Article Text
Abstract
Background It is vigorously debated whether pneumococcal polysaccharide vaccination (PPV) reduces risk of acute coronary syndrome (ACS) events in patients with community-acquired pneumonia (CAP).
Methods Clinical data were prospectively collected on a population-based cohort of adults presenting with CAP in Edmonton (Alberta, Canada). Multivariable Cox models and propensity matching were used to examine the association between PPV status and ACS events within 90 days of pneumonia. Sensitivity analyses related to PPV administration (before pneumonia vs after) and duration of benefit (90 days vs 1 year) were conducted to rule out confounding.
Results Overall, 6171 patients were included; mean age 59 (SD 21) years, 53% male subjects, 18% had ischaemic heart disease and 2738 (44%) were hospitalised. Within 90 days of pneumonia, ACS events occurred in 175 (3%) patients and most were non-fatal (162 (93%)). In multivariable analyses, PPV exposure was associated with a 58% reduction in ACS events (12 vs 16 events per 100 patient-years, adjusted HR (aHR) 0.42 (0.27 to 0.66)) and results were nearly identical with propensity matching (aHR 0.46 (0.28 to 0.73)). However, indepth sensitivity analyses, with some with large assumptions, could not refute the existence of a small protective benefit of PPV.
Conclusion Even after extensive adjustment using clinical data, the authors observed that PPV exposure was associated with a 60% reduction in ACS events among patients with pneumonia. Sensitivity analyses demonstrated that these findings, at least in part, were probably a result of confounding, most likely the ‘healthy-vaccinee’ effect. Previous observational studies using administrative data suggesting a very large protective benefit of PPV on ACS events may have been heavily confounded.
- Pneumococcal vaccination
- acute coronary events
- clinical trials
- epidemiology
- public health
Statistics from Altmetric.com
Introduction
Community-acquired pneumonia (CAP) is a leading cause of death and hospitalisations among older patients.1 ,2 Hospitalisation for pneumonia is associated with up to eightfold increases in the risk of acute myocardial infarction3 and many ‘pneumonia-related deaths’ are related to non-infectious complications including acute coronary syndrome (ACS) events.4 Explanations for the increase in cardiovascular events observed include (but are not limited to) rupture of atherosclerotic plaques triggered by the systemic inflammatory response or increased physiologic stress, alterations in thrombogenecity and global tissue hypoxaemia.3
Thus, it has been hypothesised that reducing pneumonia severity (or preventing it altogether) should reduce cardiovascular events in high-risk older patients.5 Observational studies of varying designs have evaluated the impact of pneumococcal polysaccharide vaccination (PPV) in reducing or ameliorating ACS events with conflicting results.6–12 Lamontagne et al demonstrated a 50% decrease in the rate of myocardial infarction associated with PPV, although this benefit was entirely restricted to remote (>2 years) rather than recent vaccination.6 Similarly, Hung et al observed 21%–38% reductions in myocardial infarction associated with the use of PPV with or without concurrent influenza vaccination.8 In contrast, several studies have found no association or an increased risk of ACS events with PPV.7 ,9–12
Most of these studies rely on administrative databases and are limited by lack of important prognostic information including clinical severity of pneumonia.6 ,8 Indeed, studies better able to control for premorbid health and functional status tend to show no association between PPV and ACS events,9–11 suggesting residual or refractory confounding related to the ‘healthy-user (or healthy-vaccinee) effect’.13 ,14 To fully explore this possibility and help reconcile previous conflicting results, we assessed the association between PPV exposure and ACS events in a large, clinically-detailed and population-based cohort of adults with pneumonia.
