undergone previous revascularization, such as percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG), and who had analyte data (N=15).

The “Validation-All CAD Set” was the union of the Validation-Native CAD Set and the Validation-Revasc CAD Set (N=222).

The “Validation-Control Set” was the subset of Validation-FAS who had a calcium score of zero and were designated a Control after inspection of CT data, and who had analyte data. (N=126 for the Validation-Control Set)

It is noted that by design, the only racial group represented in the study was White. Therefore, race-based sub-populations were not defined.

A. Study Endpoints

For the GLOBAL Pilot Discovery Cohort, there were four primary endpoints in the analysis: (1) Native CAD; (2) All CAD (Native or Revascularization); (3) 50% Stenosis without Revascularization; (4) 50% Stenosis or Revascularization. All analyses were applied to all primary endpoints.

B. Statistical Hypothesis

The null hypothesis of no association, between the metabolite or lipid and the endpoint, was tested against the two-sided alternative that association exists.

C. Multiple Comparisons and Multiplicity

False discovery rate (FDR) q-values were calculated (Benjamini and Hochberg, 1995). Associations with FDR q<0.05 were considered preliminary associations. In some circumstances, test results with raw p<0.05 were reported as well.

D. Missing Data

Endpoint data was not imputed. Potential covariates with more than 5% missing data were excluded. Potential covariates with less than 5% missing data were imputed to the mean.

Metabolites with more than 10% missing data were excluded from the main analyses. Missing values for metabolites and lipids with less than 10% missing were imputed to the observed minimum after normalization.

E. Analysis of Subgroups

The first and third primary endpoints were addressed using a subset of the FAS. Specifically the Native CAS Set and the Control Set were considered to the exclusion of the Revasc. CAD Set. For the purposes of discovery, further subsets were created on the basis of participants' fasting status, where patients were categorized as Fasting if they had not eaten for eight or more hours. The remainder, either known not to be fasted, or with unknown fasting status were categorized as ‘Non-Fasting’. See FIG. 50.

I. Demographic and Baseline Characteristics

The baseline and demographic characteristics of patients in the pilot study were tabulated. Continuous variables were summarized by the mean and standard error; binary variables were summarized by the count and percentage.

Table 27 shows general patient characteristics for the Discovery Set by clinical group (Revasc CAD vs. Native CAD vs. Control). A Kruskall-Wallis test was performed to investigate homogeneity of continuous measures; a Pearson's chi-squared test was conducted for binary measures; unadjusted p-values are reported.

Table 28 shows general patient characteristics for the Validation Set by clinical group (Revasc CAD vs. Native CAD vs. Control). A Kruskall-Wallis test was performed to investigate homogeneity of continuous measures; a Pearson's chi-squared test was conducted for binary measures; unadjusted p-values are reported.

TABLE 27
All ControlsNative CADRevasc CADP-value
N33836644
Age
mean (SE)53.8 (0.57) 58.02 (0.54) 59.55 (1.40)  3.93E−07
SBP
mean (SE)129.62 (0.91)  132.6 (0.92) 128.09 (2.32)  0.0550
DBP
mean (SE)79.03 (0.56) 79.73 (0.58) 75.12 (1.57)  0.0402
Male
N (%)151 (44.67)195 (53.28)30 (68.18)0.0037
Hypertension
N (%)172 (51.04)244 (66.85)40 (90.91)1.61E−08
Dyslipidemia
N (%)184 (55.42)259 (71.15) 43 (100.00)4.95E−10
Diabetes (Any)
N (%)25 (7.40) 54 (14.75)10 (22.73)8.00E−04
Type I Diabetes
N (%) 1 (0.30) 1 (0.27)0 (0.00)0.9377
Type II Diabetes
N (%)24 (7.10) 53 (14.48)10 (22.73)6.00E−04
Current Smoker
N (%) 39 (11.54) 67 (18.31) 5 (11.36)0.0331
Former Smoker
N (%) 84 (24.85)130 (35.52)21 (47.73)5.00E−04
Chest Pain
N (%)221 (65.38)212 (57.92)30 (68.18)0.0850
Angina
Equivalent
N (%)126 (37.28)122 (33.33)19 (43.18)0.3115
Shortness of Breath
N (%) 74 (21.89) 71 (19.40) 8 (18.18)0.6635
Family History of CAD
N (%)179 (52.96)223 (60.93)28 (63.64)0.0710
Fasting
N (%)120 (35.50)207 (56.56) 44 (100.00)8.28E−18
Statin
N (%)111 (32.84)184 (50.27)38 (86.36)1.28E−12
Niacin
N (%) 5 (1.48) 4 (1.09)3 (6.82)0.0164
Fibrate
N (%)12 (3.55)21 (5.74)3 (6.82)0.3254
Ezetimibe
N (%) 6 (1.78)11 (3.01) 8 (18.18)7.97E−08
Fish Oil
N (%)26 (7.69) 50 (13.66) 5 (11.36)0.0388
Bile Acid Sequestrant
N (%) 3 (0.89) 4 (1.09)0 (0.00)0.7705
Aspirin
N (%) 98 (28.99)157 (42.90)34 (77.27)3.15E−10
Clopidogrel
N (%) 7 (2.07)11 (3.01)15 (34.09)5.21E−22
Vitamin K Antagonist
N (%) 7 (2.07)17 (4.64)2 (4.55)0.1629
Nitrate
N (%) 7 (2.07)16 (4.37)11 (25.00)5.57E−11
Beta Blocker
N (%)114 (33.73)140 (38.25)34 (77.27)1.68E−07
ACE Inhibitor
N (%) 63 (18.64)106 (28.96)24 (54.55)3.13E−07

