Prognostic Significance of Ventricular Arrhythmias in 13 444 Patients With Acute Coronary Syndrome: A Retrospective Cohort Study Based on Routine Clinical Data (NIHR Health Informatics Collaborative VA‐ACS Study)

The data sets generated or analyzed or both during this study are not publicly available owing to ethical restrictions.

The study database was approved by the London‐South East Research Ethics Committee (reference number 16/HRA/3327). The requirement for individual patient consent was waived. This analysis was approved by the National Institute for Health Research (NIHR) Health Informatics Collaborative steering committee.

The NIHR Health Informatics Collaborative database consists of routinely collected data, currently from 10 collaborating National Health Service Trusts. For this study, data were collected from 5 NHS trusts (Imperial College Healthcare, University College Hospital, Oxford University Hospital, King’s College Hospital, and Guy’s and St Thomas’ Hospital), all tertiary centers with emergency departments. As previously described, the cohort inclusion criterion was any patient who had their troponin level measured for any reason (n=257 948) at each of the 5 academic centers between 2010 (2008 for University College Hospital) and 2017.^{19, 20, 21} For patients with >1 hospital episode, only the first episode in which troponin was measured was eligible. There were no other exclusion criteria. Patients meeting inclusion criteria were followed up using routinely collected data until death or censoring on April 1, 2017. Life status was ascertained using routinely collected data on the National Health Service Personal Demographics Service, which incorporates national death registry information and local notifications. For the readmissions outcome, only readmissions to the same hospital as the initial presentation were available.

Diagnoses were assigned on the basis of International Classification of Diseases, Tenth Revision (ICD‐10) codes. We classified patients as having had an ACS or CA on the basis of the ICD‐10 codes listed in the supplemental material (Tables S1–S3). ACS included the diagnoses of STEMI, non–ST‐segment–elevation myocardial infarction, and unstable angina.^{22, 23} A patient was classified as having had VA if they had the ICD‐10 codes for reentry VA, ventricular tachycardia (VT) or VF, and flutter. Comorbidities, risk factors, and implantation of a device was also determined using ICD‐10 codes. For adjusted analyses, peak troponin was used. As the troponin assay used differed between study sites, a normalized value of peak troponin as a multiple of the assay upper limit of normal was used. If patients had both VT and VF, they were assigned to the VF group for subgroup analysis.

A small proportion of cases had missing data for covariates of interest. These data were expected to be missing at random, and given the small proportion (1%), these were handled with listwise deletion for the adjusted analysis. To avoid introducing unnecessary bias, the unadjusted analyses included all cases, including those with missing data in the covariate fields.

Descriptive statistics are displayed as medians (interquartile ranges) for continuous variables and numbers (percentages) for categorical variables. Kaplan–Meier plots were used to display cumulative mortality. The log‐rank test was used to compare survival curves. In the Kaplan–Meier analyses, pairwise comparisons between groups were performed with the Benjamini and Hochberg correction for multiple testing (Figure 4).²⁴ Cox proportional hazards regression modeling was used to estimate mortality hazard ratios (HRs) in patients with and without VA/CA. To model nonlinear relationships, we used restricted cubic splines for Cox regression. Preliminary investigation suggested that 4 knots should be used to model troponin level in the restricted cubic spline analyses. Log transformation was applied to peak troponin because of the positive skew of troponin values. We performed multivariable analyses adjusting for sex, ethnicity, hemoglobin level, white cell count, platelet count, sodium level, potassium level, creatinine level, peak troponin (multiple of the assay upper limit of normal), family history of cardiovascular disease, current smoker, diabetes, hypertension, hypercholesterolemia, heart failure, previous ischemic heart disease, previous myocardial infarction, atrial fibrillation, aortic stenosis, chronic kidney disease, malignancy, obstructive lung disease, type of ACS, percutaneous coronary intervention (PCI), coronary artery bypass graft, and implantation of device. The proportional hazards assumption was tested using Schoenfeld residuals. There was a nonsignificant relationship between the Schoenfeld residuals and time for all of the univariate analyses. There was evidence of nonproportionality for some covariates in the multivariable analyses (Table S4). Recent work suggests that virtually all real‐world clinical data sets will violate the proportional hazards assumptions if sufficiently powered and that statistical tests for the proportional hazards assumption may be unnecessary.²⁵ In line with these recommendations, the HR from our Cox models should be interpreted as a weighted average of the true HRs during the follow‐up period. Negative binomial regression was performed to compare number of episodes of VA during follow‐up. Patients with any missing characteristic data were excluded from analysis of adjusted Cox proportional hazards models. Statistical analyses were performed with R 4.0.0 statistical package (R Core Team, Vienna, Austria).

