Tesi sul tema "Personalized predictive medicine"

About one in eight women in the United States will develop breast cancer over the course of her lifetime. Moreover, patient-to-patient variability in disease progression continues to complicate clinical decisions in diagnosis and treatment for breast cancer patients. Early detection of tumors is a key factor influencing patient survival, and advancements in diagnostic and imaging techniques has allowed clinicians to spot smaller sized lesions. There has also been an increase in premature treatments of non-malignant lesions because there is no clear way to predict whether these lesions will become invasive over time. Patient variability due to genetic polymorphisms has been investigated, but studies on variability at the level of cellular activity have been extremely limited. An individual’s biochemical milieu of cytokines, growth factors, and other stimuli contain a myriad of cues that pre-condition cells and induce patient variability in response to tumor progression or treatment. Circulating white blood cells called monocytes respond to these cues and enter tissues to differentiate into monocyte-derived macrophages (MDMs) and osteoclasts that produce cysteine cathepsins, powerful extracellular matrix proteases. Cathepsins have been mechanistically linked to accelerated tumor growth and metastasis. This study aims to elucidate the variability in disease progression among patients by examining the variability of protease production from tissue-remodeling macrophages and osteoclasts. Since most extracellular cues initiate multiple signaling cascades that are interconnected and dynamic, this current study uses a systems biology approach known as cue-signal-response (CSR) paradigm to capture this complexity comprehensively. The novel and significant finding of this study is that we have identified and predicted donor-to-donor variability in disease modifying cysteine cathepsin activities in macrophages and osteoclasts. This study applied this novel finding to the context of tumor invasion and showed that variability in tumor associated macrophage cathepsin activity and their inhibitor cystatin C level mediates variability in cancer cell invasion. These findings help to provide a minimally invasive way to identify individuals with particularly high remodeling capabilities. This could be used to give insight into the risk for tumor invasion and develop a personalized therapeutic regime to maximize efficacy and chance of disease free survival.

Advanced information technologies promise a massive influx of individual-specific medical data. These rich sources offer great potential for an increased understanding of disease mechanisms and for providing evidence-based and personalized clinical decision support. However, the size, complexity, and biases of the data pose new challenges, which make it difficult to transform the data to useful and actionable knowledge using conventional statistical analysis. The so-called “Big Data” era has created an emerging and urgent need for scalable, computer-based data mining methods that can turn data into useful, personalized decision support knowledge in a flexible, cost-effective, and productive way. The goal of my Ph.D. research is to address some key challenges in current clinical deci-sion support, including (1) the lack of a flexible, evidence-based, and personalized data mining tool, (2) the need for interactive interfaces and visualization to deliver the decision support knowledge in an accurate and effective way, (3) the ability to generate temporal rules based on patient-centric chronological events, and (4) the need for quantitative and progressive clinical predictions to investigate the causality of targeted clinical outcomes. The problem statement of this dissertation is that the size, complexity, and biases of the current clinical data make it very difficult for current informatics technologies to extract individual-specific knowledge for clinical decision support. This dissertation addresses these challenges with four overall specific aims: Evidence-Based and Personalized Decision Support: To develop clinical decision support systems that can generate evidence-based rules based on personalized clinical conditions. The systems should also show flexibility by using data from different clinical settings. Interactive Knowledge Delivery: To develop an interactive graphical user interface that expedites the delivery of discovered decision support knowledge and to propose a new visualiza-tion technique to improve the accuracy and efficiency of knowledge search. Temporal Knowledge Discovery: To improve conventional rule mining techniques for the discovery of relationships among temporal clinical events and to use case-based reasoning to evaluate the quality of discovered rules. Clinical Casual Analysis: To expand temporal rules with casual and time-after-cause analyses to provide progressive clinical prognostications without prediction time constraints. The research of this dissertation was conducted with frequent collaboration with Children’s Healthcare of Atlanta, Emory Hospital, and Georgia Institute of Technology. It resulted in the development and adoption of concrete application deliverables in different medical settings, including: the neuroARM system in pediatric neuropsychology, the PHARM system in predictive health, and the icuARM, icuARM-II, and icuARM-KM systems in intensive care. The case studies for the evaluation of these systems and the discovered knowledge demonstrate the scope of this research and its potential for future evidence-based and personalized clinical decision support.

