CONTEXTE : La France est caractérisée par l’existence de fortes disparités socioéconomiques et géographiquesde l’état de santé, le plus souvent objectivées par des indicateurs objectifs de morbidité et de mortalité. Ladistribution territoriale des mesures de santé perçue comme la qualité de vie liée à la santé (QdVLS) reste malconnue en population générale. Une meilleure connaissance de la répartition et des déterminants – en particuliercontextuels - de ces indicateurs subjectifs permettrait de mieux comprendre leur signification par rapport auxindicateurs objectifs et d’apprécier l’intérêt spécifique de leur suivi en population générale.OBJECTIFS : Les objectifs de cette recherche étaient [1] d’étudier l’existence de disparités spatiales de QdVLSdans la population française et d’analyser leur évolution dans le temps, [2] d’étudier les déterminants de laQdVLS à la fois individuels et contextuels dans le cadre d’une analyse multiniveau, et [3] d’évaluer lesassociations écologiques entre QdVLS et mortalité ultérieure à cinq ans d’intervalle.MATERIEL ET METHODES : Les données issues de deux enquêtes transversales nationales représentatives ont étéexploitées : l’enquête Décennale 2003 de l’Insee pour l’ensemble des travaux menés (N=22 743 [1 et 3] ; N=16 732 [2]) et l’enquête Sofres 1995 pour l’analyse de l’évolution temporelle de la QdVLS (N=3 243 [1]). Lequestionnaire utilisé dans les deux enquêtes était le SF-36. [1] Des modèles de régression linéaire multiple àeffets fixes avec recherche d’interactions ont été réalisés pour l’analyse de l’évolution temporelle. [2] L’analysemultiniveau des déterminants contextuels de la QdVLS s’appuyait sur des modèles à effets mixtes, afind’explorer une chaine causale incluant des déterminants aux niveaux individuel, du ménage, de l’unité urbaine etrégional, d’ordre démographique, socioéconomique ou intégrant la notion contextuelle plus complexed’attractivité (taux migratoire, désindustrialisation). [3] Les données de mortalité étaient issues de la statistiquenationale des décès élaborée annuellement par le CépiDc. Des modèles de régression binomiale négative ont étéréalisés pour l’analyse des associations écologiques au niveau régional entre QdVLS en 2003 et mortalitéultérieure (court terme [2003-2005] ; à 5 ans [2007-2009]) et en stratifiant sur le sexe, l’âge et les causesspécifiques de décès.RESULTATS PRINCIPAUX : [1] Une diminution significative de la QdVLS était observée entre 1995 et 2003affectant tous les groupes sociodémographiques et suggérant la possibilité d’un accroissement des disparités pourles catégories les plus fragiles de la population. [2] De fortes disparités régionales de QdVLS étaient retrouvées,persistant après ajustement sur les caractéristiques socioéconomiques individuelles. L’analyse multiniveaupermettait d’identifier des processus de médiation impliquant les variables contextuelles de désindustrialisation,le taux d’accroissement migratoire, le taux d’abstention aux élections, et les comportements liés à la santé. Desinteractions inter-niveaux et intra-régionales étaient identifiées. [3] Des associations écologiques significativesétaient retrouvées au niveau régional entre QdVLS et mortalité à cinq ans d’intervalle, persistant aprèsajustement sur le niveau socioéconomique. Des relations spécifiques étaient observées après stratification surl’âge, le genre, les causes spécifiques de décès ; le caractère prédictif de la QdVLS variait selon le délai plus oumoins court entre mesure de la QdVLS et mortalité. / BACKGROUND: Wide social and geographical disparities are reported in France for morbidity and mortalityindicators. Less is known regarding the spatial distribution in general population of self-rated health (SRH) andhealth-related quality of life (HRQoL). Improving the knowledge of the contextual determinants of HRQoLwould help towards a better understanding of their meaning and interest in general population when it comes tocompare with classical objective indicators.OBJECTIVES: The objectives were [1] to assess existing spatial disparities of HRQoL in French generalpopulation and to investigate their evolution in time, [2] to determine individual and contextual determinants ofHRQoL and [3] to explore the ecological associations between HRQoL and subsequent mortality five years later.METHODS: Data were drawn from two large representative cross sectional surveys: the Insee Decennial HealthSurvey led in 2003 (N=22 743 [study 1 and 3] ; N= 16 732 [2]) and the Sofres health survey led in 1995(N=3243 [1]).The MOS SF-36 questionnaire was used in both surveys. [1] Fixed effects linear models combinedwith interaction tests were used for assessing time trends. [2] Mixed effects linear models were used for themultilevel