The NIHR Health Informatics Collaborative data set was registered with ClinicalTrials.gov, NCT03507309.

A total of 14 468 patients had a diagnosis of ACS, 13 444 of whom survived to hospital discharge (Figure 1). Of the 14468 patients with ACS, 461 (3.2%) also had a diagnosis of a VA, whereas 508 (3.5%) suffered a CA. Of those surviving to discharge, 376 (2.8%) had a concurrent diagnosis of a VA, whereas 280 (2.1%) had a CA. The baseline characteristics grouped by presence or absence of VA are displayed in Table 1 and by type of ACS in Table S5. At a median follow‐up of 3.42 years (interquartile range, 1.69–5.10 years), there were 2480 (18.4%) deaths. In the subgroup of patients with STEMI (n=4091), 3448 (84%) underwent PCI. Of those who underwent PCI, 3066 (89%) underwent PCI on the same day as the first troponin measurement.

Considering all patients with ACS, concurrent VA was associated with a significant increase in in‐hospital mortality (unadjusted HR, 1.96 [95% CI, 1.56–2.45]; adjusted HR, 1.89 [95% CI, 1.49–2.40]; Figure 2; Table 2). However, in patients who survived to hospital discharge, concurrent VA had no effect on all‐cause mortality overall (unadjusted HR, 0.9 [95% CI, 0.7–1.16]; adjusted HR, 1.03 [95% CI, 0.80–1.33]; Figure 3A), nor when stratified by age groups (Figure S1) or in the subgroups of non–ST‐segment–elevation myocardial infarction and STEMI (Figure 4).

In the subgroup of patients with unstable angina (n=2224), VA at index presentation was associated with increased long‐term all‐cause mortality (unadjusted HR, 2.18 [95% CI, 1.20–3.96]; adjusted HR, 2.11 [95% CI, 1.11–3.99]), although the number of patients were small, with only 31 patients in this group having VA at the index presentation (Figure 4). In this unstable angina subgroup, 1011 (45%) patients without VA had PCI or a coronary artery bypass graft, whereas in the VA group the number was 20 (64.5%). This was significantly lower than in the myocardial infarction subgroup where 8039 (73.9%) had PCI in the group without VA and 283 (82%) had PCI in the VA group (P<0.0001 and P<0.02, respectively).

In patients who had a diagnosis of ACS who survived to hospital discharge, there remained a 36% increase in long‐term mortality (unadjusted HR, 1.31 [95% CI, 1.01–1.70]; adjusted HR, 1.36 [95% CI, 1.04–1.78]; Figure 3B).

VA at index presentation significantly increased the risk of VA at follow‐up (unadjusted HR, 5.38 [95% CI, 3.3–8.8]; adjusted HR, 4.15 [95% CI, 2.42–7.09]; adjusted incident rate ratio, 5.54 [95% CI, 2.76–11.13]; Figure 5A) but did not affect the composite end point of death, VA, or CA (unadjusted HR, 1.12 [95% CI, 0.90–1.4]; adjusted HR, 1.24 [95% CI, 0.98–1.57]). Of patients with VA during ACS, 4.8% had subsequent VA during follow‐up, whereas 0.9% of patients who did not have VA during the index ACS admission had subsequent VA. Only a small minority of subsequent VA were in the context of an ACS (3.3% in VA group and 7.6% in the no VA group). CA at index presentation also significantly increased the risk of VA at follow‐up (unadjusted HR, 3.35 [95% CI, 1.64–6.84]; adjusted HR, 2.60 [95% CI, 1.23–5.48]; adjusted incident rate ratio, 1.98 [95% CI, 0.73–5.35]; Figure 5B).

