The association between troponin level and mortality in patients admitted to hospital with acute stroke (NIHR Health Informatics Collaborative TROP-STROKE study)

Introduction

Acute stroke accounts for significant morbidity and mortality globally. The role of troponin for risk stratification in stroke is unclear.

Purpose

The aims of this study were to assess the relationship between peak troponin and mortality in patients with ischaemic stroke, haemorrhagic stroke, or subarachnoid haemorrhage and to compare this with the predictive value of first troponin or dynamic troponin change.

Methods

A retrospective cohort study was carried out using the National Institute for Health Research Health Informatics Collaborative Cardiovascular dataset of all consecutive patients who had a troponin measured at five hospitals between 2010 and 2017. Patients with at least one troponin measurement and a primary diagnosis of ischaemic stroke, haemorrhagic stroke or subarachnoid haemorrhage during a hospital admission were included. The main exposure variables were first and peak troponin, and dynamic troponin change, and the main outcome was all-cause mortality. Results were analysed using multivariable adjusted restricted cubic spline Cox regression. Survival analyses were adjusted for troponin assay, assay sensitivity (standard or highly sensitive), number of troponin measurements, age, sex, C-reactive protein level, white blood cell count, platelet count, haemoglobin, estimated glomerular filtration rate, angiography during admission, acute coronary syndrome during admission, and cardiovascular history (history of diabetes mellitus, myocardial infarction, heart failure, hypertension, stroke or atrial fibrillation). Receiver Operator Characteristic (ROC) curves were used to assess the predictive value of each exposure variable.

Results

4,712 patients were included in the analysis (ischaemic stroke: 3,346; haemorrhagic stroke: 718; subarachnoid haemorrhage: 648). Peak troponin was above the upper limit of normal in 47.4% of ischaemic stroke patients, 52.8% of haemorrhagic stroke patients, and 57.1% of subarachnoid haemorrhage patients. Patients with elevated peak troponin were older and had more cardiovascular risk factors.

A direct positive relationship was seen between peak troponin level and mortality hazard ratio in all three stroke subtypes (Figure 1). This relationship was consistent when considering dynamic troponin fold change for ischaemic or haemorrhagic stroke. For all three stroke subtypes, there was no added predictive value of peak troponin or dynamic troponin change over first troponin in predicting mortality (Figure 2).

Conclusions

A positive peak troponin was associated with increased mortality in patients presenting with ischaemic stroke, haemorrhagic stroke, or subarachnoid haemorrhage. Overall, serial troponin measurements may not improve mortality prediction beyond a single measurement. These findings may have implications for risk stratification of patients with acute stroke syndromes.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIHR Imperial Biomedical Research Centre, as part of the NIHR Health Informatics Collaborative, and 2) British Heart Foundation

The prognostic significance of troponin level in patients with malignancy (NIHR Health Informatics Collaborative TROP-MALIGNANCY study)

Background

Cardiac troponin is commonly raised in patients with malignancy and may aid clinicians in risk prediction. The prognostic significance of raised troponin in these patients with known malignancies remains unclear.

Purpose

We sought to investigate the relation between troponin and mortality in a large, well characterised cohort of patients undergoing cardiac troponin testing with a concomitant malignancy.

Methods

A retrospective cohort study was carried out using the National Institute for Health Research Health Informatics Collaborative Cardiovascular dataset of all consecutive patients who had a troponin measured at five hospitals between 2010 and 2017. Patients with a primary inpatient diagnosis of malignancy who had at least one cTn measurement during their hospital stay were identified. Patients were classified into solid tumour or haematological malignancy subgroups. Survival analyses were performed using multivariate Cox regression analyses and Kaplan-Meier plots. Cox regression analyses were adjusted for age, gender, C-reactive protein, haemoglobin, platelet count, white cell count, acute coronary syndrome, diabetes mellitus, heart failure, chronic kidney disease, chronic obstructive pulmonary disease, atrial fibrillation and angiography. The peak cTn level (highest level measured), standardised to the upper limit of normal (ULN), was used for all analyses.

Results

5571 patients undergoing troponin testing had a primary diagnosis of malignancy and comprised of twenty-one different cancer types. 4649 patients were diagnosed with solid tumours and 922 patients were diagnosed with haematological malignancies. Patients with raised troponin had a higher burden of cardiovascular comorbidities compared to patients with a troponin level below the ULN. The median follow-up in the cohort was 14 months (interquartile range 2–39 months). At 1-year follow-up, 2495 (42%) of patients died.