analysis, exploring a causal pathway including individual and macrolevel factors (household, urbanunit and region) assessing demographics, socioeconomics, and features related to the notion of areaattractiveness (deindustrialization, net migration rates). [3] Mortality data were drawn from the French nationalstatistics of mortality (CepiDc-Inserm). Negative binomial regression models were performed to identifyecological associations at the region level between HRQoL recorded in 2003 and subsequent mortality (shortterm [2003-2005]; 5-years later [2007-2009]), stratifying on age, gender and specific causes of death.MAIN RESULTS: [1] A significant decrease in HRQoL was observed between 1995 and 2003, affecting allsociodemographic categories and suggesting likely widening disparities in the most fragile categories. [2]Regional HRQoL disparities were found, persisting after adjusting on socioeconomic individual characteristics.Multilevel analysis showed some evidence for mediation involving contextual factors like deindustrialization,net migration rates, voter abstention rate and health-related behaviors. Cross-level interactions were found aswell. [3] Significant ecological associations were identified at the region level between HRQoL and mortalityfive years later, persisting after adjusting on deprivation. Specific relationships were observed after stratifying onage, gender, specific causes of death; the predictive ability of HRQoL for mortality was varying depending onthe mortality period considered for analysis.CONCLUSIONS: Our results highlight the interest in assessing HRQoL at the population level and in exploringthe contextual determinants at play. Systematic inclusion of validated and multidimensional HRQoLquestionnaires should be supported in national surveys, so as to improve our knowledge of long term temporaltrends in HRQoL, to promote an increased use of contextual multilevel analyses using such data, and eventuallyto help better identifying sub-groups at risk and optimizing public health interventions.

Antecedents: No hi ha gaire coneixement sobre el neurodesenvolupament de preescolars i la seva susceptibilitat enfront a factors ambientals. Objectius: Avaluar les respostes del neurodesenvolupament en nens i les seves característiques psicomètriques, i, si factors ambientals primerencs (ex., duració de la lactància materna i mares que fumen) poden influir tals respostes. Mètodes: Dues cohorts prospectives des del naixement en població general (Menorca (N=421) i Ribera d'Ebre (N=79)) van se seguides fins als 4 anys d'edat durant un període de dos anys (2001-2003). Els nens van ser avaluats per tres psicòlegs i els seus respectius mestres per les funcions neuropsicològiques (MCSA), els comportaments de dèficit d'atenció i d'hiperactivitat (TDAH-DSM-IV) i la competència social (CPSCS); junt amb l'administració (en persona) a les mares de qüestionaris generals.Resultats: Les respostes van mostrar característiques psicomètriques acceptables i els patrons neuropsicològics del TDAH eren consistents amb altres troballes sobre TDAH. La lactància materna de llarga durada estava associada amb una millora de totes les àrees comportamentals avaluades. Fumar durant l'embaràs estava associat a puntuacions cognitives més baixes.Conclusions: Avaluar el neurodesenvolupament a preescolars sans és factible i necessari per investigar efectes primerencs de factors ambientals i aplicar polítiques preventives de salut pública. / Background: Little is known about neurodevelopment among preschoolers and its susceptibility to environmental factors.Objectives: Assess neurodevelopmental outcomes and their psychometric characteristics in children; and, if early environmental factors (i.e., duration of breastfeeding and maternal smoking) influence the neurobehavioral outcomes. Methods: Two prospective population-based birth cohorts (Menorca (N=421) and Ribera d'Ebre county (N=79)) were followed up at the age of 4 years during a two year period (2001-2003). Children were assessed by three psychologists and their respective teachers for neuropsychological functions (MCSA), inattention-hyperactivity behaviors (ADHD-DSM-IV) and social behavior (CPSCS); in addition to maternal in person general questionnaires. Results: Outcomes showed acceptable psychometric characteristics and ADHD neuropsychological patterns were consistent with other ADHD findings. Long-term breastfeeding was associated with the improvement of all behavioral areas assessed. Maternal smoking during pregnancy was associated with lower cognitive scores.Conclusions: Assessing neurodevelopment in healthy preschoolers is feasible and necessary to investigate early effects of environmental factors and apply public health preventive policies.

