MBVol29n3

 

Embed or link this publication

Description

Revista de la Real acadèmia de medicina de les Illes Balears. Volumen 29, número 3, 2014

Popular Pages


p. 1

VOLUM 29 NÚM. 3 SETEMBRE - DESEMBRE 2014 Medicina Balear PUBLICACIÓ DE LA REIAL ACADÈMIA DE MEDICINA DE LES ILLES BALEARS Usefulness of Bayesian networks in epidemiological studies Consumo de alcohol y riesgo de accidentes de tráfico. Aspectos preventivos The influence of organochlorine compound exposure on the physiological development of children Respiratory tract infections caused by Human Coronavirus (HCoVs) in Balearic Islands, 2014 Bioethical analysis of transgenic animals and genetically modified organisms (GMO) La prevención primaria del cáncer de cervix: las vacunas frente al virus del papiloma humano Varón de 50 años, VIH positivo, con trombopenia y rectorragias www.medicinabalear.org

[close]

p. 2

Medicina Balear www.medicinabalear.org Medicina Balear , òrgan de la Reial Acadèmia de Medicina de les Illes Balears, va aparèixer el 1986 amb l’objectiu de donar curs a les inquietuds científiques i fomentar l’esperit d’investigació dels professionals de la sanitat balear i amb la pretensió suplementària de projectar en la societat temes d’interès sanitari. Medicina Balear publica en català, castellà o anglès treballs originals, articles de revisió, cartes al director i altres escrits d’interès relacionats amb les ciències de la salut i presta particular atenció als treballs que tinguin per àmbit les Illes Balears i altres territoris de la conca mediterrània occidental. La revista sotmet els originals a la revisió anònima per al menys dos experts externs (peer review). El material científic publicat a Medicina Balear resta protegit per drets d’autor. Medicina Balear no és responsable de la informació i opinions dels autors. Aquesta obra -llevat que s’indiqui el contrari en el text, en les fotografies o en altres il·lustracions- és subjecta a la llicència de Reconeixement-NoComercial-SenseObraDerivada 3.0 Espanya de Creative Commons; http://creativecommons.org/licenses/by-nc-nd/3.0/es/. Així, doncs, s’autoritza al públic en general a reproduir, distribuir i comunicar l’obra sempre que se’n reconegui l’autoria i l’entitat que la publica i no se’n faci un ús comercial ni cap obra derivada. Medicina Balear es troba incorporada a la Biblioteca Digital de les Illes Balears, de la Universitat de les Illes Balears, i està inclosa en les bases de dades següents: Latindex (catàleg), Dialnet, Índice Médico Español, DOAJ, Imbiomed EDITA Reial Acadèmia de Medicina de les Illes Balears www.ramib.org Campaner, 4, baixos. 07003 Palma de Mallorca Tel. 971 72 12 30 Email: info@ramib.org Pàgina web: http://www.ramib.org  Dipòsit Legal: PM 486 - 95 eISSN: 2255 - 0569 Disseny i maquetació Intelagencia Publicitat - www.intelagencia.es - intelagencia@intelagencia.es

[close]