TABLE 28
All ControlsNative CADRevasc CADP-value
N12620715
Age
mean (SE)49.48 (0.93)  60.83 (0.59)  68.6 (2.05) 3.35E−22
SBP
mean (SE)127.25 (1.55)  131.46 (1.10)  131.8 (4.16)  0.0276
DBP
mean (SE)77.33 (0.98)  78.99 (0.77)  78.87 (3.16)  0.2091
Male
N (%)46 (36.51)138 (66.67)  15 (100.00)1.36E−09
Hypertension
N (%)51 (40.48)132 (63.77) 13 (86.67)9.44E−06
Dyslipidemia
N (%)65 (53.72)152 (74.88)  14 (100.00)1.33E−05
Diabetes (Any)
N (%)12 (9.52) 44 (21.36) 4 (26.67)0.0134
Type I Diabetes
N (%)1 (0.79)3 (1.45)1 (6.67)0.1954
Type II Diabetes
N (%)11 (8.73) 41 (19.81) 3 (20.00)0.0244
Current Smoker
N (%)20 (15.87)28 (13.53)0 (0.00)0.238
Former Smoker
N (%)25 (19.84)78 (37.68) 8 (53.33)6.00E−04
Chest Pain
N (%)96 (76.19)110 (53.14)  7 (46.67)7.78E−05
Angina Equivalent
N (%)46 (36.51)53 (25.60) 5 (33.33)0.1037
Shortness of Breath
N (%)28 (22.22)30 (14.49) 7 (46.67)0.0038
Family History of CAD
N (%)53 (42.06)129 (62.32)  6 (40.00)8.00E−04
Fasting
N (%)126 (100.00)207 (100.00) 15 (100.00)NA
Statin
N (%)39 (30.95)125 (60.39) 13 (86.67)2.28E−08
Niacin
N (%)1 (0.79)3 (1.45)0 (0.00)0.7872
Fibrate
N (%)4 (3.17)7 (3.38)1 (6.67)0.7797
Ezetimibe
N (%)1 (0.79)5 (2.42)0 (0.00)0.4745
Fish Oil
N (%)8 (6.35)23 (11.11)1 (6.67)0.3251
Bile Acid Sequestrant
N (%)0 (0.00)0 (0.00)0 (0.00)NA
Aspirin
N (%)28 (22.22)99 (47.83) 9 (60.00)4.91E−06
Clopidogrel
N (%)1 (0.79)2 (0.97) 4 (26.67)3.16E−11
Vitamin K Antagonist
N (%)6 (4.76)3 (1.45)1 (6.67)0.1432
Nitrate
N (%)3 (2.38)10 (4.83)  3 (20.00)0.0084
Beta Blocker
N (%)36 (28.57)77 (37.20)12 (80.00)4.00E−04
ACE Inhibitor
N (%)26 (20.63)59 (28.50) 9 (60.00)0.0039
II. Exploratory Data Analyses for Metabolites

Sample preparation and mass spectrometry analyses were conducted by Metabolon, Inc. The raw data contained a total of 1088 analytes, measured for 1096 pilot study participants.

Of the 1088 analytes (including unnamed metabolites and complex lipids), 481 named metabolites had less than 10% missing data. All 1096 patients had less than 10% missing data for these metabolites. Statistical analyses were therefore applied to 481 analytes and 1096 patients. The data was normalized in advance of receipt. A logarithm (base 2) transformation was applied and histograms were created to show the distribution of expression by analyte (data not shown).

The metabolomics data were generated in multiple batches; however, a principal components analysis (PCA) showed no evidence of any systematic site effects.

