This study was conducted under London South East Research Ethics Committee approval (reference 18/LO/2048) granted to the King’s Electronic Records Research Interface (KERRI); COVID-19 work was reviewed with expert patient input on a virtual committee with Caldicott Guardian oversight.

We conducted a cohort study of consecutive adult patients (age > 18 y) admitted with COVID-19 to King’s College Hospital NHS Foundation Trust (comprising King’s College Hospital and Princess Royal University Hospital), between 1st March and 30th June 2020. All patients had a positive RT-PCR antigen test for SARS-CoV-2. Only patients admitted to hospital for ≥ 24 h were included. A subset of this cohort has been reported previously [3, 16].

Structured and unstructured data were extracted from the electronic health record (EHR) using previously described natural language processing (NLP) informatics tools belonging to the CogStack ecosystem [17], DrugPipeline [18], MedCAT [19], and MedCATtrainer [20]. Clinician case review was used for additional validation (Additional file 1: Methods).

CV risk factors were defined as a recorded clinical diagnosis of hypertension, diabetes mellitus, or self-reported smoking status, in the absence of documented CVD. Diabetes mellitus was defined by a clinician diagnosis documented in the electronic health record and extracted based on relevant SNOMED CT UK extension and children terms (S-44054006, S-73211009, S-46635009) encompassing both type 1 and type 2 diabetes mellitus, as previously described [19] and validated [16]. Pre-existing established CVD was defined as a previous record of ≥ 1 of the following diagnoses: myocardial infarction (MI), heart failure, myocarditis, pericarditis, endocarditis, atrial or ventricular arrhythmia, and severe valvular heart disease [21]. Additional details are provided in Additional file 1: Methods. CV risk factors and pre-existing CVD were categorised as present if they had been recorded in the EHR at any time up to the day of admission (or including the day of admission, when recorded as a pre-existing condition). Data were also collected for age, sex, ethnicity, body mass index (BMI), non-CV comorbidities (asthma, chronic obstructive pulmonary disease [COPD], chronic kidney disease [CKD]), previous venous thromboembolism [VTE] comprising deep vein thrombosis [DVT] or pulmonary embolism [PE]), CV drug therapy (Additional file 1: Methods) and clinical examination and routinely collected blood results on admission. High sensitivity cardiac troponin T (hs-cTnT) plasma levels were defined as normal when below the 99th percentile of normal values, i.e., 14 ng/L.

The primary outcome was in-hospital mortality, with cause of death ascertained from death certification. Secondary outcomes included any CV complication related to COVID-19 and incident VTE. A CV complication was defined as a new CVD diagnosis or decompensation of pre-existing CVD recorded in the EHR on presentation or at any time during admission. CV complications were based on clinician diagnoses incorporating all available clinical information, including where relevant, echocardiography and coronary angiography (additional details in Additional file 1: Methods). Hospital admission date was used as the start of follow-up. Outcomes were ascertained through to death, discharge, or 31st July 2020, whichever was earlier.

Patient data are reported as frequency (%), mean (SD) or median (IQR), as appropriate. Patient characteristics were compared across 3 groups: patients with pre-existing established CVD (CVD), CV risk factors without CVD (RF-CVD), and no CVD or CV risk factors, using the Chi-squared goodness of fit or Fisher’s exact test (categorical variables), one-way analysis of variance (continuous variables) or Kruskal–Wallis/Wilcoxon rank-sum tests (for non-normally distributed data). Bonferroni correction was used for individual comparisons. Missing blood biomarkers (< 25% missing data) were imputed using the multiple imputation approach by chain equations [22].

Cumulative incidence plots displaying the probability of in-hospital mortality and discharge were constructed based on a competing risks analysis. To evaluate the association between patient group and mortality, we used Cox proportional hazards regression, with admission date as the start of follow-up and in-hospital mortality as the dependent variable. Unadjusted, demographic adjusted (age, sex, ethnicity), and fully adjusted models (including non-CV comorbidities and medications on admission) were performed. Age was modelled as a categorical variable to allow for potentially non-linear association (< 40, 40–49, 50–59, 60–69, 70–79, 80+ years). Comorbidities were modelled as binary variables. The reference group comprised patients without pre-existing CVD or CV risk factors. The proportional hazard assumption was examined graphically and using formal tests, as described by Grambsch [23]; no major deviations from this assumption were observed.

