requirements for the degree of Doctor of Philosophy
When a patient presents to a hospital with symptoms of cardiovascular disease, one of the first courses of action is to estimate the patient’s risk of an adverse outcome. The process of categorizing patients by risk level, known as riskstratification, is an essential step in assigning appropriate therapy. Riskstratification models, which aid clinicians in this task, consist of feature sets that are combined by an algorithm to yield a score. In addition to the performance of the model, a key factor in model de- velopment is clinician acceptance of the score. One way to bolster clinician acceptance is to choose parsimonious feature sets to be used in risk scores that are convenient to integrate into the clinical workflow. A second consideration is establishing clinician trust in the model predictions. This is particularly important when using models that are difficult to explain to clinicians and when it is not straightforward to iden- tify failure modes for the model. Providing clinicians with a measure of how much to trust a given prediction from a model is one way to encourage the use of models that are difficult to interpret. In this thesis, we consider the problem of developing clinically useful risk models using real clinical data. We begin by discussing how to choose clinical variables in a data-driven fashion in the context of acute coronary syn- drome. We present a risk score that can accommodate a variable number of inputs and demonstrate that it has superior performance to the Global Registry of Acute Coronary Events (GRACE) risk score, particularly on the difficult to risk stratify low-risk patients (AUC 0.754 vs. 0.688 for the GRACE score, p < 0.007). We then discuss the development of a risk score for aortic stenosis (AS) using both data-driven feature selection and expert opinion. We show that the model performs well on pa- tients with moderate to severe aortic stenosis (AUC 0.74), as well as on the difficult to risk stratify low gradient severe AS subgroup (2-5 year hazard ratios ≥ 3.3, p < 0.05). Finally, we develop a method to identify unreliable predictions in clinical risk models and show, using the GRACE dataset, that we can identify subgroups of poor model performance to aid in bolstering clinician trust of risk models.
Most riskstratification methods use expert opinion to identify a fixed number of clinical variables that have prognostic significance. In this study our goal was to develop improved metrics that utilize a variable number of input parameters. We first used Bootstrap Lasso Regression (BLR) – a Machine Learning method for selecting important variables – to identify a prognostic set of features that identify patients at high risk of death 6-months after presenting with an Acute Coronary Syndrome. Using data derived from the Global Registry of Acute Coronary Events (GRACE) we trained a logistic regression model using these features and evaluated its performance on a development set (N = 43,063) containing patients who have values for all features, and a separate dataset (N = 6,363) that contains patients who have missing feature values. The final model, Ridge Logistic Regression with Variable Inputs (RLRVI), uses imputation to estimate values for missing features. BLR identified 19 features, 8 of which appear in the GRACE score. RLRVI had modest, yet statistically significant, improvement over the standard GRACE score on both datasets. Moreover, for patients who are relatively low-risk (GRACE≤87), RLRVI had an AUC and Hazard Ratio of 0.754 and 6.27, respectively, vs. 0.688 and 2.46 for GRACE, (p < 0.007). RLRVI has improved discriminatory performance on patients who have values for the 8 GRACE features plus any subset of the 11 non-GRACE features. Our results demonstrate that BLR and data imputation can be used to obtain improved riskstratification metrics, particularly for patients who are classified as low risk using traditional methods.
