Risk prediction models for cardiovascular disease and overall mortality

Abstract

Prediction or prognostication is at the core of modern evidence-based medicine. Prediction of overall mortality and cardiovascular disease can be improved by a systematic evaluation of measurements from large-scale epidemiological studies or by using nested sampling designs to discover new markers from omics technologies.

In study I, we investigated if prediction measures such as calibration, discrimination and reclassification could be calculated within traditional sampling designs and which of these designs were the most efficient. We found that is possible to calculate prediction measures by using a proper weighting system and that a stratified casecohort design is a reasonable choice both in terms of efficiency and simplicity.

In study II, we investigated the clinical utility of several genetic scores for incident coronary heart disease. We found that genetic information could be of clinical value in improving the allocation of patients to correct risk strata and that the assessment of a genetic risk score among intermediate risk subjects could help to prevent about one coronary heart disease event every 318 people screened.

In study III, we explored the association between circulating metabolites and incident coronary heart disease. We found four new metabolites associated with coronary heart disease independently of established cardiovascular risk factors and with evidence of clinical utility. By using genetic information we determined a potential causal effect on coronary heart disease of one of these novel metabolites.

In study IV, we compared a large number of demographics, health and lifestyle measurements for association with all-cause and cause-specific mortality. By ranking measurements in terms of their predictive abilities we could provide new insights about their relative importance, as well as reveal some unexpected associations. Moreover we developed and validated a prediction score for five-year mortality with good discrimination ability and calibrated it for the entire UK population.

In conclusion, we applied a translational approach spanning from the discovery of novel biomarkers to their evaluation in terms of clinical utility. We combined this effort with methodological improvements aimed to expand prediction measures in settings that were not previously explored. We identified promising novel metabolomics markers for cardiovascular disease and supported the potential clinical utility of a genetic score in primary prevention. Our results might fuel future studies aimed to implement these findings in clinical practice.