Following the pioneering work of Allenby and Ginter (1995) and Lenk et al.(1994); we propose in Section 2 a mixed effect model allowing for fixed and random effects as possible statistical solution to the problems mentioned above. Parameter estimation using a new, efficient variant of a Markov Chain Monte Carlo method will be discussed in Section 3 together with problems of model comparison techniques in the context of random effect models. Section 4 presents an application of the former to a brand-price trade-off study from the Austrian mineral water market. (author's abstract) / Series: Forschungsberichte / Institut für Statistik

A modelagem de predição de altura comumente exige um amplo conjunto de dados para a etapa de construção e ajuste. Ainda que este tipo de conjunto de dados tenha uma estrutura hierárquica natural, organizada pelas diferentes fazendas, talhões, parcelas, e etc., os modelos de regressão clássicos não consideram a possível variação dos parâmetros, entre os diversos grupos hierárquicos. Os modelos de efeitos mistos, em compensação, podem suportar essa variação, assumindo alguns dos parâmetros dos modelos como sendo estocásticos, além de mostrarem potencial com a possibilidade de diminuição de amostras. Esta técnica permite que a variação interindividual seja explicada considerando parâmetros de efeitos fixos (comuns à população) e parâmetros de efeitos aleatórios (específicos para cada indivíduo). Logo, é natural esperar que em povoamentos florestais com alta variação entre indivíduos, o modelo de efeito misto tenha desempenho superior ao modelo de efeito fixo. Por esta razão, os plantios de Tectona grandis L.f. podem ser considerados como uma população interessante para a modelagem de efeitos aleatórios, uma vez que tal espécie apresenta heterogeneidade de crescimento, sensibilidade à fertilidade e acidez do solo, e a maioria dos seus plantios estabelecidos no Brasil são seminais. Desta maneira este trabalho verifica o ajuste de modelos de efeitos mistos aplicados aos dados de altura total em plantios comerciais de Tectona grandis L.f, localizados no estado do Mato Grosso, com o objetivo na redução do número de amostras quando comparado ao modelo de efeitos fixos. Após a seleção do modelo linear de efeito fixo mais apropriado, testou-se quais dos coeficientes tem efeito aleatório nos diferentes agrupamentos dos dados. Em seguida, selecionou-se o grupo onde o desempenho do modelo de efeito misto em termos de ajuste e predição foi o melhor possível. Por fim, foi verificado a capacidade preditiva dos modelos ajustados por meio de processos de simulação e validação cruzada. Os resultados mostraram que o modelo misto calibrado fornece predições mais confiáveis do que a parte fixa. Este benefício ocorre mesmo ao longo das gradativas diminuições do número de árvores disponíveis para ajuste dentro conjunto de dados teste separados para a calibração do modelo misto. É possível concluir que o modelo calibrado ajustado por talhão, ao invés da parcela, propicia pouca perda de precisão. / Height prediction modeling commonly requires a broad set of data for the construction and adjustment step. Although this type of data set has a natural hierarchical structure, organized by the different farms, plots, plots, etc., the classical regression models do not consider the possible variation of the parameters among the hierarchical groups. The mixed effects models, in compensation, can support this variation, assuming some of the parameters of the models as being stochastic, besides showing potential with the possibility of sample reduction. This technique allows the interindividual variation to be explained considering parameters of fixed effects (common to the population) and parameters of random effects (specific for each individual). Therefore, it is natural to expect that in forest stands with high variation among individuals, the mixed effect model performs better than the fixed effect model. For this reason, the plantations of Tectona grandis L.f. can be considered as an interesting population for the modeling of random effects, since this species presents possible heterogeneity of growth since it is sensitive to the fertility and acidity of the soil, and most of its plantations established in Brazil are seminal. This work verifies the adjustment of mixed effects models applied to total height data in commercial plantations of Tectona grandis L.f, located in the state of Mato Grosso, with the objective of reducing the number of samples when compared to the fixed effects model. After selecting the most appropriate linear model of fixed effect, we tested which of the coefficients have random effect in the different groupings of the data. Then, we selected the group where the performance of the mixed effect model in terms of fit and prediction was the best possible. Finally, the predictive capacity of the adjusted models was verified through simulation and cross-validation processes. The results showed that the calibrated mixed model provides more reliable predictions than the fixed part. This benefit occurs even along the gradual decreases in the number of trees available to fit into separate set of test data for the calibration of the mixed model. It is possible to conclude that the calibrated model adjusted by stand, instead of the plot, provides little loss of precision.

