Risk of myocardial infarction and stroke following bloodstream infection: a population-based self-controlled case series

Introduction

Bacteraemias, or bloodstream infections (BSIs), are common, life-threatening infections with a reported 30-day mortality ranging from 15% to 30%1–3 and are responsible for hundreds of thousands of deaths each year worldwide.1 Previous work from our group has shown that 3–7% of people with BSI actually die from ischaemic heart disease or cerebrovascular disease, depending on the infecting organism and timing of death.4

The relationship between BSI and cardiovascular disease (CVD) is multifactorial and is likely due to (1) endothelial dysfunction, atherosclerotic plaque destabilisation and thrombosis (eg, type 1 myocardial infarction (MI) provoked by inflammation-mediated activation of the coagulation cascade and platelets5) and (2) supply-demand imbalance (eg, type 2 MI provoked by sepsis/hypotension6).

To date, associations between acute infections and incident MI and stroke have been reported, most notably for influenza and community-acquired pneumonia.7 8 However, most studies used designs susceptible to bias and/or did not use laboratory-confirmed infections. More recent work used laboratory confirmation and study designs less susceptible to confounding (eg, self-controlled case series—SCCS) and found clear associations between respiratory tract infections and the risk of MI.9 For BSI, cohort data has shown an increased risk for MI or stroke within 30 days of BSI.10 11 Further evidence of the short-term risk of MI with Staphylococcus aureus BSI (SAB),12 using the SCCS design, which inherently controls for confounding better than standard observational studies.

Determining the risk and timing of CVD events after common causes of bacteraemia (particularly Escherichia coli, given its ubiquity and paucity of data about CVD risk) is important given the global burden of these infections1 and potentially preventable/treatable nature of CVD events. To test our hypothesis that BSI and its associated inflammation were associated with an increased risk of CVD events, we conducted a population-based study using the SCCS design and laboratory-confirmed BSI.

Methods

Results

We identified 50 450 individuals with an ICD-10 code for MI and 56 890 with an ICD-10 code for stroke (online supplemental figure 2). Of these, 1000 (2.0%) and 1290 (2.3%) individuals respectively also had at least one community-associated BSI during their observation period and were included in the primary analysis. People with MI were 57% male with a median (IQR) age of 77 years (69–84). People with stroke were 50% male with median age of 78 years (69–84) (tables 1 and 2).

Risk of MI

Of 1000 identified MIs, 90 occurred during the risk periods of interest and 820 occurred during baseline time. The remaining 90 MIs occurred on day 0, which were excluded from further analysis due to issues ascertaining the temporality of exposure and event. Total observation time during risk periods was 82 937 days, and during baseline time was 2 927 115 days. The relative incidence of MI during days 1–7 following a community-associated BSI was 9.67 (95% CI 6.54 to 14.3). The risk declined over time and returned to baseline after 28 days (table 3 and figure 1).

Table 3

View inline•Open as popup

Crude and age-adjusted incidence rate ratio (IRR) for first myocardial infarction and first after community-associated bloodstream infection

Figure 1

Request permissions

Incidence rate ratios for myocardial infarction and stroke after community-associated bloodstream infection (primary analysis).

Secondary analyses

Hospital-associated BSI were associated with similar, although attenuated, increased risks of CVD (table 4 and table 5). The relative incidence of MI in the first 7 days was 4.43 (95% CI 2.37 to 8.25), and of stroke was 1.56 (95% CI 0.82 to 2.95) (figure 2). Stroke risk was much higher for infection with S. aureus compared with E. coli (13.88 (6.78 to 28.41) vs 1.55 (0.76 to 3.14)), respectively, for risk up to 7 days (table 3 and online supplemental figure 3). The effect of gender on the risk of MI and stroke was not statistically significant (table 5).

Table 4

View inline•Open as popup

Crude and age-adjusted incidence rate ratio (IRR) for first myocardial infarction and first stroke after hospital-associated bloodstream infection

Table 5

View inline•Open as popup

Age-adjusted incidence rate ratio (IRR) for myocardial infarction and stroke in the first 7 days after community-associated bloodstream infection—subgroup and sensitivity analyses

Figure 2

Request permissions

Incidence rate ratios for myocardial infarction and stroke after hospital-associated bloodstream infection (secondary analysis). MI, myocardial infarction.

