Modelos preditivos estimam o risco de eventos ou agravos relacionados à saúde e podem ser utilizados como ferramenta auxiliar em tomadas de decisão por gestores e profissionais de saúde. Algoritmos de machine learning (ML), por sua vez, apresentam potencial para identificar relações complexas e não-lineares presentes nos dados, com consequências positivas na performance preditiva desses modelos. A presente pesquisa objetivou aplicar técnicas supervisionadas de ML e comparar sua performance em problemas de classificação e de regressão para predizer respostas de interesse para a saúde pública e a medicina. Os resultados e discussão estão organizados em três artigos científicos. O primeiro apresenta um tutorial para o uso de ML em pesquisas de saúde, utilizando como exemplo a predição do risco de óbito em até 5 anos (frequência do desfecho 15%; n=395) para idosos do estudo \"Saúde, Bem-estar e Envelhecimento\" (n=2.677), segundo variáveis relacionadas ao seu perfil demográfico, socioeconômico e de saúde. Na etapa de aprendizado, cinco algoritmos foram aplicados: regressão logística com e sem penalização, redes neurais, gradient boosted trees e random forest, cujos hiperparâmetros foram otimizados por validação cruzada (VC) 10-fold. Todos os modelos apresentaram área abaixo da curva (AUC) ROC (Receiver Operating Characteristic) maior que 0,70. Para aqueles com maior AUC ROC (redes neurais e regressão logística com e sem penalização) medidas de qualidade da probabilidade predita foram avaliadas e evidenciaram baixa calibração. O segundo artigo objetivou predizer o risco de tempo de vida ajustado pela qualidade de vida de até 30 dias (frequência do desfecho 44,7%; n=347) em pacientes com câncer admitidos em Unidade de Terapia Intensiva (UTI) (n=777), mediante características obtidas na admissão do paciente à UTI. Seis algoritmos (regressão logística com e sem penalização, redes neurais, árvore simples, gradient boosted trees e random forest) foram utilizados em conjunto com VC aninhada para estimar hiperparâmetros e avaliar performance preditiva. Todos os algoritmos, exceto a árvore simples, apresentaram discriminação (AUC ROC > 0,80) e calibração satisfatórias. Para o terceiro artigo, características socioeconômicas e demográficas foram utilizadas para predizer a expectativa de vida ao nascer de municípios brasileiros com mais de 10.000 habitantes (n=3.052). Para o ajuste do modelo preditivo, empregou-se VC aninhada e o algoritmo Super Learner (SL), e para a avaliação de performance, o erro quadrático médio (EQM). O SL apresentou desempenho satisfatório (EQM=0,17) e seu vetor de valores preditos foi utilizado para a identificação de overachievers (municípios com expectativa de vida superior à predita) e underachievers (município com expectativa de vida inferior à predita), para os quais características de saúde foram comparadas, revelando melhor desempenho em indicadores de atenção primária para os overachievers e em indicadores de atenção secundária para os underachievers. Técnicas para a construção e avaliação de modelos preditivos estão em constante evolução e há poucas justificativas teóricas para se preferir um algoritmo em lugar de outro. Na presente tese, não foram observadas diferenças substanciais no desempenho preditivo dos algoritmos aplicados aos problemas de classificação e de regressão analisados. Espera-se que a maior disponibilidade de dados estimule a utilização de algoritmos de ML mais flexíveis em pesquisas de saúde futuras. / Predictive models estimate the risk of health-related events or injuries and can be used as an auxiliary tool in decision-making by public health officials and health care professionals. Machine learning (ML) algorithms have the potential to identify complex and non-linear relationships, with positive implications in the predictive performance of these models. The present research aimed to apply various ML supervised techniques and compare their performance in classification and regression problems to predict outcomes of interest to public health and medicine. Results and discussion are organized into three articles. The first, presents a tutorial for the use of ML in health research, using as an example the prediction of death up to 5 years (outcome frequency=15%; n=395) in elderly participants of the study \"Saúde, Bemestar e Envelhecimento\" (n=2,677), using variables related to demographic, socioeconomic and health characteristics. In the learning step, five algorithms were applied: logistic regression with and without regularization, neural networks, gradient boosted trees and random forest, whose hyperparameters were optimized by 10-fold cross-validation (CV). The area under receiver operating characteristic (AUROC) curve was greater than 0.70 for all models. For those with higher AUROC (neural networks and logistic regression with and without regularization), the quality of the predicted probability was evaluated and it showed low calibration. The second article aimed to predict the risk of quality-adjusted life up to 30 days (outcome frequency=44.7%; n=347) in oncologic patients admitted to the Intensive Care Unit (ICU) (n=777), using patients\' characteristics obtained at ICU admission. Six algorithms (logistic regression with and without regularization, neural networks, basic decision trees, gradient boosted trees and random forest) were used with nested CV to estimate hyperparameters values and to evaluate predictive performance. All algorithms, with exception of basic decision trees, presented acceptable discrimination (AUROC > 0.80) and calibration. For the third article, socioeconomic and demographic characteristics were used to predict the life expectancy at birth of Brazilian municipalities with more than 10,000 inhabitants (n=3,052). Nested CV and the Super Learner (SL) algorithm were used to adjust the predictive model, and for evaluating performance, the mean squared error (MSE). The SL showed good performance (MSE=0.17) and its vector of predicted values was used for the identification of underachievers and overachievers (i.e. municipalities showing worse and better outcome than predicted, respectively). Health characteristics were analyzed revealing that overachievers performed better on primary health care indicators, while underachievers fared better on secondary health care indicators. Techniques for constructing and evaluating predictive models are constantly evolving and there is scarce theoretical justification for preferring one algorithm over another. In this thesis no substantial differences were observed in the predictive performance of the algorithms applied to the classification and regression problems analyzed herein. It is expected that increase in data availability will encourage the use of more flexible ML algorithms in future health research.

