Association Between Use of Statins and Mortality Among Patients Hospitalized With Laboratory-Confirmed Influenza Virus Infections: A Multistate Study

Abstract

(See the editorial commentary by Walsh, on pages 1–3.)

Background. Statins may have anti-inflammatory and immunomodulatory effects that could reduce the risk of mortality from influenza virus infections.

Methods. The Centers for Disease Control and Prevention’s Emerging Infections Program conducts active surveillance for persons hospitalized with laboratory-confirmed influenza in 59 counties in 10 states. We analyzed data for hospitalized adults during the 2007–2008 influenza season to evaluate the association between receiving statins and influenza-related death.

Results. We identified 3043 patients hospitalized with laboratory-confirmed influenza, of whom 1013 (33.3%) received statins and 151 (5.0%) died within 30 days of their influenza test. Patients who received statins were more likely to be older, male, and white; to suffer from cardiovascular, metabolic, renal, and chronic lung disease; and to have been vaccinated against influenza that season. In a multivariable logistic regression model, administration of statins prior to or during hospitalization was associated with a protective odds of death (adjusted odds ratio, 0.59 [95% confidence interval, .38–.92]) when adjusting for age; race; cardiovascular, lung, and renal disease; influenza vaccination; and antiviral administration.

Conclusions. Statin use may be associated with reduced mortality in patients hospitalized with influenza.

The role of statin drugs in the reduction of serum lipids has been well documented. More recently, evidence suggesting that statins may positively affect many organ systems and disease states independent of lipid reduction has emerged [1, 2]. As inhibitors of 3-hydroxy-3-methylglutaryl coenzyme A reductase, statins inhibit the synthesis of products of the mevalonate pathway, such as isoprenoids and geranyl-geranyl pyrophosphate [1, 3, 4]. They have been shown to modify intercellular interactions and cellular chemotaxis of the immune system and reduce the release of cytokines and acute-phase proteins. Recent review articles, evaluating mostly observational studies, suggest a potential beneficial role for statins in the treatment of sepsis and community-acquired pneumonia [4, 5].

Influenza viruses are potent inducers of many biological mediators of inflammation [6]; several proinflammatory cytokines have been positively correlated with the symptoms of clinical illness of influenza [7–9], and cytokine dysregulation is regarded as a major contributor to the severe pathophysiologic changes seen in human disease caused by the avian H5N1 and the 1918 pandemic influenza viruses [10–15]. Although statins have a theoretical benefit in downregulating the immune response associated with influenza virus infection, only a few studies have investigated their effect on influenza, and these have found mixed results. The first published study [16] made use of a healthcare research database of several moderate-sized health maintenance organizations in New Mexico. They found that recipients of moderate doses of statins (≥4 mg/day) had a 40% lower mortality from pneumonia and influenza than nonusers of statins. Another study that looked at the association between statins and pneumonia and influenza hospitalizations in elderly patients using administrative data from 10 influenza seasons in Ontario, Canada, found a statistically significant effect against influenza mortality, although the magnitude of the protective effect of statins was small (odds ratio [OR], 0.87) [17]. A third study by Fleming et al looked at the effect of statins on the incidence of acute respiratory infections diagnosed largely in outpatient settings and did not find a benefit attributable to statin use [18].

Given the inherent limitations of administrative data in these studies, specifically the difficulty in precisely ascertaining exposure and outcome, we sought to evaluate associations between statin medication administration and mortality in hospitalized patients by utilizing data from a population-based surveillance system that collects information about individuals hospitalized with laboratory-confirmed influenza.

METHODS

This study was conducted using data from the Centers for Disease Control and Prevention’s (CDC) Emerging Infections Program (EIP) influenza hospitalization surveillance system. The EIP network performs active, population-based surveillance for persons hospitalized with laboratory-confirmed influenza in 59 counties in 10 states (California, Colorado, Connecticut, Georgia, Maryland, Minnesota, New Mexico, New York, Oregon, and Tennessee), representing 7% of the US population. Cases were ascertained through review of hospital laboratory lists or admission or infection-control logs. Data on demographic and clinical characteristics were collected from the medical record of each patient. Influenza vaccine information was obtained from the medical chart, state vaccination registry, primary care provider, or patient or their proxy. A patient was considered vaccinated against influenza if the vaccine was received >2 weeks prior to admission. Statin use prior to hospitalization was obtained from the hospital chart: we defined patients who had a statin medication mentioned in their admission history and physical as taking statins prior to hospitalization. Statin use during hospitalization was obtained from review of medication administration records in the hospital chart: we considered patients who had any record of statin administration at any time during their hospitalization as having received statins during their hospitalization. We did not attempt to collect statin dose or frequency either prior to or during hospitalization.