In an unadjusted subgroup analysis of VA type, patients with VF at the time of ACS had reduced all‐cause mortality compared with patients with VT or no VA (Figure S2); however, after adjusting for confounding factors, this association was no longer seen (unadjusted HR, 0.60 [95% CI, 0.4–0.9]; adjusted HR, 0.82 [95% CI, 0.54–1.24]).

In 14 468 patients with ACS, concurrent VA was associated with increased in‐hospital mortality. Patients surviving to discharge had increased risks of recurrent VA. This finding is in agreement with a previous study of 2033 patients, where VT in the context of ACS was associated with increased recurrent VTs.¹⁵ The individuals who develop VA in the context of ACS may represent a subgroup of patients who have intrinsically lower thresholds to developing VA in response to any extrinsic triggers such as ACS and are therefore more likely to have subsequent episodes of VA. Genetic factors in particular may make patients more susceptible to VA.^{4, 5} A genome‐wide association study has identified a locus, rs2824292 at 21q21, that is associated with VF in the context of myocardial infarction.⁵ In addition, acquired factors such as burden and distribution of myocardial scarring may predispose certain individuals to recurrent VA.²⁶

In the 13 444 patients with an ACS who survived to discharge, VA was not associated with adverse long‐term prognosis. This finding is broadly in agreement with previous data^{9, 10, 12, 13, 16, 17} but provides data from a cohort that is both more recent and larger in size. In contrast, in a study of patients with non–ST‐segment–elevation myocardial infarction, part of the early glycoprotein IIb/IIIa inhibition in non–ST‐segment elevation acute coronary syndrome (EARLY ACS) trial, VA was found to be associated with both increased 30‐day and 1‐year mortality.²⁷ The difference in findings may be attributed to differences in the definitions of VA. The EARLY ACS study included VF or VT that was sustained for >30 seconds and/or causing hemodynamic compromise, whereas nonsustained VT could also have been included in our study. In addition, EARLY ACS used a 48‐hour landmark period, whereas our study primarily analyzed patients who survived to discharge. Other possible factors include the time period studied (enrollment for their study was between 2004 and 2008) or the high‐risk population enrolled in EARLY ACS (2 or more criteria out of age ≥60, elevated creatine kinase MB or troponin, ST‐segment depression, or transient elevation) versus the “all‐comers” population in our study. Hai et al found that monomorphic VT was an independent predictor of all‐cause mortality in patients with ACS who survived to discharge, but nonmonomorphic VT was not.¹⁵ In adjusted analyses, we found no difference between VT versus VF; however, in our study we were unable to differentiate between polymorphic and monomorphic VT.

We hypothesize that competing causes of death in a relatively high‐risk population may explain the increase in recurrent VA without increase in mortality in the group of patients with VA during the index admission. Unfortunately, we were unable to test this hypothesis as we were unable to ascertain the causes of death in this data set. The trials of early ICD after myocardial infarction with high‐risk features for VA show a similar phenomenon. Both trials demonstrated a significant reduction in arrhythmic death in the ICD group but did not show an overall reduction in all‐cause mortality attributed to more nonarrhythmic deaths in the ICD arms.^{28, 29}

Because of the limitations of ICD‐10 coding, we were unable to accurately differentiate shockable (VT or VF) versus nonshockable (bradyarrhythmia or pulseless electrical activity) rhythms during CA. However, previous data from a similar cohort suggested that 89.6% of patients who have out‐of‐hospital CA in the context of ACS have VT or VF as the presenting rhythm,¹ which provide the basis for our assumption that the vast majority of CAs at the time of ACS in our study are attributed to VAs.