Introdução: A prostatectomia radical é uma forma definitiva de tratamento do câncer localizado da próstata. Aproximadamente 40% dos pacientes tratados pela prostatectomia radical apresentaram progressão da doença, mesmo com a doença confinada ao órgão. Acurácia previsão do risco de recorrência é útil para se considerar uma terapia adjuvante ou alguma forma de tratamento investigacional. Observa-se, atualmente, que fatores anatomopatológicos e clínicos são insuficientes para o estadiamento preciso do câncer da próstata. Alguns estudos têm ressaltado a influência dos processos de adesão celular como pré-requisitos para invasão tumoral e metástase, sendo as moléculas CD44 e a família Ezrin-Radixin-Moesin (ERM), citadas como facilitadoras destes processos. Objetivos: Avaliar a expressão do complexo protéico CD44/ERM, no câncer da próstata, bem como sua associação com os fatores prognósticos e sobrevida nesta neoplasia. Material e Métodos: Foram analisados retrospectivamente os blocos de parafina de 100 pacientes, portadores de câncer localizado da próstata comprovado por biópsia transretal da próstata, submetidos à prostatectomia radical, no período de 1995 a 2000, do Departamento de Anatomia Patológica do Hospital Geral de Fortaleza e Laboratório Biopse. As lâminas de todos os casos selecionados para este estudo foram coradas com hematoxilina e eosina e revisadas. A preparação das lâminas foi feita pela técnica de Tissue Microarray. A análise imunohistoquímica foi realizada, utilizado-se para controle positivo um tecido sabidamente com expressão da proteína a ser pesquisada. Foram realizadas duas lâminas de controle negativo. A primeira delas foi feita com a retirada do anticorpo primário e substituição por soro bovino fetal. O segundo controle negativo foi realizado com a retirada da reação do anticorpo secundário e substituição por soro contendo imunoglobulina do mesmo idiotipo do anticorpo primário, mas da mesma espécie. Os dados obtidos foram avaliados através do Teste do Qui-quadrado ou do Teste Exato de Fisher. Foram elaboradas curvas de sobrevivências, segundo o método de Kaplan-Meier. Curvas de sobrevivência foram comparadas através do Teste Log-Rank. Resultados: Os níveis de expressão imunohistoquímica de Ezrin,Radixin, Moezin e CD44V6 no câncer da próstata, foram respectivamente 48,8%, 16,3%, 20,9% e 70,9%. Houve uma diminuição da expressão do Radixin e Moesin quando comparados com os valores de PSA pré-operatório. Esta diminuição da expressão foi estatisticamente significante (p< 0,008 e p< 0,044) para os pacientes portadores de PSA> 10ng/ml no pré-operatório. A expressão do Ezrin teve significância estatística com relação à redução global da sobrevida dos pacientes quando comparado com as outras proteínas de adesão. Conclusão: A expressão da proteína Ezrin no câncer da próstata parece ser fator prognóstico independente na redução da sobrevida global, podendo vir a ser um importante marcador molecular em câncer da próstata / Introduction: Radical prostatectomy is one form of definite treatment of clinically localized prostate cancer. Approximately 40% of the patients treated by radical prostatectomy presented disease progression, even with the disease confined to the organ. Predictive accuracy of the risk of recurrence is useful to be considered as an adjuvant therapy or in some form of investigational treatment. To date, anatomopathological and clinical factors are observed as insufficient for the precise staging of prostate cancer. Some studies have emphasized the influence of cellular adhesion processes as a prerequisite for tumor invasion and metastasis, with the molecules CD44 and the ERM (Ezrin/Radixin/Moesin) family cited as facilitators of these processes. Objectives: To evaluate the expression of CD44/ERM complex protein in prostate cancer, as well as its association with prognostic and survival factors in this neoplasm. Material and Methods: Retrospectively analyzed were the paraffin blocks of 100 prostate cancer patients, proven by transrectal ultrasound-guided prostate biopsy, submitted to radical prostatectomy in the period from 1995 to 2000 at the Department of Pathological Anatomy of the General Hospital of Fortaleza and a private laboratory (Fortaleza, Brazil). The slides of all cases selected for this study were stained with hematoxylin and eosin and reviewed. The preparation of the slides was done by tissue microarray technique. Immunohistochemical analysis was carried out using a tissue known to have expression to the protein being studied as positive control. Two slides of negative control were prepared. The first slide was prepared with the omission of the primary antibody and substitution by fetal bovine serum. The second negative control was prepared with the omission of the secondary antibody response and substitution by serum containing immunoglobulin of the same idiotype of the primary antibody, but of the same species. The obtained data were evaluated through chi-square test or Fishers exact test. Survival curves were prepared following the Kaplan-Meier method. Survival curves were compared by the log-rank test. Results: The immunohistochemical expression levels of Ezrin, Radixin, Moezin and CD44V6 in prostate cancer were respectively 48.8%, 16.3%, 20.9% and 70.9%. There was a reduction of Radixin and Moesin expression when compared with preoperative PSA values. This reduction of expression was statistically significant (p <0.008 and p <0.044) for the patients with preoperative PSA> 10ng/ml. Ezrin expression had statistical significance in relation to reduction of overall patient survival when compared with other adhesion proteins. Conclusion: Ezrin protein expression in prostate cancer seems to be an independent prognostic factor in the reduction of overall survival, being able to become an important molecular marker in prostate cancer