1. Markusse IM, de Vries-Bouwstra JK, Han K, van der Lubbe PA, Schouffoer AA, Kerstens PJ, et al. Feasibility of tailored treatment based on riskstratification in patients with early rheumatoid arthritis. Arthritis Res Ther. 2014;16:430. 2. Smolen JS, Landewé R, Breedveld FC, Buch M, Burmester G, Dougados M,
A useful riskstratification algorithm places more people at the extremes rather than in the middle cate- gories of the risk distribution, thus providing clini- cians with clear guidance for action. The present evidence suggests that using information on job strain is unlikely to improve identification of indivi- duals at the highest risk of developing CHD. Our find- ings resemble those of previous studies that have examined the utility of adding other important risk markers for CHD (e.g. socio-economic position, caro- tid intima-media thickness and C-reactive protein) to
Thus, preoperative characterization and riskstratification of indeterminate adnexal masses are unmet clinical needs. A validated scoring system that standardizes imaging reports and categorizes the risk of malignant neoplasm in these women would be useful as a triage test to decide whether surgery is appropriate and, if so, the extent of surgery required. This could potentially reduce unnecessary or overextensive surgery. In the literature, various scoring systems have been developed based on clinical, biochemical (eg, cancer antigen 125 [CA 125] or human epididymis protein 4 [HE 4] levels), and ultrasonographic criteria. 10,11
Differentiated thyroid cancer (DTC) accounts for 90% of all thyroid cancers [ 1 ]. It has an excellent prognosis, and long-term survival is sustained in the vast majority of the patients as a result of traditional treatment by surgery and radioactive iodine (RAI) [ 2 ]. Regular patient fol- low-ups are consequently adapted according to the risk of disease recurrence or persistence [ 3 – 5 ]. Most oncological studies have evaluated recurrence-free survival and long-term remis- sion, however, due to its excellent prognosis, long-term survival is particularly interesting in DTC. Numerous studies have identified multiple factors correlated to overall survival rates. Two particularly influential factors that impact overall survival are RAI and total thyroidec- tomy [ 6 , 7 ]. Furthermore, age over 45 years old and male sex have been found to be significant factors that predict overall survival in patients with low-risk DTC, as well as TNM in poorly differentiated thyroid cancers [ 8 – 11 ]. However, studies evaluating long-term disease-specific survival have not focused on the role of thyroglobulin, or on that of the new riskstratification system suggested by the ATA in the prediction of the overall survival. To the best of our knowledge, the impact of pre- and post-ablation thyroglobulin and that of the initial ATA riskstratification on long-term disease-specific survival were rarely studied.
Interstitial lung disease (ILD) is the most common pulmonary manifestation of rheumatoid arthritis (RA), occurring in *10% of patients. It has emerged in recent series as a key prognos- tic factor including survival [ 1 ]. RA-ILD shares some genetic and phenotypic similarities with other fibrotic diseases including idiopathic pulmonary fibrosis, supporting the use of the same drugs in these conditions [ 2 , 3 ]. Of great interest, the INBUILD trial recruited a broad range of progressive fibrosing ILD, including patients with RA; it showed that RA patients who received nintedanib had a slower rate of progression of ILD than those who received placebo [ 3 ]. Nev- ertheless, the big challenge for rheumatologists is now the risk-stratification of RA patients for ILD. Chest high-resolution computed tomography (HRCT) is the gold standard for RA-ILD diagnosis, but costs and ionizing radiation may limit its use in clinical practice. Thus, circulat- ing biomarkers could aid in this risk-stratification, as recently reported in systemic sclerosis (SSc)-associated ILD [ 4 – 6 ]. Indeed, circulating lung epithelial-derived surfactant protein D (SPD), CCL-18 and Krebs von den Lungen-6 glycoprotein (KL-6) were identified as relevant diagnostic and prognostic markers of SSc-ILD. Our objective was to evaluate the merit of these 3 circulating markers for the diagnosis and the progression of RA-ILD.
Cox proportional hazards models were used to identify factors associated with an increased risk of death and hospi- talization (cardiovascular and HF). The follow-up time for each patient was calculated from the date of their ﬁrst eval- uation to the date of reaching death or hospitalization for cardiovascular cause or to the date of their most recent eval- uation. Odds ratios and hazard ratios (HRs) are presented with their 95% con ﬁdence intervals. In order to account for potential confounding, models were adjusted ﬁrstly for age, NT-proBNP (either as a log-transformed continuous var- iable or as a categorical variable using the thresholds detailed earlier 22 ), and glomerular ﬁltration rate and
Predictors of events
The clinical and echocardiographic characteristics of the patients who remained asymptomatic and those who experienced an event are listed in table 1. No clinical parameters (age, sex, risk factors, blood pressure) allowed signiﬁcant distinction between the two groups. By contrast, patients who developed events had markedly reduced aortic valve area and systemic arterial compliance and had signiﬁcantly higher valvulo-arterial impedance and larger left atrium than those remaining asymp- tomatic. Although conventional indices of global LV systolic function (LV volumes and ejection fraction) were statistically similar in the two groups, the LV longitudinal myocardial deformation was impaired in the group of patients who developed an event. The response to exercise was more often abnormal in these patients. With multivariable Cox regression analysis, four parameters emerged as independently associated with the combined end-point: peak aortic jet velocity (p¼0.04), LV longitudinal myocardial deformation (p¼0.03), valvulo-arte- rial impedance (p¼0.001) and indexed left atrial area (p¼0.007) (table 2). Using receivereoperator characteristic curve analysis (ﬁgure 2), a peak aortic jet velocity $4.4 m/s, a LV longitudinal myocardial deformation #15.9%, a valvular-arterial impedance
inﬂuence of treatments on platelet phosphoACC re- mains to be determined in this high-risk population. Numerous studies have reported that high on- treatment platelet reactivity and platelet activation indices reveal high-risk patients in CAD population (20). Particularly, increased circulating activated platelets, evidenced by monocyte- or neutrophil- platelet aggregates or P-selectin expression, can be detected in high-risk patients versus normal subjects. However, this increase is marginal and involves a small proportion of the total platelet pool (21 –23) . In our study, AMPK-ACC signaling activation was observed in circulating platelets. The absence of protein kinase C substrate phosphorylation, a sign of platelet activation (24), supports the theory that increased phosphoACC and platelet activation are unrelated. Even if thrombin is the principal agonist of platelet AMPK activation ex vivo (4), we found no association between ThG markers and phosphoACC. Therefore, platelet phosphoACC is not caused by thrombin-induced platelet activation in patients. Thus, our ﬁndings suggest that platelets of high-risk CAD patients exhibit altered metabolic phenotypes characterized by activation of AMPK-ACC signaling, independent of platelet activation state.