p. 3

Medicina Balear Publicació quadrimestral de ciències de la salut de la Reial Acadèmia de Medicina de les Illes Balears Director Macià Tomàs Salvà, Reial Acadèmia de Medicina de les Illes Balears (RAMIB) CONSELL EDITORIAL Subdirector Secretari de la publicació Editor científic Assessors editorials Redactor en cap Vocals A. Arturo López González, RAMIB Joan March Noguera, RAMIB Marta Couce Matovelle, Case Western Reserve University José A. Guijarro Pastor, AEMET · Jaume Rosselló Mir, UIB J. L. Olea Vallejo, RAMIB Antoni Aguiló Pons, Universitat de les Illes Balears · Bartolomé Burguera González, Cleveland Clinic (Ohio) · Amador Calafat Far, Socidrogalcohol · Carlos Campillo Artero, Universitat Pompeu Fabra · Valentín Esteban Buedo, Conselleria de Sanitat, Generalitat Valenciana · Carmen González Bosch, Universitat de València · Miguel A. Limon Pons, Institut Menorquí d’Estudis · Virgili Páez Cervi, Bibliosalut · Lucio Pallarés Ferreres, Hospital Son Espases, Ibsalut · Ignacio Ricci Cabello, University of Oxford · Guillermo Sáez Tormo, Universitat de València · M a Teófila Vicente Herrero, IUNICS CONSELL CIÉNTIFIC Mª José Anadón Baselga ( Universidad Complutense de Madrid ), Miquel Capó Martí ( Universidad Complutense de Madrid ), Antonio Coca Payeras ( Universitat de Barcelona ), James Drane ( Edinboro University ), Leopoldo Forner Navarro ( Universitat de València ), Alexandre García-Mas, ( Universitat de les Illes Balears ), Antoni Gelabert Mas ( Universitat Autònoma de Barcelona ), Joan Grimalt Obrador (Consell Superior d’Investigacions Científiques, CSIC) , Federico Hawkins Carranza ( Universitat Complutense de Madrid ), Joan Carles March Cerdà (Escuela Andaluza de Salud Pública, EASP ), Gabriel Martí Amengual ( Universitat de Barcelona ), Jasone Monasterio Aspiri ( Universitat Autònoma de Barcelona ) Rosa Pulgar Encinas ( Universidad de Granada ), Ciril Rozman ( Universitat de Barcelona ). Amb la col·laboració de Conselleria de Presidència www.medicinabalear.org

[close]

p. 4



[close]

p. 5

VOLUM 29 NÚM. 3 SETEMBRE - DESEMBRE 2014 Medicina Balear PUBLICACIÓ DE LA REIAL ACADÈMIA DE MEDICINA DE LES ILLES BALEARS www.medicinabalear.org SUMARI EDITORIAL Un bicentenari digne de commemoració: la “Toxicologie générale” de 1814 Miquel Àngel Limon Pons 8 ORIGINALS Usefulness of Bayesian networks in epidemiological studies P. Fuster-Parra, P. Tauler, M. Bennasar, A. Aguiló 10-17 18-24 Consumo de alcohol y riesgo de accidentes de tráfico. Aspectos preventivos Mª Teófila Vicente Herrero, Miguel Ruiz-Flores Bistuer, Daniel Bozzini, Luisa Capdevila García, Mª Victoria Ramírez Iñiguez de la Torre, Mª Jesús Terradillos García, Ángel Arturo López González The influence of organochlorine compound exposure on the physiological development of children Joan O. Grimalt, Maties Torrent, Jordi Sunyer 25-36 ORIGINAL BREU Respiratory tract infections caused by Human Coronavirus (HCoVs) in Balearic Islands, 2014 Jordi Reina, Carla López-Causape, María Busquets, Carmen Morales 37-39 ARTICLES ESPECIALS Bioethical analysis of transgenic animals and genetically modified organisms (GMO) 41-50 Miguel Capó Martí, Ricardo Roa-Castellanos, María José Anadón Baselga, James Drane La prevención primaria del cáncer de cervix: las vacunas frente al virus del papiloma humano Javier Cortés Bordoy, Ana Forteza Valades, Gabriel Ferret Fuchs 51-58 ESTUDI DE CASOS Varón de 50 años, VIH positivo, con trombopenia y rectorragias Inmaculada González Sayago, Paula Carrillo García, Elena Delgado Mejía, Javier Murillas Angoiti, Bartolomé Colom Oliver, Manuel del Río Vizoso 59-64 LLIBRES Cáncer colorrectal. Diagnóstico precoz y tratamiento de Susan L. Gearhart y Nita Ahuja (eds) Joan March Noguera 65 Neuroética. Cuando la materia se despierta, de Kathinka Evers Miguel Capó Martí, Ricardo Roa-Castellanos 66 eISSN 2255-0569