Subjects were included in the surveillance system if they (1) were ≥18 years at the time of hospital admission, (2) resided in the surveillance area, (3) were admitted to a surveillance area hospital between 1 October 2007 and 30 April 2008, (4) were admitted within 14 days of a positive influenza test, and (5) had evidence of a positive influenza test by at least 1 of the following methods: viral culture, immunofluorescence antibody staining (direct or indirect), reverse-transcriptase polymerase chain reaction, a commercially available rapid diagnostic test for influenza, serologic testing, or a positive, unspecified influenza test noted in the medical chart. Cases were excluded from the surveillance system if the infection was nosocomially acquired (illness onset >3 days after hospital admission) or if the initial hospitalization occurred >14 days after the positive influenza test. Patients were consented when contacted for vaccination history by study personnel if required by site-specific institutional review boards, and those who refused to participate were removed from the analysis data set. The CDC’s EIP conducted an audit for the 2007–2008 influenza season to capture any additional deaths that occurred outside the hospital. The audit was performed more than a year after the last hospital discharge in order to ensure complete death ascertainment. Patients who died >30 days after their influenza test were analyzed as having a nonevent. The 30-day cutoff was chosen in order to capture deaths attributable to influenza and due to secondary bacterial infections or exacerbation of chronic illnesses that can occur up to 2 weeks after influenza virus shedding has ended [19]. Other cutoffs were examined (7 days, 14 days, 21 days); patients who died after the cutoff were analyzed as nonevents.

Statistical analyses were conducted using SAS software (version 9.1.3; SAS Institute). Descriptive frequencies, χ² tests, and Wilcoxon rank-sum tests were performed to identify demographic and clinical characteristics associated with statin use among patients hospitalized with laboratory-confirmed influenza. Bivariate logistic regression was used to model associations between mortality and dependent variables. Pearson correlation statistics between statin use and other dependent variables were tested. The effect of statin administration prior to or during hospitalization was assessed using multivariable logistic regression; statin use was our a priori variable of interest. Variables that reached statistical significance in bivariate regression and additional variables that we thought could influence the outcome variable were also included in the initial model. Backward deletion methods (P < .05 to stay) identified which variables remained in the preliminary model. Schwarz criterion was used to assess model fit, and variables were placed back into the model if fit improved regardless of whether they met the P value criteria to stay. The covariates in the final model were included in models that examined the 7-day, 14-day, and 21-day mortality thresholds.

The protocol was reviewed and approved by the institutional review boards of the CDC and the local EIPs. The protocol was viewed as an extension of surveillance and not research in Maryland and Minnesota.

RESULTS

During the 2007–2008 influenza season, 3071 patients with laboratory-confirmed influenza were admitted to hospitals in the EIP surveillance areas. Twenty-eight surveillance subjects were excluded from this study because they did not consent to the study (n = 18) or the EIP site did not acquire human investigations committee approval (n = 10).

The median age for the remaining 3043 patients was 70.4 (interquartile range [IQR], 51.0–82.9), and 56.0% were female. Of those with known vaccination status, over half were vaccinated for influenza (57.1%), and 33.3% were given statin medications prior to or during hospitalization. Demographic characteristics, underlying conditions, vaccination history, and receipt of antivirals are presented in Table 1 for the entire population, for those on statins prior to or during hospitalization, and for those who never received statins prior to or during hospitalization.

Table 1.