A modelagem de predição de altura comumente exige um amplo conjunto de dados para a etapa de construção e ajuste. Ainda que este tipo de conjunto de dados tenha uma estrutura hierárquica natural, organizada pelas diferentes fazendas, talhões, parcelas, e etc., os modelos de regressão clássicos não consideram a possível variação dos parâmetros, entre os diversos grupos hierárquicos. Os modelos de efeitos mistos, em compensação, podem suportar essa variação, assumindo alguns dos parâmetros dos modelos como sendo estocásticos, além de mostrarem potencial com a possibilidade de diminuição de amostras. Esta técnica permite que a variação interindividual seja explicada considerando parâmetros de efeitos fixos (comuns à população) e parâmetros de efeitos aleatórios (específicos para cada indivíduo). Logo, é natural esperar que em povoamentos florestais com alta variação entre indivíduos, o modelo de efeito misto tenha desempenho superior ao modelo de efeito fixo. Por esta razão, os plantios de Tectona grandis L.f. podem ser considerados como uma população interessante para a modelagem de efeitos aleatórios, uma vez que tal espécie apresenta heterogeneidade de crescimento, sensibilidade à fertilidade e acidez do solo, e a maioria dos seus plantios estabelecidos no Brasil são seminais. Desta maneira este trabalho verifica o ajuste de modelos de efeitos mistos aplicados aos dados de altura total em plantios comerciais de Tectona grandis L.f, localizados no estado do Mato Grosso, com o objetivo na redução do número de amostras quando comparado ao modelo de efeitos fixos. Após a seleção do modelo linear de efeito fixo mais apropriado, testou-se quais dos coeficientes tem efeito aleatório nos diferentes agrupamentos dos dados. Em seguida, selecionou-se o grupo onde o desempenho do modelo de efeito misto em termos de ajuste e predição foi o melhor possível. Por fim, foi verificado a capacidade preditiva dos modelos ajustados por meio de processos de simulação e validação cruzada. Os resultados mostraram que o modelo misto calibrado fornece predições mais confiáveis do que a parte fixa. Este benefício ocorre mesmo ao longo das gradativas diminuições do número de árvores disponíveis para ajuste dentro conjunto de dados teste separados para a calibração do modelo misto. É possível concluir que o modelo calibrado ajustado por talhão, ao invés da parcela, propicia pouca perda de precisão. / Height prediction modeling commonly requires a broad set of data for the construction and adjustment step. Although this type of data set has a natural hierarchical structure, organized by the different farms, plots, plots, etc., the classical regression models do not consider the possible variation of the parameters among the hierarchical groups. The mixed effects models, in compensation, can support this variation, assuming some of the parameters of the models as being stochastic, besides showing potential with the possibility of sample reduction. This technique allows the interindividual variation to be explained considering parameters of fixed effects (common to the population) and parameters of random effects (specific for each individual). Therefore, it is natural to expect that in forest stands with high variation among individuals, the mixed effect model performs better than the fixed effect model. For this reason, the plantations of Tectona grandis L.f. can be considered as an interesting population for the modeling of random effects, since this species presents possible heterogeneity of growth since it is sensitive to the fertility and acidity of the soil, and most of its plantations established in Brazil are seminal. This work verifies the adjustment of mixed effects models applied to total height data in commercial plantations of Tectona grandis L.f, located in the state of Mato Grosso, with the objective of reducing the number of samples when compared to the fixed effects model. After selecting the most appropriate linear model of fixed effect, we tested which of the coefficients have random effect in the different groupings of the data. Then, we selected the group where the performance of the mixed effect model in terms of fit and prediction was the best possible. Finally, the predictive capacity of the adjusted models was verified through simulation and cross-validation processes. The results showed that the calibrated mixed model provides more reliable predictions than the fixed part. This benefit occurs even along the gradual decreases in the number of trees available to fit into separate set of test data for the calibration of the mixed model. It is possible to conclude that the calibrated model adjusted by stand, instead of the plot, provides little loss of precision.