III. Prediction Modeling for Primary Endpoints

Methods. Patients in the Discovery-FAS Set were categorized according to whether they had fasted for at least eight hours. By this criteria, a total of 377 participants were Fasted and 371 were Non-Fasted. Association testing, with adjustment for age and gender was conducted for the four primary endpoints, and nominal associations were defined in three ways as follows:

    • 1 Significant in Fasting and Non-Fasting combined
    • 2 Significant in Fasting and Non-Fasting independently
    • 3 Significant in Fasting alone

It is emphasized that, at this stage, ‘significant’ pertains to any association with raw, unadjusted p<0.05.

In this way, twelve scenarios were considered as follows:

    • a) Atherosclerosis in Native CAD—AnCAD
      • c. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • c. Independently Significant in Fasting and Non-Fasting—[Figure (not displayed)]
      • c. Significant in Fasting—[Figure (not displayed)]
    • b) Atherosclerosis in All CAD (including revascularization)—AaCAD
      • c. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • c. Independently Significant in Fasting and Non-Fasting—[Figure (not displayed)]
      • c. Significant in Fasting—[Figure (not displayed)]
    • c) 50% stenosis in Native CAD—SnCAD
      • c. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • c. Independently Significant in Fasting and Non-Fasting—[Figure (not displayed)]
      • c. Significant in Fasting—[Figure (not displayed)]
    • d) 50% stenosis in ALL CAD (including revascularization)—SaCAD
      • c. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • c. Independently Significant in Fasting and Non-Fasting—[Figure (not displayed)]
      • c. Analytes Significant in Fasting—[Figure (not displayed)].

When more than 9 variables had p<0.05, Age and Gender were added to the variables, and gradient boosting (see below) was applied to select 9 predictors.

Twelve prediction models were obtained by generalized linear (logistic) regression as follows. When fewer than nine variables had p<0.05, Age and Gender were added to the variables, and the full model was fitted. Otherwise, the nine variables selected by gradient boosting variables were combined with Age and Gender in a generalized linear (logistic) model.

Gradient boosting is an approach to determine a regression function that minimizes the expectation of a loss function. (Freidman J H (2001) and Friedman J H (2002)) It is an iterative method, in which the negative gradient of the loss function is calculated, a regression model is fitted, the gradient descent step size is selected, and the regression function is updated. The gradient is approximated by means of a regression tree, which makes use of covariate information, and at each iteration the gradient determines the direction in which the function needs to move, in order to improve the fit to the data.

The loss function was assumed Bernoulli, due to the binary nature of the primary endpoints. A learning rate (λ) was introduced to dampen proposed moves and to protect against over-fitting. The optimal number of iterations, given by T, was determined by 5-fold cross-validation. The minimum number of observations in each terminal node was 10. Two-way interactions were allowed. Random sub-sampling, without replacement, of half of the observations was applied to achieve variance reduction in gradient estimation.

For current purposes, 50 rounds of gradient boosting were run for each scenario, and the nine variables most often showing highest estimated relative influence were taken forwards to generalized linear modeling.

The twelve models were used to generate probability predictions for each patient in the Validation-FAS. For each model, the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated for the range of predicted probability thresholds. A Receiver Operating Characteristic (ROC) curve was generated to plot sensitivity as a function of (1-specificity). The optimal classification threshold was determined on the basis of accuracy, defined as the proportion of correct predictions. In addition, the Area Under the Curve (AUC) and accuracy was estimated (Tables 27, 28, 29, 30 for the four primary endpoints, respectively).

The performance of model-based predictions were compared to the performance of probability predictions obtained by Diamond-Forrester scoring. (Diamond and Forrester (1979)).

Detailed Results for Native CAD

The results show that the Diamond-Forrester score provides poor prediction of the GLOBAL phenotypes (FIGS. 34, 38, 42, 46). The estimates of AUC and accuracy for prediction of Native CAD indicate that performance is no better than assigning all patients as ‘at risk’ of disease, by which 62% of predictions in the Validation Set for Native CAD (Validation-Native CAD plus Validation-Control) are correct, and 64% of predictions in the Validation Set for All CAD (Validation Native CAD plus Validation-Revasc. CAD plus Validation-Control), are correct.

Metabolomics Model

I. Atherosclerosis in Native CAD—A nnCAD

    • a. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 83 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 29 provides a list of the 83 metabolomic variables for [Figure (not displayed)].