Methods
Subjects
From 2000 to 2002, all patients >17 years of age with pneumonia evaluated at seven emergency departments and six hospitals serving the Edmonton region (Alberta, Canada) were enrolled in a prospective population-based clinical registry.15–17 Briefly, all 6874 registries were managed according to a validated clinical CAP pathway based on the Pneumonia Severity Index (PSI). CAP was defined as the presence of radiographic evidence as determined by the treating physician and at least two of the following signs or symptoms: cough (productive or non-productive), pleurisy, shortness of breath, temperature >38°C and crackles or bronchial breathing on auscultation. Patients with tuberculosis, cystic fibrosis, who were immunocompromised, were pregnant or nursing, or who had been hospitalised in the previous 10–14 days were excluded. The Health Research Ethics Board of the University of Alberta approved the study (Pro00004999).
Data collection
Trained research nurses collected data15 ,17 ,18 with data collection more limited for outpatients compared with our inpatient cohort. Data collected included age, sex, comorbidities, prescription medications and premorbid functional status based on patient or proxy interview. The PSI, a well-validated tool to predict risk of 30-day mortality and used for risk adjustment, was also calculated on all patients at the site of care.
Exposure
PPV status was collected by trained staff masked to all study hypotheses.18 ,19 Patients were considered exposed if they received the 23-valent PPV any time before presenting with pneumonia. Receipt of vaccination was ascertained through multiple avenues including patient and proxy interview, medical record review, contact with primary care physicians and records from regional office of community health. There were no data indicating when the vaccine was administered—only if it was considered current or not (ie, administered within the last 5 years). Patients who received PPV during admission or at discharge were excluded from the main analysis, as the antibody response requires several weeks to manifest.20 ,21
Outcomes
Our primary outcome was a composite end point of ACS events including myocardial infarction or unstable angina or death attributed to ACS within 90 days of CAP presentation. Research nurses prospectively collected all clinically diagnosed ACS events in the emergency department and during hospitalisation. Thereafter, ACS events were ascertained by linking patients to comprehensive provincial healthcare administrative databases.22 Hospitalisations with a primary diagnostic code of ACS events were identified using the International Classification of Diseases, Ninth and Tenth Revision Clinical Modification codes (ICD-9-CM—410, 411; ICD-10-CM—I20.0, I21, I22, I146.x) while death attributed to ACS was ascertained from the provincial vital statistics file.23–25 The quality and validity of the provincial administrative databases are routinely checked both provincially and federally with processes to resolve data issues where identified.26
Statistical analysis
Kaplan–Meier curves were used to display outcomes over time. We created a correlation matrix and to avoid collinearity we excluded variables that were highly (greater than ρ 0.6) correlated with PPV exposure; only coexposure to influenza vaccine was highly correlated with PPV (ρ=0.81) and so both covariates could not be analysed together. Our primary analysis was a multivariable Cox proportional hazard model to estimate the association between PPV status and outcomes adjusted for pneumonia severity based on the PSI; comorbidities including chronic obstructive pulmonary disease, diabetes, ischaemic heart disease (IHD); functional status, smoking status and cardiovascular and other medications.
We also completed a propensity (to receive PPV) score analysis using methods that we, and others, have previously described17 ,27 ,28 based on variables present before pneumonia onset that could be associated with a decision to administer PPV. Since the PSI can only be determined at the time of pneumonia, it cannot influence a remote decision regarding vaccination and it should not be part of the propensity score.17 To ensure comparability and better control for confounding, we conducted a matched analysis using a 5-digit greedy algorithm that matched the propensity score of each patient exposed to PPV to a patient who did not have PPV.28 Multivariable Cox models, matched on propensity score and adjusted for pneumonia severity with the PSI, were used to evaluate our outcomes.17 In all models, patients were followed from time of admission to the event of interest, death, coverage termination or 31 March 2006 (end of the study). All first order interaction terms and proportional hazard assumptions were considered and none achieved clinical significance (p>0.1).