Esta tese investiga os efeitos agudos da poluição atmosférica no pico de fluxo expiratório (PFE) de escolares com idades entre 6 e 15 anos, residentes em municípios da Amazônia Brasileira. O primeiro artigo avaliou os efeitos do material particulado fino (PM2,5) no PFE de 309 escolares do município de Alta Floresta, Mato Grosso (MT), durante a estação seca de 2006. Modelos de efeitos mistos foram estimados para toda a amostra e estratificados por turno escolar e presença de sintomas de asma. O segundo artigo expõe as estratégias utilizadas para a determinação da função de variância do erro aleatório dos modelos de efeitos mistos. O terceiro artigo analisa os dados do estudo de painel com 234 escolares, realizado na estação seca de 2008 em Tangará da Serra, MT. Avaliou-se os efeitos lineares e com defasagem distribuída (PDLM) do material particulado inalável (PM10), do PM2,5 e do Black Carbon (BC) no PFE de todos os escolares e estratificados por grupos de idade. Nos três artigos, os modelos de efeitos mistos foram ajustados por tendência temporal, temperatura, umidade e características individuais. Os modelos também consideraram o ajuste da autocorrelação residual e da função de variância do erro aleatório. Quanto às exposições, foram avaliados os efeitos das exposições de 5hs, 6hs, 12hs e 24hs, no dia corrente, com defasagens de 1 a 5 dias e das médias móveis de 2 e 3 dias. No que se refere aos resultados de Alta Floresta, os modelos para todas as crianças indicaram reduções no PFE variando de 0,26 l/min (IC95%: 0,49; 0,04) a 0,38 l/min (IC95%: 0,71; 0,04), para cada aumento de 10g/m3 no PM2,5. Não foram observados efeitos significativos da poluição no grupo das crianças asmáticas. A exposição de 24hs apresentou efeito significativo no grupo de alunos da tarde e no grupo dos não asmáticos. A exposição de 0hs a 5:30hs foi significativa tanto para os alunos da manhã quanto para a tarde. Em Tangará da Serra, os resultados mostraram reduções significativas do PFE para aumentos de 10 unidades do poluente, principalmente para as defasagens de 3, 4 e 5 dias. Para o PM10, as reduções variaram de 0,15 (IC95%: 0,29; 0,01) a 0,25 l/min (IC95%: 0,40 ; 0,10). Para o PM2,5, as reduções estiveram entre 0,46 l/min (IC95%: 0,86 to 0,06 ) e 0,54 l/min (IC95%: 0,95; 0,14). E no BC, a redução foi de aproximadamente 0,014 l/min. Em relação ao PDLM, efeitos mais importantes foram observados nos modelos baseados na exposição do dia corrente até 5 dias passados. O efeito global foi significativo apenas para o PM10, com redução do PFE de 0,31 l/min (IC95%: 0,56; 0,05). Esta abordagem também indicou efeitos defasados significativos para todos os poluentes. Por fim, o estudo apontou as crianças de 6 a 8 anos como grupo mais sensível aos efeitos da poluição. Os achados da tese sugerem que a poluição atmosférica decorrente da queima de biomassa está associada a redução do PFE de crianças e adolescentes com idades entre 6 e 15 anos, residentes na Amazônia Brasileira. / This thesis investigates the acute effects of air pollution on peak expiratory flow (PEF) of schoolchildren between the ages of 6 and 15, living in Brazilian Amazon municipalities. The first article evaluated the effects of fine particulate matter (PM2.5) on PEF of 309 schoolchildren in the municipality of Alta Floresta, Mato Grosso (MT), during the dry season in 2006. Mixed effect models were estimated for the whole sample and stratified by the time of the day children attended school, and also by the presence of asthma symptoms. The second article describes the strategies used to determine the random error variance function of mixed effect models. The third one analyzes the data of the panel study with a sample of 234 schoolchildren carried out in Tangará da Serra, MT, during the dry season in 2008. Linear effects and the ones with distributed lag (PDLM) of inhalable particulate matter (PM10), PM2.5 and Black Carbon (BC) were assessed for the whole sample and stratified by age. In all three articles, the mixed effect