TABLE 29
glutamateserylleucine1-linoleoyl-GPC (18:2)
acisoga3-methoxytyrosine1-methylguanosine
threonateprolylhydroxyproline12 13-DiHOME
uratevalerylcarnitine (C5)O-sulfo-L-tyrosine
mannosecaproate (6:0)erucamide
oleic ethanolamidetigloylglycineinositol 1-phosphate
(l1P)
cysteine-glutathione disulfideguanidinosuccinateisoleucylvaline
pyroglutamylglutamineisobutyrylglycine (C4)gamma-tocopherol
valylleucineglycocholenate sulfate*1-
eicosenoylglycerophosp
hocholine (20:1n9)*
butyrylcarnitine (C4)o-cresol sulfatetyrosylglutamine
cytidineN-acetylthreonineindolepropionate
palmitoyl ethanolamideleucylglycinegamma-glutamylvaline
phenylalanylvaline2-hydroxybutyrate (AHB)2-aminoadipate
hydroxybutyrylcarnitine*leucylaspartateaspartate
1-1-arachidoyl-GPC (20:0)N6-
nonadecanoylglycerophosphocholinecarbamoylthreonyladenosine
(19:0)
glycineN6-methyladenosinemethyl glucopyranoside
(alpha + beta)
propionylglycine (C3)hexanoylcarnitine (C6)myo-inositol
pseudouridinevalylisoleucinealpha-ketobutyrate
ADSGEGDFXAEGGGVR*beta-alanineS-adenosylhomocysteine
(SAH)
2-hydroxyhippurate (salicylurate)1-linoleoyl-GPE (18:2)*1-oleoylglycerol (18:1)
alpha-glutamyltyrosinegamma-glutamylglutamatetartronate
(hydroxymalonate)
fucose3-hydroxy-2-ethylpropionate3-
methylglutarylcarnitine-2
glucuronateadenine1-methylurate
3-methylglutarylcarnitine-1xylitolN-acetyl-beta-alanine
xanthineN2 N2-dimethylguanosinehistidyltryptophan
12-HETEmethyl indole-3-acetate1-oleoyl-GPC (18:1)*
glucosehomostachydrine*3-hydroxydecanoate
salicylatephenylacetylglutamine—
Of the 83 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of eight metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 30 provides the relative influence of the eight metabolomic variables, in combination with age and gender, for the Metabolomics Model of [Figure (not displayed)]. FIG. 35 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 53 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 30
VariableRelative InfluenceDirection of Change
valylleucine28.88Decreased
glutamate14.47Elevated
acisoga14.25Elevated
urate10.39Elevated
glucuronate9.26Elevated
age6.68Elevated
fucose6.18Elevated
Butyrylcarnitine (C4)4.72Elevated
mannose4.46Elevated
Male gender0.70Present§
§The term “present” conveys that male gender was taken into account in the prediction model, with ‘relative influence’ denoting the association of male gender with the outcome (i.e., ASCAD or the presents of a coronary atherosclerotic plaque).
    • b. Independently Significant in Fasting and Non-Fasting—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 4 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 31 provides a list of the 4 metabolomic variables for [Figure (not displayed)].

TABLE 31
acisogao.cresol.sulfateCysteine.glutathione.disulfide
threonate——

Of the 4 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)]; a panel of all four metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 32 provides the relative influence of the four metabolomic variables, in combination with age and gender, for the Metabolomics Model of [Figure (not displayed)]. FIG. 36 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 53 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 32
VariableRelative InfluenceDirection of Change
acisoga40.74Elevated
age20.77Elevated
cysteine.glutathione.disulfide18.87Decreased
threonate12.67Decreased
o-cresol.sulfate4.49Elevated
Male gender2.45Present
    • c. Significant in Fasting—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 34 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 33 provides a list of the 34 metabolomic variables for [Figure (not displayed)].

TABLE 33
xylitolvalylleucinecysteine-glutathione
disulfide
3-carboxy-4-methyl-N-acetylleucinethreonate
5-propyl-2-
furanpropanoate
(CMPF)
serylleucinealpha-glutamyltyrosinefucose
phenylalanylvaline4-androsten-3alphaadenosine
17alpha-diol
monosulfate 2
12-HETEinositol 1-phosphate (I1P)valylisoleucine
glycocholenate1-docosahexaenoyl-GPC*phenylalanylserine
sulfate*(22:6)*
oleic ethanolamide2-hydroxyhippurategamma-tocopherol
(salicylurate)
acisogasalicylatepalmitoyl
ethanolamide
leucylglycinephenylalanylglycinehydroquinone sulfate
N-glycylphenylalaninepropionylglycine (C3)
acetylphenylalanine
o-cresol sulfate3-methoxytyrosine—
histidyltryptophanadenine—

Of the 34 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of eight metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 34 provides the relative influence of the eight metabolomic variables, in combination with age and gender, for the Metabolomics Model of [Figure (not displayed)]. FIG. 37 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 53 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 34
Direction
VariableRelative Influenceof Change
N-acetylphenylalanine22.27Elevated
age18.18Elevated
valylleucine17.61Decreased
xylitol8.07Elevated
2-hydroxyhippurate6.97Elevated
(salicylurate)
N-acetylleucine6.15Elevated
serylleucine6.15Decreased
fucose6.06Elevated
glycylphenylalanine4.97Decreased
Male gender3.56Present

II. Atherosclerosis in All CAD (inc revasc)—AaCAD

    • a. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 92 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 41 provides a filtered list of the 92 metabolomic variables for [Figure (not displayed)].