The primary goal of artificial intelligence research is to develop a machine capable of learning to solve disparate real-world tasks autonomously, without relying on specialized problem-specific inputs. This dissertation suggests that such machines are realistic: If No Free Lunch theorems were to apply to all real-world problems, then the world would be utterly unpredictable. In response, the dissertation proposes the information-maximization principle, which claims that the optimal optimization methods make the best use of the information available to them. This principle results in a new algorithm, evolutionary annealing, which is shown to perform well especially in challenging problems with irregular structure. / text

Trajectoires développementales de l’IMC durant l’enfance: Une étude longitudinale sur 8 ans. Introduction : L’obésité infantile, origine de nombreux problèmes de santé, représente un grand défi en santé publique. Récemment, l’importance d’étudier l’évolution du surpoids durant l’enfance ainsi que les facteurs de risques précoces pour l’obésité a été reconnue. Les trajectoires développementales d’indice de masse corporelle (IMC) chez les jeunes représentent une approche innovatrice qui nous permet de mieux comprendre cette problématique importante. Objectifs: 1) Identifier des trajectoires développementales distinctes de groupes d’enfants selon leur IMC durant l’enfance, et 2) Explorer les facteurs de risques précoces qui prédisent l’appartenance de l’enfant à la trajectoire d’IMC le plus élevé Hypothèses: 1) On s’attend à retrouver un groupe d’enfants qui suit une trajectoire d’IMC élevée durant l’enfance. 2) On s’attend à ce que certaines caractéristiques de la mère (ex : tabac pendant la grossesse et IMC élevé), soient associées à l’appartenance de l’enfant au groupe ayant la trajectoire «IMC élevé ». Méthodes: Estimation des trajectoires développementales d’IMC d’enfants, dans un échantillon populationnel (n=1957) au Québec (ELDEQ). Les IMC ont été calculés à partir de données fournies par les mères des enfants et recueillis chaque année sur une durée de 8 ans. Des données propres à l’enfant sa mère, ainsi que socioéconomiques, ont étés recueillies. Une régression logistique multinomiale a été utilisée pour distinguer les enfants avec un IMC élevé des autres enfants, selon les facteurs de risques précoces. Les programmes PROC TRAJ (extension de SAS), SPSS (version 16), et SAS (version 9.1.3) ont été utilisés pour ces analyses. Résultats: Trois trajectoires d’IMC ont étés identifiées : IMC « bas-stable » (54,5%), IMC « modéré » (41,0%) et IMC « élevé et en hausse » (4,5%). Le groupe « élevé et en hausse » incluait des enfants pour qui l’IMC à 8 ans dépassait la valeur limite pour l’obésité. Les analyses de régression logistique ont révélé que deux facteurs de risques maternels étaient significativement associés avec la trajectoire “en hausse” par rapport aux deux autres groupes : le tabac durant la grossesse et le surpoids maternel. Conclusions: Des risques d’obésité infantile peuvent êtres identifiés dès la grossesse. Des études d’intervention sont requises pour identifier la possibilité de réduire le risque d’obésité chez l’enfant en ciblant le tabac et le surpoids maternelle durant la grossesse. Mots clés: Indice de masse corporelle (IMC), obésité infantile, trajectoires développementales de groupe, facteurs de risque précoce, étude populationnelle, tabac pendant la grossesse, obésité maternelle. / Developmental Trajectories of Body Mass Index in Early Childhood: An 8-Year Longitudinal Study. Introduction: Childhood obesity has become one of the greatest Public Health challenges this century, affecting not only developed nations, but increasingly low- and middle-income countries as well. Estimating developmental trajectories of Body Mass Index (BMI) during early childhood represents an innovative approach towards a better understanding of the development of this health problem. Objective: To identify groups of children with distinct developmental trajectories of Body Mass Index (BMI) between the ages of five months and eight years, and to identify early-life risk factors that distinguish children in an atypically elevated BMI trajectory group. Methods: Group-based developmental trajectories of BMI were estimated from annual maternal assessments (5 months to 8 years) in a large population sample (n=1957). Measures of height and weight, as well as family and child characteristics were obtained yearly from mothers. Multivariate logistic regression was used to distinguish children with elevated BMI from other children, using pre and early post-natal risk factors. Results: Three trajectories of BMI were identified: low-stable BMI (54.5%), moderate BMI (41.0%) and high-rising BMI (4.5%). The high-rising group included children whose BMI, at eight years of age, exceeded the cut-off value for obesity. Multinomial logit regression analyses revealed that two maternal risk factors were significantly associated with the high-rising BMI trajectory group as compared to both the low and moderate groups: smoking during pregnancy and maternal overweight. Conclusions: Antecedents of childhood obesity can be identified during pregnancy. Intervention studies are needed in order to test the possibility that targeting maternal smoking and maternal obesity during pregnancy would reduce the risk of childhood obesity in the offspring. Keywords: Body Mass Index (BMI), child obesity, Group-based developmental trajectories, early life predictors, population-based study, maternal smoking, maternal obesity.