To investigate the association between CV complications and prognosis we performed logistic regression models with patient groups stratified by the presence or absence of CV complications as an independent categorical variable. For secondary outcomes of CV complications or VTE, logistic regression models were constructed with (1) patient group and (2) individual CV risk factors or CVDs as binary predictor variables. Unadjusted, demographic-adjusted, and fully adjusted regression models were performed as above. When individual CVDs were examined, myocarditis and pericarditis were excluded, due to their low prevalence. Information regarding hyperlipidaemia diagnoses were incomplete. A sensitivity analysis was performed where all patients who were prescribed statin therapy in the absence of diagnosed CVD were reclassified as RF-CVD. Furthermore, as BMI was missing in > 30% of patients, primary analyses were performed without adjustment for BMI. However, sensitivity analyses were performed restricted to patients with BMI data available including (1) adjustment for BMI as a continuous variable (fully adjusted model), (2) adjustment for obesity as a categorical variable (defined as BMI ≥ 30 kg/m²), and (3) reclassifying patients with obesity without diagnosed CVD as RF-CVD. An additional sensitivity analysis was performed among individuals with outcome data available i.e. those who were discharged or died (excluding current in-patients). Analyses were performed using STATA/IC (v16.1; StataCorp LLC, TX).

Between 1st March and 30th June 2020, 1,721 patients were admitted with COVID-19 (median age 71 years [IQR 56–83], 56.6% male). Of these, 349 (20.3%) had CVD, 888 (51.6%) had RF-CVD, and 484 (28.1%) had neither. Patients with CVD were older than patients with RF-CVD or neither but had a similar distribution of sex (Table 1).

CVD was more prevalent with increasing age, while RF-CVD was most common between 50–70 years (Additional file 3: Fig. S1A). Individuals from non-White ethnic groups had a higher prevalence of RF-CVD whereas CVD was more prevalent in the White group (Additional file 3: Fig. S1B). The most frequent CVD diagnoses were arrhythmia (86.2% atrial fibrillation), heart failure, and previous MI, respectively (Table 1). 119 (34.1%) patients with CVD had more than one CVD diagnosis. Rates of non-cardiovascular comorbidities were highest in patients with CVD, followed by RF-CVD (Table 1).

On admission, 83% of patients with hypertension were taking an antihypertensive agent and 73% of patients with atrial fibrillation were on oral anticoagulation. Rates of ACEI or ARB and betablocker use in heart failure patients were 47% and 62% respectively. In patients with a previous MI, rates of antiplatelet, beta-blocker and statin use were 68%, 64% and 65% respectively.

Physiological parameters and blood biomarkers are displayed in Additional file 2: Table S1. There were few clinically significant differences in physiological observations between groups, with the exception of a higher mean systolic blood pressure in patients with RF-CVD. Among blood biomarkers, C-reactive protein values were highest in patients with RF-CVD, but similar between patients with CVD and the group with neither CVD nor risk factors (Additional file 2: Table S1). Renal function was progressively worse across groups, with the lowest eGFR in patients with CVD.

Overall, 742 (43.1%) patients had at least one hs-cTnT measurement. Among patients with at least one hs-cTnT measurement, elevated values (> 14 ng/L) were observed in 133/147 (90.5%) patients with CVD, 311/409 (76.0%) patients with RF-CVD and 123/186 (66.1%) patients with no CVD or CV risk factors.

In-hospital outcomes are displayed in Table 2. Overall, 438 (25.5%) patients died and 1246 (72.4%) were discharged alive. 37 (2.1%) patients were in hospital at study close. The median length of hospitalisation for patients discharged was 9 (IQR 4–17) days and was longer for patients with CVD than those without (11 [IQR 5–19] vs. 7 [IQR 3–16] days, p < 0.001). Among patients who died, finalised death certificates were available in 382 patients. COVID-19 related pneumonia or acute respiratory distress syndrome was reported as the direct cause of death in 302 (79.1%) patients and as an indirect cause of death in 68 (17.8%) patients due to complications associated with COVID-19. This included 20 patients who died due to a CV cause: stroke (n = 8), massive PE (n = 3), decompensated heart failure (n = 4), myocardial infarction (n = 4) and acute limb ischaemia (n = 1). Only 12 deaths (3.1%) deaths were not attributed to COVID-19, e.g., malignancy or advanced dementia.