A exenteração pélvica é método efetivo no tratamento de tumores pélvicos localmente avançados. As cirurgias mais conservadoras, com preservação funcional dos esfíncteres e reconstrução continente dos tratos intestinal e urinário podem melhorar a qualidade de vida e estimular os pacientes a aceitar a cirurgia. O objetivo deste estudo foi avaliar os resultados da exenteração pélvica no tratamento dos tumores pélvicos localmente avançados em relação à preservação dos esfíncteres e fatores associados ao prognóstico. Analisou-se retrospectivamente os fatores relacionados à preservação dos esfíncteres bem como os fatores associados ao prognóstico em pacientes submetidos à exenteração pélvica. Dos 96 pacientes submetidos à exenteração pélvica, preservou-se pelo menos um dos esfíncteres em 36 (37,5%). Na década de 1990 a taxa de preservação esfincteriana foi significativamente maior do que na década de 1980 (47,6 versus 18,2%) (p = 0,005). As variáveis independentemente relacionadas à preservação de esfíncter foram tratamento realizado na década de 1990 e tumor de origem coloproctológica. A taxa de complicação pós-operatória não foi influenciada pela preservação dos esfíncteres (p = 0,276). Não se observou diferença estatisticamente significativa nas taxas de morbidade entre as décadas de 1990 e 1980 (55,6% versus 75,8%; p = 0,075). Na década de 1990 houve redução da taxa de mortalidade pós-operatória em relação à década de 1980 (9,5% versus 27,3%; p = 0,023). Em nove pacientes, as margens de ressecção estavam microscopicamente comprometidas (R1) e, em cinco macroscopicamente comprometidas (R2). As margens de ressecção não foram influenciadas pelo tipo de cirurgia (p = 0,104), nem pela preservação dos esfíncteres (p = 0,881). A taxa de sobrevida livre de doença em cinco anos foi de 40,5%. Observou-se associação da recorrência com perda de peso (p = 0,006), índice de Karnofsky (p = 0,035) e a topografia do tumor (p = 0,027). No modelo multivariado, a perda de peso e os tumores de origem ginecológica foram as variáveis independentes para recorrência. Pacientes portadores de tumores ginecológicos ou com perda de peso foram considerados de alto risco para recorrência e os pacientes portadores de tumores não ginecológicos e sem perda de peso, de baixo risco. O grupo de alto risco apresentou chance de recorrência cerca de sete vezes maior do que o de baixo risco. A sobrevida livre de doença em 5 anos para os grupos de baixo e de alto risco foram, respectivamente, de 78,0% e 21,2%. As variáveis associadas ao óbito foram a idade superior a 60 anos (p = 0,007), a perda de peso (p = 0,004), radioterapia pré-operatória (p = 0,043), década de trata mento (p = 0,050) e preservação de esfíncter (p=0,026). No modelo multivariado as variáveis associadas ao óbito foram tratamento realizado na década de 1980, a idade superior a 60 anos e a perda de peso. Com os resultados deste estudo podemos concluir que houve aumento significante da preservação dos esfíncteres na década de 1990 sem aumento da freqüência de margens cirúrgicas comprometidas nem prejuízo na sobrevida dos pacientes submetidos à exenteração pélvica com preservação dos esfíncteres / Pelvic exenteration (PE) is an effective method for treating locally advanced pelvic tumors. More conservative surgeries, preserving sphincters and continent reconstruction of the intestinal and urinary tract, which could contribute to a better quality of life and encourage patients to accept this procedure. The objective of this study was to evaluate the results of PE in the treatment of locally advanced pelvic tumors, mainly considering sphincter preservation and factors associated to the prognosis. Between 1980 and 2000, 96 PE were performed. Factors related to sphincter preservation as well as factors associated to prognosis were respectively analyzed. Of the 96 patients treated with pelvic exenteration, at least one sphincter in 36 patients was preserved (37.5%). In the 1990\'s, the sphincter preservation rate was significantly higher than in the 1980\'s (47.6 vs. 18.2 %) (p = 0.005). Independent variables related to the sphincter preservation were decades from the realization of surgery 1990\'s and coloproctological tumors. The postoperative complication rate was not influenced by sphincter preservation (p = 0.276). Statistically, there was no differentiation between the morbidity rates during the 1980\'s and 1990\'s (55.6% versus 75.8%, p = 0.075). In the 1990\'s, there was a reduction in the post-operative mortality rate compared to the 1980s\' rate (9.5% versus 27.3%; p = 0.023). In nine patients, the resection margins were compromised microscopically (R1) and in five patients, macroscopically compromised (R2). The resection margins were not influenced by the type of surgery (p = 0.104), nor by the preservation of sphincters (p = 0.881). Disease free survival at five years was 40.5%. Among the clinical variables, there was an association between recurrence and weight loss (p = 0.006) and the Karnofsky index (p = 0.035). The topography of the tumor showed links with recurrence (p = 0.027). In the multivariable model, the independent variables related to recurrence were weight loss and gynecological tumors. Patients with gynecological tumors or with weight loss were considered high risk for recurrence, while patients with no gynecological tumors and without weight loss were considered low risk. The high risk group showed 7 times more chance of recurrence than the low risk group. Survival rates of patients, who remained disease-free, after 5 years, for the low and high risk group were 78.0% and 21.2% respectively. Death was linked to ages over 60 (p = 0.007), weight loss (p = 0.004), pre-surgery radiotherapy (p = 0.043), decades from the realization of surgery (p = 0.050) and the sphincter preservation (p = 0.026). The independent variables related to death were treatments in the 1980\'s, ages over 60 and weight loss. Taking into account the results in this research, we conclude that there was a significant increase of sphincter preservation during the 1990\'s and neither the type of surgery nor sphincter preservation were associated to a higher number of surgeries with compromised margins allows pelvic exenteration to be performed with sphincter preservation, without harming survival rates

