A New Risk Chart for Acute Myocardial Infarction by a Innovative

Algoritm

Federico Licastro

, Manuela Ianni

, Roberto Ferrari

, Gianluca Campo

, Massimo Buscema

Enzo Grossi

and Elisa Porcellini

Department of Experimental, Diagnostic and Specialty Medicine, University of Bologna,

Via S. Giacomo 14, 40126 Bologna, Italy

Azienda Ospedaliera Sant'Orsola Malpighi, Seragnoli Institute, University of Bologna,

Via Massarenti, 9, 40138, Bologna, Italy

Department of Cardiology, University of Ferrara, Corso Giovecca 203, 44100, Ferrara, Italy

Semeion Research Center of Communication Science, Via Sersale 117, 00128 Roma, Italy

Bracco Foudation, via Folli 50, 20134 Milan, Italy

Keywords: Acute Myocardial Infarction (AMI), Artificial Neural Network (ANN), Twist Algorithm, Risk Chart.

Abstract: Acute myocardial infarction (AMI) is complex disease; its pathogenesis is not completely understood and

several variables are involved in the disease.. The aim of this paper was to assess: 1) the predictive capacity

of Artificial Neural Networks (ANNs) in consistently distinguishing the two different conditions (AMI or

control). 2) the identification of those variables with the maximal relevance for AMI. Genetic variances in

inflammatory genes and clinical and classical risk factors in 149 AMI patients and 72 controls were

investigated. From the data base of this case/control study 36 variables were selected. TWIST system, an

evolutionary algorithm able to remove redundant and noisy information from complex data sets, selected 18

variables. Fitness, sensitivity, specificity, overall accuracy of the association of these variables with AMI

risk were investigated. Our findings showed that ANNs are useful in distinguishing risk factors selectively

associated with the disease. Finally, the new variable cluster, including classical and genetic risk factors,

generated a new risk chart able to discriminate AMI from controls with an accuracy of 90%. This approach

may be used to assess individual AMI risk in unaffected subjects with increased risk of the disease such as

first relative with positive parental history of AMI.

1 INTRODUCTION

Morbidity and mortality of coronary heart disease

(CHD) are high and acute myocardial infarction

(AMI) is the major clinical complication of CHD

(Yusuf S

, 2001, Levi F,

2002).

AMI is a multi-factorial disease with a complex

and incompletely defined pathogenesis. In fact,

genetic, clinical and phenotypic factors are involved

in the clinical history of the disease.

Knowledge regarding risk factors predisposing to

AMI is still incomplete. It is known that many CHD

events occur in individuals with at most one risk

factor among those included in the Framingham risk

assessment.

Different investigations confirmed that

biomarkers of inflammation, such as increased blood

homocysteine (Cummings DM, 2006)

C-reactive

protein (Ridker PM, 2000) and cytokine levels

(Packard RR, 2008; Zhang C, 2008; Andersson J,

2010; Pamukcu B, 2010 and Biasillo G, 2010) could

be considered new risk factors for CHD and AMI.

Genetic variations, represented by single

nucleotide polymorphisms (SNPs) in the promoter

region of several genes regulating metabolic and

immune functions from case/control studies were

found to be associated with an increased risk of AMI

(Licastro F, 2010 and Ianni M, 2012).

Moreover, recent genome-wide association

(GWA) studies have contributed to the discovery of

new SNPs associated with CHD and AMI (Patel RS,

2011).

However, a single gene variant has a limited

contribution to the total genetic load of AMI.. This

252

Licastro F., Ianni M., Ferrari R., Campo G., Buscema M., Grossi E. and Porcellini E..

A New Risk Chart for Acute Myocardial Infarction by a Innovative Algoritm.

DOI: 10.5220/0005183102520259

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2015), pages 252-259

ISBN: 978-989-758-068-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

situation may partially explain the contradictory

results of genetic association studies using the

candidate gene SNP approach in AMI (Kullo IJ,

2007; Hamsten A, 2008; Chiappelli M, 2005).

Therefore it is important to introduce new

statistical methods to approach relationship of risk

factors with AMI.