The risk of both MI and stroke was higher for individuals with higher C-reactive protein (CRP) concentrations (table 5 and online supplemental figure 4). The relative incidence of MI in the first 7 days increased from 4.76 (1.73 to 13.1) with a peak CRP of<100 mg/L to 21.54 (9.57 to 48.5) for those with CRP>300 mg/L. Numerically lower risk was observed for both aspirin and statin use, particularly for MI, but CIs were wide and overlapped (table 5 and online supplemental figure 5). The relative incidence of MI in the first 7 days was 7.75 (4.65 to 12.9) for people prescribed statins compared with 14.2 (7.58 to 26.6) in those who were not.

The results were robust to multiple changes of definitions and assumptions in the sensitivity analyses (table 5).

Discussion

In this population-based study, we found community-associated BSI to be associated with incident MI and stroke. Risk was highest in the first 7 days after BSI and largely normalised 28-days after infection. Risk was clearly associated with the magnitude of the inflammatory response, whereby patients with the highest CRP concentrations were at the greatest risk of CVD events. Intriguingly, our results also suggest a possible protective effect of aspirin and statins. These findings were robust to a range of sensitivity analyses.

These findings are consistent with earlier studies reporting an increased risk of MI and stroke following BSI.10–12 Building on this work, we show that risk is mediated by the magnitude of the inflammatory response to BSI as well as infecting organism. The magnitude of associations we report is smaller than the previous largest study of community-associated BSI. Dalager-Pederson et al10 reported an adjusted relative risk of 17.7 for MI and 25.8 for stroke in the first 30-days after BSI. However, their use of a cohort design and choice of control groups with minimal adjustment risks significant residual confounding, therefore potentially inflating estimates of risk. The only previous study using SCCS of SAB by Corrales-Medina et al12 reported an IRR for MI of 35.3 and 7.9 during the first 2 and 14 days after SAB respectively. The discrepancy with our findings may be due to our inclusion of a separate day 0 risk period to account for the uncertainty about which event (ie, BSI or MI/stroke) happened first. Including a separate day 0 risk period is recommended to avoid bias and inflation of effect estimates in SCCS and is discussed in more detail in the methods.14 Furthermore, in this study, only 11 MIs occurred during the risk period (82% of which occurred within 2 days), which limits the robustness of these findings.

The discrepancy in risk of stroke between S. aureus and E. coli BSI is likely explained by differential rates of endocarditis. Endocarditis is a common cause of community-associated SAB and has an established risk of stroke due to embolism from left-sided valvular infection. We did not have echocardiographic or coding data available to accurately ascertain the incidence of endocarditis in our cohort. However, previous work using case-control designs has shown similar increased risk of stroke with SAB, with the highest risks reported in patients with endocarditis (adjusted RR: 117, specific BSI organisms unclear).10 11

The relationship between hospital-associated BSI and CVD events was less clear. Elevated baseline risk in already hospitalised patients due to comorbidities in combination with fewer events may limit the power of our study.

The mechanism of BSI-associated cardiovascular events is likely multifactorial. Inflammation triggered by infection leads in some to atherosclerotic plaque disruption, endothelial injury and increased risk of thrombosis through activation of the coagulation system and platelets.5 7 This is compounded by haemodynamic changes and mitochondrial dysfunction that occur during severe infection, causing macrocirculatory and microcirculatory dysfunction leading to a mismatch between oxygen delivery and consumption.15 We clearly demonstrate a link between the magnitude of the inflammatory response and risk of CVD events. Patients with a CRP exceeding 300 mg/L had over four times the risk of MI and three times the risk of stroke compared with patients with CRP<100 mg/L. The magnitude of the inflammatory response in BSI is complicated and influenced by age, comorbidities, pathogen virulence, host-pathogen interaction and treatment factors. The smaller effect sizes previously reported for viral infections compared with BSI are likely explained by greater systemic inflammation precipitated by bacterial compared with viral infection. Together, these findings strengthen the hypothesis that infection-associated inflammatory events are proportional to the magnitude of the inflammatory response.

