Comparative Effectiveness of High-Dose Versus Standard-Dose Influenza Vaccine Among Patients Treated by Maintenance Hemodialysis

Abstract

Rationale & Objective:

Studies among patients on maintenance dialysis suggest that standard-dose influenza vaccine (SDV) may not prevent influenza-related outcomes. Little is known about the comparative effectiveness of SDV versus high-dose influenza vaccine (HDV) in this population.

Study Design:

Cohort study using data from the United States Renal Data System.

Setting & Participants:

507,552 adults undergoing in-center maintenance hemodialysis between the 2010/11 and 2014/15 influenza seasons.

Exposures:

SDV and HDV.

Outcomes:

All-cause mortality, hospitalization due to influenza or pneumonia, and influenza-like illness (ILI) during the influenza season.

Analytic Approach:

Patients were eligible for inclusion in multiple yearly cohorts, thus our unit of analysis was the influenza patient-season. To examine the relationship between vaccine dose and effectiveness outcomes, we estimated risk differences and risk ratios using propensity score weighting of Kaplan-Meier functions, accounting for a wide range of patient- and facility-level characteristics. For non-mortality outcomes, we used competing risk methods to account for the high mortality rate in the dialysis population.

Results:

Within 225,215 influenza patient-seasons among adults ≥65 years, 97.4% received SDV and 2.6% received HDV. We observed similar risk estimates for HDV and SDV recipients for mortality (risk difference, −0.08%; 95% CI, −0.85% to 0.80%), hospitalization due to influenza or pneumonia (risk difference, 0.15%; 95% CI, −0.69% to 0.93%), and ILI (risk difference, 0.00%; 95% CI, −1.50% to 1.08%). Our findings were similar among adults <65, as well as within subgroups defined by influenza season, age group, dialysis vintage, month of influenza vaccination, and vaccine valence.

Limitations:

Residual confounding and outcome misclassification.

Conclusions:

The HDV does not appear to provide additional protection beyond the SDV against all-cause mortality or influenza-related outcomes for adults undergoing hemodialysis. The additional cost and side-effects associated with HDV should be considered when offering this vaccine. Future studies of HDV and other influenza vaccine strategies are warranted.

Introduction

Influenza causes substantial morbidity and mortality in patients with end-stage renal disease (ESRD). Patients with ESRD have an impaired innate and adaptive immune system, including defects in complement activation and B- and T-cell function,^1–3 which contribute to increased risks for severe complications from influenza or influenza-related hospitalizations, mortality, and healthcare costs compared to the general population.^4–7 The Centers for Disease Control and Prevention (CDC) has long recommended that patients with ESRD receive annual influenza vaccination.^8,9

Currently, several types of influenza vaccines are available in the United States, including the standard-dose and high-dose influenza vaccines (HDV). The trivalent—and more recently quadrivalent—inactivated, standard-dose, seasonal influenza vaccines have been administered as standard practice in dialysis clinics. In 2009, the United States Food & Drug Administration licensed a trivalent HDV for use among adults ≥65 years to induce higher antibody responses and provide better protection from influenza than the standard-dose vaccine (SDV).¹⁰ The HDV contains the same three strains as SDV but more antigen (60 versus 15 μg per strain) than standard vaccines. The use of HDV among patients on maintenance hemodialysis has increased slowly over time.¹¹

An accumulating body of literature demonstrates that the HDV is more effective than SDVs in preventing influenza-related medical encounters, hospitalizations, and mortality in the general population of adults ≥65 years.^12–18 Yet, some studies in healthy, older adults have not demonstrated a benefit of HDV.^19,20 In the dialysis population, a recent observational study by Miskulin et al. reported that HDV was more effective than SDV in preventing all-cause hospitalization during the 2016/17 influenza season but not the 2015/16 season.²¹ However, this study was limited by small sample size, adjustment for a small set of potential confounders, and the absence of influenza-related outcomes, warranting additional examination of the effectiveness of the HDV in the dialysis population.