Orientador: Juliany Gomes Quitzan / Resumo: O carcinoma de células escamosas (CCE) é uma neoplasia epitelial originada dos queratinócitos da pele de cães. Sua etiologia está relacionada à exposição a raios solares ultravioletas, o que o coloca como uma das neoplasias mais frequentes em países tropicais de clima quente. O CCE cutâneo tem comportamento invasivo local, baixa capacidade de metástase e frequentemente encontra-se associado à ceratose actinica. Neste estudo, objetivou-se avaliar o efeito do tratamento pré-operatório com piroxicam (Px) na expressão de COX-2 e Ki67, indicadores de inflamação e proliferação celular, respectivamente, em cães acometidos por CCE cutâneo. Além disso, o intervalo livre de doença (ILD) por um período de pelo menos 180 dias após excisão cirúrgica associada a regime de baixa dose diária com Px (0.3mg/kg) e ciclofosfamida (CYC; 15mg/m2) foi determinado nesta população. Não houve diferença estatística significativa na expressão de COX-2 após o tratamento pré-cirúrgico com Px; no entanto, houve diminuição no índice proliferativo marcado com Ki67 (P<0.05). Não foi encontrada alteração significativa no ILD no grupo tratado com Px e CYC (160 dias), quando comparado com o grupo controle retrospectivo (145 dias), tratado somente com ressecção cirúrgica. Não houve diferença estatística no ILD quando analisados o grau histológico, índice mitótico, tempo de evolução, metástase de linfonodos e o comprometimento das margens cirúrgicas. Adicionalmente, cães com estadiamento T4, independente do tratam... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: Squamous cell carcinoma (SCC) is an epithelial neoplasm that arises from skin keratinocytes of the dogs. Its etiology is related to ultraviolet sunlight, which puts it as one of the most frequent neoplasm in tropical hot weather countries. Cutaneous SCC have a locally invasive, low metastatic behavior, and is often associated with actinic keratosis. This study aimed to evaluate the effect of pre-operative treatment with piroxicam (Px) on COX-2 and Ki67 expression, indicators of inflammation and cell proliferation, respectively. Besides that, the evaluation of disease-free interval (DFI) of these animals for at least 180 days after surgical excision associated with a daily low-dose treatment with Px (0.3mg/kg) and cyclophosphamide (CYC; 15mg/m2). There was no statically significant difference between COX-2 expression before and after treatment with Px; However, there was a significant decrease on Ki67 proliferative index (P<0.05). No significance was found in DFI when comparing the group treated with Px and CYC (160 days) and the control retrospective group (145 days), treated only with surgical resection. There was no statically difference on DFI when accessing histological grade, mitotic index, evolution time, lymph node metastasis, and incomplete surgical margins (P>0.05). Additionally, dogs with T4 stage, independent of the treatment, were 3.2 and 4.8-fold more likely to develop an early recurrence when comparing with T3 and T2, respectively. The results of this study demo... (Complete abstract click electronic access below) / Mestre

The most important prognostic factor in breast cancer is the presence or absence of metastases in axillary lymph nodes. Frozen section and touch imprint cytology are conventional intra-operative methods used in the detection of metastatic breast cancer with varying sensitivities and specificities. The limitation of these methods led to the development of alternative molecular diagnostic tests, such as GeneSearch, a commercial real-time quantitative Polymerase Chain Reaction (RT-qPCR) assay that allows for an intra-operative diagnosis of metastatic breast cancer. When the GeneSearch assay was discontinued, Metasin was developed as an in-house RT-qPCR replacement assay. Metasin targets the epithelial cell marker cytokeratin 19 (CK19) and the breast marker mammaglobin (MGB) mRNA to confirm the presence or absence of metastatic disease, whilst the reference gene porphobilinogen deaminase (PBGD) acts as a positive control for the performance of the assay. The optimised assay can produce a result within 32 minutes allowing it to be used in the intra-operative setting to detect metastatic breast cancer in sentinel lymph nodes. 154 archived lymph node homogenates that were previously analysed by both GeneSearch and histology in parallel were used to validate Metasin. Out of 154 cases, 148 showed concordance with both GeneSearch and Metasin with 111 cases being negative and 37 cases being positive. There were six discordant cases, four in which only Metasin detected metastases and two in which only GeneSearch picked up metastases. Out of the four Metasin-only positive cases, three were found to be positive on histology after deeper levels were cut in the slices sent for histological assessment. Therefore, one case could not be shown histologically to be positive for metastases. There were two cases that were missed by Metasin but picked up by GeneSearch. One case was positive on histology and the second case negative for histology. The error rate for Metasin was 3.89%. The sensitivity and specificity of the Metasin assay were found to be 95% and 98% respectively, and the positive and negative predictive values were 90% and 98% respectively. These results are comparable to those of GeneSearch. Metasin had an assay time of less than 45 minutes and was operated by biomedical scientists. The results of the validation process were deemed acceptable for the assay to be run live and used in the clinical setting. Metasin continues to provide breast cancer patients at Princess Alexandra Hospital with all the advantages that a molecular intra-operative diagnostic service provides.

