부산대학교 도서관

Hypothesis: Risk of death estimation can optimize the management of ambulatory patients with heart failure (HF). Objectives: To determine the chronobiology of death in these patients, and develop a mortality predictive model that improves the estimation of risk. Methods: Prospectively, demographic, clinical, biochemical and echocardiographic variables of two ambulatory HF patients cohorts were collected. The UIC-Sant Pau cohort consists of 1233 patients and they were controlled at the Hospital de la Santa Creu i Sant Pau HF unit from 2003 till 2008. The REDINSCOR cohort consists of 1293 patients from the REDINSCOR registry (National network of HF in Spain). Prospective observational study conducted with patients of 18 different Spanish hospitals that were included between 2007 and 2011. Firstly, we did a retrospective analysis to describe the distribution of patients death according to the circadian pattern, day of the week, monthly and seasonal (chronobiology) from the UIC - Sant Pau cohort. Subsequently, with 653 patients of the same cohort (excluding patients included simultaneously in the REDINSCOR cohort), we defined the UIC Sant Pau reduced cohort as a derivation cohort to develop predictive models of death at 1 and 4 years. We applied traditional statistical methods (Cox regression) and methods based on the use of Classification and regression tree analysis (CART) as an alternative to traditional statistical methods. We compared the predictive performance of the models obtained by Cox regression and CART analysis. Finally, we performed two types of validation of the obtained predictive models: an internal cross-validation and an external validation, by applying the models obtained to the sample of validation ("REDINSCOR cohort"). Results: In the UIC Sant Pau cohort patients did not show circadian or dependent on the day of the week mortality peaks. There is a marked mortality peak during the winter season. Predictive models obtained by Cox regression and analysis of CART showed a very similar discriminatory capacity. Prediction of death at 1 year CART’s model uses only three variables (NT-proBNP, age and clinical score) to classify patients into groups of high, intermediate and low risk of mortality. It has remarkable predictive accuracy (AUC: 0.72; 95% CI: 0,67-0,77), reproduced when the model is applied to the validation cohort (AUC: 0.67; 95% CI: 0,62-0,71) (p=0.12). CART's prediction model of death at 4 years, uses the NT-proBNP, clinical score and sodium variables, to discriminate patients in the three risk groups. Its predictive accuracy is good (AUC: 0.70; 95% CI: 0,66-0,74), and is also reproduced when applied to the validation cohort (AUC: 0.66; 95% CI: 0,63-0,70) (p=0.23). Conclusions: The mortality of ambulatory HF patients does not follow the circadian pattern observed in the general population for cardiovascular events, but a seasonal rhythm with higher mortality in winter is detected. We propose new predictive mortality models at 1 and 4 years, easy to implement and interpret. That may help to advance in the study, management, and prognosis of ambulatory patients with heart failure.