[close]

p. 6



[close]

p. 7

VOLUME 29 NUMBER 3 SEPTEMBER - DECEMBER 2014 Medicina Balear SCIENTIFIC JOURNAL OF THE ROYAL ACADEMY OF MEDICINE OF THE BALEARIC ISLANDS www.medicinabalear.org CONTENTS EDITORIAL A bicentennial worthy to be commemorated: The 1814 edition of “Toxicologie générale” Miquel Àngel Limon Pons 8 ORIGINAL ARTICLES Usefulness of Bayesian networks in epidemiological studies P. Fuster-Parra, P. Tauler, M. Bennasar, A. Aguiló 10-17 18-24 Alcohol and traffic accidents risk in Spain. Preventive aspects The influence of organochlorine compound exposure on the physiological development of children Joan O. Grimalt, Maties Torrent, Jordi Sunyer Mª Teófila Vicente Herrero, Miguel Ruiz-Flores Bistuer, Daniel Bozzini, Luisa Capdevila García, Mª Victoria Ramírez Iñiguez de la Torre, Mª Jesús Terradillos García, Ángel Arturo López González 25-36 SHORT ORIGINAL Respiratory tract infections caused by Human Coronavirus (HCoVs) in Balearic Islands, 2014 Jordi Reina, Carla López-Causape, María Busquets, Carmen Morales 37-39 SPECIAL ARTICLES Bioethical analysis of transgenic animals and genetically modified organisms (GMO) 41-50 Miguel Capó Martí, Ricardo Roa-Castellanos, María José Anadón Baselga, James Drane Primary prevention of cervical cancer: vaccines against human papillomavirus Javier Cortés Bordoy, Ana Forteza Valades, Gabriel Ferret Fuchs 51-58 CASE STUDIES 50-year-old, VIH positive male, with thrombocytopenia and rectal bleeding Inmaculada González Sayago, Paula Carrillo García, Elena Delgado Mejía, Javier Murillas Angoiti, Bartolomé Colom Oliver, Manuel del Río Vizoso 59-64 BOOKS Cáncer colorrectal. Diagnóstico precoz y tratamiento of Susan L. Gearhart and Nita Ahuja (eds) Joan March Noguera 65 Neuroética. Cuando la materia se despierta, of Kathinka Evers Miguel Capó Martí, Ricardo Roa-Castellanos Medicina Balear 2013; 26 (2); 5-6 66 eISSN 2255-0569

[close]