Characteristics of Adults Hospitalized With Laboratory-Confirmed Influenza by Statin Use and Nonstatin Use Prior to or During Hospitalization

	Adults Hospitalized During the 2007–2008 Influenza Season
Characteristic:	Total (N = 3043)	Use of Statins Prior to or During Hospitalization (n = 1013)	No Statins Prior to or During Hospitalization (n = 2030)	P Valuea
Age in years				<.0001
Median	70.4	75.9	63.5
(IQR)	(51.0–82.9)	(64.7–83.8)	(45.0–82.4)
Sex				<.0001
Female	1703	514 (50.8)	1189 (58.6)
Male	1339	498 (49.2)	841 (41.4)
Unknownb	1	1 (0)	0
Race and ethnicity				<.0001
Hispanic	175	45 (4.4)	130 (6.4)
White	1907	697 (68.8)	1210 (59.6)
Black	535	132 (13.0)	403 (19.9)
Asian	84	33 (3.3)	51 (2.5)
Hawaiian/Pacific Islander	5	2 (0.2)	3 (0.2)
American Indian/Native Alaskan	3	1 (0.1)	2 (0.1)
Multiracial	11	2 (0.2)	9 (0.4)
Non-Hispanic, race unknownb	6	2 (0.2)	4 (0.2)
Unknown race and ethnicityb	317	99 (9.8)	218 (10.7)
Admission from LTCF
Yes	401	118 (11.7)	283 (13.9)	.067
No	2613	890 (87.9)	1723 (84.9)
Unknownb	29	5 (0.5)	24 (1.2)
Underlying conditionsc
Cardiovascular disease	1336	669 (66.0)	667 (32.9)	<.0001
Chronic metabolic disease	1083	526 (51.9)	557 (27.4)	<.0001
Chronic lung disease	747	308 (30.4)	439 (21.6)	<.0001
Renal disease	481	231 (22.8)	250 (12.3)	<.0001
Asthma	477	124 (12.2)	353 (17.4)	.0002
Vaccination status
Vaccinated	1513	629 (62.1)	884 (43.6)	<.0001
Not vaccinated	1139	297 (29.3)	842 (41.5)
Unknownb	391	87 (8.6)	304 (15.0)
Antivirals
Initiated within 48 h of admit	773	276 (27.3)	497 (24.5)	.17
No antivirals within 48 h of admit	2064	676 (66.7)	1388 (68.4)
Unknownb	206	61 (6.0)	145 (7.1)

Data are no. (%) except where indicated.

Abbreviations: IQR, interquartile range; LTCF, long-term care facility.

a

χ² test for categorical variables; Wilcoxon rank-sum test for continuous variable.

b

Unknowns were excluded from χ² analysis for sex, race, admission from LTCF, vaccination status, and antivirals.

c

Underlying conditions are not mutually exclusive.

The majority of statin users were on statins both prior to and during hospitalization (76%). One hundred eleven patients (11%) were on statins only during hospitalization, and 132 patients (13%) were on statins only prior to hospitalization.

Patients who received statins were more likely to be older, male, and white; to have cardiovascular, metabolic, lung, or renal disease; and to have been vaccinated against influenza. Patients with asthma were less likely to have received statins prior to or during hospitalization.

Among patients hospitalized with laboratory-confirmed influenza, 151 (5.0%) died within 30 days of their influenza test. Thirty-nine percent of the deaths occurred after discharge from the hospital. However, the majority of deaths occurred shortly after discharge; the median number of days until death for deaths occurring after discharge was 7 (IQR, 3–15). Unadjusted, the association between statin use and death did not reach statistical significance (OR, 0.71 [95% confidence interval {CI}, .49–1.03]).

Collinearities between statin use prior to or during hospitalization and other dependent variables were examined, and major collinearities were not identified. Age, race, previous influenza vaccination, asthma, cardiovascular disease, renal disease, and lung disease reached statistical significance in bivariate analysis and were included in the logistic regression model. In addition, variables that we thought could influence the outcome variable (sex, antivirals administered within the first 48 hours of hospitalization, and metabolic disease) were also included in the model.

In our model, statin use was significantly associated with a decreased odds of death (adjusted odds ratio [AOR], 0.59 [95% CI, .38–.92]) while controlling for age, white race (black as referent), cardiovascular disease, lung disease, renal disease, vaccination status, and antiviral use initiated within 48 hours of hospitalization (Table 2). White race, age, cardiovascular disease, lung disease, and renal disease were all associated with an increase in odds of death. Model fit improved with the inclusion of influenza vaccine and antivirals, even though these variables were not statistically associated with death. This model also resulted in a more conservative estimate of the effect of statins on mortality (when vaccination and antiviral statuses were removed, the AOR decreased to 0.44 [95% CI, .29–.65]). Asthma, chronic metabolic diseases, and sex did not remain in the model.

Table 2.