The Auto Contractive Map algorithm (AutoCM)

has already been used to explore the concomitant

presence and the associations among several genetic

and phenotypic variables with AMI in a multi factor

network (Licastro F, 2010 ). Individual phenotype

biomarkers can vary widely as a function of time

and may be different among subjects as result of

gender, concomitant diseases, metabolic disorders,

dietary intake and other environmental variables;

therefore, to understand natural processes and

recreate those processes may be useful to employ

machine learning systems able to manage highly non

linear interactions (Penco S, 2005; Lisboa PJC,

2002; Grossi E, 2007).

Artificial neural networks (ANNs) function are

able to reconstruct the imprecise rules which may be

underlying a complex set of data (testing) (Coppedè

F, 2013). In recent years ANNs have been used

successfully in medicine, for example they have

been used to investigate the predictive values of risk

factors related to the conversion from amnestic mild

cognitive impairment to AD (Tabaton M, 2010), to

differentiate fronto-temporal dementia from AD

(Franceschi M, 2011) to identify genetic variants

essential to differentiate sporadic amyotrophic

lateral sclerosis cases from controls (Penco S, 2008;

Buscema M, 2012), or maternal risk for Down

syndrome child (Coppedè F, 2010).

In this pilot study we applied ANNs to

investigate genetic, clinic and phenotypic markers in

149 patients with AMI and 72 healthy subjects

selected from a previously described investigation

(Licastro F, 2010). The present study was performed

by using supervised ANNs to develop a predictive

model able to distinguish AMI patients from healthy

controls on the basis of classical risk factors and

gene variants involved in the disease.

Moreover, TWIST algorithm (Buscema M,

2004) was applied to the genetic, clinical and

phenotypic dataset in order to select relevant

variables associated with AMI risk. Sensitivity and

specificity of this selected group of variables in

relation with AMI risk was then tested. No previous

study has investigated the relationship among clinic,

phenotypic and genetic polymorphisms by ANNs

and TWIST in AMI.

The aim of this study was to investigate whether

these innovative mathematical approaches might

identify key new variables to better assess AMI risk

and describe a new risk chart for the disease.

2 MATERIALS AND METHODS

2.1 Subjects

149 consecutive patients with clinical diagnosis of

AMI (mean age = 71 ± 12; 70% male and 29%

female) from the Cardiology Unit of Ferrara

University Hospital were enrolled. Each patient met

diagnostic criteria for AMI based on

electrocardiography changes and standard laboratory

findings confirmed by echocardiography and

coronary angiography.

Controls consisted of 72 healthy subjects (mean

age = 75 ± 5; 49% male and 51% female) belonging

to a longitudinal population study, called “Conselice

study of brain aging” (Ravaglia G, 2001)

All

controls did not show cardiovascular or

inflammatory diseases at the beginning of the follow

up (1999-2000) and were still free of these

pathological conditions at the end of the follow up

(2004-2005).

The research protocol was approved by relevant

institutional review boards, all participants gave

written and informed consent and the investigation is

consistent with the principles outlined in the

Declaration of Helsinki.

2.2 SNP Detection

Genomic DNA from peripheral blood leukocytes of

AMI and healthy controls was obtained by a method

described elsewhere (Grimaldi LM, 2000). Genetic

determination of polymorphism in promoter regions

of IL-1β-511 C/T (rs16944), IL-6 -174 C/G

(rs=1800795), IL-10 -1082 G/A (rs1800896), TNF-

α -308 G/A (rs1800629), ACT -51 G/T (rs1884082),

VEGF -2578 C/A (rs699947) and HMGCR -911

C/A (rs3761740) genes was performed by Real

Time-PCR method. SNP-specific primers and

probes were designed according to the TaqMan

genotyping assay by ABI (Foster City, CA, USA)

and the assays were performed in 25 µl total volume

on Stratagene MX3000P following manufacturer’s

instructions (Licastro F, 2010; Chiappelli M, 2005).

IFNγ (rs2430561) and IL-6 (rs1800795) genotypes

were assayed by Real Time-PCR using the following

allele specific modified LNA primers (Latorra D,

2003).