Figure 1 shows Kaplan-Meier plots for patients stratified by troponin level. Patients with a troponin level ≥1xULN had a higher risk of death compared to patients with a troponin level <1xULN (Figure 1A). A similar trend was shown in cancer subtypes (Figure 1B, C). Raised troponin was an independent predictor of mortality in all patients with malignancy (adjusted hazard ratio 1.66, 95% confidence interval [CI] 1.52–1.81), in solid tumours (adjusted hazard ratio 1.63, 95% CI 1.48–1.81) and in haematological malignancy (adjusted hazard ratio 1.75, 95% CI 1.44 to 2.13) when compared to patients with a troponin level <1xULN.

Conclusion

A raised troponin was associated with an increased mortality risk in patients with malignancy regardless of cancer subtype. Stratification of mortality risk using troponin may help guide clinicians in making management decisions for patients with malignancy.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIHR Imperial Biomedical Research Centre, as part of the NIHR Health Informatics Collaborative, and 2) British Heart Foundation

Developing informatics infrastructure to curate datasets using electronic health record data from five hospitals for translational cardiovascular research

Introduction

It has been challenging for researchers to access granular electronic health record (EHR) data at scale. One emerging prospect is to use big data to traverse the translational spectrum from an early discovery phase to a later implementation phase.

Purpose

To create a research-ready dataset to support translational research in cardiovascular medicine, using routinely-collected EHR data from multiple hospitals. As an early discovery phase study, we estimated the effect of invasive versus non-invasive management on the survival of patients with non-ST elevation myocardial infarction (NSTEMI) aged 80 years or older (SENIOR-NSTEMI Study). As a later implementation phase study, we 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

Using Microsoft SQL we developed a dataset of 257948 consecutive patients who had a troponin measured between 2010 and 2017 at five hospitals. We extracted phenotypically detailed data, including demographics, blood tests, procedural data, and survival status.

For the SENIOR-NSTEMI Study, eligible patients were 80 years or older who were diagnosed with NSTEMI. We estimated mortality hazard ratios comparing invasive with non-invasive management. For the TROP-RISK Study, we modelled the relation between peak troponin level and all-cause mortality using multivariable adjusted restricted cubic spline Cox regression analyses.

Results

For the SENIOR-NSTEMI Study, 1500 patients with NSTEMI were included who had a median age of 86 (interquartile range (IQR) 82–89) years of whom (845 [56%]) received non-invasive management. During a median follow-up of 3 (IQR 1.2–4.8) years, the adjusted cumulative five-year mortality was 36% in the invasive and 55% in the non-invasive group (hazard ratio 0.68, 95% confidence interval 0.55–0.84).

For the TROP-RISK Study, during a median follow-up of 1198 days (IQR 514–1866 days), 55850 (21.7%) deaths occurred. There was an unexpected inverted U-shaped relation between troponin level and mortality in acute coronary syndrome (ACS) patients (n=120049) (Figure 1A). 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 (Figure 1B).

Conclusion

Routine EHR data can be aggregated across multiple sites to create highly granular datasets for research. The SENIOR-NSTEMI Study showed a survival advantage of invasive compared with non-invasive management of elderly patients with NSTEMI, who were underrepresented in previous trials. The inverted U-shaped relationship between troponin and mortality in ACS patients in the TROP-RISK Study demonstrates that assembling sufficiently large datasets can cast light on patterns of disease that are impossible to adequately define in single centre studies.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): 1) NIHR Imperial Biomedical Research Centre, as part of the NIHR Health Informatics Collaborative, and 2) British Heart Foundation

The association between age, troponin level, and mortality in patients hospitalised with acute pulmonary embolism (NIHR Health Informatics Collaborative TROP-PE study)

Background

A positive cardiac troponin (cTn) is an independent predictor of short-term mortality in individuals presenting with acute pulmonary embolism (PE). However, there is limited evidence regarding the impact age has on the association between cTn levels and mortality in patients with PE.

Purpose

The aim of our study was to investigate the relationship between cTn level, age, and all-cause mortality, in hospitalised patients diagnosed with an acute PE.

Methods

A retrospective cohort study using the National Institute for Health Research Health Informatics Collaborative Cardiovascular dataset of all consecutive patients who had a troponin measured at five hospitals between 2010 and 2017. Patients admitted to hospital with a primary diagnosis of PE with at least one cTn measurement were included. We modelled the relation between peak troponin level and all-cause mortality using multivariable adjusted restricted cubic spline Cox regression analyses. Effect estimates were adjusted for age, gender, high-sensitivity troponin assay, C-reactive protein, haemoglobin, platelet count, white cell count, creatinine, sodium, potassium, diabetes, hypertension, hypercholesterolaemia, acute coronary syndrome, atrial fibrillation, heart failure, acute kidney injury, chronic kidney disease, obstructive lung disease, inflammatory disorders, pneumonia and malignancy. The peak cTn level (highest level measured), standardised to the upper limit of normal (ULN), was used for all analyses.