We did not have detailed data regarding symptoms before presentation or treatment, and so we cannot accurately determine which factor(s) are most important in determining the magnitude of systemic inflammation. However, given previous associations with mortality, this is clearly an area of research that needs further exploration.4 Earlier and more effective antimicrobial treatment will likely attenuate the inflammatory response; however, it is unclear if antibiotics given earlier in the patient pathway, such as prehospital, make much difference to outcomes compared with standard of care.14 Therefore, trials of immunomodulatory therapies targeting the host response to BSI and other severe infections in combination with the usual pathogen-directed therapy (ie, antibiotics) are warranted. Anticoagulation, antiplatelet drugs and statins should also be considered for trials. Our data suggest a possible protective effect of these drugs, particularly when considering these are prescribed to individuals with the highest CVD risk who would be expected to be most at-risk following BSI. However, in patients hospitalised with COVID, clinical trials of aspirin and simvastatin have been disappointing,16 17 and trials of therapeutic anticoagulation have been mixed.18 However, a recent meta-analysis of trials of statins in patients with COVID reported reduced mortality, suggesting some promise for this approach.19 These contrast with the clear benefits seen of immunomodulatory therapies such as corticosteroids and IL-6 receptor antagonists, suggesting immunomodulatory approaches may be preferred.20 21 Furthermore, over half of the patients included here were already taking antiplatelet drugs (with a similar number taking a statin), which limits eligibility for an antiplatelet trial as well as increasing the risk of major bleeding events if anticoagulants were given concurrently.

Our study provides useful data to design and power treatment trials, including CRP thresholds to guide participant enrichment to reduce heterogeneity and improve sensitivity, and suggests that CVD events should be collected as important secondary endpoints in BSI clinical trials. Furthermore, our data should prompt extra vigilance for CVD events in patients with BSI, particularly those with comorbidities and very elevated CRP concentrations, as they are likely to be at the highest risk of adverse outcomes.

Key strengths of our study are the use of laboratory-confirmed BSI to accurately confirm culprit organism and timing of specimen collection combined with the use of the SCCS methodology. This study design eliminates residual time-invariant confounding between people who do and do not have BSI, which is inherent in cohort studies. Furthermore, use of troponin results to pinpoint timing of MI and multiple sensitivity analyses make our findings robust and very unlikely to be due to chance. Excluding day 0 risk avoids reverse causation. In reality, the infection that leads to a positive blood culture and increased risk of CVD events will have been developing for days/weeks. This makes day 0 reverse causation biologically implausible and likely leads to an underestimation of CVD risk following BSI. However, we included a separate day 0 risk period to reduce bias and inflation of effect estimates. Furthermore, from a pragmatic perspective, it would be very difficult to design trials or give treatment to mitigate risk in this period before BSI diagnosis is confirmed, as it typically takes 24–48 hours to isolate and identify the culprit organism. The major limitation of our study is the reliance on clinical coding for diagnoses of MI and stroke with potential misclassification without independent assessment of clinical histories, ECG and neuroimaging data to confirm these diagnoses. The use of routinely collected troponin results and multiple sensitivity analyses mitigate this for MI, but similar blood-based biomarkers are not routinely used for stroke, and so an analogous approach could not be undertaken. Therefore, these data should be treated with more caution. Furthermore, the lack of physiological data or more detailed coding did not allow the distinction between type 1 and type 2 MI. However, the association with ischaemic stroke and that these were clinically coded diagnoses suggest a large proportion were type 1 MIs and all were of clinical significance (and not just a troponin leak associated with sepsis). Similarly, the lack of ECG data makes it impossible to determine what proportion of ischaemic strokes are due to new-onset atrial fibrillation precipitated by BSI. Further mechanistic work to delineate the precise pathophysiology of the associations we report are outside the scope of this study and would require prospective data collection. Detailed in-patient treatment data were also not available, so we cannot fully ascertain if antiplatelet and statin medications offer any protection for CVD events or if patients not taking antiplatelet drugs were anticoagulated.

In conclusion, we found compelling evidence of an increased risk of MI and stroke in the first 2 weeks following BSI, which was clearly associated with the magnitude of the inflammatory response. Our findings contribute to the growing body of knowledge of infection-associated CVD events and suggest that targeting the host response to infection may be a fruitful area for future research and improving patient outcomes.