models were adjusted by time trend, temperature, humidity and personal characteristics. The models also considered the adjustment of the residual autocorrelation and of the random error variance function. Regarding the exposures, its effects were evaluated in 5hs, 6hs, 12hs and 24hs, on the current day, with lags of 1 to 5 days and moving averages of 2 and 3 days. According to results in Alta Floresta, the models for all the children indicated reductions in the PEF varying from 0.26 l/min (CI95%: 0.49; 0.04) to 0.38 l/min (CI95%: 0.71; 0.04), for each increase of 10g/m3 on PM2.5. Significant effects of pollution were not observed in the group of asthmatic children. The 24-hour exposure presented significant effects in the group of students who attended school in the afternoon and in the group of non-asthmatic ones. The exposure from midnight to 5:30 A.M. was significant both to students who attended school in the morning and the ones who studied in the afternoon. In Tangará da Serra, the results showed significant reductions on the PEF for increases of 10 units of pollutants, mainly for lagged exposures of 3, 4 and 5 days. For PM10, the reductions varied from 0.15 (CI95%: 0.29; 0.01) to 0.25 l/min (CI95%: 0.40; 0.10). For PM2.5, the reductions ranged from 0.46 l/min (CI95%: 0.86 to 0.06) to 0.54 l/min (CI95%:0.95; 0.14). And for BC, the reduction was about 0.014 l/min. In relation to PDLM, more important effects were noticed in models based on the exposure of the current day until 5 past days. The global effect was significant only for PM10, with PEF reduction of 0.31 l/min (CI95%: 0.56; 0.05). This approach also indicated significant lagged effects for all pollutants. In the end, this study observed that the children between 6 and 8 years old were the most vulnerable to pollution effects. These findings in the thesis suggest that air pollution due to biomass burning is associated to PEF reduction in children and teenagers between the ages of 6 and 15, living in the Brazilian Amazon.

Esta tese investiga os efeitos agudos da poluição atmosférica no pico de fluxo expiratório (PFE) de escolares com idades entre 6 e 15 anos, residentes em municípios da Amazônia Brasileira. O primeiro artigo avaliou os efeitos do material particulado fino (PM2,5) no PFE de 309 escolares do município de Alta Floresta, Mato Grosso (MT), durante a estação seca de 2006. Modelos de efeitos mistos foram estimados para toda a amostra e estratificados por turno escolar e presença de sintomas de asma. O segundo artigo expõe as estratégias utilizadas para a determinação da função de variância do erro aleatório dos modelos de efeitos mistos. O terceiro artigo analisa os dados do estudo de painel com 234 escolares, realizado na estação seca de 2008 em Tangará da Serra, MT. Avaliou-se os efeitos lineares e com defasagem distribuída (PDLM) do material particulado inalável (PM10), do PM2,5 e do Black Carbon (BC) no PFE de todos os escolares e estratificados por grupos de idade. Nos três artigos, os modelos de efeitos mistos foram ajustados por tendência temporal, temperatura, umidade e características individuais. Os modelos também consideraram o ajuste da autocorrelação residual e da função de variância do erro aleatório. Quanto às exposições, foram avaliados os efeitos das exposições de 5hs, 6hs, 12hs e 24hs, no dia corrente, com defasagens de 1 a 5 dias e das médias móveis de 2 e 3 dias. No que se refere aos resultados de Alta Floresta, os modelos para todas as crianças indicaram reduções no PFE variando de 0,26 l/min (IC95%: 0,49; 0,04) a 0,38 l/min (IC95%: 0,71; 0,04), para cada aumento de 10g/m3 no PM2,5. Não foram observados efeitos significativos da poluição no grupo das crianças asmáticas. A exposição de 24hs apresentou efeito significativo no grupo de alunos da tarde e no grupo dos não asmáticos. A exposição de 0hs a 5:30hs foi significativa tanto para os alunos da manhã quanto para a tarde. Em Tangará da Serra, os resultados mostraram reduções significativas do PFE para aumentos de 10 