We sought to compare the clinical effectiveness of HDV vs. SDV in preventing all-cause mortality, hospitalization due to influenza or pneumonia, and influenza-like illness (ILI) among patients with chronic kidney failure undergoing maintenance hemodialysis. As the healthy-user effect and frailty are well-documented challenges of observational studies of influenza vaccine effectiveness in populations with poor health, we used robust methods to account for possible confounding. We also used competing risk methods to account for the high mortality rate in the dialysis population.

Methods

Data Source

We utilized the United States Renal Data System (USRDS) to identify patients with chronic kidney disease receiving in-center maintenance hemodialysis during the years 2009–2015. USRDS is a national registry of patients with ESRD who are eligible for Medicare coverage from the Centers for Medicare and Medicaid Services (CMS), and it contains routinely-reported clinical information from dialysis centers, as well as Medicare claims for inpatient, outpatient, and physician diagnoses and procedures.²²

Study Design and Patients

We constructed yearly cohorts for five individual influenza seasons: 2010/11, 2011/12, 2012/13, 2013/14, 2014/15. The index dates were anchored on the date of influenza vaccination (Figure 1), which was required to be administered prior to the start date of the influenza season. Eligible patients included adults ≥18 years with ESRD and Medicare as a primary insurance payer who had started hemodialysis at least 9 months prior to the index date, to allow for a 3-month Medicare enrollment period prior to the 6- month baseline period. We further required continuous hemodialysis receipt for the 3 months immediately prior to vaccination. Baseline covariates were ascertained during the 6-month baseline period prior to the index date. The follow-up period began on the start date of the influenza season. The start and end dates of each influenza season were defined using national influenza surveillance data from the CDC.^23–27 Specifically, the start and end of each season were defined as the midpoint of the first week during which >10% and <10%, respectively, of national culture isolates were positive for influenza (Table S1). The primary analysis was conducted in adults ≥65 years; a secondary analysis was conducted in adults <65 years. Patients could be eligible for inclusion in multiple yearly cohorts.

Figure 1.

Study design schematic of inclusion criteria and follow-up time for each yearly cohort of adult patients undergoing maintenance hemodialysis. The index date was anchored on the date of influenza vaccination, which was required to be administered prior to the start date of the influenza season. Follow-up began on the start date of the influenza season.

Exposure Assessment

Influenza vaccines were classified as HDV, SDV, or unknown (Table S2), based on outpatient, physician, and inpatient procedure coding in billing claims for influenza vaccine or influenza vaccine administration. We searched for Current Procedural Terminology (CPT), Healthcare Common Procedure Coding System (HCPCS), and International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes assigned between August 1 and the end of each influenza season. We assumed that multiple codes within a 7-day period were duplicate codes for the same vaccination, and therefore, two codes for the same vaccine dose were collapsed, whereas two codes for a known and unknown dose were categorized as the known dose. Codes within 7 days for different known doses were considered separate doses. Codes spaced by greater than 7 days were also considered separate doses.

Outcome Assessment

We examined three primary clinical outcomes: all-cause mortality, the first occurrence of hospitalization for influenza or pneumonia, and the first occurrence of ILI. All-cause mortality was assessed using the CMS ESRD Death Notification Form. As laboratory-confirmed influenza status is not available in the claims data, we employed two levels of claims-based influenza definitions from diagnosis coding: 1) hospitalization for influenza or pneumonia; and 2) inpatient or outpatient influenza-like illness (ILI) (see Table S3 for code lists). We performed a sensitivity analysis of ILI with a more narrow definition, as done previously.^28,29

Covariates

Our analyses included a wide range of covariates identified during the 6-month baseline period (Figure 1, Table S4). Covariates included demographic characteristics (e.g., age, sex, race, dual-eligible for Medicaid, region, year), clinical characteristics (e.g., cause of ESRD, years on dialysis), dialysis facility characteristics (e.g., affiliation, type, profit status, size), and several comorbidities and procedures captured using ICD-9-CM and HCPCS codes. To account for potential confounding by frailty^30,31 and the healthy-user bias,^32,33 we also included covariates on preventive health services (e.g., other vaccinations, health screenings), healthcare utilization (e.g., skilled nursing days, hospital days), and frailty indicators (e.g., oxygen use, mobility aids). We categorized the timing of vaccine administration as vaccination in August or September versus October through the start of influenza season.