Breast cancer (BC), Colorectal Cancer (CRC) and Non-Small Cell Lung Cancer (NSCLC) are among the most commonly diagnosed solid tumors, and occupy the first places in the mortality rankings. Compared to an old fashioned one-size-fits-all approach, precision medicine offers the possibility to accurately choose the most appropriate therapeutic strategy, that fits the patients not only from the clinical (age, comorbidities) but also from a molecular point of view. A genetic and biological understanding of the tumor, integrated with a weighted analysis of results can help the clinician in designing a therapeutic pathway that, ideally from the start, gives the patients the best response rates. The aim of my research is to evaluate the markers that have the greatest impact on the prediction of therapy response. Mutational analysis revolutionized the NSCLC treatment paradigm and, consequently, improved the prognosis. EGFR mutated patients benefit from target therapy with tyrosine kinase inhibitors. A fluid and longitudinal monitoring of mutational status is becoming a key factor in disease management. Firstly we extracted circulating free DNA (cfDNA) from the plasma of 30 patients with EGFR-mutated NSCLC and assessed mutational status with real-time PCR. We then monitored such mutation during target therapy in 19 patients. The liquid biopsy had a sensitivity of 60% in confirming the tissue mutation. Patients whose EGFR mutation was not detectable on plasma had a longer Progression free survival (PFS) and Overall survival (OS). Next step will be assessing if cfDNA analysis allows early detection of resistance mutation such as T790M. Next part of my research focused on luminal BC, working partially retrospectively on data from a phase III study of 90 ER-positive, HER2 negative locally advanced breast cancer patients that were randomly assigned 1:1 to receive Let 2,5 mg daily and metronomic oral Cyc 50 mg daily with (arm B; n=45) or without (arm A, n=45) sorafenib 400 mg/bid daily for six months as neoadjuvant treatment. The predictive role of Ki67, SUV variations and metabolic response and its changes with regards to clinical response and survival was analyzed. The serum of 32 patients was analyzed via Luminex Multiplex Panel technology. 38 analytes (cytokines and growth factors) were simultaneously measured according to arm of treatment and time of sample collection (before and after treatment). Patients were divided into groups according to response to therapy (RECIST). Then we investigated a possible link between chemotherapy-induced RNA disruption and survival/progression. Analysis were performed on 40 biopsies taken at baseline and 15 days after the beginning of the neoadjuvant therapy. The RNA for each sample or subdivided sample was then assessed using the RNA Disruption Assay. The maximum RNA disruption Index (RDI) value for each patient at day 15 was used for all analyses. Finally, We investigated the discordance of mutational status between primary and metastatic site in colorectal cancer.Patients with metastatic CRC who underwent surgery of both primary and metastasis were retrospectively evaluated, and mutational status assessment of K-RAS, N-RAS, BRAF and PIK3CA was performed on 21 patients. Median DFS was 20.5 months (95% CI 9.9-29.6) in patients with concordance in mutational status versus 10.4 months (95% CI 6.1-not reached) in patients with discordance (p=0.01) and median OS was 35.9 months (95% CI 26.3-not reached) in patients with concordance in mutational status 25.6 months (95% CI 6.6-not reached) in patients with discordance (p=0.038). In conclusion discordance seems related to clinical outcome. Overall my results show that new strategies and technologies allow the researchers and the clinicians to strive for a better and more complete understanding of solid tumors complex evolution, an integrated and focused approach to the early disease could become the future of disease management.

Background Suboptimal health status (SHS) is an intermediate health status between ideal health and diseases. It is characterized by chronic fatigue, perception of health complaints and a cluster of physical symptoms lasting for more than three months. SHS is recognised as a subclinical, reversible stage of chronic diseases. Objectives Study I. To investigate the prevalence of SHS in a cross-sectional study. Study II. To screen transcriptomic biomarkers for SHS in a case-control study. Study III. To screen metabolomics biomarkers for SHS in a case-control study. Materials and Methods Study I. A cross-sectional study was conducted from September 2017 to November 2017. SHS questionnaire-25 was used to assess the SHS levels of the participants. Study II. The RNA sequencing (RNA-Seq)-based transcriptome analysis was firstly conducted on buffy coat samples collected from 30 participants with SHS and 30 age- and sex-matched healthy controls. Study III. The liquid chromatography-mass spectrometry (LC-MS)- based untargeted metabolomics analysis was conducted on plasma samples collected from 50 SHS participants and 50 age- and sex-matched healthy controls. Result In Study I, a total of 4839 Chinese university students enrolled in this study. The prevalence of SHS was 8.39%, with the prevalence of 6.57% among males and 9.60% among females. The multivariate logistic regression results showed that SHS was significantly associated with age (Odd ratio (OR) = 1.193, P = 0.019), female (OR = 1.437, P = 0.002), sleep duration (OR = 0.609, P < 0.001), insomnia symptoms (OR = 1.238, P < 0.001), anxiety symptoms (OR = 1.025, P = 0.019), and depression symptoms (OR = 1.082, P < 0.001). In study II, transcriptome analysis identified a total of 46 differentially expressed genes, in which 22 transcripts were significantly increased and 24 transcripts were decreased in the SHS group. Gene Ontology and Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis revealed that several biological processes were related to SHS, such as ATP-binding cassette transporter and neurodegeneration. A combination of transcripts can distinguish SHS individuals from the healthy controls with a sensitivity of 83.3%, a specificity of 90.0%, and an area under the receiver operating characteristic curve (AUC) of 0.938. In study III, metabolomics analysis identified a total of 24 significantly altered metabolites as the candidate biomarkers for SHS. Pathway analysis revealed that sphingolipid metabolism, taurine metabolism, and steroid hormone biosynthesis are the disturbed metabolic pathways related to SHS. A combination of four metabolic biomarkers (sphingosine, pregnanolone, taurolithocholate sulfate, cervonyl carnitine) can distinguish SHS individuals from the controls with a sensitivity of 94.0%, a specificity of 90.0%, and an AUC of 0.977. Conclusion SHS is prevalent in Chinese university students. Older age, female, insomnia, depression, and anxiety symptoms are risk factors for SHS, while longer sleep duration is a protective factor for SHS. Blood transcripts and metabolites are valuable biomarkers for SHS identification. These findings suggest the potential utility of SHS-related transcriptomic and metabolomic biomarkers for the Predictive, Preventive, and Personalized Medicine (PPPM) of chronic diseases.