Crude and Adjusted Odds Ratios of Death During Hospitalization and Within 30 Days of Hospital Discharge Among Patients Hospitalized With Laboratory-Confirmed Influenza (N = 3043)

	Died (n = 151)	OR (95% CI)
Predictor variables	No. (%)	Unadjusted	Adjustedf
Received statins prior to or during admission	40 (26.5)	0.71 (.49–1.03)	0.59 (.38–.92)
Agea	…	1.04 (1.03–1.05)	1.02 (1.01–1.04)
Maleb	67 (44.4)	1.02 (.73–1.41)	…
Whitec	129 (95.6)	6.40 (2.81–14.58)	3.18 (1.34–7.53)
Cardiovascular disease	101 (66.9)	2.71 (1.92–3.83)	1.64 (1.02–2.62)
Chronic metabolic diseaseb	53 (35.1)	0.98 (.69–1.38)	…
Chronic lung disease	64 (42.4)	2.38 (1.70–3.32)	1.98 (1.30–3.01)
Renal disease	39 (25.8)	1.93 (1.32–2.82)	1.83 (1.13–2.96)
Asthmab	14 (9.3)	0.54 (.31–.94)	…
Influenza vaccine receiptd	86 (72.3)	2.02 (1.34–3.04)	1.26 (.79–2.00)
Antivirals initiated within 48 h of admite	30 (23.3)	0.80 (.53–1.22)	0.79 (.49–1.28)

P < .05 to remain in model.

Abbreviations: CI, confidence interval; OR, odds ratio.

a

Age as a continuous variable.

b

Did not remain in final model.

c

Black race as referent, other race and ethnicities excluded (n = 135).

d

Unknown vaccine receipt excluded (n = 119).

e

Unknown antiviral initiation excluded (n = 129).

f

Adjusted model (n = 98).

There were too few patients that had received statins prior to hospitalization only or received statins only during their hospitalization for us to evaluate the impact of statins on mortality separately for these 2 groups. We did evaluate the relationship between statin use and mortality in the subgroup of patients who had all received statins prior to hospitalization (the same population represented in Table 2 but with patients who did not start taking statins until they were hospitalized excluded [n = 111]) and found similar results (AOR, 0.58 [95% CI, .37–.92]). We also looked at the association between statins and mortality by different lengths of time between influenza test and date of death and found protective results for all 4 time periods in the adjusted models (Table 3). The point estimate (OR) was slightly farther from 1 for the narrower lengths of time, but the CIs overlapped for all 4 periods.

DISCUSSION

Our study found that statins were associated with a decrease in odds of dying among cases hospitalized with laboratory-confirmed influenza when adjusted for age, race, cardiovascular disease, chronic lung disease, renal disease, influenza vaccine receipt, and initiation of antivirals within 48 hours of admission. To our knowledge, this is the first published observational study that evaluates the relationship between statin use among patients hospitalized with laboratory-confirmed influenza and death.

Our findings confirm and add to those of Frost et al and Kwong et al. Frost et al’s matched cohort and case-control study [16] used health maintenance organization encounter data to evaluate the relationship between statin therapy and influenza mortality and combined codes for unspecified pneumonia and influenza (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes 486–487), whereas our study used a surveillance system for laboratory-confirmed influenza hospitalizations, giving our study less chance for misclassification of disease outcomes. The cohort study published by Kwong et al [17] showed that being prescribed a statin was statistically associated with a decrease in pneumonia hospitalizations and 30-day pneumonia mortality, although the effects were minimal (ORs of 0.92 and 0.84, respectively). Kwong et al’s study combined pneumonia and influenza diagnoses into a single outcome (ICD-9 codes 480–487 and ICD-10 codes J10–J18) and did not capture information on use of statin medication during hospitalization; this lack of specificity in determining exposure and outcome may account for the differences in the magnitude of the protective effect of statins compared to our study. A more recent study published by Fleming et al showed no benefit of statin use on incidence of acute respiratory infections in primary care [18]; however, their study defined statin users by prescription records and did not assess prescription adherence. We make an assumption with our data that self-reported statin use during an admission history by the patient is an accurate assessment of prescription adherence. In addition, we assume that documentation of statin use from medical administration records during the hospital stay is accurate. A more critical difference between our studies may be the disease end points chosen. In Fleming’s study, the disease end points included influenza-like illness, acute bronchitis, pneumonia, and upper respiratory infections but did not include more severe sequelae as a result of influenza infection. Statins may be more beneficial at preventing severe disease outcomes, such as death, but may not play a role in reducing incidence of infection or minor illness. Other published studies evaluating the role of statins in decreasing mortality from sepsis and community-acquired pneumonia [4, 5] suggest that our findings are biologically plausible.