ANewRiskChartforAcuteMyocardialInfarctionbyaInnovativeAlgoritm

253

IFN(+874)(rs2430561):

Primers F:

5’TTTATTCTTACAACACAAAATCAAATC+T-3’,

5’-TTTATTCTTACAACACAAAATCAAATC+A-3’,

Primer R

5’-TGTGCCTTCCTGTAGGGTATTATTA-3’

Il-6(-174)(rs1800795):

Primer F

5’-TCCCCCTAGTTGTGTCTTGC+C-3’,

5’-TCCCCCTAGTTGTGTCTTGC+G-3’,

Primer R

5’-AATCCCACATTTGATATAAATCTTTGT-3’

RT-PCR was performed in 96 well plates using

Stratagene MX3000P platform. Reaction volume

included a SYBR Green PCR Master Mix with the

enzyme, Mg

and dNTPs (ABI, Fosteer City, CA,

USA) PCR primers and genomic DNA (0.5ng/µl)

was of 25µl . A start of 10 min at 95°C was followed

by 40 cycles at 95°C for 15 s and 60°C for 60 s.

2.3 Plasma Lipid Profile Detection

Plasma levels of total cholesterol, triglycerides and

HDL were measured by commercial clinical

laboratory assay.

2.4 Artificial Neural Networks (ANNs)

ANNs are adaptive models for the analysis of data;

these algorithms are inspired to the functioning

processes of the human brain (Grossi E, 2007) and

they are able to modify their internal structure in

relation to a function objective. The adaptive feature

is fundamental in case of complex data set in which

non linearity prevails.

The ANNs are particularly suited for solving

problems of the non linear type, being able to

reconstruct the approximate rules put into a certain

set of data. In this study, we applied supervised

ANNs to data network in which the result of the

processing (the output desired) is already defined.

Supervised ANNs calculate an error function that

measures the distance between the desired fixed

output (target) and their own output, and adjust the

connection strengths during the training process to

minimize the result of the error function. The

learning constraint of the supervised ANNs tests its

own output to overlap that of a determined target.

The general form of these ANNs is: y = f(x,w∗),

where we constitutes the set of parameters which

best approximate the function. The ANNs used in

the study are characterized by the law of learning

and topology. The laws of learning identify

equations which translate the ANNs inputs into

outputs, and rules by which the weights are modified

to minimize the error or the internal energy of the

ANN. In this study, we have used as a standard

model the Back Propagation standard (BP-FF)

(Rumelhart DE, 1982), belonging to a very large

family of ANNs defined by different interconnected

layers of nodes. These are characterized by a non

linear function, which can be differentiated and

limited, and has a linear combination of the

activations coming from the previous layer in the

input. The function is generally a sigmoid type.

The fundamental equation that characterizes the

activation of a single node and therefore, the transfer

of the signal from one layer to another is showed in

Fig 1.

Results obtained with these ANNs have also

been compared with a model of linear statistic such

as, the Linear Discriminant Analysis (LDA;

Software SPSS) (Tabaton M, 2010).

Fig 1: Single node activated value equation xj = node

activation value; f = sigmoidal function; W= vector of

weights arriving at j-n node; s = weights and nodes layer;

xi = Input nodes at J-n node.

2.5 Twist Algorithm

TWIST (Training With Input Selection and Testing)

is a new evolutionary algorithm (Buscema M, 2004)

able to generate two subsets of data with a very

similar probability density of distribution and with

the minimal number of effective variables for pattern

recognition.

Consequently, in the TWIST algorithm every

individual of the genetic population will be defined

by two vectors of different lengths:

1) the first one, showing which records (N) have

to be stored into the subset A and which ones have

to be stored into the subset B;

2) the second one, showing which inputs (M)

have to be used into the two subsets and which one

have to be deleted.

TWIST has already been applied to medical data

base with promising results (Penco S, 2005; Tabaton

M, 2010; Coppedè F, 2010; Buscema M, 2005;

Lahner E, 2008; Buri L, 2010; Street ME, 2008;

Buscema M, 2010; Rotondano G, 2011; Pace F,

2010). TWIST consists of a population of

Multilayer Perceptrons. The “reverse strategy” used

HEALTHINF2015-InternationalConferenceonHealthInformatics

254

in this algorithm tends to generate two subsets with

the same probability density function, and this is

exactly the gold standard of every random

distribution criterion (Buscema M, 2005).