p. 8

eISSN 2255-0569 EDITORIAL Un bicentenari digne de commemoració: la “Toxicologie générale” de 1814 A bicentennial worthy to be commemorated: The 1814 edition of “Toxicologie générale” Miquel Àngel Limon Pons Doctor en periodisme. Acadèmic corresponent de la Reial Acadèmia de Medicina de les Illes Balears Una efemèrides gens banal es commemora dins l’any 2014. Té el doble caràcter d’efemèrides de la història de la Medicina i, alhora, dels annals bibliogràfics de matèria científica que es relacionen amb les Illes Balears, i fins amb la vida de la nostra Reial Acadèmia de Medicina. El 1814, un ambiciós i intel·ligent químic i metge instal·lat a París, en el melic de la ciència i la universitat mundials d’aleshores, presentava a les principals autoritats acadèmiques del país el manuscrit d’una obra que havia de ser cridada a reformar i refundar una branca científica que encara llavors es movia en la mera especulació i les contradiccions. Era la Toxicologia, una de les múltiples àrees de coneixement de les ciències de la salut que encara no havien conegut el salt a la investigació sistemàtica, a l’experimentació de laboratori i a la reformulació dels principis fonamentals de la mateixa. Qui havia de ser autor d’una semblant capgirada a favor Mateu Orfila, cap als 30 anys d’edat d’una especialitat científica nomia Mateu Orfila Rotger (1787-1853), segurament el primer menorquí en assolir una personalitat intel·lectual i científica moderna; d’altra banda, membre corresponent de la nostra docta corporació mèdica de les Balears. Tot s’havia endegat aquells dies en què Orfila, de París estant, se sostenia impartint privadament cursos de química, de botànica, d’anatomia i de medicina legal. Ho feia en un petit laboratori privat, gairebé casolà, del carrer de Foin de Saint-Jacques. Amb major o menor dedicació, ho va mantenir durant tres o quatre anys, si més no fins l’1 de març de 1819 en què rebé el seu cobejat ingrés als cenacles universitaris, ara ja amb la condició de professor titular de la Sorbona. Un dia d’abril de 1813 succeí que Orfila explicava al seu auditori una lliçó sobre les propietats de l’àcid arseniós. Com li agradava de fer, relacionava les propietats i descrivia les reaccions i precipitacions que una solució d’aquesta substància donava per l’acció de diferents reactius. Ell hi mantenia, però, les explicacions llibresques que la tradició anterior a ell oferia al respecte. Per casualitat, en aquells moments, durant una classe, Orfila hi disposava d’una escudelleta de cafè. I així, se li va ocórrer d’afegirhi una mica de la solució arsenical al beuratge. Aleshores, ell esperava que les reaccions fossin les mateixes que es donaven quan es feia amb solucions aquoses. Però no. En les barreges de cafè i dissolució arsenical, cap de les reaccions tòpiques no es complien com en aigua. En paraules del mateix Orfila a la seva autobiografia, «em vaig procurar brou, llet, vi, infusions diverses…; en un mot, molts dels líquids que s’empren sovint en l’economia domèstica. El nombre d’experiments que vaig fer amb aquestes substàncies 8 Medicina Balear 2014; 29 (3): 8-9

[close]

p. 9

Un bicentenari digne de commemoració: la “Toxicologie générale” de 1814 fou molt gros; i tots em demostraren que els verins, si no tots, en la seva major part, no podien ésser reconeguts seguint els mètodes usuals quan es troben barrejats amb líquids d’origen animal o vegetal […]». Regirant, de seguida, a les principals biblioteques científiques de París, i llegint i rellegint tot el que d’important s’havia escrit fins llavors sobre toxicologia, portaren Les novetats científiques que la recerca d’Orfila aportava Orfila a concloure així: «Idò, la toxicologia no existeix». eren múltiples. Direm, tanmateix, que, abans de la puEn efecte, amb base i exblicació de l’obra, els verins perimentació científica moeren cercats en els conducderna i sistemàtica no existes digestius de la persona tia encara. Tot era per fer. que n’havia ingerit alguna Acte seguit, el jove científic dosi. Orfila, però, lluny de es lliurà a la investigació limitar-se a aquesta localitsobre un semblant cúmul zació, investigà la presència d’interrogants, de dubtes i dels tòxics en altres òrgans, d’ardor investigadora. Recom ara el fetge o el cervell. clòs a una casa de camp a Hi aplicava procediments Villeneuve-le-Roi, endeganous per a descobrir-los, i ria una febril experimentael seu ús metòdic havia de ció al llarg de tot l’estiu de permetre de revelar-ne míni1813. Un a un, feia proves mes quantitats. A París, es i més proves amb verins féu proverbial que la investii líquids de tota mena de gació essencial del Dr. Orfila procedència. En els quali havia comportat el sacrifici derns de notes, hi anotava de més de 4000 cans, als els processos i hi registraquals havia aplicat els tòxics va els resultats. El rigor i la i les dosis per després anasistemàtica ho presidí tot litzar els òrgans interns de amb absoluta escrupolosil’animal. En definitiva, aquetat metòdica. Fou així que lla era una nova toxicologia: durant l’hivern de 1813 i la una toxicologia científica i primavera de 1814 ja es moderna, arrels bàsiques va considerar en òptimes de la toxicologia forense condicions d’oferir a l’editor actual. Com ell mateix curà de la Facultat de Medicina d’escriure en el pròleg de de París, M. Crochard, un la primera edició, «un gran primer volum de l’obra que nombre de fets que han de anomenà Traité des poisons servir de fonament a la totirés des regnes minéral, xicologia eren abans descoPortada de la primera edició de la “Toxicologie générale” vegetal et animal ou toxiconeguts o mal estudiats». logie générale. L’any 1815, li feia a mans el segon i darrer tom del treball que havia En paraules de Josep Sureda i Blanes, va ser així que de marcar fita en els annals de la ciència europea in- Orfila entrà en el camí que el portaria «a la glòria cientíternacional. Orfila, amb ella, acabava d’establir els fo- fica del seu temps». I a guanyar-se per a si —afegim— naments de la toxicologia científica moderna. Hi havia un grau d’immortalitat permanent en la història de la rebut, a més, la sanció elogiosa de l’Institut de França, ciències modernes amb evident abast internacional, de manera que el treball va assolir un èxit fenomenal i orgull avui de les Illes Balears. fulgurant en tots els principals nuclis acadèmics i cien- tífics d’Europa. En vida d’Orfila, se’n ferien cinc edicions en llengua francesa; i les traduccions a l’anglès i l’alemany són del mateix 1814. L’any 1821, finalment, el traduí al castellà el Dr. Larra, l’insigne i poc recorregut intel·lectual espanyol afrancesat, pare del periodista romàntic Mariano José de Larra, Fígaro. Medicina Balear 2014; 29 (3): 8-9 9