In-hospital mortality was greatest among patients with CVD (37.3%), intermediate in patients with RF-CVD (25.7%), and lowest among patients with neither (16.5%). Figure 1 displays cumulative incidence plots of the probability of in-hospital death or discharge over time for each group. For the overall cohort, there was a positive association between CVD and in-hospital mortality in unadjusted (HR 2.17 [95% CI 1.64–2.87], p < 0.001), and demographic-adjusted models (adjusted HR 1.45 [95% CI 1.09–1.94], p = 0.012). In fully adjusted models additionally accounting for non-CV comorbidities and baseline medications, there was a positive trend (aHR 1.36 [95% CI 0.97–1.92], p = 0.076). This effect was principally driven by a prognostic association in patients under 70 years of age (Fig. 2A), whereas the effect of CVD was smaller and non-statistically significant among patients aged 70 years and older (Fig. 2B). RF-CVD conferred an increased risk of mortality for the overall cohort in unadjusted analyses (HR 1.51 [95% CI 1.17–1.95], p = 0.002), but not in demographic-adjusted (aHR 1.17 [95% CI 0.90–1.53], p = 0.233) or fully adjusted models (aHR 1.13 [95% CI 0.85–1.51], p = 0.388). RF-CVD were not associated with mortality in patients older or younger than 70 years of age (Fig. 2A, B).

The main findings were unchanged in a sensitivity analysis with patients prescribed statin therapy without diagnosed CVD reclassified as RF-CVD (n = 64 patients reclassified, Additional file 3: Fig. S2A, B). BMI data were available in 1031 patients (60% total cohort). Sensitivity analyses in this subset adjusting for BMI as a continuous variable, (Additional file 3: Fig. S3A, B) or obesity as a categorical variable (Additional file 3: Fig. S4A, B) demonstrated similar effects to the main analysis. In addition, when obesity was included as a CV risk factor, 83 individuals with a BMI ≥ 30 kg/m² were reclassified as RF-CVD (from the no RF or CVD category). Effect estimates remained robust with only marginally wider confidence intervals (Additional file 3: Fig. S5A, B). Finally, a sensitivity analysis excluding current in-patients also showed similar findings to the main analysis. (Additional file 3: Fig. S6A, B).

Cardiovascular complications occurred in 431 (25.0%) patients, with two-thirds occurring in patients with CVD (n = 230, 65.9%). Patients with RF-CVD also had a higher CV complication rate than patients with neither CVD or CV risk factors (16.9% vs. 10.5%, Bonferroni adjusted p < 0.001). The most frequent CV complications were cardiac arrhythmias (84.7% atrial fibrillation), followed by acute heart failure (distinct from myocarditis) and acute MI, respectively (Table 2). Among patients presenting with an acute MI, 3 patients displayed ST elevation MI (STEMI) and underwent emergency percutaneous coronary intervention. Two additional patients underwent coronary angiography, one patient was diagnosed with spontaneous coronary artery dissection and one patient diagnosed with myocarditis. The remaining cases of acute MI were clinically considered to represent non-ST elevation or type 2 MIs [24]. The incidence of clinician-diagnosed myocarditis was low (0.7%). When arrhythmia-related complications were excluded, 59% of complications occurred in patients with CVD, 33% in patients with RF-CVD, and 8% in patients with neither.

In patients with CVD, the majority of CV complications represented exacerbations or decompensation of underlying CVD, rather than a new presentation, e.g., 86% of myocardial infarctions occurred in individuals with a previous myocardial infarction (Additional file 3: Fig. S7). Among specific CVDs and risk factors, pre-existing AF was associated with the highest adjusted odds of having any CV complication, followed by previous MI (Fig. 3A). For non-arrhythmia related CV complications, the highest adjusted odds were seen in patients with a previous myocardial infarction (Fig. 3B).

When CV complications were defined by cardiac biomarker elevation, in the subset of patients with a hs-cTnT measurement (n = 742), the presence of troponin elevation (Additional file 3: Fig. S8A) or troponin-elevation greater than 10 × normal (Additional file 3: Fig. S8B) were both associated with increased odds of in-hospital mortality across groups.

