Cardiovascular diseases (CVDs) are the main cause of mortality worldwide, counting for a third of world demises. Consequently, early detection and underlying factors of these pathologies can play a critical role in successful treatments. Many researchers have applied machine learning (ML) for mortality risk estimation in CVDs. However, this is difficult due to their complex and multifactorial nature and the lack of large, unbiased data collections. This thesis holds statistical analysis results and a binary classification model for CVDs mortality prediction based on the ESCARVAL-RISK study, a large cohort study (54,678 patients) running from January 2008 until December 2012. This study faces highly imbalanced classes that may lead to classification models with low specificity and sensitivity. This work proposes several ways to balance classes, including hyperparameter optimization and sample techniques tested over 15 different classification algorithms to overcome the problem. While the specificity is low, the proposed approach using SHapley Additive exPlanations (SHAP) identifies factors that may be optimal targets for intensified preventive interventions. / Kardiovaskulära sjukdomar är den främsta dödsorsaken i världen och står för en tredjedel av alla dödsfall i världen. Därför kan tidig upptäckt och underliggande faktorer för dessa sjukdomar spela en avgörande roll för framgångsrika behandlingar. Många forskare har tillämpat maskininlärning (ML) för uppskattning av dödlighetsrisker vid hjärt- och kärlsjukdomar. Detta är dock svårt på grund av deras komplexa och multifaktoriella natur och bristen på stora, opartiska datainsamlingar. Denna avhandling innehåller statistiska analysresultat och en binär klassificeringsmodell för att förutsäga dödligheten i hjärt- och kärlsjukdomar baserat på ESCARVAL-RISK-studien, en stor kohortstudie (54 678 patienter) som pågick från januari 2008 till december 2012. I studien finns mycket obalanserade klasser som kan leda till klassificeringsmodeller med låg specificitet och känslighet. I detta arbete föreslås flera sätt att balansera klasserna, inklusive optimering av hyperparametrar och provtagningstekniker som testats över 15 olika klassificeringsalgoritmer för att lösa problemet. Även om specificiteten är låg identifierar den föreslagna metoden med hjälp av SHapley Additive exPlanations(SHAP) faktorer som kan vara optimala mål för intensifierade förebyggande insatser.

Selected methods of mortality analysis focused on adults and the oldest age-groups Abstract Questions about human life span, longevity and mortality in general are natural to almost everyone. This Doctoral Thesis deals with one central question - whether some limit of human life span or of its improvements exists. It is rather a methodological work, therefore its aim is to introduce not only relevant theories but above all the methods usable in the mortality analysis focused on adults or the oldest-old. At the beginning the most important theories and opinions of scientist dealing with mortality are introduced. In the first half of the analytical part mainly the traditional and basic approaches are included. The theme of life tables is opened by an analysis of its construction in the Czech Republic, together with its possible modifications. As a result the independent macro code for the SAS software is attached in the electronic Appendix. This macro enables to calculate the unknown parameters of selected mortality laws by the method of weighted non-linear least squares and to produce the smoothed and extrapolated values of mortality rates. Using the individual life durations, life tables according to education attainment were constructed (also attached in the electronic Appendix). In the second half of the...

L’objectif de cette étude était d'étudier les différences de mortalité régionales en Ukraine, les structures par causes de décès et les tendances à différentes échelles spatiales. Tout d'abord, des contrastes géographiques nets ont été mis en évidence : pour les hommes, entre l'ouest et l'est de l'Ukraine, et pour les femmes - entre l'ouest et le sud-est. L’étude des disparités régionales en fonction de l’âge révèle que les régions les plus avancées sont caractérisées par une diminution de la mortalité chez les personnes les plus âgées. Dans les grandes villes, pour la plupart des causes de décès (à l'exception du cancer et des maladies infectieuses) vivre dans une grande métropole est un avantage pour survivre. Le cas de la ville de Slavoutytch qui avait particulièrement souffert de l'accident de Tchernobyl a fait l’objet d’une analyse spécifique. Deuxièmement, une attention particulière a été portée aux schémas régionaux de mortalité par causes médicales de décès et a montré que les régions ukrainiennes se trouvent à différents stades de la lutte contre les maladies dégénératives. Enfin, les particularités de la dynamique de deux dernières décennies ont été étudiées. Pendant cette période l'espérance de vie a connu une forte baisse, suivie d’une stagnation, puis d’un début de reprise. Ces évolutions se sont accompagnées d’une transformation des schémas régionaux de mortalité par cause. Ces changements, qu’ils soient majeurs ou plus modestes n’ont pas été réellement identifiés par les pouvoirs publics et aucune avancée sanitaire décisive ne s’est produite dans aucune des régions de l'Ukraine. Il n'y a eu ni stratégies, ni mesures ou réformes efficaces mises en œuvre pour permettre une amélioration substantielle de la santé de la population. / The aim of this study was to investigate regional mortality differences in Ukraine, cause-of-death patterns and trends at different spatial scales. First of all, general contrasts were established: for males between the west and east of Ukraine, while for females between the west and south-east. Study of regional disparities in respect of different age groups revealed that regions that succeeded the most are those characterized by decreased mortality at older ages. As for big cities, for the most of cases of death (except cancer and infectious diseases) living in a big metropolis is an advantage to surviving. The special case of the city of Slavutych that suffered the most from Chernobyl accident was studied. Second, special attention was given to regional patterns of mortality for different medical cause of death, and which showed that Ukrainian regions are at different stages in the fight against man-made and degenerative diseases. Lastly, peculiarities of dynamics over last two decades were investigated. During this period life expectancy has experienced a sharp decrease, followed by stagnation, and recovery. This evolution was accompanied by transformation of regional cause-specific mortality patterns. These changes, larger and smaller, were not actually accounted for by public authorities and no important breakthroughs happened in any Ukraine region. There were no efficient strategies, measures or reforms implemented that would allow substantial improvements in the health of the population.