In personalized medicine, two important tasks are predicting disease risk and selecting appropriate treatments for individuals based on their baseline information. The dissertation focuses on providing improved risk prediction for ordinal outcome data and proposing score-based test to identify informative markers for treatment selection. In Chapter 1, we take up the first problem and propose a disease risk prediction model for ordinal outcomes. Traditional ordinal outcome models leave out intermediate models which may lead to suboptimal prediction performance; they also don't allow for non-linear covariate effects. To overcome these, a continuation ratio kernel machine (CRKM) model is proposed both to let the data reveal the underlying model and to capture potential non-linearity effect among predictors, so that the prediction accuracy is maximized. In Chapter 2, we seek to develop a kernel machine (KM) score test that can efficiently identify markers that are predictive of treatment difference. This new approach overcomes the shortcomings of the standard Wald test, which is scale-dependent and only take into account linear effect among predictors. To do this, we propose a model-free score test statistics and implement the KM framework. Simulations and real data applications demonstrated the advantage of our methods over the Wald test. In Chapter 3, based on the procedure proposed in Chapter 2, we further add sparsity assumption on the predictors to take into account the real world problem of sparse signal. We incorporate the generalized higher criticism (GHC) to threshold the signals in a group and maintain a high detecting power. A comprehensive comparison of the procedures in Chapter 2 and Chapter 3 demonstrated the advantages and disadvantages of difference procedures under different scenarios.
Biostatistics

The human genome is a source of information for researchers that study complex diseases with the perspective of a better understanding of the pathologies and the development of new therapeutic strategies. Starting from the beginning of the current century, a growing number of technologies devoted to DNA sequencing have emerged, generally referred to as Next Generation Sequencing (NGS) technologies. NGS gradually decreased the cost of sequencing a human genome to around US$1000, enabling the use of these technologies for clinical and research purposes, such as Genome-wide association studies (GWAS). GWAS studies have enlightened the presence of disease- associated loci, in particular variants that could be used to evaluate the risk of an individual to develop a disease. Unfortunately, different sources of errors are able to impair the interpretation and use of NGS data: on the one hand, we have noise related to the process of DNA sequencing and read alignment errors, which could lead to false positive calls or artifacts. On the other hand, variants could be poor predictors for the manifestation of their associated disease. Nowadays the challenge of genomic data interpretation has driven the research towards the development of methods for the analysis and interpretation of genomic variations, eventually predicting the probability of a patient to develop a definite disease. A fair evaluation of these tools is essential to understand the applicability of the presented methods in clinical practice. The Critical Assessment of Genome Interpretation (CAGI) has been developed with the aim of defining the current state of art in terms of methods for predicting the impact of genomic changes at molecular and phenotype levels. CAGI is a community-driven experiment in which different prediction methods, developed by a set of invited groups, are evaluated on a common dataset. Unfortunately, no common guidelines were given to evaluate the tools presented in CAGI experiments, this has made the comparison between different CAGI experiments cumbersome, since different mathematical indexes and scripts have been used to evaluate the involved methods. My PhD project has been focused on the development of software for the assessment of machine learning methods in regression and multiple phenotype challenges. This tool is based on state of the art assessment principles, derived from literature or previous CAGI experiments. This software is available as an R package and has been used to repeat or perform new assessments on a wide range of CAGI experiments. The knowledge acquired during the development of this project was used to evaluate two CAGI 5 challenges: Pericentriolar Material 1 (PCM1) and Intellectual Disability (ID) panel. The experience I have acquired, through the development of all previously mentioned works, has led the improvement and assessment of a machine learning method. In particular, I have developed a software for the prediction of cholesterol levels, based on genotype data. Eventually I have tested the reliability of this method. This tool was the milestone in a project founded by the Italian Ministry of Health.
Il genoma umano è una risorsa ricca di informazioni per i ricercatori che si dedicano allo studio delle patologie complesse. L’obiettivo di questo genere di ricerche è giungere ad una migliore comprensione di queste malattie e quindi sviluppare nuove strategie terapeutiche per la cura dei pazienti affetti. Dall’inizio di questo secolo, un numero crescente di tecnologie per il sequenziamento del DNA sono state sviluppate, sono conosciute come tecnologie “Next Generation Sequencing” (NGS). Le tecnologie NGS hanno gradualmente diminuito il costo del sequenziamento di un genoma umano fino a circa 1000 dollari, ciò ha consentito l’utilizzo di questi strumenti nella pratica clinica e nella ricerca, in particolare negli studi di associazione genome-wide o “Genome-wide association studies” (GWAS). Questi lavori hanno portato alla luce l’associazione di alcune varianti con alcune patologie o caratteri complessi. Queste varianti potrebbero essere utilizzate per valutare il rischio che un individuo sviluppi una particolare patologia. Sfortunatamente diverse sorgenti di errore sono in grado di ostacolare l’uso e l’interpretazione dei dati genomici: da una parte abbiamo il rumore legato al processo di sequenziamento e gli errori di allineamento delle reads. Dall’altra parte gli SNP non sempre possono essere utilizzati in modo affidabile per predire l’insorgenza della malattia a cui sono stati associati. Il Critical Assessment of Genome Interpretation è stato organizzato con l’obiettivo di definire lo stato dell’arte nei metodi che stimano l’effetto di variazioni genetiche a livello molecolare o fenotipico. Negli anni il CAGI ha dato vita a più competizioni in cui diversi gruppi di ricerca hanno testato i loro metodi di predizione su diversi dataset condivisi. L’assenza di linee generali su come condurre la valutazione delle performance dei predittori, ha reso difficile un confronto fra metodi sviluppati in edizioni diverse del CAGI. In questo contesto, il progetto di dottorato si è focalizzato nello sviluppo di un software per la valutazione di metodi di apprendimento automatici basati sulla regressione o la predizione di fenotipi multipli. Questo strumento si fonda su criteri di analisi della performance, derivanti dalla letteratura e da precedenti esperimenti del CAGI. Questo software è stato sviluppato in R ed utilizzato per ripetere o valutare ex novo la qualità dei predittori in un gran numero di esperimenti del CAGI. Le conoscenze acquisite durante lo sviluppo di questo progetto, sono state utilizzate per valutare due competizioni del CAGI 5: la Pericentriolar Material 1 (PCM1) e il Pannello per le Disabilità Intellettive (ID). L’esperienza derivante dal completamento dei lavori precedentemente elencati, ha guidato lo sviluppo e il miglioramento delle prestazioni di un metodo predittivo. In particolare è stato sviluppato un software per la predizione dei livelli di colesterolo, basato su dati genotipici, di cui è stata testata la validità con criteri matematici allo stato dell’arte. Questo strumento è stato la pietra portante di un progetto fondato dal Ministero della Salute Italiano.