[close]

p. 10

eISSN 2255-0569 ORIGINAL Usefulness of Bayesian networks in epidemiological studies Utilidad de la redes bayesianas en los estudios epidemiológicos P. Fuster-Parra, P. Tauler, M. Bennasar, A. Aguiló Research Group on Evidence, Lifestyles & Health. Research Institute on Health Sciences (IUNICS) Universitat Illes Balears, Palma de Mallorca, Spain Corresponding author Pilar Fuster-Parra Edifici Anselm Turmeda Ctra. Valldemossa km 7,5 - 07122 Palma de Mallorca E-mail: pilar.fuster@uib.es Recibido: 12 – V – 2014 Aceptado: 23 – VI - 2014 doi: 10.3306/MEDICINABALEAR.29.03.10 Abstract Introduction: Bayesian networks are a form of statistical modelling, which has been widely used in fields like clinical decision, systems biology, human immunodeficiency virus (HIV) and influenza research, analyses of complex disease systems, interactions between multiple diseases and, also, in diagnostic diseases. The present study aimed to show the usefulness of Bayesian networks (BNs) in epidemiological studies. Material and Methods: 3,993 subjects (men 1,758, women 2,235) belonging to the public productive sector from the Balearic Islands (Spain), which were active workers, constitute the data set. Results: A BN was built from a dataset composed of twelve relevant features in cardiovascular disease epidemiology. Furthermore, the structure and parameters were learnt with GeNIe 2.0 tool. Taking into account the main topological properties some features were optimized, obtaining a hypothesized scenario where the likelihoods of the different features were updated and the adequate conclusions were established. Conclusions: Bayesian networks allow us to obtain a hypothetical scenario where the probabilities of the different features are updated according to the evidence that is introduced. This fact makes Bayesian networks a very attractive tool. Keywords: Bayesian networks, model averaging, cardiovascular lost years, cardiovascular risk score Resumen Introducción: Las redes Bayesianas son una forma de modelización estadística, las cuales han sido ampliamente utilizadas en campos como la decisión clínica, biología de sistemas, virus de inmunodeficiencia humana (VIH) e investigación en influenza, análisis de sistemas de enfermedades complejos, interacciones entre múltiples enfermedades y, también, en enfermedades de diagnóstico. Este estudio tiene como objetivo mostrar la utilidad de las redes Bayesianas en estudios epidemiológicos. Material y Métodos: 3,993 individuos (hombres 1,758, mujeres 2,235) pertenecientes al sector productivo público de las Islas Baleares (España), los cuales eran trabajadores activos, constituyen la base de datos. Resultados: Una red Bayesiana se ha obtenido a partir de una base de datos compuesta de doce características relevantes de la epidemiología de la enfermedad cardiovascular. Por otra parte, la estructura y los parámetros se han obtenido con la herramienta Genie 2.0. Teniendo en cuenta las principales propiedades topológicas algunas características fueron optimizadas. Conclusiones: Las redes Bayesianas permiten obtener un escenario hipotético donde las probabilidades de las diferentes características se van actualizando de acuerdo con la evidencia introducida. Este hecho hace de las redes Bayesianas una herramienta muy atractiva, además permite establecer diversas conclusiones. Palabras clave: Redes bayesianas, modelo promediado, años cardiovasculares perdidos, escala de riesgo cardiovascular Introduction Bayesian networks (BNs)16, 25 also referred to as causal networks or beliefs networks, are a form of statistical modelling which allow us to obtain a graphical network describing the dependencies and conditional independencies from empirical data. They have proven to be a promising tool for discovering relationships9, they capture the way an expert understands the relationships among all the features 6 and, even, they have been used in data analysis 8. The origins of BN modelling lie within the data mining and machine learning literature5, 13. BNs are a kind of probabilistic graphical model (PGM)18, which combine graph theory (to help in the representation and resolution of complex problems) and probability theory (as a way of representing uncertainty). A PGM is defined as a graph where nodes represent random variables4, 12 and arcs represent dependencies between such variables11, 24. A PGM is called a BN when the graph connecting its variables is a directed acyclic graph (DAG). The graphical representation of BNs captures the compositional structure of the relations and the general aspects of all probability distributions factorized according to that structure12. 10 Medicina Balear 2014; 29 (3): 10-17