Results

1,477 patients with at least one cTn measurement and a diagnosis of acute PE were included. During a median follow-up of 34.8 months, there were 290 (19.6%) deaths. Elevated cTn (>1xULN) was associated with a hazard ratio (HR) of 3.29 (95% confidence interval [CI] 1.95–5.53) for 30-day mortality and 2.12 (95% CI 1.63–2.75) for 3-year mortality. Higher cTn levels were progressively associated with a higher mortality risk, reaching a maximum HR of 2.59 (95% CI 1.64–4.09) at 141xULN (Figure 1). Younger patients (<55 years), compared with those aged over 55, had the highest 3-year HR associated with a positive cTn of 2.94 (95% CI 1.48–5.82) despite having the lowest troponin levels (mean 7.01xULN) on admission (Figure 2).

Conclusion

Elevated cTn, at all ages, is associated with an increased mortality risk in patients presenting with PE, with increasing cTn levels conferring a progressively worse long-term prognosis. Elevated cTn, no matter how small, needs to be taken seriously, particularly in young patients with an acute PE.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): 1) NIHR Imperial Biomedical Research Centre, as part of the NIHR Health Informatics Collaborative, and 2) British Heart Foundation

Coronary haemodynamics associated with left ventricular hypertrophy in aortic stenosis and hypertension

Background

Left ventricular hypertrophy (LVH) occurs in both aortic stenosis (AS) and hypertension (HT) due to an increase in afterload. However, in AS there is an increase in resting coronary flow (per gram of LV) while in HT it is reduced.

Wave intensity analysis (WIA) is a well-established method of characterising and quantifying the energies that drive coronary flow. Energies propagating from the proximal vessel (aorta and systemic arteries) interact with energies travelling from the distal end (myocardial microcirculation). WIA allows the separation of these energies into the waves that drive cyclic changes in coronary flow.

Purpose

We aimed to compare coronary flow patterns in LVH secondary to AS with coronary flow patterns in LVH secondary to HT.

Methods

Thirty-one participants were recruited (mean age 63, 18 female), 10 with LVH and severe AS, 11 with LVH and HT, and 10 with no LVH and no AS. Participants underwent invasive pressure and Doppler velocity measurements in each of the left coronary arteries and echocardiography. We applied WIA.

Results

Mean resting coronary flow per gram of LV tissue (Fig. 1) was increased in participants with LVH secondary to AS (1.62±0.60 ml/min/g) and reduced in participants with LVH secondary to HT (0.49±0.27 ml/min/g), compared to participants with no LVH and no AS (1.47±0.73 ml/min/g).

We observed marked differences between the magnitudes of the waves driving coronary flow in the three groups (Fig. 2). Forward and backward travelling waves are shown above and below the line respectively. Waves causing acceleration of coronary forward flow are shown as black and waves causing deceleration are shown in white.

Wave 6, the backwards decompression wave (BDW), is particularly important for myocardial perfusion. The BDW corresponds to the diastolic 'suction' of blood down the coronary arteries during myocardial relaxation.

The energy of the BDW was increased in LVH secondary to AS (31.1x103 W m–2 s–2) but was reduced in LVH secondary to HT (12.3x103 W m–2 s–2) (p<0.05), compared to participants with no LVH and no AS (14.3x103 W m–2 s–2).

The energy of the BDW correlated with LV cavity pressure (r=0.84, p<0.001) and diastolic time (r=−0.62, p<0.001) only in LVH secondary to AS participants. In contrast, the BDW correlated with LV mass (r=−0.49, p=0.03) in participants with LVH secondary to HT and with no LVH and no AS, but not in participants with LVH secondary to AS.

Conclusions

In hypertension, LVH is associated with reduced mean coronary flow and reduced myocardial “suction” during diastole, presumably by the reduction in myocardial compliance associated with diastolic dysfunction.

However, in AS the large pressure gradient between the LV cavity and the aorta results in a large contractile force which is generated in systole and then released in diastole. This large diastolic force overwhelms any local impairment caused by the hypertrophied myocardium and contributes to high resting coronary flow in AS.

Funding Acknowledgement

Type of funding sources: None.