Ce travail présente un dispositif de surveillance épidémiologique novateur en santé environnementale fondée sur des évènements de santé déclarée, au plus près de la nature des plaintes exprimées par les populations : des symptômes et une dégradation de la qualité de vie. Ce dispositif est proposé à partir d’une réflexion conceptuelle reposant sur une revue de la littérature sur les théories psychométriques et le modèle transactionnel du stress et d’une application de terrain portant sur la population (n=1 495) riveraine du site industriel chimique de Salindres (Gard). Ces travaux permettent de vérifier les principales propriétés métriques d’indicateurs de santé déclarée mesurés, par les instruments MOS SF-36 et SCL-90-R, dans le cadre d’une population exposée à une pollution environnementale. Ils vérifient également l’adéquation et la sensibilité de ces indicateurs pour étudier les facteurs de risque environnementaux qui se réfèrent aux représentations cognitives des risques, comme la perception sensoriels des stimuli émis par les sites industriels. Ils montrent, en outre, que la disponibilité de références nationales portant sur la mesure de ces indicateurs leur confère un critère de qualité essentiel d’interprétation des résultats. Ces travaux apportent enfin des éléments d’appréciation de l’acceptabilité de la mesure des indicateurs de santé déclarée dans la population, des recommandations sur la communication pour la mise en œuvre d’un tel dispositif et des pistes de réflexion pour la gestion. / This work presents an innovative epidemiological surveillance system in environmental health based on self-reported health indicators, closer to the complaints of the local population leaving around a source of environmental pollution: symptoms and loss of the quality of life. This approach is built up from a conceptual framework based on the psychometric theory and the transactional model of stress. A field application of this approach was carried out the population neighboring (n=1 495) the chemical industrial area of Salindres, Gard. This study verifies that self-reported health indicators measured with MOS SF-36 and SCL-90-R can be used to study populations exposed to multifactorial environmental risks. This study confirms that these indicators are relevant and sensitive when applied to environmental risk factors referring to the cognitive representations of risks, such as the sensorial perception of stimuli emitted by the industries. This work shows, besides, that the availability of national references for these indicators confers interpretability to these indicators, an essential quality criterion. This work discusses qualitatively the acceptability of measuring indicators of self-reported health in the population, proposes recommendations on communication for the setting of that kind of approach, and opens up avenues on management orientations.

Submitted by Cássia Santos (cassia.bcufg@gmail.com) on 2015-11-12T12:16:48Z No. of bitstreams: 2 Tese - Sabrina Sgambatti - 2015.pdf: 2847998 bytes, checksum: 069ae0ef61790cc3dd2564c3e7fa9120 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2015-11-12T12:21:22Z (GMT) No. of bitstreams: 2 Tese - Sabrina Sgambatti - 2015.pdf: 2847998 bytes, checksum: 069ae0ef61790cc3dd2564c3e7fa9120 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-11-12T12:21:22Z (GMT). No. of bitstreams: 2 Tese - Sabrina Sgambatti - 2015.pdf: 2847998 bytes, checksum: 069ae0ef61790cc3dd2564c3e7fa9120 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2015-07-17 / Background. Anticipating the introduction of the 10-valent pneumococcal conjugate vaccine (PCV10) on childhood National Immunization Program (NIP), an active population-based surveillance on pneumonia hospitalizations was conducted as a baseline, enabling a vaccination impact study. The objectives of the present research were: (i) to assess the reliability of the Hospital Information System of the Unified Health System (SIH-SUS) as a data source for assessing PCV10 impact on pneumonia; (ii) to measure the impact of vaccination with PCV10 in reducing the incidence of clinical and X-Ray confirmed pneumonia, in children residing in Goiânia municipality. Methods. In this study, we conducted an active prospective population-based surveillance on pneumonia in the post PCV10 vaccination period (2011-2013), in all 17 pediatric hospitals of Goiânia, with similar methodology used in the previous pneumonia surveillance during the pre vaccination period (2007-2009). Children aged 2-35 months of age, admitted to hospitalization with suspected diagnosis of pneumonia, were elegible for the survey. Clinical pneumonia and X-Ray confirmed pneumonia were the outcomes. The intervention was the PCV10, introduced in June 2010 in Goiania. Probabilistic