The incidence of VTE was higher in patients with CVD versus RF-CVD or neither (18.3% vs. 10.9% and 10.7% respectively, p < 0.001 for each). However, among CVDs and CV risk factors, hypertension (in the absence of established CVD) was the only independent CV predictor of VTE. Patients with previous VTE (n = 166, 48% anticoagulated at admission) had the highest rate of new (incident) VTE (49.4% vs. 8.4% with no previous VTE p < 0.001), and previous VTE was the strongest predictor of incident VTE, including adjustment for baseline anticoagulation use (Fig. 3C).

The prevalence of CVD in our study was similar to other large hospital cohorts [4, 25]. We corroborate previous reports showing that a history of CVD is associated with greater risk of COVID-19-related mortality [4‐6]. Interestingly, we identify an interaction with age, wherein the increased mortality risk is mainly apparent in people below 70 years of age whereas it is not statistically significant in people aged 70 and over. Reasons why older individuals do not also manifest higher CVD-related mortality warrant further investigation but may be related to a higher competing risk of non-CVD-related mortality due to frailty, non-CV comorbidities and immunosenescence, such that CVD has relatively minor additional prognostic effect. Similar age-dependent mortality effects have been reported in other studies [6, 7].

CV risk factors such as hypertension (56%) and diabetes (35%) were more prevalent than established CVD in our in-patient cohort, similar to other UK studies [4]. There are conflicting data regarding the prognostic impact of common CV risk factors in COVID-19. In hospitalised cohorts such as the UK ISARIC study, diabetes had a marginal independent effect on mortality risk, similar to our findings [4]. The majority of patients in ISARIC had uncomplicated diabetes and hypertension was not assessed. In the population-based OpenSAFELY study, diabetes was independently associated with a higher mortality, whereas hypertension was not independently associated with mortality [6]. Another large UK population-based study reported a 1.8-fold higher mortality risk for patients with type 2 diabetes after adjustment for relevant factors [26]. These divergent findings may be reconciled by considering that mortality rates in population-based studies reflect the risk of infection as well as risk of mortality once infected. For example, higher mortality risk associated with diabetes in population-based studies was suggested to be partly related to the level of glycaemic control [7]. We hypothesise that there may be an association between diabetes and infection risk. Previous studies have demonstrated that individuals with diabetes are at increased risk of serious infections [27] and poor glycaemic control has been associated with serious infections and hospital admission [28], although this association has not yet directly been shown for SARS-CoV-2. Nevertheless, even if patients with diabetes are more likely to be admitted, the current study and previous reports [4] suggest that mortality in this group could to a significant degree relate to the co-existence of CVD or other complications.

It should be acknowledged, however, that there may be interactions among CV risk factors and other variables that affect mortality risk e.g., ethnicity [7] and the effective treatment of risk factors [7].

A greater mortality risk associated with pre-existing CVD as compared to CV risk factors without CVD raises questions about the potential mechanisms underlying the higher risk. It has been proposed by several authors that endothelial dysfunction may be a major contributor to severe COVID-19 [29, 30]. Accordingly, pre-existing endothelial dysfunction may increase the likelihood of developing severe endothelial and vascular impairment with COVID-19. Such a mechanism would not readily explain the differential risk between established CVD and CV risk factors since both conditions are associated with endothelial dysfunction (and the majority of patients with CVD have CV risk factors). An alternative possibility is that patients with pre-existing established CVD are more prone to develop further cardiac injury and dysfunction which, in combination with pulmonary and right heart problems that represent the major manifestations of severe COVID-19, leads to life-threatening illness.

To explore this possibility, we analysed CV complications. The use of semi-automated pipelines to capture all clinician-diagnosed CV events minimised selection and indication bias. With this, our data demonstrate a high frequency of CV events overall (25%), with the majority (73%) representing arrhythmias, mostly atrial fibrillation, with rates comparable to smaller studies [5, 31, 32]. Other complications included acute MI and acute heart failure, with few clinically-diagnosed cases of myocarditis (0.7%).

Patients with pre-existing CVD had higher rates of CV complications than those with CV risk factors without CVD or patients without either CVD or CV risk factors. They also had higher rates of VTE but, whereas CVD was an independent predictor for incident CV complications, it was not a predictor of VTE. Importantly, we found that a high proportion of COVID-19 related CV complications (mainly cardiac) represent exacerbated or destabilised pre-existing CVD, rather than new presentations. Taken together, these findings suggest that the detrimental impact of pre-existing CVD on COVID-19 severity and mortality may be mediated mainly by increased cardiac problems rather than systemic vascular abnormalities such as VTE. In support of this idea, we found that myocardial injury as assessed by troponin elevation was most prevalent in patients with pre-existing CVD and strongly associated with mortality.