TABLE 41
acisogatyrosylglutamineN6-
carbamoylthreonyladenosine
Glutamatexanthine2-linoleoyl-GPC* (18:2)*
Threonatebeta-alanine3-methyl-2-oxobutyrate
Mannoseisobutyrylglycine (C4)methyl glucopyranoside
(alpha + beta)
Urate3-methylglutarylcarnitine-1serylleucine
cysteine-glutathione disulfidevalerylcarnitine (C5)caproate (6:0)
oleic ethanolamide1-linoleoyl-GPC (18:2)N-methyl proline
pyroglutamylglutaminehexanoylcarnitine (C6)laurylcarnitine (C12)*
butyrylcarnitine (C4)2-hydroxybutyrate (AHB)o-cresol sulfate
Cytidine1-arachidoyl-GPC (20:0)gamma-
glutamylglutamate
hydroxybutyrylcarnitine*guanidinosuccinateN-acetyl-beta-alanine
alpha-glutamyltyrosinefucose1-
eicosenoylglycerophosphocholine
(20:1n9)*
2-hydroxyhippurate (salicylurate)phenylacetylglutamineN-acetylglycine
Valylleucine3-methylglutarylcarnitine-2seryltyrosine
propionylglycine (C3)glycohyocholate4-guanidinobutanoate
1-N6-methyladenosineS-methylcysteine
nonadecanoylglycerophosphocholine
(19:0)
GlycineN2 N2-dimethylguanosineisoleucylvaline
12-HETEgamma-glutamylvalineadenine
pseudouridineleucylaspartate1-methylurate
Salicylate2-hydroxyoctanoatexylitol
Glucosealpha-ketobutyratephenylalanylalanine
ADSGEGDFXAEGGGVR*glycocholenate sulfate*O-sulfo-L-tyrosine
1-linoleoyl-GPE (18:2)*valylisoleucineerucamide
Phenylalanylvalinehomostachydrine*pregnanediol-3-
glucuronide
Tigloylglycinemethyl indole-3-acetate3-hydroxy-2-
ethylpropionate
Glucuronateleucylglycinepyridoxal
palmitoyl ethanolamideN-acetylthreonine1-oleoyl-GPC (18:1)*
1-oleoylglycerol (18:1)2-hydroxydecanoate2prime-deoxyuridine
12 13-DiHOME1-methylguanosinethreonylphenylalanine
3-methoxytyrosineprolylhydroxyproline2-aminoadipate
2-linoleoyl-GPE* (18:2)*prolylglycine—

Of the 92 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of eight metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 36 provides the relative influence of the eight metabolomic variables combined with age and gender for the Metabolomics Model of [Figure (not displayed)]. FIG. 39 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 54 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 36
VariableRelative InfluenceDirection of Change
valylleucine26.79Decreased
acisoga16.87Elevated
glutamate13.93Elevated
urate9.74Elevated
glucuronate8.74Elevated
mannose7.13Elevated
age6.24Elevated
12-HETE5.03Decreased
Valerylcarnitine (C5)4.81Elevated
Male gender0.72Present
    • b. Independently Significant in Fasting and Non-Fasting—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 6 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 37 provides a list of the 6 metabolomic variables for [Figure (not displayed)].

TABLE 37
threonatethreonatecysteine-glutathione
disulfide
o-cresol1-glucose
sulfatenonadecanoylglycerophosphocholine
(19:0)

Of the 6 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of all six metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 38 provides the relative influence of the six metabolomic variables, in combination with age and gender, for the Metabolomics Model of [Figure (not displayed)]. FIG. 40 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 54 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 38
Direction
VariableRelative Influenceof Change
acisoga39.32Elevated
age17.31Elevated
1.nonadecanoylglycerophosphocholine12.00Decreased
(19:0)
cysteine-glutathione disulfide10.91Decreased
threonate10.71Decreased
glucose6.64Elevated
Male gender2.06Present
o-cresol sulfate1.05Elevated
    • c. Significant in Fasting —[Figure (not displayed)]
      • i. Of the 481 analytes measured, 48 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 39 provides a list of the 48 metabolomic variables for [Figure (not displayed)].