There are several limitations to this study. First, our cohort of patients may not be representative of all persons hospitalized with influenza because patients not tested for influenza would not have been detected by the EIP surveillance system. In addition, the limited sensitivity of rapid tests also leads to case underascertainment. An earlier evaluation of the EIP pediatric surveillance system estimated the sensitivity of the EIP system for patients <18 years hospitalized with influenza to be 38% [20]; the EIP network has not duplicated this type of evaluation in older age groups. A more recent evaluation by the Oregon EIP site conducted during the 2006–2007 influenza season investigated the demographic characteristics of adult patients hospitalized with influenza and pneumonia (ICD-9-CM codes 480–487) who were not tested for influenza during their hospital stay compared with patients with laboratory-confirmed influenza reported to the Oregon EIP (Oregon Public Health Division, unpublished data). The age and sex distribution and the presence of chronic conditions of the 2 groups of patients were not statistically different, suggesting that patients whose influenza virus infection may have gone undetected may not differ significantly from patients reported to the EIP hospitalization surveillance system. Even if patients not captured by our surveillance system differed from those reported to the EIP system, we do not have any reason to suspect that their exposures to statin medications would have systematically differed.

Second, our study relied mostly on data available from chart review, limiting the quality of information regarding statin use that we were able to capture, which may have been an issue for the deaths identified as occurring after discharge. All EIP sites reviewed local vital statistics records to determine if any of the hospitalized patients died within 30 days following hospital discharge in an attempt to minimize bias from underascertainment of deaths. However, this practice introduced another potential limitation into our study in that we could not determine exposure to statins after patients were discharged from the hospital; more than one-third of our deaths occurred after discharge. We classified these patients as statin users or nonusers based on their history of statin use prior to or during hospital admission.

In Table 3, we presented data that showed the OR between statins and death was smallest when the time between influenza test and death was shortest (7 days). One must interpret our analysis cautiously because multiple hypothesis testing was performed, which could inflate type 1 error. Furthermore, this trend is not statistically significant because the CIs overlap; however, it may be worth exploring in future studies. Deaths occurring >1 week after the diagnosis of influenza may be more likely due to other causes, creating a misclassification bias that drives the estimate of the OR closer to the null hypothesis.

Table 3.

Crude and Adjusted Odds Ratios for Mortality Within 7 Days, 14 Days, 21 Days, and 30 Days From Influenza Test Date Among Patients Hospitalized With Laboratory-Confirmed Influenza (N = 3043)

	Died (n = 151)	OR (95% CI)
Weeks from positive test date to death	No. (%)	Unadjusted	Adjusteda
Within 7 days	70 (46)	0.54 (.30–.96)	0.46 (.23–.90)
Within 14 days	110 (73)	0.55 (.35–.87)	0.51 (.30–.88)
Within 21 days	134 (89)	0.67 (.45–1.00)	0.60 (.37–.97)
Within 30 days	151 (100)	0.71 (.49–1.03)	0.59 (.38–.92)

Abbreviations: OR, odds ratio; CI, confidence interval.

a

All models adjusted for age, white race (black as referent), cardiovascular disease, lung disease, renal disease, vaccination status, and antiviral initiation within 48 hours.

The third limitation of our study, also related to our use of chart review alone, was that it was difficult to ascertain the underlying functional health status of these patients, which may potentially confound the relationship between statins and severity of influenza. Patients on statins may have better access to healthcare, making statin use a marker for a healthier lifestyle [21]. Since receipt of influenza vaccine could be considered a marker for a healthier status and removing vaccination status from the model influenced the association between statins and death, we retained vaccination status in the model to control for potential confounding between vaccination and statin use and to improve model fit. Also, patients who have better access to care may be less likely to delay seeing their provider and may be more likely to be hospitalized at an earlier, less severe stage of illness than those without access to care who delay seeking care until their symptoms are more severe. However, our statin users were older, appeared to have more underlying conditions, and had longer lengths of stay than nonstatin users, suggesting that their baseline health status may not be better than that of nonstatin users.