Currently the mechanism of thromboembolic risk in COVID-19 remains unclear. Potential mechanisms that have been suggested include vascular endothelial dysfunction, abnormal complement and coagulant pathway activation, and abnormal platelet activation [33]. The fact that there was no independent association between new-onset VTE and established CVD may partly reflect higher rates of antiplatelet or anticoagulant therapy in this group. Another confounding factor could be undiagnosed VTE [10]. Ongoing clinical trials examining anticoagulation strategies and additional pathophysiology studies will provide further insights into this question.

The high incidence of cardiac arrhythmias, mostly atrial fibrillation, observed in this study may have multiple precipitants, such as myocardial ischaemia, increased sympathetic tone, inflammation (systemic as well as myocardial), and electrolyte imbalance. Our finding of a low incidence of myocarditis is consistent with several other reports and a recent review of autopsy cases [34, 35].

Our finding that a large proportion of CV complications represented destabilised pre-existing CVD, supports the importance of identifying CVD in patients presenting to hospital with COVID-19 (including new diagnoses) and maintaining evidence-based CV care, alongside disease-specific treatment for COVID-19, including, for example, continuation of ACE inhibitors or ARBs in individuals with an indication [16]. The high incidence of arrhythmias may warrant more systematic electrocardiographic screening of hospitalised patients, particularly as the detection of new-onset atrial fibrillation is an indication for anticoagulation to reduce the risk of stroke and systemic thromboembolism. The low rates of STEMI in our cohort, matches reports of declining admission rates for STEMI during the pandemic [36], suggesting that clinical deterioration and CV complications in patients with COVID-19 may be less frequently due to STEMI. Additionally, given the age and comorbidity profile of patients hospitalised with severe COVID-19, this population has a high risk of type 2 myocardial infarction (i.e., a mismatch between oxygen supply and demand, without acute atherothrombotic plaque disruption [24]), in the presence or absence of underlying CAD. Since this may have implications for triage and treatment protocols, a low threshold for biomarker (troponin) assessment in patients with pre-existing CVD should be considered.

Our analysis was limited to individuals who required hospital admission and is therefore only generalisable to this population. This was a retrospective study of prospectively entered data in the EHR. Although this study assessed CV risk factors and established CVD as separate entities, a limitation of this approach is that there is a continuum and that individuals with risk factors may have undiagnosed CVD. Nevertheless, the presence of overt diagnosed CVD does appear to distinguish this group in terms of outcomes. During the early stages of the pandemic, echocardiography and coronary angiography were only performed in selected cases in keeping with recommendations to avoid unnecessary cardiac imaging (in order to reduce transmission of the virus, protect healthcare professionals, and conserve personal protective equipment [37]). A small minority of patients were still in hospital and were censored at the study end-date (2.1%). However, a sensitivity analysis in patients who were either discharged or died revealed similar findings (Additional file 3: Fig. S6). Our selection of cardiovascular risk factors was based on those with highest prevalence, most reliably reported, and to further explore findings in preceding large UK population studies [4, 7]. We did not have robust information regarding dyslipidaemia, however a sensitivity analysis reclassifying patients prescribed statin therapy (with no known CVD) as RF-CFD, as a crude measure of hyperlipidaemia or high CV risk, showed similar effect estimates to the main analysis. In addition, there was a significant amount of missing data for BMI, however effect estimates were robust in sensitivity analyses accounting for BMI and obesity, including obesity as a RF-CVD. We considered this approach more appropriate than multiple imputation, because underweight and overweight individuals, may be more likely to have their BMI recorded, thus contradicting the required missing at random assumption [38].

Finally, our multivariable analyses adjusted for patient characteristics and the presence or absence of several comorbidities, however measures of control (e.g., blood pressure control for hypertension or HbA1c for diabetes) were not assessed and may impact the risk of death. Additionally, we cannot exclude residual confounding due to unmeasured non-cardiac comorbidities, such as malignancy, hepatic or non-vascular neurological diseases.