L’immunothérapie se présente aujourd’hui comme une alternative de choix dans le traitement des cancers. Son objectif est d’amplifier la réponse immunitaire contre les cellules tumorales tout en préservant les cellules normales. Les travaux antérieurs de mon équipe ont démontré qu'une réponse immunitaire antitumorale a lieu dans les carcinomes bronchiques non à petites cellules (CBNPC) et que des lymphocytes T cytotoxiques (CTL) spécifiques peuvent contribuer à la régression de la tumeur. Les travaux de mon équipe ont démontré aussi que l’interaction de l’intégrine CD103, souvent exprimée sur les lymphocytes infiltrant la tumeur (TIL), avec son ligand, le marqueur des cellules épithéliales E-Cadhérine, à la surface des cellules tumorales, est nécessaire à la polarisation des granules cytotoxiques et leur exocytose pour déclencher la lyse de la cellule cible. L’objectif principal de mon projet de thèse est de déterminer la contribution réelle des lymphocytes T CD8+/CD103+ infiltrant les tumeurs épithéliales dans la réponse CTL antitumorale et le rôle de CD103 dans la régulation de leurs fonctions effectrices in situ. Dans un premier temps, j’ai caractérisé, sur le plan transcriptionel et phénotypique, les TIL de CBNPC humains. Mes résultats ont montré que les lymphocytes T CD8+/CD103+ présentent une signature moléculaire caractéristique des cellules T mémoires résidentes dans les tissues (TRM), avec une expression des récepteurs CD69 et CD45RO. Mes résultats ont montré aussi que cette population lymphocytaire co-Exprime les récepteurs inhibiteurs PD-1 et Tim-3. Dans un deuxième temps, j’ai étudié la fonctionnalité des TIL CD8+/CD103+ et le rôle de CD103 dans leur activité cytotoxique anti-Tumorale. Mes résultats ont d’abord indiqué que les lymphocytes T CD103+ sont plus sensibles à la mort cellulaire induite par activation (AICD) que les TIL CD103-, et qu’ils expriment le granzyme B et CD107a suite à une activation spécifique. De plus, ils sont capables d’exercer une activité cytotoxique spécifique à l’encontre des cellules tumorales autologues suite à la neutralisation de l’interaction de PD-1 avec son PD-L1, et que des anticorps anti-CD103 bloquants inhibe cette fonction. Ensuite, j’ai analysé l’impact de l’expression de CD103 à la surface des TIL sur la survie de patients atteints de CBNPC de stade 1. Mes résultats ont révélé que cette intégrine favorise l’infiltration des TIL dans les régions tumorales épithéliales et qu’une forte expression de CD103 sur les TIL corrèle avec une amélioration de la survie des patients. Enfin, J’ai examiné le rôle de CD103 dans cette fonction et dans la réponse immunitaire antitumorale in vivo. Mes résultats préliminaires ont montré une croissance tumorale retardée des tumeurs LL2 transfectées avec l’E-Cadhérine et CCL5 greffées dans les souris CD103-WT. De plus l’inhibition de cette croissance corrèle avec une infiltration plus importante des tumeurs avec des lymphocytes T CD8+/CD103+. Ces résultats suggèrent un rôle important de la coexpression de CCL5 et d’E-Cadhérine par la tumeur dans le recrutement et la rétention des CTL au site tumoral. L’ensemble de ces travaux est en faveur du rôle important de CD103 dans la régulation de l’immunité T CD8 dans les tumeurs épithéliales et de l’utilité des anticorps neutralisants anti-PD-1 et anti-Tim-3 pour inverser l'épuisement de cette population lymphocytaires CD8+/CD103+. / Today Immunotherapy is clearly an alternative choice in the treatment of cancers. Its main objective is to enhance the cytotoxic immune response against tumor cells while preserving normal cells. We have previously demonstrated that there is an antitumor immune response in the Non-Small-Cell lung carcinoma (NSCLC) and cytotoxic T lymphocytes (CTL) contribute to NSCLC tumor regression. We further showed that the CD103 integrin interaction (oftenly expressed on tumor infiltrating lymphocytes (TIL)) with its ligand, the epithelial cell marker E-Cadherin, expressed at the surface of tumor cells, is necessary for the polarization and exocytosis of TIL cytotoxic granules and to trigger the lysis of the tumor target cells. The main purpose of my thesis project is to determine the actual role/ contribution of CD8+/CD103+ T lymphocytes (infiltrating the epithelial tumors) in the regulation of antitumor CTL response and to study the role of CD103 in the regulation of their in situ effector functions. Firstly, TIL infiltrating human NSCLC were characterized at transcriptional and phenotypic level. My results show that CD8+/CD103+ T lymphocytes have a molecular signature characteristic of memory T cells resident in tissues (MRT), with expression of CD69 receptors and CD45RO. My results also showed that this cell population co-Expresses the inhibitory receptors, PD-1 and Tim-3.In a second step, I studied the functionality of CD8+/CD103+ TIL and the role of CD103 in the regulation of anti-Tumor cytotoxic activity. My results have first indicated that CD103+ TIL are more sensitive to activation induced cell death (AICD) than TIL-CD103- and CD103+ TIL express granzyme B and CD107a