In addition, when the “reverse strategy” is

applied, two fitness indicators are generated: the

accuracy on the subset B after the training on the

subset A, and the accuracy on the subset A after the

training on the subset B. But only the lower

accuracy of the two is saved as the best fitness of

each individual of the genetic population rather than

an average of the two or the higher of the two. This

criterion increases the statistical probability that the

two sub-samples are equally balanced during the

genetic evolution because of the quasi logarithmic

increase of the optimization process. We have also

demonstrated experimentally (Buscema M, 2013)

that when there is no information in a dataset, the

behaviors of the TWIST algorithm, the Training and

Testing Random Splitting and the K-Fold Cross

Validation are absolutely equivalent. Therefore,

TWIST does not code noise to reach optimistic

results (Buscema M, 2013).

Each ANN has to learn a subset of the global

dataset and has to be tested with another subset of

the dataset in a blind way. By this application the

fitness function of TWIST is re-programmed: the

population of Multilayer Perceptrons is exchanged

with a population of simple K Nearest Neighbour

(KNN), based on Euclidean metric. Basic kNN

algorithm is able to find the Euclidean similarity

between two samples that is training set and testing

set. So kNN is a suitable cost function for Twist

optimization. ANNs for classification are applied in

a second step, when the two subsamples (Training

and Test set) are already defined.

This change makes TWIST faster and more

oriented to discover explicit similarities between

input variables and classes (AMI and controls).

2.6 Training and Testing Protocol

Training and testing validation protocols consisted

of the following steps:

1) Division of the data set in to two sub-samples:

subset A and subset B. In the first run subset A was

used as the Training Set and the subset B as the

Testing Set.

2) Application of ANN trained on the Training Set.

In this phase the ANN learns to associate the input

variables with those that are indicated as targets.

3) At the end of training phase all the values

adaptively created by the artificial neural networks

and predefined as setting parameters are freezed and

kept apart for the testing phase

4) The Testing Set, which has not be seen before,

was then shown to the classificator that expressed an

evaluation based on the previously training; this

operation was performed for each input vector; each

result (output vector) was not communicated to the

classificator.

5) The ANN was evaluated only in reference to the

generalization ability acquired during the Training

phase.

6) In a second run an identical ANN is applied to

subset B which was used as training subset and then

to subset A which used as a testing subset.

2.7 The Receiver-Operating

Characteristic Curves (ROCs) and

Areas under Curves (AUC)

Sensitivity and specificity along with positive and

negative predictive values with 95 % confidence

intervals for each strategy were estimated by AUC.

ROCs were calculated and compared for clinical and

statistical rules with a nonparametric approach using

a paired design (Delong ER, 1988). Odds ratio was

used for computing the association of variables with

the selected outcome. Chi-squared test was used to

assess the statistical significance of differences

among proportions. All p values involve hypothesis

tests against a two-sided alternative. Differences

were considered significant at a 0.05 probability

level (Buri L, 2010).

3 RESULTS

The original data set included the following

variables: Male, Female, Age<50 years, Age=50

years, Age>50 years, High BMI, Diabetes, High

cholesterol, Low HDL, High triglycerides, AMI,

Controls, CC genotype IL-1ß, CT genotype IL-1ß,

TT genotype IL-1ß, GG genotype ACT, GT genotype

ACT, TT genotype ACT, GG genotype IL-6, GC

genotype IL-6, CC genotype IL-6, CC genotype

HMGCR, CA genotype HMGCR, AA genotype

HMGCR, GG genotype IL-10, GA genotype IL-10,

AA genotype IL-10, CC genotype VEGF, CA

genotype VEGF, AA genotype VEGF, TT genotype

IFN-γ, TA genotype IFN-γ, AA genotype IFN-γ, GG

genotype TNF-α, GA genotype TNF-α, AA genotype

TNF-α.

The application of the TWIST algorithm allowed

the selection of 18 variables (italic variables) from a

previously established larger data set with 36

ANewRiskChartforAcuteMyocardialInfarctionbyaInnovativeAlgoritm

255

Figure 2: Linear correlation index in AMI.

variables (all variables).

The linear correlation indexes of 36 variables in

AMI patients has been reported in figure 2 where

variables appears more associated (up the line) or

less associated (down the line) with the AMI .

Tables 1 showed the results regarding sensitivity,

specificity and overall accuracy according AUC

analysis regarding association with AMI by using

the variables data set after the TWIST selection. By

10 different applications, the mean sensitivity

resulted 89.9%, mean specificity 85.4%, mean

overall accuracy 87.6.