TABLE 39
12-HETEN-acetylphenylalanine1-arachidonylglycerol
alpha-glutamyltyrosineN-acetylleucinePyroglutamylvaline
salicylate4-androsten-3alpha 17alpha-phenylalanyltryptophan
diol monosulfate 2
2-hydroxyhippurate (salicylurate)o-cresol sulfatemethyl indole-3-acetate
acisogaphenylalanylvalineHistidyltryptophan
3-carboxy-4-methyl-5-propyl-2-leucylglycine4-ethylphenyl sulfate
furanpropanoate (CMPF)
threonatephenylalanylglycine1-myristoylglycerol
(14:0)
glycocholenate sulfate*propionylglycine (C3)inositol 1-phosphate
(I1P)
xylitolmannitol1-
nonadecanoylglycerophosphocholine
(19:0)
1-docosahexaenoyl-GPC* (22:6)*serylleucineGlucose
phenylalanylserinehydroquinone sulfateN-stearoyltaurine
3-methoxytyrosineadenosineValylisoleucine
oleic ethanolamide2-hydroxydecanoatebeta-alanine
cysteine-glutathione disulfidetyrosylglutamineN-acetylglycine
glycylphenylalanineN-octanoylglycineAllantoin
valylleucineadeninePhenylalanylphenylalanine

Of the 48 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of seven metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 40 provides the relative influence of the seven metabolomic variables, in combination with age and gender, for the Metabolomics Model of [Figure (not displayed)]. FIG. 41 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 54 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 40
Direction
VariableRelative Influenceof Change
age21.21Elevated
valylleucine20.76Decreased
N-acetylphenylalanine18.59Elevated
2-hydroxyhippurate12.20Elevated
(salicylurate)
N-acetylleucine6.10Elevated
12-HETE5.96Decreased
xylitol5.40Elevated
glycylphenylalanine5.39Decreased
Male gender4.40Present

III. 50% stenosis in Native CAD—SnCAD

    • a. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 49 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 41 provides a list of the 49 metabolomic variables for [Figure (not displayed)].

TABLE 41
threonateserotonin (5HT)5alpha-androstan-3alpha
17beta-diol disulfate
N-acetylglycinexanthine1-stearoyl-GPC (18:0)
glycerate2-oleoyl-GPE* (18:1)*serine
isobutyrylglycine (C4)4-guanidinobutanoateacisoga
valerylcarnitine (C5)leucylleucinemannose
fumaratecholatevalylleucine
1-propionylglycine (C3)gamma-tocopherol
nonadecanoylglycerophosphocholine
(19:0)
tartronate (hydroxymalonate)glycocholate3-ethylphenylsulfate
2-hydroxyhippurateN-octanoylglycineglutamate
(salicylurate)
1-arachidoyl-GPC (20:0)glycoursodeoxycholatesphingosine 1-phosphate
threitolisovalerylglycinecarnitine
N-(2-furoyl)glycinepregnanediol-3-glucuronidearabonate
tigloylglycine5alpha-androstan-3betacyclo(leu-pro)
17beta-diol monosulfate 2
salicylatearabinoseindoleacetylglutamine
N-acetylthreonine1-linoleoyl-GPE (18:2)*prolylglycine
xylonate5-HETE—
xylosehydroquinone sulfate—

Of the 49 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of eight metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 42 provides the relative influence of the eight metabolomic variables, in combination with age and gender, for the Metabolomics Model of [Figure (not displayed)]. FIG. 43 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 55 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 42
VariableRelative InfluenceDirection of Change
Age37.04Elevated
valerylcarnitine (C5)14.32Elevated
N-acetylthreonine10.39Elevated
tigloylglycine8.72Decreased
2-hydroxyhippurate7.06Elevated
(salicylurate)
glycerate6.42Decreased
salicylate5.67Decreased
threonate5.58Decreased
tartronate (hydroxymalonated);4.25Elevated
Male gender0.55Present
    • b. Independently Significant in Fasting and Non-Fasting —[Figure (not displayed)]
      • i. Of the 481 analytes measured, 2 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 43 provides a list of the 2 metabolomic variables for [Figure (not displayed)].

TABLE 43
N-acetylglycine3-ethylphenylsulfate—

Of the 2 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of both variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 44 provides the relative influence of the two metabolomic variables in combination with age and gender for the Metabolomics Model of [Figure (not displayed)]. FIG. 44 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 55 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 44
VariableRelative InfluenceDirection of Change
age67.33Elevated
N-acetylglycine14.67Decreased
3-ethylphenylsulfate12.88Elevated
Male gender5.12Elevated
    • c. Significant in Fasting—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 28 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 45 provides a filtered list of the 28 metabolomic variables for [Figure (not displayed)].