Brugada syndrome: Diagnosis, risk stratiﬁcation and management 193
Figure 3. Indication for implantation of an implantable cardioverter deﬁbrillator (ICD), according to the risk of sudden cardiac death
(SCD). Modiﬁed from Refs. [1,2] . The risk of SCD is represented with a red arrow. The red boxes represents an indication for ICD implantation;
the green box represents a patient who should not be implanted, according to the latest guidelines; the orange boxes are not addressed in
my study concerns the strategies of temporal stratification in astro- physics and archaeology’s images. i plan to take into consideration a problematic issue which is transversal to contemporary scientific disci- plines: the representation of the stratification of temporal layers. Whilst the disciplines chosen for this research, astrophysics and archaeology, are very different in terms of their subject of study, they do have a com- mon point which is that a large number of their objects are invisible to the naked eye. more precisely, my objective is to compare the strategies of visual representation of the temporal layers stratified in these invisi- ble objects.
We look at the eﬀects of the radial density stratification on the stability of the q-vortex, a commonly accepted model for aircraft trailing vortices. It has been demonstrated that the 2D Lamb–Oseen vortex develops a Rayleigh– Taylor instability when its core is heavier than the surrounding fluid (Joly, Fontane & Chassaing 2005, Sipp et al. 2005). The underlying mechanism relies on baroclinic vorticity generation due to any misalignment between the density gradient and the centripetal acceleration field. The instability is triggered provided that the density decreases radially somewhere in the vortex core. This mechanism is also active in the 3D trailing vortex and aﬀects its stability characteristics due to the addition of an axial component in the acceleration field.
L’incidence des cheminements institutionnels
Pour comprendre l’effet de la stratification scolaire sur les inégalités d’accès à l’université, on peut avancer une seconde piste d’interprétation, complémentaire de celle que nous venons de développer concernant les différences de conditions de scolarisation et d’apprentissage. Elle est dérivée des travaux américains et français sur les effets des « cheminements institutionnels » et des établissements scolaires fréquentés sur l’accès à l’université et en particulier aux établissements « d’élite ». Depuis les années 1980 aux États-Unis (Hill, 2008; Letendre, 1996) et plus récem- ment en France (Buisson-Fenet et Draelants, 2013; Van Zanten, 2009), ces travaux démontrent la production continuelle de mécanismes qui « permettent d’instaurer des clôtures institutionnelles » au sein du système éducatif, et donnent ainsi la possibilité à certains élèves « d’accéder progressivement à des espaces éducatifs protégés des effets de la massification scolaire » (Draelants, 2013, p. 4). Outre l’analyse de « barrières structurelles » (sélection économique ou méritocratique) permettant de protéger l’accès à certaines universités prestigieuses, certains travaux « indiquent que le système d’aspirations se construit au cours des “enchaînements” entre établissements et types d’établissements qui dessinent les carrières insti- tutionnelles d’élèves » (Draelants, 2013, p. 4). Ainsi, ces études ont notamment cherché à comprendre les mécanismes par lesquels se formaient les aspirations aux études supérieures, par-delà les effets de l’origine sociale des élèves. Les travaux de Draelants en Belgique montrent qu’en plus de la trajectoire scolaire antérieure de l’élève et de son origine socioculturelle, les aspirations à poursuivre des études supérieures sont modelées par un « effet établissement » (en particulier les pratiques d’information des écoles sur les études universitaires).