after specific activation. Furthermore, CD103+ TIL are able to exert a specific cytotoxic activity against autologous tumor cells following the neutralization of PD-1- PD-L1 interaction, and that of anti-CD103 antibody inhibits this blocking function. After, I analyzed the impact of the expression of CD103 on the surface of TIL on the survival of patients with NSCLC stage 1. My results revealed that this integrin promotes the infiltration of TIL in epithelial tumor regions and a strong expression of CD103 on TIL correlates with improved patient survival. Finally, I examined the role of CD103 in this function and the antitumor immune response in vivo. My preliminary results showed a tumor growth delay of LL2 tumors transfected with E-Cadherin and CCL5 grafted in CD103-WT mice. Furthermore inhibition of growth correlates with a higher tumor infiltrating with CD8+/CD103+ T lymphocytes. These results suggest an important role of the coexpression of CCL5 and E-Cadherin by the tumor in the recruitment and retention of CTL at the tumor site. The whole work supports the role of CD103 in regulating the CD8 T cells-Mediated immune response in epithelial tumors and the usefulness of anti-PD-1 neutralizing and anti-Tim-3 for reversing the depletion of this lymphocyte population CD8+ / CD103+.

Os melanomas correspondem a 7% de todas as neoplasias malignas em cães e são principalmente encontrados em cavidade oral e lábios, correspondendo a 33% dos tumores de boca, possuem um prognóstico ruim devido ao fato de serem diagnosticados tardiamente, por sua grande capacidade de invasão local e formação de metástases, além de altas taxas de recidiva após o tratamento cirúrgico. A Galectina-3 (Gal-3) é uma proteína responsável por diversas funções fisiológicas como adesão, apoptose, angiogênese, proliferação e diferenciação. Em medicina veterinária existem poucos estudos relacionando à expressão da Gal-3 com prognóstico e a progressão da neoplasia. Realizamos imuno-histoquímica para Gal-3 em 27 melanomas orais caninos que foram avaliados de maneira semiquantitativa e quantitativa, e comparamos os resultados obtidos com a sobrevida, outros marcadores prognósticos (Ki67, índice mitótico e atipia nuclear), expressão de proteínas relacionadas à apoptose (BCL2 e CASP3) e parâmetros histopatológicos (grau de pigmentação e tipo histológico). Detectamos alta expressão de Gal-3 em melanomas com maior sobrevida pós-cirúrgica e uma alta expressão nuclear de Gal-3 em melanomas com menor sobrevida pós-cirúrgica. Além disso, houve correlação entre as expressões de Gal-3 e BCL2, assim como entre atipia nuclear e sobrevida pós-cirúrgica. É sabido que a Gal-3 é capaz de formar heterodímeros com a BCL2 no citoplasma para atuar na evasão da morte por apoptose, impedindo a liberação da citocromo C. Já no núcleo, a Gal-3 induz à parada do ciclo celular, reduzindo a taxa da proliferação. Apesar do reduzido número amostral devido à dificuldade nos acompanhamentos clínicos nossos dados permitem sugerir que a Gal-3 é possui potencial para ser um marcador prognóstico de sobrevida em casos de melanomas orais caninos. Novos estudos devem ser realizados afim de confirmar nossas observações e elucidar o papel da Gal-3 nesta neoplasia. / Melanomas are almost 7% of all malignant neoplasms in dogs. They are mainly found in the oral cavity and lips, corresponding to 33% of tumors of the oral cavity. They carry a poor prognosis because of late diagnoses, local invasiveness, high metastatic and recurrence rates after surgical treatment. Galectin-3 (Gal-3) is a protein with a variety of biological roles such as in adhesion, apoptosis, angiogenesis, proliferation and differentiation. In veterinary medicine, there are few studies comparing the expression of Gal-3 with prognosis and tumor progression. We performed immunohistochemistry for Gal-3 in 27 canine oral melanomas and evaluated the immunolabelling both semi-quantitatively and quantitatively. The results were compared with survival, other prognostic markers (Ki67, mitotic index and nuclear atypia), expression of proteins related to apoptosis (BCL2 and CASP3) and histopathological parameters (degree of pigmentation and histological type). We detected higher expression of Gal-3 in cases of melanoma that presented longer post-surgical survival and a higher nuclear expression of Gal-3 in dogs with melanoma that had shorter post-surgical survival. In addition, there was correlation between the Gal-3 and BCL2 expressions, as well as between nuclear atypia and post-surgical survival. It is known that Gal-3 is able to form heterodimers with BCL2 in the cytoplasm leading to evasion of apoptosis, through preventing mitochondrial cytochrome C release. Nuclear Gal-3 induces the cell cycle arrest, reducing the proliferation rate. Despite the small sample size due to the difficulty in clinical follow-up, our data suggest that Gal-3 has the potential to be a prognostic marker for survival in cases of canine oral melanomas. Further studies should be performed to confirm our observations and elucidate the role of Gal-3 in this neoplasm.