TABLE 45
leucylleucinevalylisoleucine7-methylguanine
asparagineglycocholenate sulfate*cyclo(leu-pro)
glyceratearabitolMethionine
threitolN-acetylglycinepropionylglycine (C3)
cholateserotonin (5HT)Serine
N-octanoylglycinexylose2-oleoyl-GPE* (18:1)*
xylonateN-acetylputrescineTigloylglycine
isobutyrylglycine (C4)arabonate3-ethylphenylsulfate
isovalerylglycinelysine—
fumarateN-(2-furoyl)glycine—

Of the 28 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of eight metabolomic variables were selected as best predictors; they were combined with age and gender in a prediction model for CAD. Table 46 provides the relative influence of the eight metabolomic variables, in combination with age and gender, for the Metabolomics Model of [Figure (not displayed)]. FIG. 45 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 55 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 46
VariableRelative InfluenceDirection of Change
age24.44Elevated
leucylleucine21.83Decreased
serotonin (5HT)11.37Elevated
N-acetylputrescine9.68Decreased
glycocholenate sulfate8.56Decreased
propionylglycine (C3)6.95Decreased
cholate6.14Decreased
asparagine5.73Elevated
3-ethylphenylsulfate4.79Elevated
Male gender0.50Present

IV. 50% stenosis in ALL CAD (inc revasc)—SaCAD

    • a. Significant in Fasting & Non-Fasting Combined—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 72 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 47 provides a list of the 72 metabolomic variables for [Figure (not displayed)].

TABLE 47
threonateprolylglycine2-hydroxydecanoate
1-linoleoyl-GPE (18:2)*N-octanoylglycineglutamate
N-acetylglycinethreitolN-acetylthreonine
glycoursodeoxycholatefumaratetaurine
2-hydroxyhippurate (salicylurate)pregnanediol-3-glucuronide1-
oleoylplasmenylethanol
amine*
salicylate1-oleoyl-GPI (18:1)*1-palmitoyl-GPE (16:0)
2-linoleoyl-GPE* (18:2)*serotonin (5HT)N-acetylglutamate
mannosexylonate13-HODE + 9-HODE
tigloylglycinecyclo(leu-pro)1-palmitoyl-GPI* (16:0)*
2-glyceratehydroquinone sulfate
linolenoylglycerophosphocholine (18:3n3)*
1-tartronate (hydroxymalonate)caprylate (8:0)
nonadecanoylglycerophosphocholine
(19:0)
2-oleoyl-GPE* (18:1)*xylose1-stearoyl-GPC (18:0)
isovalerylglycineglycohyocholateglycochenodeoxycholate
isobutyrylglycine (C4)glucosep-cresol sulfate
N-(2-furoyl)glycinexanthine12-HETE
glycocholatecyclo(L-phe-L-pro)5-hydroxyindoleacetate
acisogabeta-alaninearabonate
4-guanidinobutanoatepyridoxate2-hydroxyoctanoate
1-arachidoyl-GPC (20:0)tartarateurate
propionylglycine (C3)1-linoleoyl-GPC (18:2)valylleucine
valerylcarnitine (C5)pyridoxalcarnitine
1-oleoylglycerol (18:1)cholate1-linoleoyl-GPI* (18:2)*
1-oleoyl-GPE (18:1)serineN-acetylputrescine
arabinosehomostachydrine*succinate

Of the 72 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of eight metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 48 provides the relative influence of the eight metabolomic variables in combination with age and gender for the Metabolomics Model of [Figure (not displayed)]. FIG. 47 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 56 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 48
Direction
VariableRelative Influenceof Change
Age18.38Elevated
glycoursodeoxycholate16.52Decreased
acisoga12.81Elevated
2-hydroxyhippurate10.33Elevated
(salicylurate)
1-linoleoyl.GPE (18:2)10.26Decreased
valerylcarnitine (C5),8.91Elevated
threonate7.13Decreased
mannose7.12Elevated
salicylate7.02Elevated
Male gender1.52Present
    • b. Independently Significant in Fasting and Non-Fasting—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 5 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 49 provides a filtered list of the 5 metabolomic variables for [Figure (not displayed)].

TABLE 49
N-acetylglycinethreonateSalicylate
2-hydroxyhippurate (salicylurate)3-ethylphenylsulfate—

Of the 5 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of all five metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 50 provides the relative influence of the five metabolomic variables in combination with age and gender for the Metabolomics Model of [Figure (not displayed)]. FIG. 48 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 56 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 50
VariableRelative InfluenceDirection of Change
Age40.82Elevated
2-hydroxyhippurate19.56Elevated
(salicylurate)
threonate14.84Decreased
salicylate12.23Elevated
Male gender7.00Present
N-acetylglycine3.13Decreased
3-ethylphenylsulfate2.42Elevated
    • c. Analytes Significant in Fasting—[Figure (not displayed)]
      • i. Of the 481 analytes measured, 40 metabolomic variables exhibited a nominal univariate association (raw p<0.05) for [Figure (not displayed)]. Table 51 provides a filtered list of the 40 metabolomic variables for [Figure (not displayed)].