Statistical Analysis

Patients could be eligible for inclusion in multiple yearly cohorts, thus our unit of analysis was the influenza patient-season. To examine the relationship between vaccine dose and effectiveness outcomes, we estimated crude and weighted cumulative risk functions using Kaplan-Meier methods, accounting for the competing risk of death (for the non-mortality outcomes of ILI and hospitalization due to influenza or pneumonia).^34–36 We addressed potential confounding due to differences in observed covariates between HDV and SDV recipients by using propensity score weighting methods. Specifically, we applied standardized mortality ratio (SMR) weights to the Kaplan-Meier estimator. We estimated the predicted probability of receiving HDV, or the propensity score (ps), using logistic regression with a priori selected potential confounders as predictors (Table S4). Age was modeled using restricted quadratic splines with four knots. SMR weights for SDV recipients were calculated as ps/(1-ps); HDV recipients remained unweighted.³⁷ We plotted the standardized mean differences of baseline covariates in the unweighted and SMR-weighted populations to determine whether weighting the population reduced imbalances of observed covariates and made the treatment groups more exchangeable.³⁸ We stratified models by influenza season, valence, age, dialysis vintage (in years), and timing of influenza vaccination; weights were created separately for each subgroup analysis. Covariate balance was assessed separately for each subgroup analysis. Patients receiving influenza vaccine prior to the beginning of the influenza season were followed from the first day of the influenza season until the first occurrence of the outcome, a competing risk event (i.e., death for non-mortality outcomes), or a censoring event (i.e., disenrollment from Medicare Parts A and B, kidney transplant, switch to peritoneal dialysis, loss to follow-up, receipt of a subsequent influenza vaccine, or end of influenza season). We calculated risk differences and risk ratios using the cumulative risk function estimates at the end of each influenza season. A nonparametric-based bootstrap was used to estimate 95% confidence intervals (CIs) for daily risk differences and risk ratios between groups. Additionally, we conducted bias analyses to estimate the strength of unmeasured confounders that would be required to mask associations between vaccine type and each primary outcome after adjusting for measured covariates.³⁹ This study using deidentified data was considered exempt from human subjects review by the Institutional Review Board at Washington University. Analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC) and R Statistical Software 3.3.⁴⁰

Results

After applying inclusion and exclusion criteria, we identified 255,281 eligible adult patients who contributed 507,552 unique influenza patient-seasons (Figure S1). In the primary analysis of adults ≥65 years (n = 225,215 influenza patient-seasons), SDV made up 97.4% (n=219,439) of the eligible vaccinations, and HDV was administered in the remaining 2.6% (n=5,776). All HDV were trivalent, whereas 76.7% of SDV was trivalent and 23.3% was quadrivalent. Tables 1 and S5 present characteristics of the study population stratified by vaccine dose. The mean age was slightly higher among HDV recipients than SDV recipients (75.8 vs. 74.6 years). HDV administration was less common among patients who were black or other race, dual-eligible for Medicaid, on dialysis for ≥3 years, or residing outside of the Midwest. HDV recipients had a higher prevalence of several comorbidities including ischemic heart disease, chronic obstructive pulmonary disease, cancer, liver disease. HDV and SDV recipients had a similar prevalence of frailty indicators. HDV recipients were more likely to receive preventive health care such as diabetic eye exams, lipid testing, and cancer screenings. Some characteristics of the HDV recipients changed over time (e.g., dialysis vintage; Table S6). The distribution of observed covariates was well-balanced after propensity-score weighting, as indicated by standardized mean differences less than 10% in the weighted population;⁴¹ Figure S2 presents these results for the primary analysis.

Table 1.

Characteristics of the 225,215 unique influenza patient-seasons among US adults ≥65 years undergoing maintenance hemodialysis, by influenza vaccine dose.