A neoplasia mamária é a doença mais entre as neoplasias em cadelas. As características clínicas dos tumores mamários caninos e sua relação com prognóstico foram discutidos, incluindo idade, raça, estagiamento clínico, diagnóstico histopatológico, hormônios e proliferação celular. Fatores clínicos prognósticos incluindo diâmetro do tumor e comprometimento linfonodal são discutidos em relação a graduação histopatológica e expressão de OCT4 (marcador para célula tronco tumoral). Avaliação imunohistoquímica dos marcadores (RE, RP, Ki-67 e OCT4) de neoplasias mamárias foi descritas. Foi observada marcação positiva para o OCT4, porém não se observou associação com fatores prognósticos clínicos. Assim como a marcação de RE, PR e Ki-67 foram observados. Houve uma forte associação entre graduação histopatológica de malignidade e tipo histológico. / Mammary neoplasms are the most common neoplasm in female dogs. The clinical features of canine mammary gland tumors and their relation to prognosis were discussed, including age, breed, staging, histopathological diagnosis, hormones, cell proliferation. Additional clinical prognostic factors including tumor size, and lymph node status are discussed in relation to graduation histopathological and OCT4 expression (cancer stem cell marker). Immunohistochemical evaluation of the markers (ER, PR, KI-67, OCT4) of the neoplastic canine mammary gland is described. We observed positive staining for OCT4, but was not associated with clinical prognostic factors. And the marking of ER, PR and Ki-67 was observed. There was a strong association between histopathological grade and histological type of malignancy.

Modelos preditivos estimam o risco de eventos ou agravos relacionados à saúde e podem ser utilizados como ferramenta auxiliar em tomadas de decisão por gestores e profissionais de saúde. Algoritmos de machine learning (ML), por sua vez, apresentam potencial para identificar relações complexas e não-lineares presentes nos dados, com consequências positivas na performance preditiva desses modelos. A presente pesquisa objetivou aplicar técnicas supervisionadas de ML e comparar sua performance em problemas de classificação e de regressão para predizer respostas de interesse para a saúde pública e a medicina. Os resultados e discussão estão organizados em três artigos científicos. O primeiro apresenta um tutorial para o uso de ML em pesquisas de saúde, utilizando como exemplo a predição do risco de óbito em até 5 anos (frequência do desfecho 15%; n=395) para idosos do estudo \"Saúde, Bem-estar e Envelhecimento\" (n=2.677), segundo variáveis relacionadas ao seu perfil demográfico, socioeconômico e de saúde. Na etapa de aprendizado, cinco algoritmos foram aplicados: regressão logística com e sem penalização, redes neurais, gradient boosted trees e random forest, cujos hiperparâmetros foram otimizados por validação cruzada (VC) 10-fold. Todos os modelos apresentaram área abaixo da curva (AUC) ROC (Receiver Operating Characteristic) maior que 0,70. Para aqueles com maior AUC ROC (redes neurais e regressão logística com e sem penalização) medidas de qualidade da probabilidade predita foram avaliadas e evidenciaram baixa calibração. O segundo artigo objetivou predizer o risco de tempo de vida ajustado pela qualidade de vida de até 30 dias (frequência do desfecho 44,7%; n=347) em pacientes com câncer admitidos em Unidade de Terapia Intensiva (UTI) (n=777), mediante características obtidas na admissão do paciente à UTI. Seis algoritmos (regressão logística com e sem penalização, redes neurais, árvore simples, gradient boosted trees e random forest) foram utilizados em conjunto com VC aninhada para estimar hiperparâmetros e avaliar performance preditiva. Todos os algoritmos, exceto a árvore simples, apresentaram discriminação (AUC ROC > 0,80) e calibração satisfatórias. Para o terceiro artigo, características socioeconômicas e demográficas foram utilizadas para predizer a expectativa de vida ao nascer de municípios brasileiros com mais de 10.000 habitantes (n=3.052). Para o ajuste do modelo preditivo, empregou-se VC aninhada e o algoritmo Super Learner (SL), e para a avaliação de performance, o erro quadrático médio (EQM). O SL apresentou desempenho satisfatório (EQM=0,17) e seu vetor de valores preditos foi utilizado para a identificação de overachievers (municípios com expectativa de vida superior à predita) e underachievers (município com expectativa de vida inferior à predita), para os quais características de saúde foram comparadas, revelando melhor desempenho em indicadores de atenção primária para os overachievers e em indicadores de atenção secundária para os underachievers. Técnicas para a construção e avaliação de modelos preditivos estão em constante evolução e há poucas justificativas teóricas para se preferir um algoritmo em lugar de outro. Na presente tese, não foram observadas diferenças substanciais no desempenho preditivo dos algoritmos aplicados aos problemas de classificação e de regressão analisados. Espera-se que a maior disponibilidade de dados estimule a utilização de algoritmos de ML mais flexíveis em pesquisas de saúde futuras. / Predictive models estimate the risk of health-related events or injuries and can be used as an auxiliary tool in decision-making by public