Around 12-15% of patients with locally advanced rectal cancer (LARC) undergo a pathologically complete response (Tumour Regression Grade 4 - TRG4) to neoadjuvant chemoradiotherapy; the remainder exhibit a spectrum of tumour regression (TRG1-3). Understanding therapy-related genomic alterations may help us better predict response, progression-free and overall survival, and also identify both novel and repurposed treatment strategies based on the underlying biology of the disease. The Northern Ireland Biobank provided 48 formalin fixed paraffin embedded (FFPE) rectal cancer biopsies and matched resections following neoadjuvant therapy (discovery cohort). These were analysed using high-throughput gene expression microarray, DNA mutational profiling and microsatellite instability profiling. Differential gene expression analysis (analysis of variance) was performed contrasting tumour regression grades in both biopsies and resections to identify predictive and therapy related features. Real time PCR was utilised for microarray validation while immunohistochemistry (IHC) was employed to measure CD56+ cell populations in an independent (validation) cohort (n=150). A NK cell-like gene expression signature was observed following long course chemoradiotherapy in a tumour regression-dependent manner. CD56+ NK cel, populations were measured by IHC and found to be significantly higher in TRG3 patients. Furthermore, it was observed that patients positive for CD56 ceils after therapy had a better overall survival (HR=0.282, 95%C,=0.109-0.729, x2=7.854, p=.OO5). In silico drug selection using QUADrATiC analysis identified clinically relevant therapeutic FDA-approved compounds based upon the NK cell-like signature. We demonstrated that identifying an independently validated predictive signature from biopsies for LARC patients treated with LCPCRT was not possible. However, we identified a novel post-therapeutic NK-like transcription signature in patients responding to neoadjuvant chemoradiotherapy. Furthermore, CD56 positive patients had better overall survival. Therefore, harnessing an NK-like response after therapy may improve outcomes for locally advanced rectal cancer patients.

Cette thèse apporte trois contributions majeures. Le premier chapitre, un article paru dans la Revue Française d’Économie n°2/vol XXXIV, propose une revue de la littérature sur les implications de la régulation des tests génétiques de prédisposition sur le marché de l'assurance santé. Nous montrons que les formes actuelles de régulation réalisent un arbitrage entre la maximisation du bien-être social ex ante et l’incitation à la prévention. Cet arbitrage est conditionné par la façon dont l’acquisition de l’information affecte les comportements de prévention et de révélation des agents, la discrimination des risques par les assureurs ainsi que la nature des contrats. Le deuxième chapitre étudie théoriquement l’impact de la reclassification sur la prévention, la décision de faire le test ainsi que sur le bien-être social dans la réglementation d’obligation de divulgation. En particulier, nous montrons qu’en fonction du coût de l’effort de prévention la valeur individuelle de l’information génétique avec reclassification peut être plus élevée que sans reclassification. De plus, nous montrons comment les préférences temporelles affectent la valeur individuelle de l’information génétique. D’après nos résultats, le bien-être social est strictement plus élevé sans reclassification qu'avec reclassification. Le dernier chapitre étudie et caractérise des contrats incitatifs pouvant être mis en œuvre pour développer la médecine personnalisée avec des traitements très efficaces, dans un contexte d'aléa moral sur l’effort fourni par la firme pour améliorer l'efficacité du médicament. Nous considérons un modèle dans lequel l'autorité de santé a trois possibilités. Il peut appliquer le même traitement (standard ou nouveau traitement) à l'ensemble de la population ou mettre en œuvre la médecine personnalisée, c'est-à-dire utiliser les informations génétiques pour proposer le traitement le plus adapté à chaque patient. Nous caractérisons d'abord le contrat de remboursement de médicaments d'une entreprise produisant un nouveau traitement avec un test génétique compagnon lorsque l'entreprise peut entreprendre un effort pour améliorer la qualité des médicaments. Ensuite, nous déterminons les conditions sous lesquelles la médecine personnalisée doit être mise en œuvre lorsque cet effort est observable et quand il ne l'est pas. Enfin, nous montrons comment la non observabilité de l'effort affecte la décision de l'autorité sanitaire de mettre en œuvre la médecine personnalisée avec des traitements hautement efficaces
This thesis includes three main contributions. The first chapter, an article published in 2019 in the “Revue Française d’Économie n°2/vol XXXIV”, provides a literature review on the implications of genetic testing regulations on the health insurance market. We show that the choice of a regulation results from a trade-off between the maximization of ex-ante social welfare and incentive to undertake prevention actions. Indeed, this trade-off depends on the way information acquisition impacts prevention and revelation behaviors of agents, as well as of its impact on insurance contract. The second chapter studies theoretically how reclassification impacts testing and prevention decision as well as social welfare in the Disclosure Duty regulation. In particular, we show that the incentives of agents to take genetic with reclassification can be higher than without reclassification according to the effort cost. In addition, we show how time preferences affect the incentive to take genetic testing. Finally, we show that the social welfare is strictly higher without reclassification than with reclassification. The last chapter studies and characterizes contracts that can be implemented to develop personalized medicine with highly effective treatment in context of moral hazard about firm effort to improve drug efficacy. It also studies how the non-observability of effort impacts the decision of a health authority to implement personalized medicine with highly effective treatments. We consider a model in which the health authority has three possibilities. It can apply either the same treatment (a standard or a new treatment) to the whole population or implement personalized medicine, i.e., use genetic information to offer the most suitable treatment to each patient. We first characterize the drug reimbursement contract of a firm producing a new treatment with a companion genetic test when the firm can undertake an effort to improve drug quality. Then, we determine the conditions under which personalized medicine should be implemented when this effort is observable and when it is not. Finally, we show how the unobservability of effort affects the conditions under which the health authority implements personalized medicine