And certainly, as with any observational study, there may have been other confounding factors that we were unable to capture through chart review. Our analysis revealed an unexpected finding that white patients had higher odds of dying than black patients. Our race data were not complete, and this may have led to some misclassification issues, or this may indicate a chance finding, selection bias, or an unidentified confounding factor.

Last, we determined from chart review whether patients received statins prior to or during hospitalization, but we did not attempt to collect sufficient data from medication records to track statin use during patients’ entire hospital stay. This was a limitation because patients who were mechanically ventilated sometime during their hospital stay would not have been able to take oral statin medications while ventilated. Thus, we cannot classify patients by whether they received statins before, during, or after a period requiring mechanical ventilation. We did not exclude these patients from our final analysis because this group also represents the sickest among patients hospitalized with laboratory-confirmed influenza. We did analyze the data set with mechanically ventilated patients removed, which resulted in a slightly stronger protective association between statins and death (AOR, 0.54 [95% CI, .31–.92]).

Although observational studies evaluating influenza vaccine effectiveness may be subject to selection bias and residual confounding, most studies confirm the benefits of vaccination for reducing mortality from influenza [22–24]. Our finding that vaccine was not effective in preventing mortality from influenza is likely explained by the timing of our study. We evaluated the impact of statins on influenza mortality in 2007–2008, a year in which the vaccine was not well matched to the circulating viruses [25, 26], so our finding that statins are more protective than vaccine may not be reproducible in a year when the vaccine provides a good match to circulating strains of virus. In addition, vaccination status may not impact mortality in the event of vaccine failure regardless of match to circulating strains; 50% of our patients were over the age of 70, an age group in which influenza vaccine is known to be less effective [27].

A recent study by McGeer et al [28] found a significant reduction in mortality among hospitalized patients treated with oseltamivir. Our study did not confirm this finding; although our crude OR for antivirals was mildly protective against death, it was not statistically significant (AOR, 0.80 [95% CI, .53–1.22]). This difference may be due to our differing definitions of antiviral use. McGeer’s study defined course of antiviral therapy as those who received ≥4 doses and continued their therapy during their hospitalization; patients on antivirals who did not meet this criteria were excluded from their study. Our study did not assess duration of antiviral therapy, and we did not require the therapy to continue during hospitalization.

Future studies evaluating the role of statins in treating influenza should examine underlying functional status, dose and duration of statin therapy, use of statins in younger age groups, and identification of the most effective class of statins. The issue of a potential benefit of statins would be best addressed in the context of randomized controlled trials (RCTs) and would allow for examination of such issues as dose response, use in younger age groups, and identifying the most effective class of statins. RCTs would also likely be the only method of assessing whether long-term prophylaxis with statins would be a worthwhile strategy in reducing morbidity and mortality from influenza.

Despite their limitations and the need for RCTs before statins can be widely promoted for the treatment of influenza, our findings suggest that statins are a promising area of exploration and could provide a useful adjunct to antiviral medications and vaccine, particularly in settings where circulating influenza virus strains are not susceptible to antiviral medications or vaccine is in short supply or not well matched to circulating viruses.

Notes

Acknowledgments.

We thank all the influenza surveillance officers of the Emerging Infections Program, specifically Pam Daily (California), Deborah Aragon (Colorado), Kim Yousey-Hindes (Connecticut), Darcy Fazio (Connecticut), Kyle Openo (Georgia), Elisabeth Vaeth (Maryland), Craig Morin (Minnesota), Emily Hancock (New Mexico), Nancy Spina (New York–Albany), Ruth Belflower (New York–Rochester), Christine Long (New York–Rochester), Melissa Powell (Oregon), Terri McMinn (Tennessee), and Sandra Gray (Tennesee), as well as our collaborators at the CDC (Tiffany D’Mello, Mackenzie Nowell, and Joyce Gyamfi), for their contributions to this study.

Financial support.

This work was funded by the Emerging Infections Program Cooperative Agreement between the ten study sites and Centers for Disease Control and Prevention.

Potential conflicts of interest.

W. S. has been an occasional consultant to Sanofi Pasteur, Novartis, Pfizer, Dynavax, and GlaxoSmithKline and is a member of the Data Safety Monitoring Board for experimental vaccine studies for Merck. All other authors report no potential conflict.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Author notes

Presented in part: 47th Annual Meeting of the Infectious Diseases Society of America, Philadelphia, PA, 30 October 2009. Presentation 706.