Sensitivity analyses
To evaluate the robustness of our study results, we conducted numerous sensitivity analyses. First, previous evidence suggests the effectiveness of PPV in reducing ACS events is influenced by proximity of vaccine receipt because antibody titre induced by PPV takes weeks to months to develop and then decreases with time.20 ,21 Thus, we replicated the main analysis in the 310 patients who received PPV at the time of hospital discharge and examined short- and long-term outcomes. Second, we extended analyses to evaluate the effect of PPV on ACS events at 1 and 5 years. Given the prevailing hypothesis that the episode of pneumonia ‘triggers’ an ACS event,3 it seems unlikely that after 90 days of recovery and convalescence, PPV would continue to affect longer term outcomes. Moreover, if pneumonia triggers ACS events, any reduction in ACS events at 1 or 5 years associated with PPV should be accompanied by a corresponding reduction in repeat (ie, ‘return’) pneumonia admissions. Thus, we also evaluated the effect of PPV in reducing repeat pneumonia admissions both at 1 and 5 years. Third, we evaluated the effect of PPV in older patients ≥65 years of age as they are routinely targeted for PPV and are at a substantial increased risk of pneumonia and cardiac events. Fourth, we included an interaction term in our models for PPV and history of IHD to evaluate the consistency of our results among those with and without IHD. Last, we evaluated the effect of PPV on hospitalisations or death due to gastrointestinal disorders or fractures as there is no biologically plausible mechanism for PPV to influence these outcomes.7 ,9–12 All analyses were conducted using SAS V.9.2 (SAS Institute Inc.).
Results
Of the 6874 cohort patients with CAP, we excluded 310 (5%) who had PPV first administered during their initial CAP presentation and 393 (6%) who could not be linked to the administrative databases, resulting in a final study cohort of 6171 patients. Mean age was 59 (SD 21), 3261 (53%) were male subjects, 1105 (18%) had a history of IHD and 2738 (44%) were treated as inpatients. As expected, pneumonia severity varied according to site of treatment with 1746 (64%) of inpatients having severe pneumonia (PSI score >90; Class IV/V) compared with only 448 (13%) in outpatients. Overall, 725 (12%) received PPV and were considered up-to-date prior to their pneumonia. Patients receiving PPV were older, more likely to have comorbidities and had a higher medication burden (table 1). Within 90 days of presenting with pneumonia, the composite outcome of ACS-related hospitalisation or mortality occurred in 175 (3%) patients; 162 (3%) had a non-fatal ACS event; and 23 (0.4%) patients died from ACS. ACS-related events were far more common in patients 65 years of age and older (144 (5%) vs 31 (0.9%), p<0.001), and overall, 642 (10%) died within 90 days.
Composite ACS events and PPV exposure: main analyses
After adjustment, PPV was associated with a substantial reduction in risk of the composite outcome (25 of 725 (3%) or 12 events per 100 person-years vs 150 of 5446 (3%) or 16 events per 100 patient years) compared with those without PPV, adjusted HR (aHR) 0.42 (0.27 to 0.66, p<0.001) (table 2, figures 1 and 2). The propensity-matched analysis (c-statistic =0.86) permitted 724 of the 725 (99.9%) patients exposed to PPV to be matched to 724 controls. In adjusted analyses, similar reductions in ACS events were observed in favour of PPV exposure (25 (3%) or 16 events per 100 person-years vs 54 (7%) or 38 events per 100 patient years; aHR 0.46 (0.28 to 0.73), p=0.001) (table 2 and figure 2).
The benefits of PPV exposure were largely restricted to reductions in non-fatal ACS-related events (aHR 0.35 (0.21 to 0.57), p<0.001) and there was no statistically significant independent association with ACS-related deaths (aHR 0.92 (0.32 to 2.63), p=0.88). Again, results of the propensity-matched analysis were nearly identical (table 2).