La Polykystose Rénale Autosomique Dominante (PKRAD) est une des pathologies héréditaires les plus fréquentes et affecte environ un individu sur 1000. Elle se caractérise par une importante variabilité clinique, notamment dans l’âge de survenue de l’insuffisance rénale terminale. Deux gènes sont en cause : le gène PKD1 situé sur le chromosome 16 (85% des cas) et le gène PKD2 situé sur le chromosome 4 (15% des cas). Les progrès majeurs dans la compréhension des mécanismes moléculaires impliqués ont permis le développement de stratégies thérapeutiques spécifiques, et de nouvelles questions surgissent : quels patients traiter ? Quand débuter les traitements ? La cohorte Genkyst, qui vise à inclure tous les patients suivis pour PKRAD dans la région Grand Ouest, nous a d’abord permis de décrire la variabilité génétique rencontrée dans la PKRAD. Nous avons ensuite démontré l’existence de fortes corrélations génotype-phénotype, en rapportant l’influence sur l’âge de survenue de l’insuffisance rénale terminale non seulement du gène en cause, mais aussi du type de mutation pour le gène PKD1. Enfin, l’analyse des données cliniques et génétiques de 1341 patients nous a permis de développer un algorithme pronostique, baptisé le PROPKD score, permettant de stratifier le risque de progression vers l’insuffisance rénale terminale. Nous espérons que ces travaux participeront à l’individualisation de la prise en charge des patients atteints de PKRAD, ce qui est un enjeu crucial à l’arrivée des nouveaux traitements
Autosomal Dominant Polycystic Kidney Disease (ADPKD) is one of the most frequent Mendelian inherited disorders, and affects approximately one individual out of 1000. ADPKD is marked by a high clinical variability, especially regarding age at end-stage renal disease (ESRD). Two genes are identified: PKD1 located on the chromosome 16 (85% of the pedigrees) and PKD2 located on the chromosome 4 (15% of the pedigrees). Substantial progress in understanding the cellular mechanisms underlying ADPKD has triggered the development of targeted therapies, and new questions are arising: which patients should be treated? When should we begin these treatments? Thanks to Genkyst cohort, which aims to include all consenting ADPKD patients from the western part of France, we first described the important allelic variability encountered in ADPKD. Secondly, we demonstrated the important influence of not only the gene involved, but also of PKD1 mutation type. Last, the analysis of clinical and genetic characteristics of 1341 patients from the Genkyst cohort allowed us to develop a prognostic algorithm, named the PROPKD score for predicting renal outcome in ADPKD. Our hope is that these works will participate in the development of individualized medicine in ADPKD, which is crucial in the context of the emerging targeted therapies

En médecine prédictive personnalisée, modéliser avec précision la maladie et les processus de soins d'un patient est crucial en raison des dépendances temporelles à long terme inhérentes. Cependant, les dossiers de santé électroniques (DSE) se composent souvent de données épisodiques et irrégulières, issues des admissions hospitalières sporadiques, créant des schémas uniques pour chaque séjour hospitalier.Par conséquent, la construction d'un modèle prédictif personnalisé nécessite une considération attentive de ces facteurs pour capturer avec précision le parcours de santé du patient et aider à la prise de décision clinique.LSTM sont efficaces pour traiter les données séquentielles comme les DSE, mais ils présentent deux limitations majeures : l'incapacité à interpréter les résultats des prédictions et à prendre en compte des intervalles de temps irréguliers entre les événements consécutifs. Pour surmonter ces limitations, nous introduisons de nouveaux réseaux neuronaux à mémoire dynamique profonde appelés Multi-Way Adaptive et Adaptive Multi-Way Interpretable Time-Aware LSTM (MWTA-LSTM etAMITA), conçus pour les données séquentielles collectées de manière irrégulière.L'objectif principal des deux modèles est de tirer parti des dossiers médicaux pour mémoriser les trajectoires de maladie et les processus de soins, estimer les états de maladie actuels et prédire les risques futurs, offrant ainsi un haut niveau de précision et de pouvoir prédictif
In personalized predictive medicine, accurately modeling a patient's illness and care processes is crucial due to the inherent long-term temporal dependencies. However, Electronic Health Records (EHRs) often consist of episodic and irregularly timed data, stemming from sporadic hospital admissions, which create unique patterns for each hospital stay. Consequently, constructing a personalized predictive model necessitates careful consideration of these factors to accurately capture the patient's health journey and assist in clinical decision-making. LSTM networks are effective for handling sequential data like EHRs, but they face two significant limitations: the inability to interpret prediction results and to take into account irregular time intervals between consecutive events. To address limitations, we introduce novel deep dynamic memory neural networks called Multi-Way Adaptive and Adaptive Multi-Way Interpretable Time-Aware LSTM (MWTA-LSTM and AMITA) designed for irregularly collected sequential data. The primary objective of both models is to leverage medical records to memorize illness trajectories and care processes, estimate current illness states, and predict future risks, thereby providing a high level of precision and predictive power