Tissue Doppler E prime velocity and E/E prime predict 19-year cardiovascular mortality in hypertension

Background

We have previously shown that tissue Doppler assessments of left ventricular (LV) diastolic function predict cardiac events in a hypertensive population over a period of 4 years (1). These out-performed traditional echocardiographic measures in a well-treated hypertensive population.

Purpose

We aimed to test whether tissue Doppler assessment of LV diastolic function would predict cardiovascular (CV) mortality in the Hypertension Associated Cardiovascular Disease sub-study of the Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT).

Methods

ASCOT was a multicentre randomised trial with a 2x2 factorial design (2). Inclusion criteria for the study included hypertension and three other CV risk factors, including male sex and age over 55. Protocols, including for echocardiography, have been detailed previously (2, 3).

This study comprised the 519 patients recruited to the St Mary's Hospital site of the ASCOT study, who were followed for 19 years with mortality flagged by the Office for National Statistics. We have used all reported deaths on or before 31st January 2019. CV deaths include deaths due to coronary heart disease (CHD), stroke and other CV causes.

Echocardiography was performed one year after blood pressure control. mean tissue Doppler E' was calculated as the average of septal, lateral and inferior wall measurements over three cycles. The ratio of the transmitral Doppler E wave velocity and the composite mean of E' was used to calculate E/E' ratio.

Statistical analysis was performed using Python including multivariable Cox proportional hazards regression. A two-sided P-value <0.05 was considered statistically significant.

Results

After a median of 19 years, 317 patients survived (mean age at baseline 60.7y, 38 female) and 202 did not (mean age 68.1y, 30 female). Twenty-three deaths were due to CHD, 11 were due to stroke, 27 were due to other CV causes, and 76 were due to cancer. Baseline characteristics were not significantly different between those who survived to follow up and those who did not.

Unadjusted analysis showed a strong association between CV mortality and E' (HR=0.74, p<0.005) and E/E' (HR=1.18, p<0.005) (Table 1). The association between CV mortality and E' was attenuated slightly but persisted after adjusting for age and sex (HR=0.83, p=0.02) and after adjusting for age, sex and systolic BP (HR=0.83, p=0.03).

The association between CV mortality and E/E' was attenuated but persisted after adjusting for age and sex (HR=1.12, p=0.01) and after adjusting for age, sex and systolic BP (HR=1.11, p=0.04).

There was a weak association between all-cause mortality and both E/E' and E', which was null after adjusting for age and sex. Figure 1 shows unadjusted Kaplan Meier survival curves for E'.

Conclusions

Tissue Doppler E' velocity and E/E' predicted 19-year cardiovascular mortality in a hypertensive population independent of age, sex and systolic blood pressure.

Funding Acknowledgement

Type of funding sources: Private grant(s) and/or Sponsorship. Main funding source(s): The HACVD substudy was supported by an investigational grant by Pfizer International, New York, NY, USA. The principal funding source for ASCOT was Pfizer, New York, NY, USA, additional funding was also provided by Servier Research Group, Paris, France, and Leo Laboratories, Copenhagen, Denmark.

Myocardial strain by 3D-speckle tracking echocardiography predicts long-term risk of cardiovascular morbidity and mortality in the general population: the Southall And Brent Revisited (SABRE) study

Background

Both left ventricular (LV) ejection fraction (EF) and Global Longitudinal Strain (GLS) by 2D-echocardiography predict mortality and cardiac events, and GLS may be superior to EF. 3D-speckle tracking echocardiography (3D-STE), a recently validated method, allows simultaneous assessment of EF, GLS and principal tangential strain (PTS), but its prognostic utility in the general population is unknown.

Purpose

We hypothesized that 3D-STE derived LV myocardial strains predict a composite of cardiac endpoints, and that GLS would be a better prognostic marker than EF. We also investigated the utility of PTS compared with GLS and EF.

Methods

A total of 529 individuals (69±6y; 76.6% male) from SABRE study, a UK-based tri-ethnic community cohort, underwent health examinations. The association between 3D-STE EF or multidirectional myocardial strains and a composite cardiac endpoints comprising coronary heart disease (fatal/non-fatal), heart failure hospitalization, new-onset arrhythmia was determined using Cox proportional hazards models with and without adjustment for potential confounders and Harrell's C statistics were calculated. Associations with cardiovascular (CV) mortality was examined as a secondary objective. The incremental value of 3D-STE EF, GLS and PTS in improving CV risk stratification by the established Framingham risk score (FRS) was investigated using a likelihood ratio test on a series of nested Cox proportional hazards models.