Characteristic	HDV n = 5,776 (2.6%)	SDV n = 219,439 (97.4%)
Mean age, y	75.8 ± 6.9	74.6 ± 7.0
Male sex	3,064 (53.0%)	111,827 (51.0%)
Race
White	4,378 (75.8%)	134,917 (61.5%)
Black	1,125 (19.5%)	71,408 (32.5%)
Other	273 (4.7%)	13,114 (6.0%)
Cause of chronic kidney failure
Diabetes	2,524 (43.7%)	103,702 (47.3%)
Hypertension	1,898 (32.9%)	71,973 (32.8%)
GN	443 (7.7%)	14,708 (6.7%)
Other	911 (15.8%)	29,056 (13.2%)
Region
Northeast	928 (16.1%)	37,869 (17.3%)
South	1,959 (33.9%)	97,158 (44.3%)
West	801 (13.9%)	39,280 (17.9%)
Midwest	2,088 (36.1%)	45,132 (20.6%)
Dual-eligible for Medicaid	1,338 (23.2%)	76,492 (34.9%)
Dialysis vintage
< 1 year	420 (7.3%)	11,394 (5.2%)
1–2 years	2,512 (43.5%)	76,237 (34.7%)
3–4 years	1,348 (23.3%)	54,441 (24.8%)
5–9 years	1,172 (20.3%)	58,560 (26.7%)
≥ 10 years	324 (5.6%)	18,807 (8.6%)
Hospitalized in last month (weeks)
0 days	5,391 (93.3%)	205,549 (93.7%)
1–6 days	257 (4.4%)	9,358 (4.3%)
≥ 7 days	128 (2.2%)	4,532 (2.1%)
Skilled nursing facility in last month	133 (2.3%)	5,911 (2.7%)
Comorbidities
Cancer	895 (15.5%)	25,393 (11.6%)
COPD	898 (15.5%)	29,177 (13.3%)
Diabetes	3,666 (63.5%)	141,026 (64.3%)
Hypertension	3,896 (67.5%)	139,110 (63.4%)
Ischemic heart disease	3,613 (62.6%)	122,555 (55.8%)
Liver disease	441 (7.6%)	10,511 (4.8%)
Frailty indicators
Ambulance / life support	1,328 (23.0%)	55,171 (25.1%)
Difficulty walking	714 (12.4%)	28,605 (13.0%)
Mobility aidsa	526 (9.1%)	22,011 (10.0%)
Skin ulcer (decubitus)	575 (10.0%)	19,792 (9.0%)
Use of oxygen	664 (11.5%)	22,370 (10.2%)
Weakness	632 (10.9%)	26,796 (12.2%)
Screening tests/Prevention
HbA1C test	3,155 (54.6%)	129,109 (58.8%)
Cancer screening	841 (14.6%)	30,543 (13.9%)
Diabetic eye exam	2,177 (37.7%)	71,798 (32.7%)
Hepatitis B vaccine/titre	1,288 (22.3%)	42,217 (19.2%)
Lipid test	1,991 (34.5%)	66,996 (30.5%)
Pneumococcal vaccine	321 (5.6%)	10,974 (5.0%)

Abbreviations: COPD, chronic obstructive pulmonary disease; HDV, high-dose vaccine; SD, standard deviation; SDV, standard-dose vaccine; HbA1c, hemoglobin A1c.

^aDefined as use of walker, wheelchair, or modified bathroom equipment.

The counts of outcome events, censoring events, competing risk events, and person-time at risk are presented for each outcome in Table 2. Among SDV recipients, the crude risks of death and hospitalization for pneumonia or influenza were similar (8.7% and 7.6%, respectively); in contrast, the risk of ILI was more than three times higher (28.1%). For each outcome, we observed similar weighted risks for HDV and SDV recipients throughout the influenza season, accounting for the competing risk of death for non-mortality outcomes (Fig 2). In the weighted analyses, we observed similar associations between vaccine dose and risk of mortality (risk difference, −0.08%; 95% CI, −0.85% to 0.80%), hospitalization due to influenza or pneumonia (risk difference, 0.15%; 95% CI, −0.69% to 0.93%), and ILI (risk difference, 0.00%; 95% CI, −1.50% to 1.08%). Results did not change appreciably in the analysis with the restricted ILI definition (Table S7). Estimates on the relative scale followed similar patterns (Table 2).

Table 2.

Estimates of the preventive effect of the high-dose vaccine (HDV) vs. standard-dose vaccine (SDV) on mortality, hospitalization due to influenza or pneumonia, and influenza-like illness measured at the end of each influenza season, accounting for the competing risk of death (for non-mortality outcomes), among adults ≥65 years.