Analyse intégrative de données génomiques et pharmacologiques pour améliorer la prédiction de la réponse aux thérapies cibléesL'utilisation de thérapies ciblées dans le contexte de la médecine personnalisée du cancer a permis d’améliorer le traitement des patients dans différents types de cancer. Cependant, alors que la décision thérapeutique est basée sur une unique altération moléculaire (par exemple une mutation ou un changement du nombre de copies d’un gène), les tumeurs montrent différents degrés de réponse. Dans cette thèse, nous démontrons que la décision thérapeutique basée sur une unique altération n’est pas optimale et nous proposons un modèle mathématique intégrant des données génomiques et pharmacologiques pour identifier de nouveaux biomarqueurs prédictifs de la réponse thérapeutique. Le modèle a été construit à partir de deux bases de données de lignées cellulaires (the Genomics of Drug Sensitivity in Cancer, GDSC and the Cancer Cell Line Encyclopedia, CCLE) et validé avec des données de lignées et des données cliniques. De plus, nous avons également développé une nouvelle méthode pour améliorer la détection des mutations somatiques à partir de données de séquençage d'exomes complets et proposons un nouvel outil, cmDetect, disponible gratuitement pour la communauté scientifique
Integrated analysis of genomic and pharmacological data to better predict the response to targeted therapiesThe use of targeted therapies in the context of cancer personalized medicine has shown great improvement of patients’ treatment in different cancer types. However, while the therapeutic decision is based on a single molecular alteration (for example a mutation or a gene copy number change), tumors will show different degrees of response. In this thesis, we demonstrate that a therapeutic decision based on a unique alteration is not optimal and we propose a mathematical model integrating genomic and pharmacological data to identify new single predictive biomarkers as well as combinations of biomarkers of therapy response. The model was trained using two public large-scale cell line data sets (the Genomics of Drug Sensitivity in Cancer, GDSC and the Cancer Cell Line Encyclopedia, CCLE) and validated with cell line and clinical data. Additionally, we also developed a new method for improving the detection of somatic mutations using whole exome sequencing data and propose a new tool, cmDetect, freely available to the scientific community

Les effets indésirables médicamenteux (EIM) ont des répercussions considérables tant sur la santé que sur l'économie. De 1,9% à 2,3% des patients hospitalisés en sont victimes, et leur coût a récemment été estimé aux alentours de 400 millions d'euros pour la seule Allemagne. De plus, les EIM sont fréquemment la cause du retrait d'un médicament du marché, conduisant à des pertes pour l'industrie pharmaceutique se chiffrant parfois en millions d'euros.De multiples études suggèrent que des facteurs génétiques jouent un rôle non négligeable dans la réponse des patients à leur traitement. Cette réponse comprend non seulement les effets thérapeutiques attendus, mais aussi les effets secondaires potentiels. C'est un phénomène complexe, et nous nous tournons vers l'apprentissage statistique pour proposer de nouveaux outils permettant de mieux le comprendre.Nous étudions différents problèmes liés à la prédiction de la réponse d'un patient à son traitement à partir de son profil génétique. Pour ce faire, nous nous plaçons dans le cadre de l'apprentissage statistique multitâche, qui consiste à combiner les données disponibles pour plusieurs problèmes liés afin de les résoudre simultanément.Nous proposons un nouveau modèle linéaire de prédiction multitâche qui s'appuie sur des descripteurs des tâches pour sélectionner les variables pertinentes et améliorer les prédictions obtenues par les algorithmes de l'état de l'art. Enfin, nous étudions comment améliorer la stabilité des variables sélectionnées, afin d'obtenir des modèles interprétables
Adverse drug reaction (ADR) is a serious concern that has important health and economical repercussions. Between 1.9%-2.3% of the hospitalized patients suffer from ADR, and the annual cost of ADR have been estimated to be of 400 million euros in Germany alone. Furthermore, ADRs can cause the withdrawal of a drug from the market, which can cause up to millions of dollars of losses to the pharmaceutical industry.Multiple studies suggest that genetic factors may play a role in the response of the patients to their treatment. This covers not only the response in terms of the intended main effect, but also % according toin terms of potential side effects. The complexity of predicting drug response suggests that machine learning could bring new tools and techniques for understanding ADR.In this doctoral thesis, we study different problems related to drug response prediction, based on the genetic characteristics of patients.We frame them through multitask machine learning frameworks, which combine all data available for related problems in order to solve them at the same time.We propose a novel model for multitask linear prediction that uses task descriptors to select relevant features and make predictions with better performance as state-of-the-art algorithms. Finally, we study strategies for increasing the stability of the selected features, in order to improve interpretability for biological applications

Nous avons développé un nouveau test de prédiction de l’efficacité de thérapies anti-cancéreuses. Ce concept se base sur la détermination d’une signature moléculaire tumorale par RT-qPCR. Cette signature est issue d’un algorithme de normalisation innovant de comparaison des données d’expression relative des cibles de la tumeur avec celles de tissus de référence. Cette normalisation offre à chaque cible de la signature un rang et un score spécifique permettant de hiérarchiser les voies pro-tumorales afin de trouver la ou les cibles dominantes responsables du développement de la tumeur. La signature comprend des cibles donnant des informations générales sur le statut et l’hétérogénéité de la tumeur, sur l’angiogenèse et la lymphangiogenèse, sur le microenvironnement tumoral et enfin sur l’activité migratoire. Une validation préclinique dans les modèles du cancer colorectal, de la prostate et du glioblastome, a montré que le test est prédictif de l’efficacité thérapeutique
We developed a new tool for prediction of cancer treatment efficacy. Our process is based on the determination of the molecular signature which is intended to provide a clinician’s decision tool helping to select which tumor signaling pathway(s) has/have to be targeted for best therapeutic effect. This signature representing a scoring obtained by RT-qPCR through a sequential normalization process of the expression level of target genes in the tumor compared to cancer type-specific references. These genes were selected because of a good knowledge of related biological functions, a correlation between expression level and aggressiveness of the tumor, the existence of a therapeutic arsenal already in clinical use. This signature is validated in a preclinical model of colorectal cancer and prostate cancer and glioblastoma. The results obtained show that the test we developed allows to identify the most important signaling pathway implicated in the disease and to choose the best drug