health officials and health care professionals. Machine learning (ML) algorithms have the potential to identify complex and non-linear relationships, with positive implications in the predictive performance of these models. The present research aimed to apply various ML supervised techniques and compare their performance in classification and regression problems to predict outcomes of interest to public health and medicine. Results and discussion are organized into three articles. The first, presents a tutorial for the use of ML in health research, using as an example the prediction of death up to 5 years (outcome frequency=15%; n=395) in elderly participants of the study \"Saúde, Bemestar e Envelhecimento\" (n=2,677), using variables related to demographic, socioeconomic and health characteristics. In the learning step, five algorithms were applied: logistic regression with and without regularization, neural networks, gradient boosted trees and random forest, whose hyperparameters were optimized by 10-fold cross-validation (CV). The area under receiver operating characteristic (AUROC) curve was greater than 0.70 for all models. For those with higher AUROC (neural networks and logistic regression with and without regularization), the quality of the predicted probability was evaluated and it showed low calibration. The second article aimed to predict the risk of quality-adjusted life up to 30 days (outcome frequency=44.7%; n=347) in oncologic patients admitted to the Intensive Care Unit (ICU) (n=777), using patients\' characteristics obtained at ICU admission. Six algorithms (logistic regression with and without regularization, neural networks, basic decision trees, gradient boosted trees and random forest) were used with nested CV to estimate hyperparameters values and to evaluate predictive performance. All algorithms, with exception of basic decision trees, presented acceptable discrimination (AUROC > 0.80) and calibration. For the third article, socioeconomic and demographic characteristics were used to predict the life expectancy at birth of Brazilian municipalities with more than 10,000 inhabitants (n=3,052). Nested CV and the Super Learner (SL) algorithm were used to adjust the predictive model, and for evaluating performance, the mean squared error (MSE). The SL showed good performance (MSE=0.17) and its vector of predicted values was used for the identification of underachievers and overachievers (i.e. municipalities showing worse and better outcome than predicted, respectively). Health characteristics were analyzed revealing that overachievers performed better on primary health care indicators, while underachievers fared better on secondary health care indicators. Techniques for constructing and evaluating predictive models are constantly evolving and there is scarce theoretical justification for preferring one algorithm over another. In this thesis no substantial differences were observed in the predictive performance of the algorithms applied to the classification and regression problems analyzed herein. It is expected that increase in data availability will encourage the use of more flexible ML algorithms in future health research.

There is some evidence to suggest that human papilloma virus (HPV) may play a causal role in head and neck carcinoma (HNSCC). The aim of this study was to investigate the prevalence of HPV DNA in HNSCC and to determine whether any correlation exists with p16 or survival. An initial pilot study of sixty formalin-fixed HNSCC was carried out in order to optimise the methodology for the PCR and immunohistochemistry. A further 84 benign lesions, 12 dysplasias and additional 80 HNSCC were also included. In the pilot study the prevalence of all HPV types was 67% of which 18% were high risk-HPV (HR-HPV) and for the combined carcinoma sample it was 59% of which 25% were HR-HPV. The overall HPV prevalence was 51% and 42% for benign lesions and dysplasias with HR-HPV accounting for 14% and 8% respectively. A total of four alpha HPV types were identified and eleven beta HPV types. Multiple HPV types co-existed in the same tissue and in some cases both alpha and beta HPV. The results may suggest that HR-HPV may play a role in a small subset of HNSCC. An association was found between HPV status and gender, age group, survival, nodal metastasis and T3 tumour size and smoking. HPV16 was predominantly present in female patients and was associated with an improved overall survival and recurrence free survival. p16 positivity varied from 76-78% in carcinomas, 51% in benign lesions and 66% in dysplasias. p16 status was not associated with disease recurrence or nodal metastasis. Positive p16 staining and high staining intensity was associated with a poorer overall survival and the male gender, an older age group, anatomic site, and T2 tumour size. Overall HPV status was not correlated with p16 expression but a correlation found between p16 and HPV16 may suggest that p16 could potentially act as a surrogate marker of HPV16. However, the lack of concordance would suggest that in isolation p16 may not be a reliable marker for HR-HPV and should not be relied upon in isolation. Our findings could suggest that HPV16 and p16 status may be independent predictors for prognosis and disease recurrence.