The results obtained with independent

application of Back Propagation feed forward

artificial neural network using 8 hidden nodes in

sequence a-b and b-a were stable and consistently

reached an average overall accuracy near to 90%.

Consequently, the tested algorithms were able to

find a good correlation between some variables and

diagnosis, after the removal of noisy attributes.

Data from this investigation suggest a new risk

chart to be applied for the prevention of AMI in

unaffected subjects.

Table 1.

Table 1 Results obtained with the application of

10 independent supervised ANNs, five with the

sequence a-b. and five with the sequence b-a (see

text for explanation) in discriminating AMI from

control status by using the 18 variables after

selection operated by the TWIST algorithm.

4 DISCUSSION

A recent investigation by using a new CHD risk

assessment model described that age, gender,

diabetes and family history of AMI in combination

with seven blood biomarkers yielded a 43% clinical

net reclassification of patients previously considered

of intermediate risk level by Framingham’s criteria

(Cross DS, 2012).

In a German population study, among 10,981

men followed up for 11 years, 378 subjects

developed AMI; current smoking, excess body

weight and physical inactivity were associated with

the disease (Li K, 2014).

A recent investigation

confirmed that smoking (66%), hypertension (50%)

and diabetes (43%) were the principal risk factors

for AMI. However, the study conclusion was that

none of these factors reached an association so solid

to be used for AMI prediction in unaffected subjects

(Juárez-Herrera Ú, 2013).

This notion was reinforced by results from an

independent investigation on 605 consecutive

patients hospitalized for a first AMI showing that the

preventive potential of a classical risk factor

based health check was limited (Mortensen MB,

2013).

Moreover above quoted studies used

conventional statistical models and therefore

required a large number of patients and controls.

ANN Sens. Spec Acc.

1 *1 (ab) 83.51 90.48 86.99

2 *2 (ab) 86.6 85.71 86.16

3 *3 (ab) 84.54 83.33 83.93

4 *4 (ab) 85.57 88.1 86.83

5 *5 (ab) 96.15 83.33 89.74

6 *1 (ba) 92.31 83.33 87.82

7 *2 (ba) 94.23 83.33 88.78

8 *3 (ba) 94.23 86.67 90.45

9 *4 (ba) 94.23 86.67 90.45

10 *5 (ba) 87.63 83.33 85.48

Mean 89.9 85.43 87.66

HEALTHINF2015-InternationalConferenceonHealthInformatics

256

In the pilot study described here, ANNs and

other new potent mathematical algorithms have been

used and the need of large number of cases was not a

limitation. By applying these new statistical

algorithms we searched for a cluster of variables

able to discriminate AMI cases form healthy

controls.

TWIST algorithm operated a selection of 18

factors. and reached an average overall accuracy

near to 90%. On the other hand, results obtained

with the same validation protocol using all 36

variables were clearly inferior with an average

overall accuracy of 74.99% (data not shown).

Present results partially confirm previously

published findings showing that SNP in pro-

inflammatory genes were associated with increased

AMI risk in a different patients set by using a

diverse non conventional statistical analysis

(Licastro F, 2011) .

In conclusion the tested algorithms were able to

find a set of variables highly associated with AMI

diagnosis in men. This cluster is comprehensive of

new SNPs which can be easily assessed in

laboratory practice. Thereafter, the new cluster of

variables might be used to better describe AMI risk

in unaffected subjects with positive familiarity of the

disease.

These findings need to be confirmed in larger

case/control or longitudinal population studies.

REFERENCES

Andersson J, Libby P and Hansson GK. 2010 ‘Adaptive

immunity and atherosclerosis’ Clin Immunol; vol. 134,

no. 1, pp. 33–46.

Biasillo G, Leo M, Della Bona R, Biasucci LM. 2010

‘Inflammatory biomarkers and coronary heart disease:

from bench to bedside and back’ Intern Emerg Med;

vol. 5, no. 3, pp. 225–233.

Buri L, Hassan C, Bersani G, Anti M, Bianco MA,

Cipolletta L, Di Giulio E, Di Matteo G, Familiari L,

Ficano L, Loriga P, Morini S, Pietropaolo V, Zambelli

A, Grossi E, Intraligi M, Buscema M, SIED

Appropriateness Working Group. 2010

‘Appropriateness guidelines and predictive rules to

select patients for upper endoscopy: a nationwide

multicenter study’ Am J Gastroenterol; vol. 105, no.