Lung cancer is the leading cause of cancer-related deaths among men and women. Non-Small Cell Lung Cancer (NSCLC) constitutes the most common type of lung cancer and is frequently diagnosed at advanced stages. In the past decade, discovery of Epidermal Growth Factor Receptor (EGFR) mutations have heralded a new paradigm of personalized treatment for NSCLC. Clinical studies have shown that molecular targeted therapies increase survival and improve quality of life in patients. Despite these advances, the realization of personalized therapies for NSCLC faces a number of challenges including the integration of clinical and genetic data and a lack of clinical decision support tools to assist physicians with patient selection. This thesis demonstrates the development of a predictive computational model for personalized therapeutic interventions in advanced NSCLC. The findings suggest that the combination of clinical and genetic data significantly improves the model’s predictive performance for tumor response than clinical data alone.

Personalized medicine offers the most effective treatment protocols to the individual Chronic Myeloid Leukemia (CML) patients. Understanding the molecular biology that causes CML assists in providing efficient treatment. After the identification of an activated tyrosine kinase BCR-ABL1 as the causative lesion in CML, the first-generation Tyrosine Kinase inhibitors (TKI) imatinib (Glivec®), were developed to inhibit BCR-ABL1 activity and approved as a treatment for CML. Despite the remarkable increase in the survival rate of CML patients treated with imatinib, some patients discontinued imatinib therapy due to intolerance, resistance or progression. These patients may benefit from the use of secondgeneration TKIs, such as nilotinib (Tasigna®) and dasatinib (Sprycel®). All three of these TKIs are currently approved for use as frontline treatments. Prognostic scores and molecularbased predictive assays are used to personalize the care of CML patients by allocating risk groups and predicting responses to therapy. Although prognostic scores remain in use today, they are often inadequate for three main reasons. Firstly, since each prognostic score may generate conflicting prognoses for the risk index and it can be difficult to know how to treat patients with conflicting prognoses. Secondly, since prognostic score systems are developed over time, patients can benefit from newly developed systems and information. Finally, the earlier scores use mostly clinically oriented factors instead of those directly related to genetic or molecular indicators. As the current CML treatment guidelines recommend the use of TKI therapy, a new tool that combines the well-known, molecular-based predictive assays to predict molecular response to TKI has not been considered in previous research. Therefore, the main goal of this research is to improve the ability to manage CML disease in individual CML patients and support CML physicians in TKI therapy treatment selection by correctly allocating patients to risk groups and predicting their molecular response to the selected treatment. To achieve this objective, the research detailed here focuses on developing a prognostic model and a predictive model for use as a personalized medicine support system. The system will be considered a knowledge-based clinical decision support system that includes two models embedded in a decision tree. The main idea is to classify patients into risk groups using the prognostic model, while the patients identified as part of the high-risk group should be considered for more aggressive imatinib therapy or switched to secondgeneration TKI with close monitoring. For patients assigned to the low-risk group to imatinib should be predicted using the predictive model. The outcomes should be evaluated by comparing the results of these models with the actual responses to imatinib in patients from a previous medical trial and from patients admitted to hospitals. Validating such a predictive system could greatly assist clinicians in clinical decision-making geared toward individualized medicine. Our findings suggest that the system provides treatment recommendations that could help improve overall healthcare for CML patients. Study limitations included the impact of diversity on human expertise, changing predictive factors, population and prediction endpoints, the impact of time and patient personal issues. Further intensive research activities based on the development of a new predictive model and the method for selecting predictive factors and validation can be expanded to other health organizations and the development of models to predict responses to other TKIs.
Thesis (Ph.D.) -- University of Adelaide, School of Computer Science, 2018

Indiana University-Purdue University Indianapolis (IUPUI)
Precision medicine refers to the delivering of customized treatment to patients based on their individual characteristics, and aims to reduce adverse events, improve diagnostic methods, and enhance the efficacy of therapies. Among efforts to achieve the goals of precision medicine, researchers have used observational data for developing predictive modeling to best predict health outcomes according to patients’ variables. Although numerous predictive models have been reported in the literature, not all models present high prediction power, and as the result, not all models may reach clinical settings to help healthcare professionals make clinical decisions at the point-of-care. The lack of generalizability stems from the fact that no comprehensive medical data repository exists that has the information of all patients in the target population. Even if the patients’ records were available from other sources, the datasets may need further processing prior to data analysis due to differences in the structure of databases and the coding systems used to record concepts. This project intends to fill the gap by introducing an interoperable solution that receives patient electronic health records via Health Level Seven (HL7) messaging standard from other data sources, transforms the records to observational medical outcomes partnership (OMOP) common data model (CDM) for population health research, and applies predictive models on patient data to make predictions about health outcomes. This project comprises of three studies. The first study introduces CCD-TOOMOP parser, and evaluates OMOP CDM to accommodate patient data transferred by HL7 consolidated continuity of care documents (CCDs). The second study explores how to adopt predictive model markup language (PMML) for standardizing dissemination of OMOP-based predictive models. Finally, the third study introduces Personalized Health Risk Scoring Tool (PHRST), a pilot, interoperable OMOP-based model scoring tool that processes the embedded models and generates risk scores in a real-time manner. The final product addresses objectives of precision medicine, and has the potentials to not only be employed at the point-of-care to deliver individualized treatment to patients, but also can contribute to health outcome research by easing collecting clinical outcomes across diverse medical centers independent of system specifications.