Results

During follow-up (median, 8y), there were 56 composite cardiac endpoints and 24 CV deaths. EF and radial strain were negatively associated, while GLS, global circumferential strain and PTS were positively associated with the composite cardiac endpoints in unadjusted models (Table 1). Associations were only marginally affected by adjustment for potential confounders although confidence intervals of the estimate increased slightly (Table 1). There was little difference in the C-statistics for EF, GLS or PTS for the composite cardiac endpoints (Table 1). Associations with CV mortality were generally weaker and only GLS showed some evidence of a positive association with CV mortality in unadjusted and adjusted models (Table 1). Compared to EF and GLS, PTS most improved the predictive value (model fit) of FRS for composite cardiac endpoints (Table 2). None of the measures convincingly improved calibration for CV mortality.

Conclusions

3D-STE-derived LV myocardial strains predicted adverse cardiac events and CV mortality in a multi-ethnic sample of the UK general population. PTS/3D-strain was an independent predictor of cardiac events with some evidence of it being a slightly better predictor than conventional indices of LV function (GLS and EF). Future prospective studies are needed to confirm and extend these findings.

Funding Acknowledgement

Type of funding sources: Foundation. Main funding source(s): The main SABRE study is supported by the Wellcome Trust and BHF.

P2444Comparison of 3D and 2D echocardiography-derived indices of left ventricular function and structure to predict long-term mortality in the general population: Southall And Brent Revisited (SABRE) study

Background

Three-dimensional echocardiography (3DE) may have advantages over two-dimensional echocardiography (2DE) for the assessment of left ventricular (LV) function and structure. However, studies comparing 3DE and 2DE-derived indices in relation to mortality are limited, particularly in the general population.

Purpose

We examined associations between 2DE and 3DE-derived LV ejection fractions (LVEF) and volumes for all-cause mortality in a population-based sample.

Methods

A total of 899 individuals (age, 69.6±6.1 years; 77.5%male) from the SABRE study, a UK-based tri-ethnic community cohort, underwent a comprehensive transthoracic echocardiography examination. 2D LVEF and volumes were calculated, and full-volume 3D LV datasets acquired over 4 sub-volumes were obtained using a matrix-array transducer and were analysed offline using Qlab advanced, v7.0. The associations between both 2D- and 3D-derived LVEF (≥55% vs. <55%), body surface area indexed end-diastolic volume (iEDV) and end-systolic volume (iESV), and all-cause mortality were determined using Cox proportional hazards models. Survival curves were constructed using the Kaplan-Meier method.

Results

Of the 899 individuals, 118 (13.1%) died over a median follow-up period of 8 years to 2018. Kaplan Meier survival estimates (Figure 1 illustrates LVEF) and Cox regression revealed that 2D and 3D LVEF, iEDV and iESV were associated with increased risk of all-cause mortality (LVEF (≥55% vs. <55%), 3DE: HR=0.53 (0.35, 0.80); 2DE: HR=0.51 (0.34, 0.75), iEDV (per 1SD increment), 3DE: HR=1.20 (1.0, 1.41); 2DE: HR=1.19 (1.0–1.41), iESV (per 1SD increment), 3DE: HR=1.27 (1.1, 1.52), 2DE: HR=1.32 (1.15, 0.1.51)). However, 3DE associations tended to be stronger in models adjusted for classical risk factors including age, sex, ethnicity, systolic blood pressure, cholesterol:HDL ratio, body mass index, antihypertensive medications, diabetes, and smoking (LVEF (≥55% vs. <55%), 3DE: HR=0.59 (0.39, 0.90); 2DE: HR=0.69 (0.46, 1.0), iEDV (per 1SD increment), 3DE: HR=1.20 (1.0, 1.41); 2DE: HR=1.10 (0.93, 1.31), iESV (per 1SD increment), 3DE: HR=1.27 (1.1, 1.52), 2DE: HR=1.20 (1.04, 0.1.39)).

Figure 1. Kaplan-Meier curves.

Conclusions

In this population-based study, both 2DE and 3DE-derived indices of LV structure and function were associated with all-cause mortality independently of classical risk factors, with some indication that strengths of association were greater for 3DE-derived indices.

Acknowledgement/Funding

SABRE is funded by BHF, Diabetes UK, the MRC and the Wellcome Trust. LA holds a scholarship grant from Imam Abdulrahman Bin Faisal University, SA

6029Invasive versus medical management for non-ST elevation myocardial infarction in the elderly (SENIOR-NSTEMI study)

Background

Trials and registry studies suggest lower long-term mortality after invasive than medical management among patients with non-ST elevated myocardial infarction (NSTEMI), but elderly patients were underrepresented.