unidades do poluente, principalmente para as defasagens de 3, 4 e 5 dias. Para o PM10, as reduções variaram de 0,15 (IC95%: 0,29; 0,01) a 0,25 l/min (IC95%: 0,40 ; 0,10). Para o PM2,5, as reduções estiveram entre 0,46 l/min (IC95%: 0,86 to 0,06 ) e 0,54 l/min (IC95%: 0,95; 0,14). E no BC, a redução foi de aproximadamente 0,014 l/min. Em relação ao PDLM, efeitos mais importantes foram observados nos modelos baseados na exposição do dia corrente até 5 dias passados. O efeito global foi significativo apenas para o PM10, com redução do PFE de 0,31 l/min (IC95%: 0,56; 0,05). Esta abordagem também indicou efeitos defasados significativos para todos os poluentes. Por fim, o estudo apontou as crianças de 6 a 8 anos como grupo mais sensível aos efeitos da poluição. Os achados da tese sugerem que a poluição atmosférica decorrente da queima de biomassa está associada a redução do PFE de crianças e adolescentes com idades entre 6 e 15 anos, residentes na Amazônia Brasileira. / This thesis investigates the acute effects of air pollution on peak expiratory flow (PEF) of schoolchildren between the ages of 6 and 15, living in Brazilian Amazon municipalities. The first article evaluated the effects of fine particulate matter (PM2.5) on PEF of 309 schoolchildren in the municipality of Alta Floresta, Mato Grosso (MT), during the dry season in 2006. Mixed effect models were estimated for the whole sample and stratified by the time of the day children attended school, and also by the presence of asthma symptoms. The second article describes the strategies used to determine the random error variance function of mixed effect models. The third one analyzes the data of the panel study with a sample of 234 schoolchildren carried out in Tangará da Serra, MT, during the dry season in 2008. Linear effects and the ones with distributed lag (PDLM) of inhalable particulate matter (PM10), PM2.5 and Black Carbon (BC) were assessed for the whole sample and stratified by age. In all three articles, the mixed effect models were adjusted by time trend, temperature, humidity and personal characteristics. The models also considered the adjustment of the residual autocorrelation and of the random error variance function. Regarding the exposures, its effects were evaluated in 5hs, 6hs, 12hs and 24hs, on the current day, with lags of 1 to 5 days and moving averages of 2 and 3 days. According to results in Alta Floresta, the models for all the children indicated reductions in the PEF varying from 0.26 l/min (CI95%: 0.49; 0.04) to 0.38 l/min (CI95%: 0.71; 0.04), for each increase of 10g/m3 on PM2.5. Significant effects of pollution were not observed in the group of asthmatic children. The 24-hour exposure presented significant effects in the group of students who attended school in the afternoon and in the group of non-asthmatic ones. The exposure from midnight to 5:30 A.M. was significant both to students who attended school in the morning and the ones who studied in the afternoon. In Tangará da Serra, the results showed significant reductions on the PEF for increases of 10 units of pollutants, mainly for lagged exposures of 3, 4 and 5 days. For PM10, the reductions varied from 0.15 (CI95%: 0.29; 0.01) to 0.25 l/min (CI95%: 0.40; 0.10). For PM2.5, the reductions ranged from 0.46 l/min (CI95%: 0.86 to 0.06) to 0.54 l/min (CI95%:0.95; 0.14). And for BC, the reduction was about 0.014 l/min. In relation to PDLM, more important effects were noticed in models based on the exposure of the current day until 5 past days. The global effect was significant only for PM10, with PEF reduction of 0.31 l/min (CI95%: 0.56; 0.05). This approach also indicated significant lagged effects for all pollutants. In the end, this study observed that the children between 6 and 8 years old were the most vulnerable to pollution effects. These findings in the thesis suggest that air pollution due to biomass burning is associated to PEF reduction in children and teenagers between the ages of 6 and 15, living in the Brazilian Amazon.