linkage was performed between the SIH-SUS database (secondary data) and the active population surveillance (primary data) for the year 2012, to measure the agreement of case identification on pneumonia hospitalization rates between both data sources. To assess the impact of PCV10, annual incidence of clinical pneumonia and X-Ray confirmed pneumonia (per 100,000 population) and respective 95% confidence interval (95%CI) was estimated for the post vaccinations period and compared to the rates obtained for the pre vaccination period. The relative risk for pneumonia and respective 95%CI were calculated based on Poisson distribution. The percentage change in rates (1-relative risk) between pre and post vaccination periods was calculated. Results. Pneumonia incidence rates obtained by the SIH-SUS were statistically similar to those obtained by active population surveillance for children 2-23meses (p = 0.184). On the PCV10 impact evaluation study, the rates of hospitalization for clinical and RXT confirmed pneumonia in children under 24 months decreased 13.1% (from 5,728/100,000 to 4,976/100,000) and 25.4% (from 2,497/100,000 to 1,862/100,000), respectively, after routine immunization. / Introdução. Antecipando a introdução da vacina pneumocócica conjugada 10-valente (PCV10) no calendário de vacinação infantil do Programa Nacional de Imunizações (PNI), um estudo de vigilância de base populacional ativa foi conduzido como linha de base, possibilitando, assim, avaliar o impacto da vacinação nas hospitalizações por pneumonia. Assim, os objetivos desta investigação foram: (i) avaliar a confiabilidade do Sistema de Informações Hospitalares do Sistema Único de Saúde (SIH-SUS) como fonte de dados para estudos de avaliação de impacto da PCV10 nas pneumonias; (ii) avaliar o impacto da vacinação com a PCV10 na redução da incidência de hospitalizações de crianças com pneumonia clínica e confirmada por Raio-X de tórax (RXT), residentes no município de Goiânia. Métodos. Neste estudo, conduzimos uma vigilância populacional prospectiva, ativa, de pneumonias no período pós vacinal (2011-2013) em 17 hospitais pediátricos de Goiânia, com metodologia similar à conduzida em estudo anterior, no período pré vacinal (2007-2009). Foram elegíveis para o estudo crianças de 2 a 35 meses de idade, admitidas com com diagnóstico inicial de pneumonia. Os desfechos foram pneumonia clínica e pneumonia confirmada por RXT. A intervenção foi a PCV10, introduzida em junho de 2010 em Goiânia. A técnica de linkage probabilístico foi utilizada para vincular o banco de dados do SIH-SUS (dados secundários) e o da vigilância populacional ativa (dados primários) referentes ao ano de 2012, e desta forma, avaliar a concordância no diagnóstico e nas taxas de hospitalização por pneumonia entre as duas fontes de dados. Para avaliar o impacto da PCV10, calculou-se a incidência anual de pneumonia clínica e confirmada por RXT (por 100.000 habitantes) e respectivos intervalos de 95% de confiança (IC95%) para o período pós vacinal, e comparou-se com as taxas do período pré vacinal. O risco relativo para pneumonia e respectivos IC95% foram calculados com base na distribuição de Poisson. O percentual de mudança entre as taxas pré e pós vacinal foi calculado como 1-risco relativo. Resultados. As taxas de pneumonia obtidas pelo SIH-SUS foram estatisticamente similares às obtidas por vigilância populacional ativa para as crianças de 2-23meses (p=0,184). No estudo de avaliação do impacto da PCV10, as taxas de hospitalização por pneumonia clínica e confirmada por RXT em crianças menores de 24 meses reduziram 13.1% (de 5,728/100,000 para 4,976/100,000) e 25.4% (de 2,497/100,000 para 1,862/100,000), respectivamente, após a vacinação de rotina. Conclusões. Dados do SIH-SUS podem ser utilizados para avaliar o impacto da PCV10 nas hospilazações por pneumonia na infância. Após 3 anos de vacinação com a PCV10 em Goiânia, observou-se significante queda das taxas de hospitalização por pneumonia clinica e confirmada por RXT em crianças alvo do PNI.

Machine learning is spearheading progress for the field of artificial intelligence in terms of providing competition in strategy games to a human opponent, be it in a game of chess, Go or poker. A field of machine learning, which shows the most promising results in playing strategy games, is reinforcement learning. The next milestone for the current research lies in a computer game Starcraft II, which outgrows the previous ones in terms of complexity, and represents a potential new breakthrough in this field. The paper focuses on analysis of the problem, and suggests a solution incorporating a reinforcement learning algorithm A2C and hyperparameter optimization implementation PBT, which could mean a step forward for the current progress.