[close]

p. 11

Usefulness of Bayesian networks in epidemiological studies BN modelling is widely used in fields like clinical decision23, systems biology7, 13, human immunodeficiency virus (HIV) and influenza research21, 26, analyses of complex disease systems14, 19, 20, interactions between multiple diseases17 and, also, in diagnostic diseases1, 2, 3, 22, 27. The aim of the present study was to show the usefulness of Bayesian networks (BNs) in epidemiological studies focused in cardiovascular risk factors. Learning Bayesian networks Learning BNs from a dataset has become an increasingly active area of research. Although, sometimes experts can create good BNs from their own experience, it can be a very tedious task for domains with large knowledge bases. Many methods (learning algorithms) from machine learning community have been developed to automate the creation of BNs using cases collected from past experience. To obtain a BN, it is necessary to determine a structure (defined by a DAG) and the conditional probabilities assigned to each node of the DAG. Therefore, to learn a BN involves two tasks: I) structural learning, that is, the identification of the topology of the BN, and II) parametric learning, that is the estimation of numerical parameters (conditional probabilities) for a network topology. Bayesian network structure learning Basically, there are two approaches to structure learning: I) search-and-score structure learning, and constraintbased structure learning. Search and score search algorithms assigns a number (score) to each Bayesian network structure, and we look for the model structure with the highest score. Constraint based search algorithms establish a set of conditional independence analysis on the data. Based on this analysis an undirected graph is generated. Using additional independence test, the network is converted into a BN. Bayesian network parameter learning In a BN the conditional probability distributions are called the parameters, obtaining a conditional probability distribution for each node of the network topology. Theoretical Background Let us consider a probabilistic model M, consisting of a set V of discrete random variables (features) and a joint probability distribution P. Let D (it is the graphical structure of the causal network) be a directed acyclic graph (DAG), whose set of vertices has a one to one correspondence with the variables of M. Two random variables X and Y in a causal network are d-separated if for all the paths between X and Y, there is an intermediate variable Z such that either i) the connection is serial (X→Z→Y or X←Z←Y) or diverging (X←Z→Y) and Z is instantiated or ii) the connection is converging (X→Z←Y), and neither Z nor any of Z’s descendants have received evidence15. D is said to be an I-map of M if every d-separation condition in D corresponds to a conditional independence relationship in M. D is a minimal I-map of M if none of its arrows can be remove without destroying its –I-mapness. A BN of the probabilistic model is defined as a DAG D that is a minimal I-map of M. The joint probability distribution factorized as a product of several conditional distributions and denotes the dependency/independency structure by a DAG, which is called the chain rule for BNs: P(X1,…,Xn )= ∏ P(Xi |Pa(Xi )) I=1 n The independencies from the graph are easily translated into the probabilistic model. The local Markov condition and the global Markov property are important characteristics of the network topology when the BN is used to make inferences (that is to predict new scenarios when new information is introduced). The local Markov condition establishes that each feature is conditionally independent of the set of all its non-descendants given the set of all its parents. The global Markov property states that any node is conditionally independent of any other node given its Markov blanket (which includes its parents, its children, and the other children’s parents (spouses). Any node in the BN would be d-separated from the nodes belonging to the nonMarkov blanket given its Markov blanket. Materials and Methods Participants Participants were active workers belonging to the public productive sector from the Balearic Islands (Spain). Subjects were invited to participate in the study during their annual work health assessment. Any worker attending the work health assessment could be included in the study. 4,300 workers were invited to participate. Among them, 3,993 subjects (men 1,758, women 2,235) agreed to participate. Participants signed informed consent prior to enrolment in the study. After acceptance, a complete family and personal medical history was recorded. The study design was in accordance with the Declaration of Helsinki and received approval from the Balearic Islands Clinical Research Ethical Committee. Medicina Balear 2014; 29 (3): 10-17 11