Purpose

To estimate the effect of invasive compared with medical management on survival in patients with NSTEMI aged ≥80 years, using routine clinical data.

Methods

We used National Institute for Health Research Health Informatics Collaborative data to identify eligible patients admitted during 2010–2017 at five tertiary centres. We compared patients who did and did not have invasive management within 3 days of their peak troponin level. To limit the effect of immortal time bias, follow-up started 3 days after peak troponin: deaths within three days were excluded. We conducted intention-to-treat analyses.

Propensity scores were derived from a logistic regression model based on pre-treatment variables: patient demographics, blood test results, cardiovascular risk factors, history of cardiovascular disease and other comorbidities. We modelled non-linear relationships using splines. Patients with high probability (based on propensity score) of medical or invasive intervention were excluded. We used Cox models to estimate hazard ratios (HR) comparing invasive with medical management. Three methods were used to control confounding; multivariable-adjusted, multivariable-adjusted additionally for continuous propensity score (primary analysis), and inverse-probability-of-treatment (IPT) weighting. Kaplan-Meier survival curves were plotted. The robustness of the results to unmeasured confounding was assessed in sensitivity analyses.

Results

The 2,239 patients (61.3% medical management) included in analyses had a median age of 85 (IQR 82–89) years. During a median follow-up of 32.1 (IQR 11.1–54.3) months, there were 1,015 (45.3%) deaths. At 3-years, cumulative survival was 78.9% and 50.3% in the invasive and medical management groups, respectively (Figure 1).

The crude HR comparing invasive with medical management was 0.34 (95% CI 0.29–0.40). The multivariable-adjusted HR was 0.44 (95% CI 0.36–0.53), was unchanged with additional adjustment for propensity score, and was 0.46 (95% CI 0.39–0.56) in the IPT-weighted model (all p<0.0001). The E-value for the point estimate was 2.91: this implies that residual confounding could explain the association if there is an unmeasured covariate with a relative risk of at least 2.91 for both mortality and undergoing invasive management. The highest mortality HR for comorbidities included in our model were aortic stenosis 1.66 (95% CI 1.28–2.14) and obstructive lung disease 1.50 (95% CI 1.16–1.94).

Figure 1. Kaplan-Meier survival curves

Conclusion

This study provides evidence that the survival advantage from invasive management may extend to elderly patients with NSTEMI. Future research should address the possibility of unmeasured confounding, including by post-admission prognostic factors that affect choice of invasive or medical management.

Acknowledgement/Funding

Funded by NIHR Imperial Biomedical Research Centre (BRC) using NIHR Health Informatics Collaborative data service, supported by OUH, GSTT & UCLH BRCs

P4345Supporting big data research in cardiovascular medicine using routinely-collected data

Background

Many of the data points required to support translational research are collected as a matter of routine, and should be available within electronic patient records. Variations in clinical and data recording practice can mean that the extraction and standardisation of this data, with the aim of producing a large-scale, research-ready dataset, presents a number of challenges.

Purpose

We set out to create a large-scale, research-ready dataset to support translational research in cardiovascular medicine, using routinely-collected data from five large university-hospital partnerships. As an initial focus, we selected those data points that would support an investigation of the relationship between test results and outcomes in acute coronary syndrome (ACS).

Methods

The National Institute of Health Research (NIHR) Health Informatics Collaborative (HIC) is a programme of infrastructure development aimed at increasing the quality and availability of routinely-collected data for collaborative, translational research. Eighteen university-hospital partnerships signed the data sharing agreement, and are working to facilitate the sharing and re-use of data across centres, for approved research purposes.

With support from the Directors of the NIHR Biomedical Research Centres (BRCs) within five of the largest partnerships, we established a clinical data collaboration, specifying a dataset and selecting an initial research question (Figure 1). The NIHR HIC team worked to extract data against this specification. With approval from an ethics committee, and from the information governance teams at each contributing centre, data was processed by one of the centres for standardisation and analysis.

Results

The specified dataset represented a longitudinal record for patients presenting with a suspected ACS, characterised by a request for a troponin test (Figure 1). The dataset included 156 data points, grouped into demographics, cardiovascular risk factor profile, emergency department attendance and inpatient episodes, blood tests, echocardiography and mortality. Data was extracted from the records of patients for whom a troponin test was requested between 2010 and 2017. A total of 257,948 records were standardised and analysed.