This dissertation looks at how new sources of information should be incorporated into medical decision-making processes to improve patient outcomes and reduce costs. There are three fundamental challenges that must be overcome to effectively use personalized medicine, we need to understand: 1) how best to appropriately designate which patients will receive the greatest value from these processes; 2) how physicians and caregivers interpret additional patient-specific information and how that affects their decision-making processes; and finally, (3) how to account for a patient’s ability to engage in their own healthcare decisions. The first study looks at how we can infer which patients will receive the most value from genomic testing. The difficult statistical problem is how to separate the distribution of patients, based on ex-ante factors, to identify the best candidates for personalized testing. A model was constructed to infer a healthcare provider’s decision on whether this test would provide beneficial information in selecting a patient’s medication. Model analysis shows that healthcare providers’ primary focus is to maximize patient health outcomes while considering the impact the patient’s economic welfare. The second study focuses on understanding how technology-enabled continuity of care (TECC) for Chronic Obstructive Pulmonary Disease (COPD) and Congestive Heart Failure (CHF) patients can be utilized to improve patient engagement, measured in terms of patient activation. We shed light on the fact that different types of patients garnered different levels of value from the use of TECC. The third study looks at how data-driven decision support systems can allow physicians to more accurately understand which patients are at high-risk of readmission. We look at how we can use available patient-specific information for patients admitted with CHF to more accurately identify which patients are most likely to be readmitted, and also why – whether for condition-related reasons versus for non- related reasons, allowing physicians to suggest different patient-specific readmission prevention strategies. Taken together, these three studies allow us to build a robust theory to tackle these challenges, both operational and policy-related, that need to be addressed for physicians to take advantage of the growing availability of patient-specific information to improve personalized medication processes.

Indiana University-Purdue University Indianapolis (IUPUI)
The marker strategy design (MSGD) has been proposed to assess and validate predictive markers for targeted therapies and immunotherapies. Under this design, patients are randomized into two strategies: the marker-based strategy, which treats patients based on their marker status, and the non-marker-based strategy, which randomizes patients into treatments independent of their marker status in the same way as in a standard randomized clinical trial. The strategy effect is then tested by comparing the response rate between the two strategies and this strategy effect is commonly used to evaluate the predictive capability of the markers. We show that this commonly used between-strategy test is flawed, which may cause investigators to miss the opportunity to discover important predictive markers or falsely claim an irrelevant marker as predictive. Then we propose new procedures to improve the power of the MSGD to detect the predictive marker effect. One is based on a binary response endpoint; the second is based on survival endpoints. We conduct simulation studies to compare the performance of the MSGD with the widely used marker stratified design (MSFD). Numerical studies show that the MSGD and MSFD has comparable performance. Hence, contrary to popular belief that the MSGD is an inferior design compared with the MSFD, we conclude that using the MSGD with the proposed tests is an efficient and ethical way to find predictive markers for targeted therapies.

Yes
Appeals have been made for eye care professionals to start prescribing anti-myopia therapies as part of their routine management of myopic children. 1–3 These calls are fuelled by two key considerations. Firstly, that interventions to slow myopia progression have shown success in randomized controlled trials (RCTs) 4–7, and secondly, appreciation that the risk of sight-threatening complications rises dose-dependently with the level of myopia. 8,9 Notwithstanding existing gaps in knowledge regarding the efficacy of current treatments (see below), these considerations argue that myopia control interventions should be widely adopted, and that they should be instigated at an early age – especially in children most at risk – in order to reduce the final level of myopia. Therefore in managing a child with myopia, an eye care professional would have to decide not only which therapy to recommend, but at what age to start treatment. In this review we discuss the future role of genetic prediction in helping clinicians treat myopia.
NIHR Senior Research Fellowship. Grant Number: SRF‐2015‐08‐005

The control of type-I error is a focal point for clinical trials. On the other hand, it is also critical to be able to detect a truly efficacious treatment in a clinical trial. With recent success in supervised learning (classification and regression problems), artificial intelligence (AI) and machine learning (ML) can play a vital role in identifying efficacious new treatments. However, the high performance of the AI methods, particularly the deep learning neural networks, requires a much larger dataset than those we commonly see in clinical trials. It is desirable to develop a new ML method that performs well with a small sample size (ranges from 20 to 200) and has advantages as compared with the classic statistical models and some of the most relevant ML methods. In this dissertation, we propose a Similarity-Principle-Based Machine Learning (SBML) method based on the similarity principle assuming that identical or similar subjects should behave in a similar manner. SBML method introduces the attribute-scaling factors at the training stage so that the relative importance of different attributes can be objectively determined in the similarity measures. In addition, the gradient method is used in learning / training in order to update the attribute-scaling factors. The method is novel as far as we know. We first evaluate SBML for continuous outcomes, especially when the sample size is small, and investigate the effects of various tuning parameters on the performance of SBML. Simulations show that SBML achieves better predictions in terms of mean squared errors or misclassification error rates for various situations under consideration than conventional statistical methods, such as full linear models, optimal or ridge regressions and mixed effect models, as well as ML methods including kernel and decision tree methods. We also extend and show how SBML can be flexibly applied to binary outcomes. Through numerical and simulation studies, we confirm that SBML performs well compared to classical statistical methods, even when the sample size is small and in the presence of unmeasured predictors and/or noise variables. Although SBML performs well with small sample sizes, it may not be computationally efficient for large sample sizes. Therefore, we propose Recursive SBML (RSBML), which can save computing time, with some tradeoffs for accuracy. In this sense, RSBML can also be viewed as a combination of unsupervised learning (dimension reduction) and supervised learning (prediction). Recursive learning resembles the natural human way of learning. It is an efficient way of learning from complicated large data. Based on the simulation results, RSBML performs much faster than SBML with reasonable accuracy for large sample sizes.