In our study, CA in the context of ACS was associated with adverse in‐hospital and long‐term prognosis, with increased recurrence of VA after discharge and increased long‐term mortality. Previous studies have primarily studied VF in the context of ACS and found increased in‐hospital mortality but no increase in long‐term mortality.^{10, 13, 16} In contrast, our study, which did find an increase in long‐term mortality, also included CA attributed to VT, and potentially a small proportion of nonshockable rhythms causing CA. This group of patients with CA during ACS do not currently meet criteria for ICD implantation under primary or secondary prevention indications if fully revascularized. Although we do not have data on the left ventricular ejection fraction of the patients in our study, we did adjust for ICD‐10 codes of heart failure to reduce the effect of this confounding factor.

Our findings may have significant implications for this subgroup of patients with ACS. Given the increase in recurrent VA and long‐term mortality in this group, ICD implantation may be of benefit. However, this concept requires further systematic study as, because of the competing causes of death, there is the possibility that an ICD may reduce arrhythmic death without reducing all‐cause mortality, by converting the mode of death from arrhythmic death to nonarrhythmic death.

In the subgroup of patients with unstable angina who survived to discharge, concurrent VA during the unstable angina episode was associated with adverse long‐term prognosis. In the era of high‐sensitivity troponin, VA in the context of unstable angina is a rare occurrence and therefore it may not be well suited to a cohort study design, and the very small sample in this study may therefore not be adequately represented in this study to make definitive conclusions.

However, in agreement with our findings, a previous study of 543 patients with unstable angina did reveal an association between VA in the context of unstable angina and in‐hospital and 6‐month all‐cause mortality.³⁰ It is possible that the adverse prognosis relates to the reduced rate of PCI in the unstable angina subgroup. Without treatment of the lesion responsible for the ACS, the “reversible” cause to which the VA is attributed to has not been truly reversed, thus predisposing to further VA and mortality. It is possible that a portion of this subgroup suffered VA caused by coronary vasospasm, which unlike STEMI and non–ST‐segment–elevation myocardial infarction, is not amenable to PCI and can be poorly treated, leaving these patients susceptible to recurrent episodes and future VA.³¹

This study benefits from data collected from a large number of patients from multiple UK hospitals. There are, however, some limitations. These real‐world data are likely to be representative of the UK population, and our findings may not be generalizable to other populations and health care systems. There can be difficulty in accounting for potential confounding factors because of the incomplete nature of, or inaccuracies in, the data. In particular, there were insufficient data on left ventricular ejection fraction, coronary anatomy, and results of revascularization to include in our models. Where an ICD was implanted, we do not have data regarding the indication, timing of implant, or therapies from the device. We included ICD‐10 codes for heart failure to adjust for this important confounder. We were also unable to accurately determine which patients underwent primary PCI. However, a large percentage of patients with STEMI had PCI on the same day as the first troponin measurement, which is likely to indicate primary PCI. We were also unable to distinguish polymorphic from monomorphic VT, accurately account for the type of CA (shockable versus nonshockable) at index presentation or follow‐up, determine the timing of arrhythmic events (eg, whether occurring only in the first 48 hours after the index event or not), or establish whether VT was sustained or nonsustained, and these factors all may have prognostic implications. It is possible that some patients with ACS were not coded as also having VAs; however, our VT/VF incidence is comparable with trials and prospective registries (3.2% in our study versus 1.5‐5.8% in others^{3, 11, 27, 32}). Given that only readmissions to the same hospital are included in this data set, we may be underestimating the incidence of secondary outcomes; however, the most important end point, all‐cause mortality, is free from this limitation. Unfortunately, we were unable to determine the cause of death in this cohort. Patients with VAs are likely to have had different medications given during the hospital admission and on discharge, and the data on medications were not available. Lastly, the ICD‐10 code we used for device implantation is not specific to the type of device; therefore, it is unknown if the device was an ICD, pacemaker, or other device.