The collaboration has been successful, and an initial version of the combined dataset has been created. The size of the dataset has yielded new insights into the relationship between test results and outcomes, and publications are in preparation. An expanded dataset of over 800 data points has been agreed for the next phase of the collaboration, and three other centres have joined.

Figure 1. NIHR HIC dataset generation

Conclusion

It is perfectly feasible – in terms of governance and technology – to re-use routinely-collected data for collaborative, translational research in cardiovascular medicine. The resulting dataset will be large and complex enough to require big data tools and techniques, and will yield the kind of insights afforded only by big data in medicine.

Acknowledgement/Funding

Funded by NIHR Imperial Biomedical Research Centre (BRC) using NIHR Health Informatics Collaborative data service, supported by OUH, GSTT & UCLH BRCs

2231HsCRP predicts mortality beyond troponin in 102,337 patients with suspected acute coronary syndrome (CRP-RISK study)

Background

The incremental long-term prognostic value of high-sensitivity C-reactive protein (hsCRP) above troponin in a large real-world cohort of unselected patients presenting with suspected acute coronary syndromes (ACS) is unknown.

Purpose

We hypothesised that a mildly elevated hsCRP is associated with mortality risk in patients with suspected ACS, independent of troponin level.

Methods

We used the National Institute for Health Research Health Informatics Collaborative data of 257,948 patients who had a troponin measured at 5 cardiac centres. We excluded patients with clinically abnormal white cell counts and hsCRP >15 mg/L to try limiting the population to those without overt infections, malignancies or systemic inflammatory conditions that may confound our analyses. Patients were divided into four hsCRP groups (<2, 2–4.9, 5–9.9 and 10–15 mg/L) and the association between hsCRP levels and all-cause mortality assessed.

Results

There were 102,337 patients included in the analysis (hsCRP <2 mg/L (n=38,390), 2–4.9 mg/L (n=27,397), 5–9.9 mg/L (n=26,957) and 10–15 mg/L (n=9,593)). Figure 1A displays cumulative mortality per hsCRP group, revealing increasing mortality with each consecutive group. Figure 1B further stratifies the groups according to dichotomised peak troponin level as positive or negative. This shows the greatest mortality for patients in the highest hsCRP group who also had a positive troponin assay (36.0% at 3 years).

In Cox regression analysis with time-dependent covariates, even mildly raised hsCRP was an independent predictor of mortality over time, after adjusting for age, gender, haemoglobin, white cell count, platelet count, creatinine and troponin positivity. There was a positive and graded relationship between hsCRP level and mortality at baseline, which remained at 3-years (hazard ratio (95% CI) of 1.32 (1.18–1.48) for those with hsCRP 2.0–4.9mg/L, and 1.40 (1.26–1.57), and 2.00 (1.75–2.28) for those with hsCRP 5–9.9 mg/L and 10–15 mg/L, respectively.

We explored whether inclusion of hsCRP could better reclassify the population into at-risk mortality groups. The association with 30-day, 1-year and 3-year mortality was assessed using three different risk models (model 1: age, gender, haemoglobin, creatinine; model 2: model 1 plus troponin (positivity versus negativity); model 3: model 2 plus hsCRP groups. For cumulative mortality at each time point, each successive model was better able to discriminate risk than its precursor (p<0.0001); such that inclusion of troponin and hsCRP gave the most robust risk discrimination. Model 3 achieved an AUROC >0.8 at 30 days, 1-year and 3-year mortality, surpassing the use of troponin on its own.

Figure 1. Kaplan-Meier mortality curves

Conclusions

These multi-centre, real-world data from a large cohort of patients with suspected ACS identify hsCRP as a clinically meaningful prognostic marker in addition to troponin levels and point to its potential utility in selecting patients for novel treatments targeting inflammation.

Acknowledgement/Funding

Funded by NIHR Imperial Biomedical Research Centre (BRC) using NIHR Health Informatics Collaborative data service, supported by OUH, GSTT & UCLH BRCs

P3592Troponin level and mortality risk in an unselected population of over 250,000 patients (TROP-RISK study)

Background

Current evidence suggests a direct relationship between the magnitude of troponin elevation and mortality, albeit over a limited range of troponin levels, and clinicians generally work under the impression that higher troponins signify higher mortality in all age groups.

Purpose

The objective was to use big data to determine the relationship between the full spectrum of troponin level and mortality in patients in whom troponin testing has been performed for clinical purposes.