TABLE 51
N-octanoylglycinesalicylateDimethylglycine
1-oleoylglycerol (18:1)7-methylguaninexylonite
isovalerylglycinelysinePhenylalanylphenylalanine
N-acetylglycineglycoursodeoxycholateValylisoleucine
2-3-indoxyl sulfateGlycerate
linolenoylglycerophosphocholine(18:3n3)*
asparagine6-oxopiperidine-2-carboxylic1-arachidonylglycerol
acid
isobutyrylglycine (C4)1-Fumarate
arachidonoylglyercophosphate
cyclo (leu-pro)2-hydroxyhippurate3-ethylphenylsulfate
(salicylurate)
cholatethreitol7-HOCA
serotonin (5HT)methionineTaurine
threonateacisogaCholesterol
N-acetylputrescinetigloylglycineArabitol
propionylglycine (C3)1-linoleoylglycerol (18:2)—
2-oleoyl-GPE* (18:1)*1-oleoyl-GPI (18:1)*—

Of the 40 metabolomic variables exhibiting a nominal univariate association for [Figure (not displayed)], a panel of eight metabolomic variables were selected as best predictors; these were combined with age and gender in a prediction model for CAD. Table 52 provides the relative influence of the eight metabolomic variables in combination with age and gender for the Metabolomics Model of [Figure (not displayed)]. FIG. 49 provides a ROC curve for the Metabolomics Model of [Figure (not displayed)]. Table 56 provides the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the range of predicted probability thresholds; Area Under the Curve (AUC) and accuracy was estimated.

TABLE 52
RelativeDirection
VariableInfluenceof Change
age15.37Elevated
cholesterol15.19Decreased
1-oleoylglycerol (18:1)15.12Elevated
acisoga14.01Elevated
2.hydroxyhippurate (salicylurate)9.47Elevated
asparagine8.18Elevated
taurine7.93Decreased
6-oxopiperidine-2-carboxylic acid7.50Elevated
propionylglycine (C3)6.66Decreased
Male gender0.56Present

For each model below, the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated for the range of predicted probability thresholds (Tables 53, 54, 55, 56). A Receiver Operating Characteristic (ROC) curve was generated to plot sensitivity as a function of (1-specificity). The optimal classification threshold was determined on the basis of accuracy, defined as the proportion of correct predictions. In addition, the Area Under the Curve (AUC) and accuracy was estimated (Tables 53, 54, 55, 56 for Native CAD, All CAD, 50% stenosis in Native CAD, and 50% stenosis in All CAD, respectively). The first row for each model indicates the performance of the maximum accuracy threshold, the optimal balance between sensitivity and specificity. Those models with a second row were optimized for a high negative predictive value (NPV).

TABLE 53
PositiveNegative
Sensi-Speci-PredictivePredictive
ModelAUCtivityficityValueValueAccuracy
DF0.451.000.000.62N/A0.62
AFNFnCAD0.820.850.610.780.710.76
0.990.090.640.920.65
AIFNFnCAD0.800.850.640.800.720.77
1.000.100.650.930.66
AFnCAD0.810.870.580.770.740.76
0.990.260.690.940.72
DF = Diamond-Forrester

TABLE 54
PositiveNegative
Sensi-Speci-PredictivePredictive
ModelAUCtivityficityValueValueAccuracy
DF0.451.000.000.64N/A0.64
AFNFaCAD0.830.930.500.770.810.78
1.000.160.680.950.69
AIFNFaCAD0.810.850.660.810.710.78
1.000.070.650.900.66
AFaCAD0.820.830.640.800.680.76
1.000.160.680.950.69
DF = Diamond-Forrester

TABLE 55
PositiveNegative
Sensi-Speci-PredictivePredictive
ModelAUCtivityficityValueValueAccuracy
DF0.450.031.001.000.780.78
SFNFnCAD0.730.210.960.640.800.79
0.950.300.290.950.45
SIFNFnCAD0.760.300.960.720.820.81
0.930.370.310.950.50
SFnCAD0.670.041.001.000.780.78
0.960.200.260.950.38
DF = Diamond-Forrester

TABLE 56
PositiveNegative
Sensi-Speci-PredictivePredictive
ModelAUCtivityficityValueValueAccuracy
DF0.450.031.001.000.740.75
SFNFaCAD0.780.340.940.660.800.78
0.960.300.330.950.47
SIFNFaCAD0.780.230.970.720.780.77
0.960.300.330.950.47
SFaCAD0.740.220.960.690.780.77
0.970.220.310.950.42
DF = Diamond-Forrester

Free full text: Click here