[close]

p. 12

Pilar Fuster-Parra et al. Epidemiological model From the dataset, and using GeNIe 2.0 tool [10], a BN structure was inferred. A Bayesian search algorithm was selected to obtain a DAG, which is a search-and-score algorithm. Figure 1 shows the obtained structure. Once the structure is obtained the parameters could be calculated. The EM (expectation-maximization) algorithm, which is an iterative method for finding maximum likelihood or maximum a posteriori (MAP) estimates of parameters in statistical models, was used to determine the parameters. A distribution probability was obtained for each node (feature) in the DAG. The resulting BN is shown in Figure 2. Results and Discussion BNs are used to make inferences [4], that is, to obtain new likelihoods of features when new information is introduced. To show it, two patterns of reasoning were selected: causal reasoning (from top to bottom), and intercausal reasoning which is close to human reasoning. In this last case the concept of Markov blanket was considered. BN Inference: Causal reasoning pattern We use this pattern when we reason from top to bottom. To illustrate this sort of reasoning we have selected four examples comparing the likelihood variations in men and women groups. In Figure 3 Physical Activity feature is instantiated to the “no practice” value (PA = no practice), Smoking feature is instantiated to the “yes value” (Smoking = yes), and Gender feature is instantiated to the “men value”. Then, it can be observed how the likelihoods of the different features change. Likelihood of Blood Pressure (BP) feature in optimal state decreased from 0.47 to 0.18; and increased states such as HTA severe, mild and moderate, from 0.01, 0.16, and 0.05 to 0.03, 0.28, and 0.11 respectively. BMI feature in Obesity TI and Overweight GII states increased its likelihoods from 0.13 and 0.19 to 0.29 and 0.32 respectively. Figure 1: Structure of a BN obtained performing a Bayesian search algorithm. From the DAG it can be observed how the different features are connected. Figure 2: BN model obtained for cardiovascular disease risk factors. From the BN likelihoods this sample shows normal triglycerides (TG), normal glucose, normal cardiovascular risk score (CVRS) and normal waist circumference (WC). 12 Medicina Balear 2014; 29 (3): 10-17