Vaccine Dose Exposure	Outcome Events	Censoring Events	Competing Risk Events (Deaths)	Person-time, y	Crude cumulative Risk, % (95% CI)	Risk Difference, % (95% CI)		Risk Ratio (95% CI)
						Crude	Weighted	Crude	Weighted
MORTALITY
SDV	15,999	203,440	NA	85,068	8.7 (8.6, 8.9)	Reference	Reference	Reference	Reference
HDV	457	5,319	NA	2,270	9.3 (8.5, 10.2)	0.58 (−0.22, 1.33)	−0.08 (−0.85, 0.80)	1.07 (0.97, 1.15)	0.99 (0.91, 1.08)
HOSPITALIZATION DUE TO INFLUENZA/PNEUMONIA
SDV	14,441	192,011	12,987	82,544	7.6 (7.5, 7.8)	Reference	Reference	Reference	Reference
HDV	463	4,943	370	2,191	9.2 (8.4, 10.0)	1.51 (0.66, 2.28)	0.15 (−0.69, 0.93)	1.20 (1.09, 1.30)	1.02 (0.92, 1.10)
INFLUENZALIKE ILLNESS
SDV	54,711	157,639	7,089	74,565	28.1 (27.9, 28.3)	Reference	Reference	Reference	Reference
HDV	1,559	4,012	205	1,967	30.0 (28.9, 31.2)	1.90 (0.54, 3.02)	0.00 (−1.50, 1.08)	1.07 (1.02, 1.11)	1.00 (0.95, 1.04)

Abbreviations: CI, confidence interval; HDV, high-dose vaccine; NA, not applicable; SDV, standard-dose vaccine.

^aRisk differences (and risk ratios) were calculated as the difference (ratio) between the SDV and HDV cumulative risk functions.

Figure 2.

Cumulative risk and 95% confidence interval estimates of mortality, hospitalization for pneumonia or influenza, and influenza-like illness, accounting for the competing risk of death (for non-mortality outcomes), pooled across five influenza seasons (2010/11, 2011/12, 2012/13, 2013/14, 2014/2015). Cumulative risk was estimated using standardized mortality ratio-weighted Kaplan-Meier functions. The scales of the y-axes differ by outcome.

Within subgroups, the risks of mortality, hospitalization for pneumonia or influenza, and influenza-like illness were generally similar between HDV and SDV recipients throughout the influenza season, apart from a few notable exceptions (Figs 3–5, Tables S8–S10). First, we observed a higher risk of hospitalization among HDV versus SDV recipients in the 2010/11 season (risk difference, 2.85%; 95% CI, 0.59% to 5.86%), but not for any other seasons. Also, for both the hospitalization and ILI outcomes, we observed higher risk among HDV versus SDV recipients who were 65–74 years, lower risk among those who were 75–84 years, and no difference in risk between those <65 or ≥85 years.

Figure 3.

The risk of mortality among patients who received the high-dose versus standard-dose influenza vaccines, by subgroup. Risk differences were calculated as the difference between weighted cumulative risk functions, with bootstrap confidence intervals. Analyses were performed in adults ≥65 years, with the exception of the age stratified analyses. Abbreviations: CI, confidence interval; HDV, high-dose vaccine; SDV, standard-dose vaccine.

Figure 5.

The risk of influenza-like illness among patients who received the high-dose versus standard-dose influenza vaccines, by subgroup. Risk differences were calculated as the difference between weighted cumulative risk functions, accounting for the competing risk of death, with bootstrap confidence intervals. Analyses were performed in adults ≥65 years, with the exception of the age stratified analyses. Abbreviations: CI, confidence interval; HDV, high-dose vaccine; SDV, standard-dose vaccine.

In the bias analyses, we quantified the possibility that our observed null associations may be explained by an unmeasured confounding variable. We observed that the exposure–confounder relative risk and the confounder–outcome relative risk must have been at least as large as 1.4, 1.5, and 1.3 for mortality, hospitalization due to influenza or pneumonia, and ILI, respectively, to have shifted the corresponding weighted upper confidence limit across the null (Tables S11–13).

