Big data analytics in cardiovascular sciences

Abstract

Introduction It has been challenging for researchers to access granular electronic health record (EHR) data at scale in England. The National Institute for Health Research (NIHR) Health Informatics Collaborative (HIC) enables the sharing of routine EHR data across NHS hospitals for research. One emerging prospect is to use big data to traverse the translational spectrum. As an example of an early discovery phase study, I assessed the effect of invasive versus non-invasive management on the survival of patients aged 80 years or older with non-ST elevation myocardial infarction (NSTEMI) (SENIOR-NSTEMI Study). As an example of a later implementation phase study, I determined the relationship between the full spectrum of troponin level and mortality in patients in whom troponin testing was performed for clinical purposes (TROP-RISK Study).

Methods Five NHS Trusts contributed data: Imperial, University College London, Oxford, King’s and Guy’s and St Thomas’. Microsoft SQL was used to develop a dataset of 257,948 consecutive patients who had a troponin measured between 2010 and 2017. Phenotypically detailed data were extracted, including patient demographics, blood tests, procedural data, and survival status. All studies conducted were retrospective cohort studies. For the SENIOR-NSTEMI Study, eligible patients were 80 years or older who were diagnosed with NSTEMI. Mortality hazard ratios were estimated comparing invasive with non-invasive management. For the TROP-RISK Study, the relation between peak troponin level and all-cause mortality was modelled using multivariable adjusted restricted cubic spline Cox regression analyses.

Results For the SENIOR-NSTEMI Study, 2672 patients with NSTEMI were included who had a median age of 85 (interquartile range (IQR) 82-89) years of whom 59.8% received non-invasive management. During a median follow-up of 2.7 (IQR 1.0-4.5) years, the adjusted cumulative five-year mortality was 40% in the invasive and 63% in the non-invasive group (hazard ratio 0.52, 95% confidence interval 0.43-0.62). For the TROP-RISK Study, during a median follow-up of 1198 days (IQR 514-1866 days), 55,850 (21.7%) deaths occurred. There was an unexpected inverted U-shaped relation between troponin level and mortality in acute coronary syndrome (ACS) patients (n=120,049). The paradoxical decline in mortality at very high troponin levels may be driven in part by the changing case mix as troponin levels increase; a higher proportion of patients with very high troponin levels received invasive management.

Conclusion Routinely collected EHR data can be aggregated across multiple sites to create highly granular datasets for research which can be used to answer research questions that cross the translational spectrum. The SENIOR-NSTEMI Study demonstrates a survival advantage of invasive compared with non-invasive management of NSTEMI patients aged 80 years or older, who were underrepresented in previous trials. In the TROP-RISK Study, the inverted U-shaped relationship between troponin level and mortality in ACS patients demonstrates that assembling sufficiently large datasets can cast light on patterns of disease that are impossible to adequately define in single centre studies.