Methods

As part of the National Institute for Health Research Health Informatics Collaborative project, all troponin values measured during the study period (2010 to 2017) were assembled from five cardiovascular centres. Troponin concentrations were standardised as a multiple of each laboratory's 99th-percentile of the upper limit of normal (ULN). All patients were followed up until death or censoring on 1st April 2017. To model the relation between peak troponin level and all-cause mortality we used restricted cubic spline Cox regression analysis. Splines were adjusted for patient age, gender, haemoglobin, creatinine, white cell count and C-reactive protein.

Results

257,948 patients underwent troponin assessment. During a median follow-up of 1,198 (IQR, 514–1,866) days, there were 55,850 (21.7%) deaths. Using multivariable-adjusted restricted cubic spline Cox regression analysis, an inverted-U shaped relationship was observed between peak troponin level and mortality in all patients (Figure 1A). Among patients who were admitted to hospital, the recorded diagnostic code was acute coronary syndrome (ACS) in 14,468 patients and non-ACS in 120,049 patients.

The revascularisation rate within 3 months was 61.0% (n=8,820) in ACS versus 4.0% (n=4,793) in non-ACS patients. There was a very different rate of revascularisation across the spectrum of troponin. The rate was only 1.4% for troponins below 1 xULN, and 6.1% between 1 and 10 xULN. Beyond 10 xULN, rate of revascularisation rose rapidly to over 85% for greater than 10,000 xULN (Figure 1B). Stratifying patients by revascularisation, the restricted cubic spline Cox regression curve showed a progressive increase in mortality within both the revascularised and non-revascularised strata, even to very high peak troponin levels (Figure 1C). Overall, revascularisation was associated with lower hazard ratios across all troponin levels. A similar pattern was seen when patients were stratified by the presence or absence of ACS diagnosis.

Figure 1. Troponin level and mortality

Conclusions

An elevated troponin, even slightly above the ULN should be taken seriously. The inverted-U shaped mortality relationship with troponin occurred because patients with the highest troponin formed a different clinical subgroup who underwent different clinical management with a high revascularisation rate. These data on troponin level and mortality may help to inform clinical practice decisions and guide future risk stratification algorithms for patients with elevated troponin.

Acknowledgement/Funding

Funded by NIHR Imperial Biomedical Research Centre (BRC) using NIHR Health Informatics Collaborative data service, supported by OUH, GSTT & UCLH BRCs

Ethnic differences in cross-sectional associations between impaired glucose regulation, identified by oral glucose tolerance test or HbA1c values, and cardiovascular disease in a cohort of European and South Asian origin

AIMS

We contrasted impaired glucose regulation (prediabetes) prevalence, defined according to oral glucose tolerance test or HbA1c values, and studied cross-sectional associations between prediabetes and subclinical/clinical cardiovascular disease (CVD) in a cohort of European and South Asian origin.

METHODS

For 682 European and 520 South Asian men and women, aged 58-85 years, glycaemic status was determined by oral glucose tolerance test or HbA1c thresholds. Questionnaires, record review, coronary artery calcification scores and cerebral magnetic resonance imaging established clinical plus subclinical coronary heart and cerebrovascular disease.

RESULTS

Prediabetes was more prevalent in South Asian participants when defined by HbA1c rather than by oral glucose tolerance test criteria. Accounting for age, sex, smoking, systolic blood pressure, triglycerides and waist-hip ratio, prediabetes was associated with coronary heart disease and cerebrovascular disease in European participants, most obviously when defined by HbA1c rather than by oral glucose tolerance test [odds ratios for HbA1c -defined prediabetes 1.60 (95% CI 1.07, 2.39) for coronary heart disease and 1.57 (95% CI 1.00, 2.51) for cerebrovascular disease]. By contrast, non-significant associations were present between oral glucose tolerance test-defined prediabetes only and coronary heart disease [odds ratio 1.41 (95% CI 0.84, 2.36)] and HbA1c -defined prediabetes only and cerebrovascular disease [odds ratio 1.39 (95% CI 0.69, 2.78)] in South Asian participants. Prediabetes defined by HbA1c or oral glucose tolerance test criteria was associated with cardiovascular disease (defined as coronary heart and/or cerebrovascular disease) in Europeans [odds ratio 1.95 (95% CI 1.31, 2.91) for HbA1c prediabetes criteria] but not in South Asian participants [odds ratio 1.00 (95% CI 0.62, 2.66); ethnicity interaction P = 0.04].

CONCLUSIONS

Prediabetes appeared to be less associated with cardiovascular disease in the South Asian than in the European group. These findings have implications for screening, and early cardiovascular prevention strategies in South Asian populations.