Sensitivity analysis
First, a large beneficial effect was observed for those who received PPV during the acute CAP episode compared with those without PPV (aHR 0.14 (0.05 to 0.44), p<0.001) (figure 2). In fact, patients who received PPV during the acute CAP event did not experience an ACS event within 2 or 4 weeks of discharge compared with 92 (1.7%) and 125 ACS events (2.3%), respectively, in the non-vaccinated group (p<0.001). Second, extension of the follow-up period did not change our results at 1 year (aHR 0.43 (0.29 to 0.65), p<0.001) or at 5 years (aHR 0.64 (0.47 to 0.87), p=0.004) (figure 2). Moreover, at 1 year, PPV was not associated with reduction in repeat pneumonia admissions (aHR 0.91 (0.69 to 1.20), p=0.50) or at 5 years (aHR 1.09 (0.89 to 1.34), p=0.39). Third, analyses restricted to only those ≥65 years of age were similar to our overall findings (aHR 0.44; 95% CI 0.28 to 0.69, p<0.001). Fourth, no clinically important interaction was observed between PPV and IHD (p>0.10) indicating results of the PPV were consistent among those with and without a history of IHD. Last, we did not observe any statistically significant effect of PPV on hospitalisations or death related to gastrointestinal disorders (aHR 0.70 (0.36 to 1.37), p=0.30) or fractures (aHR 1.28 (0.51 to 3.21), p=0.60).
Discussion
In this observational study of over 6000 CAP patients, 3% of patients suffered ACS events within 90 days of presentation and rates were much higher (6%) in the older people. PPV exposure was associated with a substantial 40%–50% reduction in fatal and non-fatal ACS events within 90 days of pneumonia driven by a significant decrease in non-fatal ACS events. The results were similar with traditional multivariable analyses or a propensity-matched analysis. Moreover, despite additional sensitivity analyses incorporating clinical information and several large assumptions, the association between PPV and ACS could still not be easily discounted. Whether the beneficial results are completely due to PPV or partially explained by intractable confounding is uncertain, but our findings suggest that most previous studies of this question have yielded very exaggerated results.
Two large, heavily cited observational studies have shown a strong association between PPV and reduction in ACS events.6 ,8 In a case control study of 43 209 patients by Lamontagne et al, those with new myocardial infarction were less likely to have received the PPV than controls (adjusted OR 0.53 (CI 0.40 to 0.70)).6 A second prospective cohort study in 36 636 patients showed PPV exposure was associated with a reduction in hospitalisation for acute myocardial infarction (aHR 0.52 (0.38 to 0.71)).8 Both analyses were limited in their adjustment for important confounding covariates. Our primary analyses, which overcome many of these limitations, confirm PPV is associated with a large reduction in ACS events and our estimate of effect is virtually identical to these previous reports.
A more critical interpretation of our findings in the context of our sensitivity analyses suggests that the observed benefits may, at least partially, be related to near impossible to control confounding. For example, a large protective effect against ACS events among patients receiving PPV during the acute pneumonia event was observed. Although this in itself is not improbable, the fact that no ACS events occurred within the first 2–4 weeks postdischarge in this group suggests some selection bias was occurring, as PPV generally requires 2–4 weeks to initiate a reasonable response.20 Moreover, the majority of our cohort were older and antibody response is known to be poor in older populations with comorbidities.29 Further, the reduction in ACS events is postulated (by others) to occur through reductions in infection;6 yet 40%–50% reductions in ACS were still observed over the longer term despite minimal reductions in rates of pneumonia.