[close]

p. 13

Usefulness of Bayesian networks in epidemiological studies Figure 3: BN where the following evidence is introduced: physical activity (PA) = no practice, Smoking = yes, and Gender = men. Triglycerides (TG) in normal state decreased its likelihood from 0.83 to 0.57. Glucose in normal state decreased its likelihood from 0.87 to 0.79. Cardiovascular lost years feature (CVLY) increased its fourth quartile from 0.22 to 0.74. To compare the likelihoods variations within the group of women, we selected women state in the Gender feature. The likelihood variations are shown in Figure 4. Figure 4: BN considering the following evidence is introduced: physical activity PA = no practice, Smoking = yes, and Gender =women. Medicina Balear 2014; 29 (3): 10-17 13

[close]

p. 14

Pilar Fuster-Parra et al. Figure 5: BN considering the following evidence: physical activity PA = no practice, Smoking = yes, Gender = men, and BP = severe. Likelihood of Blood Pressure (BP) feature in optimal state increased from 0.47 to 0.57; and decreased states such as HTA mild and moderate, from 0.16 and 0.05 to 0.13 and 0.03 respectively; but also normal state decreased from 0.16 to 0.12. BMI feature in Obesity TI and Overweight GII states increased its likelihoods from 0.13 and 0.19 to 0.15 and 0.21 respectively, being higher in men than in women. Triglycerides (TG) in normal state increased its likelihood from 0.83 to 0.88. Glucose in normal state decreased its likelihood from 0.87 to 0.89. Cardiovascular lost years feature (CVLY) increased its fourth quartile from 0.22 to 0.36; and its third quartile from 0.24 to 0.35. Results indicated that men were at increased cardiovascular disease risk compared to women under conditions of smoking and no practice of physical activity. We also considered another hypothetical situation characterized by a severe hypertension (BP). The likelihood changes are shown in Figure 5. Cardiovascular lost years feature (CVLY) increased its likelihood in fourth quartile from 0.22 to 0.96 and Cardiovascular risk score (CVRS) decreased its low state from 0.92 to 0.69. On the other hand, results obtained in women when blood pressure was instantiated to the highest value are shown in Figure 6. In this last situation, Cardiovascular lost years feature (CVLY) increased its likelihood in fourth quartile from 0.22 to 0.83. And Cardiovascular risk score (CVRS) preserved its low state in 0.92. Taking into account these results, men were revealed again as the gender with higher cardiovascular disease risk. BN Inference: Intercausal reasoning pattern We refer to intercausal reasoning, which constitutes a very common pattern in human reasoning, when different causes of the same effect can interact. Using this reasoning pattern, the following two examples were considered: maximizing CVLY in first quartile state and maximizing CVLY in fourth quartile state. To illustrate this way of reasoning the following features from the Markov blanket of CVLY were considered: CVRS, BP, HDL and Smoking. Figure 7 shows the likelihood variation when Smoking feature is instantiated to non-smoker, Cardiovascular risk score (CVRS) is instantiated to low value, Blood pressure (BP) is instantiated to optimal value and HDL is instantiated to low value. Under these instantiations the following changes can be observed: Gender in women state increases from 0.56 to 0.90, showing that under this situation the likelihood of being a woman is higher. Age in 35-44 state increased its likelihood from 0.46 to 0.59. The likelihood of practising physical activity increased from 0.48 to 0.83. BMI in normal weight state increased from 0.44 to 0.68. 14 Medicina Balear 2014; 29 (3): 10-17

[close]

p. 15

Usefulness of Bayesian networks in epidemiological studies Figure 6: BN where the following evidence is introduced: physical activity PA = no practice, Smoking = yes, Gender =women, and BP = severe. Triglycerides in normal value increased its likelihood from 0.83 to 0.97. Glucose in normal value increased its likelihood from 0.87 to 0.97. Waist circumference in normal value increased its likelihood from 0.54 to 0.72, and Cardiovascular lost years in first quartile value increased its likelihood from 0.29 to 0.91. Medicina Balear 2014; 29 (3): 10-17 15

[close]

Comments

no comments yet