6, pp. 1327-1337.

Buscema M, Breda M, Lodwick W. 2013’Training With

Input Selection and Testing (TWIST) algorithm: a

significant advance in pattern recognition performance

of machine learning’ Journal of Intelligent Learning

Systems and Applications, vol. 5, pp. 29-38.

Buscema M, Grossi E, Capriotti M, Babiloni C, Rossini P.

2010 ‘The I.F.A.S.T. Model Allows the Prediction of

Conversion to Alzheimer Disease in Patients with

Mild Cognitive Impairment with High Degree of

Accuracy’ Curr Alzheimer Res; vol. 7, no. 2, pp. 173-

187.

Buscema M. 2004 ‘Genetic Doping Algorithm (GenD):

Theory and Application’ Expert Syst; vol. 21, no. 2,

pp. 63-79.

Buscema M, Grossi E, Intraligi M, Garbagna N, Andriulli

A, Breda M. 2005 ‘An Optimized Experimental

Protocol Based on Neuro-Evolutionary Algorithms.

Application to the Classification of Dyspeptic Patients

and to the Prediction of the Effectiveness of Their

Treatment’ Artif Intell Med; vol. 34, no. 33, pp. 279-

305.

Buscema M, Penco S and Grossi E. 2012 ‘A Novel

Mathematical Approach to Define the Genes/SNPs

Conferring Risk or Protection in Sporadic

Amyotrophic Lateral Sclerosis Based on Auto

Contractive Map Neural Networks and Graph Theory’

Neurol Res Int. 478560.

Chiappelli M, Tampieri C, Tumini E, et al. Porcellini E,

Caldarera CM, Nanni S, Branzi A, Lio D, Caruso M,

Hoffmann E, Caruso C, Licastro F. 2005 ‘Interleukin-

6 gene polymorphism is an age-dependent risk factor

for myocardial infarction in men’ Int J Immunogenet;

vol. 32, no. 6, pp. 349–353.

Coppedè F, Grossi E, Migheli F, Migliore L. 2010

‘Polymorphisms in Folate-Metabolizing Genes,

Chromosome Damage, and Risk of Down Syndrome

in Italian Women: Identification of Key Factors Using

Artificial Neural Networks’ BMC Med Genomics; vol.

3, no. 42.

Coppedè F, Grossi E, Buscema M, Migliore L. 2013

‘Application of artificial neural networks to investigate

one-carbon metabolism in Alzheimer's disease and

healthy matched individuals’ PLoS One; vol. 8, no. 8,

pp. e74012.

Cross DS, McCarty CA, Hytopoulos E, Beggs M, Nolan

N, Harrington DS, Hastie T, Tibshirani R, Tracy RP,

Psaty BM, McClelland R, Tsao PS, Quertermous T.

2012 ‘Coronary risk assessment among intermediate

risk patients using a clinical and biomarker based

algorithm developed and validated in two population

cohorts’ Curr Med Res Opin; vol. 28, no. 11, pp.

1819-1830.

Cummings DM, King DE, Mainous AG, Geesey ME.

2006 ‘Combining serum biomarkers: the association

of C-reactive protein, insulin sensitivity, and

homocysteine with cardiovascular disease history in

the general US population’ Eur J Cardiovascular Prev

Rehabil; vol. 13, no. 2, pp. 180–185.

Delong ER , Delong DM and Clarke-Pearson DL. 1988

‘Comparing the areas under two or more receiver

operating characteristic curves: a nonparametric

approach’ Biometrics; vol. 44, no. 3, pp. 837-45.

Franceschi M, Caffarra P, Savarè R, Cerutti R, Grossi E;

Tol Research Group. 2011 ‘Tower of London test: a

comparison between conventional statistic approach

and modelling based on artificial neural network in

differentiating fronto-temporal-dementia from

ANewRiskChartforAcuteMyocardialInfarctionbyaInnovativeAlgoritm

257

Alzheimer’s disease’ Behav Neurol; vol. 24, no. 2, pp.

149-158.