Suite aux progrès techniques et méthodologiques dans le secteur de la modélisation, l’apport de ces approches est désormais reconnu par l’ensemble des acteurs de la recherche clinique et pourrait avoir un rôle clé dans la recherche sur les maladies progressives. Parmi celles-ci les études pharmacométriques (PMX) sont rarement utilisées pour répondre aux hypothèses posées dans le cadre d’études dites de confirmation. Parmi les raisons évoquées, les analyses PMX traditionnelles ignorent l'incertitude associée à la structure du modèle lors de la génération d'inférence statistique. Or, ignorer l’étape de sélection du modèle peut aboutir à des intervalles de confiance trop optimistes et à une inflation de l’erreur de type I. Pour y remédier, nous avons étudié l’apport d’approches PMX innovantes dans les études de choix de dose. Le « model averaging » couplée à un test du rapport de « vraisemblance combiné » a montré des résultats prometteurs et tend à promouvoir l’utilisation de la PMX dans les études de choix de dose. Pour les études dites d’apprentissage, les approches de modélisation sont utilisées pour accroitre les connaissances associées aux médicaments, aux mécanismes et aux maladies. Dans cette thèse, les mérites de l’analyse PMX ont été évalués dans le cadre de la maladie de Parkinson. En combinant la théorie des réponses aux items à un modèle longitudinal, l’analyse PMX a permis de caractériser adéquatement la progression de la maladie tout en tenant compte de la nature composite du biomarqueur. Pour conclure, cette thèse propose des méthodes d’analyses PMX innovantes pour faciliter le développement des médicaments et/ou les décisions des autorités réglementaires. / In the mid-1990, model-based approaches were mainly used as supporting tools for drug development. Restricted to the “rescue mode” in situations of drug development failure, the impact of model-based approaches was relatively limited. Nowadays, the merits of these approaches are widely recognised by stakeholders in healthcare and have a crucial role in drug development for progressive diseases. Despite their numerous advantages, model-based approaches present important drawbacks limiting their use in confirmatory trials. Traditional pharmacometric (PMX) analyses relies on model selection, and consequently ignores model structure uncertainty when generating statistical inference. The problem of model selection is potentially leading to over-optimistic confidence intervals and resulting in a type I error inflation. Two projects of this thesis aimed at investigating the value of innovative PMX approaches to address part of these shortcomings in a hypothetical dose-finding study for a progressive disorder. The model averaging approach coupled to a combined likelihood ratio test showed promising results and represents an additional step towards the use of PMX for primary analysis in dose-finding studies. In the learning phase, PMX is a key discipline with applications at every stage of drug development to gain insight into drug, mechanism and disease characteristics with the ultimate goal to aid efficient drug development. In this thesis, the merits of PMX analysis were evaluated, in the context of Parkinson’s disease. An item-response theory longitudinal model was successfully developed to precisely describe the disease progression of Parkinson’s disease patients while acknowledging the composite nature of a patient-reported outcome. To conclude, this thesis enhances the use of PMX to aid efficient drug development and/or regulatory decisions in drug development.

No description available.

This study aimed at evaluating the agreement of spaceborne Light Detection and Ranging (lidar) ICESat-2 canopy height with Airborne Laser Scanning (ALS) derived canopy height to inform about the performance of ICESat-2 canopy height metrics and understand its uncertainties and utilities. The agreement was assessed for different forest types, physiographic regions, a range of percent canopy cover, and diverse disturbance histories. Results of this study suggest that best agreements are found using strong beam data collected at night for canopy height retrieval using ICESat-2. The ICESat-2 showed great potential for estimating canopy heights, particularly in evergreen forests with high canopy cover. Statistical models were developed using fixed-effects and mixed-effects modeling approaches to predict ALS canopy height metrics using ICESat-2 parameters and other attributes. Overall, ICESat-2 showed good agreement with ALS canopy height and showed its predictive ability to characterize canopy height. The outcome of this study will help the scientific community understand the capabilities and limitations of ICESat-2 canopy heights; the study also provides a new approach to obtain wall-to-wall ALS standard canopy height maps at landscape level.

Cette thèse est consacrée aux méthodes statistiques d’extrapolation dans les essais de recherche de dose en pédiatrie. Dans un premier temps, nous réalisons une revue systématique de la littérature sur le sujet. Elle met en évidence la nécessité de proposer de nouvelles méthodes pour la conception des études d’escalade de dose chez l’enfant. Nous apportons des réponses à cette problématique en exploitant l’information disponible chez l’adulte. Dans une première série de travaux, nous étudions l’intérêt de la prédiction des paramètres pharmacocinétiques (PK) en pédiatrie à l’aide de méthodes d’extrapolation : l’allométrie et la maturation. Cette évaluation est réalisée à partir de données PK chez l’adulte et l’enfant pour la méfloquine. Faisant appel aux paramètres prédits, nous développons une approche pour choisir les temps de prélèvements (design) d’une étude PK. Nous recommandons un design obtenu par optimisation grâce à la méthode de D-optimalité en utilisant le logiciel PFIM. Ce design est ensuite validé à l’aide de simulations sur différents modèles. Une seconde série de travaux nous amène à proposer