Discussion

We conducted a large comparative study of the effectiveness of the high-dose versus standard-dose influenza vaccines among adults with chronic kidney failure undergoing maintenance hemodialysis in the US. In the overall population, we observed similar risks of all-cause mortality, hospitalization due to influenza or pneumonia, and ILI among patients who received HDV compared with those who received SDV. Our findings were generally consistent across subgroups of influenza season, age group, years on dialysis, timing of influenza vaccination, and valence. We observed differences between HDV and SDV in a few subgroups, but the patterns were not consistent across influenza seasons or age groups. Associations between a potential unmeasured confounder and our three outcomes would have required associations that are stronger than our measured confounder-outcome associations to have masked preventive effects.

Consideration of the results of comparative effectiveness studies of preventive therapies requires careful thought about the tradeoffs of benefits versus costs. Due to the substantial influenza burden in the dialysis population, even small improvements in vaccine effectiveness could be sufficient for recommending a specific type of annual influenza vaccination in this population. However, increased risks of side effects and increased costs must also be considered when weighing the cost-benefit of implementing a new vaccination strategy. Our results suggest that the HDV does not provide additional benefit beyond the SDV on a population-level. Additionally, the excess cost of the HDV is about $33 per dose,⁴² incurring an additional cost burden of $16.9 million under the assumption that all 511,000 patients on dialysis would be vaccinated with HDV.⁴³ HDV is also more reactogenic, causing more frequent side effects, although most of these adverse events are minor (e.g., injection-site reactions, systemic adverse events, and gastrointestinal events).^{12,19,44–48} Safety data on the HDV is limited to the healthy adult population; a comparison of the safety of HDV versus SDV in the ESRD population is needed.

Previous observational studies have demonstrated that SDVs are ineffective or minimally effective at reducing all-cause mortality or influenza-related hospitalizations in the ESRD population,^29,49 despite reasonably robust antibody responses regarded as protective in 30% to 90% of vaccine recipients.^50–55 Our work extends these findings by clarifying that the HDV does not provide additional protection beyond the SDV against all-cause mortality or influenza-related outcomes, even though the HDV elicits greater antibody responses than SDV in patients with ESRD.⁵⁶ These observations highlight the importance of post-licensure clinical studies that extend beyond immunogenicity-related outcomes to incorporate clinical outcomes.

Our conclusions differ somewhat from an observational study using data from a not-for-profit dialysis provider in the US.²¹ Similar to our study, Miskulin et al. reported similar rates of all-cause mortality among HDV versus SDV recipients. In contrast with our study, Miskulin et al. reported lower rates of all-cause hospitalization among HDV versus SDV recipients, though their finding was restricted to the 2016/2017 influenza season and not the 2015/2016 season. The 95% confidence intervals for the preventive effect of HDV (hazard ratio, 0.93; 95% CI, 0.86–1.00) included the null value, whereas the effect estimate moved down and away from the null in the subgroup analysis restricted to patients aged ≥65 years (hazard ratio, 0.88; 95% CI, 0.79–0.97). Other stratified analyses suggested similar protection among HDV and SDV recipients. Our results are not directly comparable with the previous study given several differences. Specifically, Miskulin et al. included later seasons (2015/2016 and 2016/17), patients undergoing peritoneal dialysis, and a smaller sample size (N=19,157). Furthermore, the previous study did not incorporate influenza-related outcomes nor did it use competing risks methodology to account for the high mortality in the dialysis population.

Perhaps most importantly, the analysis by Miskulin et al. accounted for a limited set of potential confounders and therefore their effect estimates may be subject to residual confounding bias. In contrast, our study accounted for potential confounding by frailty^30,31 and the healthy-user bias^32,33 through adjustment for a comprehensive set of individual-level and facility-level potential confounders. It is well-established that observational studies of the effectiveness of preventive interventions such as influenza vaccination in elderly or sick populations can be highly subject to confounding by frailty and healthcare-seeking behavior, as those who receive the treatment generally have higher anticipation of benefit by providers (e.g., longer predicted life expectancies), or stronger personal motivation to receive care (e.g. better access to healthcare, higher socioeconomic status, less frailty).^49,57–60 Often, this confounding is difficult to control and results in artificially protective adjusted effect estimates due to residual confounding. We were concerned that this phenomenon might also be present when comparing HDV -- a newer, more expensive, specialized vaccine — to the standard vaccine. However, that is not what we observed. Instead, older individuals receiving HDV tended to be sicker, and the crude effect estimate was above the null, suggesting that the confounding was not due to the healthy-user phenomenon. This is supported by our previous work, where we demonstrated that comparisons of HDV to SDV among adult patients with ESRD are less subject to confounding by the healthy user bias than comparisons of vaccinated vs unvaccinated persons.⁶¹ However, despite our attempts, the possibility of unmeasured confounding remains.