One possible explanation for our findings may be related to confounding by the healthy-user or healthy-vaccinee effect,13 ,14 ,17 ,30 whereby more healthy or health-seeking patients are administered PPV compared with non-PPV patients.31 Indeed, patients receiving PPV were more likely to receive influenza vaccination, cholesterol lowering therapy and had stopped smoking, which is consistent with the healthy-user. However, PPV users were also significantly older and had more comorbidities (eg, diabetes, IHD), which should have increased the risk of an ACS event. Although detailed and prespecified, our adjustments for markers of the healthy-user effect17 ,32 and sensitivity analyses could not entirely refute a beneficial effect of PPV on postpneumonia ACS, although our findings suggest that if there is a benefit it is far smaller than demonstrated by others. Indeed, there was no evidence of confounding for non-cardiovascular events such as major gastrointestinal bleeding or fractures. A second possible explanation for our findings is that reductions in ACS events may not be entirely related to reduced occurrence of pneumonia. Possible alternative mechanisms include reductions in inflammation, and atherogenesis and acute plaque rupture through modulation of the antibody response and cross reactivity with low-density lipoproteins.29 ,33
There are several limitations to our work. First, the exact timing of the PPV was unknown except for patients who received the PPV inhospital. However, the expected duration of protection from a dose of PPV is considered at least 5–10 years34 and we were able to confirm in all patients that their PPV was ‘up to date’ (ie, administered within the last 5 years prior to admission). We acknowledge that the protective effect of the PPV may be stronger in those who recently received the vaccine compared with those who received it more remotely, which may have affected our results. Indeed, the strongest effect of PPV on ACS was observed in those who received the vaccine inhospital and may be interpreted by some as representing a dose–response relationship, although our analysis also suggests selection bias. Moreover, we could not ascertain the PPV status after patients were discharged from hospital or emergency departments. However, it is unlikely that PPV soon after the episode of CAP would have had time to affect outcomes within weeks and for longer term outcomes this lack of exposure information would tend to bias to the null. Second, only 12% of our cohort had received the PPV and this may have reduced our power for subgroup analyses and affected the precision of our results. Third, the ACS-related events that occurred during hospitalisation were based on clinical assessments and not based on standard surveillance and definitions (eg, ECG done as indicated, troponins drawn based on symptoms) while postdischarge events were based on (admittedly validated23–25) administrative data coding algorithms. Both approaches would tend to overlook (or not code) minor and less symptomatic events and differential misclassification based on PPV status is unlikely. Last, given the high degree of collinearity, we could not examine both PPV and seasonal influenza vaccination in the same models, although all of our hypotheses related to the PPV. Like the PPV, the influence of influenza vaccination on occurrence of ACS is controversial with studies showing both benefit11 ,35 ,36 and no effect12 because of methodological considerations similar to those we have elaborated on herein.
In conclusion, although our primary analyses suggest that PPV is associated with a significant 40%–50% reduction in the risk of ACS events and confirm some earlier studies, we believe these results should be interpreted with caution. Although PPV may be associated with a small but clinically important reduction in ACS events, it is certainly implausible that the magnitude of benefit is as large as found by us and reported by others.6 ,8 Indeed, more indepth sensitivity analyses suggest the presence of refractory confounding, which may explain, at least in part, the association between PPV and ACS events. Like most observational studies demonstrating an unexpected or unanticipated large benefit,37 previous reports suggesting a very large benefit of PPV exposure on reducing acute cardiovascular events have substantially overestimated its cardioprotective benefits and underestimated how difficult it is to control confounding using administrative data and without clinical information.
References
Footnotes
Funding DTE receives salary support from Alberta Heritage Foundation for Medical Research (AHFMR) and the Canadian Institutes for Health Research (CIHR). SRM receives salary support from AHFMR and holds an endowed chair in patient health management. TJM received Grants-in-aid from Capital Health; and unrestricted grants from Abbott Canada, Pfizer Canada and Janssen-Ortho Canada. The study sponsors played no role in study design or conduct; collection, analysis, interpretation of data; writing of the report; or in the decision to submit the paper for publication. All authors declare that DTE, JJ, JMS, TJM, SRM have no non-financial interests that may be relevant to the submitted work.
Competing interests None.
Ethical approval Ethical approval for the study was obtained from The Health Research Ethics Board of the University of Alberta approved the study (approval number pro00004999).
Provenance and peer review Not commissioned; internally peer reviewed.