Grimaldi LM, Casadei VM, Ferri C, Veglia F, Licastro F,

Annoni G, Biunno I, De Bellis G, Sorbi S, Mariani C,

Canal N, Griffin WS, Franceschi M. 2000

‘Association of early-onset Alzheimer's disease with

an interleukin-1alpha gene polymorphism’ Ann

Neurol; vol. 47, no. 3, pp. 361-365.

Grossi E, Mancini A and Buscema M. 2007 ‘International

experience on the use of artificial neural networks in

gastroenterology’ Dig Liver Dis; vol. 39, no. 3, pp.

278-285.

Hamsten A and Eriksson P. 2008 ‘Identifying the

susceptibility genes for coronary artery disease: from

hyperbole through doubt to cautious optimism’ J

Intern Med; vol. 263, no. 5, pp. 538–552.

Ianni M, Callegari S, Rizzo A, Pastori P, Moruzzi P,

Corradi D, Porcellini E, Campo G, Ferrari R, Ferrario

MM, Bitonte S, Carbone I, Licastro F. 2012 ‘Pro-

inflammatory genetic profile and familiarity of acute

myocardial infarction’ Immun Ageing; vol. 9, no. 1,

pp. 14.

Juárez-Herrera Ú, Jerjes-Sánchez C, RENASICA II

Investigators. 2013 ‘Risk factors, therapeutic

approaches, and in-hospital outcomes in Mexicans

with ST-elevation acute myocardial infarction: the

RENASICA II multicenter registry’ Clin Cardiol; vol.

36, no. 5, pp. 241-248.

Kullo IJ and Ding K. 2007 ‘Mechanisms of disease: The

genetic basis of coronary heart disease’ Nat Clin Pract

Cardiovasc Med; vol. 4, no. 10, pp. 558–569.

Lahner E, Intraligi M, Buscema M, Centanni M, Vannella

L, Grossi E, Annibale B. 2008 ‘Artificial Neural

Networks in the Recognition of the Presence of

Thyroid Disease in Patients with Atrophic Body

Gastritis’ World J Gastroenterol; vol. 14, no. 4, pp.

563-568.

Latorra D, Campbell K, Wolter A, Hurley JM. 2003

‘Enhanced allele-specific PCR discrimination in SNP

genotyping using 3' locked nucleic acid (LNA)

primers’ Hum Mutat; vol. 22, no. 1, pp. 79-85.

Levi F, Lucchini F, Negri E, La Vecchia C. 2002 ‘Trends

in mortality from cardiovascular and cerebrovascular

diseases in Europe and other areas of the world’

Heart; vol. 88, no. 22, pp. 119–124.

Li K, Monni S, Hüsing A, Wendt A, Kneisel J, Groß ML,

Kaaks R. 2014 ‘Primary preventive potential of major

lifestyle risk factors for acute myocardial infarction in

men: an analysis of the EPIC-Heidelberg cohort’ Eur J

Epidemiol; vol. 29, no. 1, pp. 27-34.

Licastro F, Chiappelli M, Porcellini E, Campo G,

Buscema M, Grossi E, Garoia F, Ferrari R. 2010

‘Gene-gene and gene - clinical factors interaction in

acute myocardial infarction: a new detailed risk chart’

Current Pharmaceutical Des; vol. 16, no. 7, pp. 783–

788.

Licastro F, Chiappelli M, Caldarera CM, Porcellini E,

Carbone I, Caruso C, Lio D, Corder EH. 2011

‘Sharing pathogenetic mechanisms between acute

myocardial infarction and Alzheimer disease as shown

by partially overlapping of gene variant profiles’ J Alz

Dis; vol. 23, no. 3, pp. 421-431.

Lisboa PJC. 2002 ‘A review of evidence of health benefit

from artificial neural networks in medical

intervention’ Neural Netw; vol. 15, no. 1, pp. 11-39.

Mortensen MB, Sivesgaard K, Jensen HK, Comuth W,

Kanstrup H, Gotzsche O, May O, Bertelsen J, Falk E.

2013 ‘Traditional SCORE-based health check fails to

identify individuals who develop acute myocardial

infarction’ Dan Med J; vol. 60, no. 5, A4629.