des recommandations pour la planification d’un essai de recherche de dose. Nous avançons d’abord des techniques pour choisir les doses à tester grâce à l’utilisation des données adultes et de l’extrapolation. Nous proposons ensuite une méthode proche de la méta-analyse pour prédire les probabilités de toxicités pour chaque dose. Enfin, nous employons la méthode de l’Effective sample size afin de construire une loi a priori lors de l’utilisation d’une estimation bayésienne. Nous validons ces recommandations sur une étude de cas en utilisant une méthode d’escalade de dose, la méthode de réévaluation séquentielle bivariée, pour laquelle nous évaluons à la fois la toxicité et l’efficacité. A partir de l’exemple de la molécule erlotinib, nous effectuons une série de simulations sur plusieurs scénarios afin d’illustrer les performances de la planification. / This thesis addresses extrapolation techniques for statistical models for dose-finding studies in pediatrics. After a litterature review on these clinical trials, we observed the need of methodological propositions for the planification of dose- finding studies in pediatrics. We deal with this issue using information from the adult population. In a first research, the objectives are to design a pharmacokinetic (PK) study by using information from adults and evaluate the robustness of the recommended design through a case study of mefloquine. Pediatric PK parameters are predicted from adult PK using extrapolation functions such as allometry and maturation. A D-optimal design for children is obtained with PFIM by assuming the extrapolated design. The robustness of the recommended design is evaluated in a simulation study with four different models and is compared to the empirical design used for the pediatric data. In a second research, we propose a global approach to conduct a pediatric dose-finding clinical trial using extrapolation from adult information. First, we extrapolate the dose-range from adults using allometry and maturation. Then, using an approach to meta-analysis, we choose the initial probabilities of toxicity for each dose. Finally, we use the effective sample size method to choose the prior distribution of parameters in a Bayesian setting. We perform a simulation study based on the molecule erlotinib to evaluate the performances of this global approach.

Le concept d’habitat, qui définit le lieu de vie des organismes par des conditions abiotiques et biotiques, est déterminant pour étudier les relations entre les organismes et leur environnement. La sélection d’habitat est le processus à travers lequel l’organisme va choisir l’habitat où il se trouve en fonction des différents habitats disponibles autour de lui. Cette sélection va dépendre d’un choix individuel, qui est propre à l’organisme (ex. son comportement), et d’un choix commun, qui est observable chez des organismes qui partagent des traits communs (ex. les individus d’une même espèce). Les modèles spécifiques de sélection d’habitat cherchent à expliquer et prédire ce choix commun, et sont notamment utilisés pour les cours d’eau dans les outils d'aide à la définition de débits écologiques. Pour les poissons de rivière, la plupart des modèles spécifiques à l’échelle du microhabitat sont peu transférables à d’autres rivières. En effet, ils sont construits à partir de données d’abondance échantillonnées dans le même site pendant quelques campagnes. Afin d’améliorer la qualité prédictive de ces modèles, j’ai développé une approche prometteuse de modélisation multi-sites et multi-campagnes permettant à la fois de considérer la réponse non linéaire de la sélection et la surdispersion des données d’abondance. A partir de suivis individuels par télémétrie, j’ai montré la pertinence des modèles de sélection spécifiques malgré la forte variabilité individuelle observée. Finalement, la sélection d’habitat étant dépendante de processus structurant les communautés et agissant à l’échelle du paysage, telle que la dispersion des individus, j’ai mis en évidence l’intérêt d’utiliser des techniques légères d’échantillonnage comme les observations par plongée pour caractériser les structures des communautés et leurs répartitions spatiales. Ces techniques permettront alors d’étudier l’influence des processus du paysage sur les modèles de sélection d’habitat / The habitat concept, which defines the place where organisms live, is composed by abiotic and biotic conditions and differs for examples between species or activities. The habitat selection is the process where organisms choose the habitat to live in function of all habitats available around them. This habitat selection depends on an individual choice related to the organism, for example its behavior and a common choice related to organisms sharing common traits as individuals from the same species. Specific habitat selection models are developed to understand and represent this common choice and used to build ecological flow tools. For freshwater fish, most of specific habitat selection models have low transferability between reaches and rivers. Indeed, they are built from abundance data and sampled in the same study reach during few numbers of surveys. In order to improve predictive quality of models, I developed an attractive modelling approach, both multi-reach and multi-survey, involving the non-linear response of habitat selection and abundance data overdispersion. Then, despite the high individual variability of habitat selection, I showed, from telemetry data, the relevance of developing specific habitat selection models. Finally, as the habitat selection is also depending on processes which influence community structures at the landscape scale (e.g. dispersal), I demonstrate the benefits of sampling methods such as snorkeling to characterize community structures and their longitudinal distributions at a large spatial scale. These techniques will allow studying the influence of landscape processes on habitat selection models.