Pace F, Riegler G, de Leone A, Pace M, Cestari R,

Dominici P, Grossi E, EMERGE Study Group. 2010

‘Is It Possible to Clinically Differentiate Erosive from

Non erosive Reflux Disease Patients? A Study Using

an Artificial Neural Networks-Assisted Algorithm’

Eur J Gastroenterol Hepatol; vol. 22, no. 10, pp.

1163-1168.

Packard RR and Libby P. 2008 ‘Inflammation in

atherosclerosis: from vascular biology to biomarker

discovery and risk prediction’ Clin Chem; vol. 54, no.

1, pp. 24–38.

Pamukcu B, Lip GY, Devitt A, Griffiths H, Shantsila E.

2010 ‘The role of monocytes in atherosclerotic

coronary artery disease’ Ann Med; vol. 42, no. 6, pp.

394–403.

Patel RS and Ye S. 2011 ‘Genetic determinants of

coronary heart disease: new discoveries and insights

from genome-wide association studies’ Heart; vol. 97,

no. 18, pp. 1463–1473.

Penco S, Grossi E, Cheng S, Intraligi M, Maurelli G,

Patrosso MC, Marocchi A, Buscema M. 2005

‘Assessment of the Role of Genetic Polymorphism in

Venous Thrombosis Through Artificial Neural

Networks’ Ann Hum Genet; vol. 69, no. 6; pp. 693-

706.

Penco S, Buscema M, Patrosso MC, Marocchi A, Grossi

E. 2008 ‘New application of intelligent agents in

sporadic amyotrophic lateral sclerosis identifies

unexpected specific genetic background’ BMC

Bioinformatics; vol. 9, pp. 254.

Ravaglia G, Forti P, Maioli F, Orlanducci P, Sacchetti L,

Flisi E, Dalmonte E, Martignani A, Cucinotta D,

Cavalli G. 2001 ‘Conselice study: a population based

survey of brain aging in a muncipality of the Emilia

Romagna region: (A.U.S.L. Ravenna). Design and

methods’ Arch Gerontol Geriatr Suppl; vol. 7, pp.

313-324.

Ridker PM, Rifai N, Pfeffer M, Sacks F, Lepage S,

Braunwald E. 2000 ‘Elevation of tumor necrosis

factor-alpha and increased risk of recurrent coronary

events after myocardial infarction’ Circulation; vol.

101, no. 18, pp. 2149–2153.

Rotondano G, Cipolletta L, Grossi E, Koch M, Intraligi M,

Buscema M, Marmo R; Italian Registry on Upper

Gastrointestinal Bleeding (Progetto Nazionale

Emorragie Digestive). 2011 ‘Artificial Neural

Networks Accurately Predict Mortality in Patients

with Non variceal Upper GI Bleeding’ Gastrointest

Endoscop; vol. 73, no. 2, pp. 218-226.

Rumelhart DE and McClelland JL. 1982 ‘An interactive

HEALTHINF2015-InternationalConferenceonHealthInformatics

258

activation model of context effects in letter perception:

Part 2. The contextual enhancement effect and some

tests and extensions of the model’ Psychol Rev; vol.

89, no.1, pp. 60-94.

Street ME, Grossi E, Volta C, Faleschini E, Bernasconi S.

2008 ‘Placental Determinants of Fetal Growth:

Identification of Key Factors in the Insulin-Like

Growth Factor and Cytokine Systems Using Artificial

Neural Networks’ BMC Pediatr; vol. 8, no. 24.

Tabaton M, Odetti P, Cammarata S, Borghi R, Monacelli

F, Caltagirone C, Bossù P, Buscema M, Grossi E.

2010 ‘Artificial Neural Networks Identify the

Predictive Values of Risk Factors on the Conversion

of Amnestic Mild Cognitive Impairment’ J Alzheimers

Dis; vol. 19, no. 3, pp. 1035-1040.

Yusuf S, Reddy S, Ounpuu S, Anand S. 2001 ‘Global

burden of cardiovascular diseases: part I: general

considerations, the epidemiologic transition, risk

factors, and impact of urbanization’ Circulation; vol.

104, no. 22, pp. 2746–2753.

Zhang C. 2008 ‘The role of inflammatory cytokines in

endothelial dysfunction’ Basic Res Cardiol; vol. 103,

no. 5, pp. 398–406.

ANewRiskChartforAcuteMyocardialInfarctionbyaInnovativeAlgoritm

259