Our findings should be interpreted within the context of several factors that affect vaccine effectiveness, including influenza virus circulation, serologic vaccine match, and seasonal influenza severity.⁶² Our study period included a range of seasonal influenza severity (one low severity, two moderate, two high seasons) as measured in the overall population of older adults in the US.⁶³ These severity metrics correlated with the observed seasonal risk of clinical outcomes in our analyses. Also, influenza A (H3N2) viruses – which are associated with lower vaccine effectiveness - were the predominant strains during all influenza seasons in our study, with the exception of 2013/14.^23–27 Compared to influenza A (H1N1) or influenza B viruses, H3N2 viruses have more frequently undergone antigenic change resulting in differences between the virus components of the influenza vaccine and circulating influenza viruses.⁶⁴ However, the serologic vaccine match – the proportion of circulating influenza strains that matched the vaccine - was relatively high during three of the four seasons with H3N2 as the predominant strain (i.e., 97%, 80%, 86%, and 40% for seasons 2010/11, 2011/12, 2012/13, and 2014/15, respectively),^23–27 indicating similarity between the influenza viruses in the vaccine and circulating in the community.

Our results are subject to several limitations. First, our observational study design did not involve randomization of the exposure and therefore our effect estimates are potentially subject to confounding by unobserved differences between exposure groups. However, we attempted to control confounding through restriction of the study population (i.e., by vaccination status and age) as well as analytic adjustment for a rich set of covariates. It is noteworthy that we did not account for individual dialysis facilities within our regression models; however, our analysis accounted for dialysis center-level characteristics – several of which were strong predictors of HDV receipt.¹¹ Second, our analysis only accounted for baseline characteristics measured prior to vaccination. However, this is supported by a previous study of influenza vaccine effectiveness among patients with ESRD that demonstrated that accounting for time-varying confounders of health status measured from both clinical and claims data did not reduce bias.⁴⁹ Third, our study design required survival until 9 months after dialysis initiation; therefore, our results may not generalize to all incident patients. Fourth, in the absence of laboratory-confirmed influenza outcomes, we used claims-based definitions. Thus, outcome misclassification is a possibility, though we used sensitivity analyses to explore the impact of the definitions on our findings. Fifth, some subgroup analyses may have had limited ability to detect a true difference between HDV and SDV given small numbers of HDV recipients. Sixth, the start date of follow-up was based on national influenza data, and therefore ignores geographic variation in the timing of influenza seasons. Lastly, our study is primarily based on administrative billing claims data which is collected for administrative and reimbursement purposes, rather than for clinical research. Thus, we were not able to account for certain important clinical factors unavailable in claims data (e.g., BMI, vascular access, laboratory measurements) that may result in confounding bias; and some of our measures may be subject to misclassification, such as vaccination ascertainment if patients paid out-of-pocket. However, we expect this misclassification to be unlikely, as patients receiving regular in-center hemodialysis have very frequent healthcare encounters in dialysis clinics, and influenza vaccination is offered to Medicare beneficiaries without a co-pay. Importantly, our use of this data is also a strength because it captures information on a wide variety of health-related diagnoses, procedures, use of durable medical equipment, frailty markers, hospitalizations, and death.

In summary, our large comparative study failed to demonstrate that HDV has superior effectiveness compared to SDV for preventing all-cause mortality and influenza-related outcomes among patients receiving maintenance hemodialysis. Given the findings of our population-level study, along with the substantially higher cost and side-effect profile of HDV compared to SDV, it appears that HDV should not conclusively be considered the standard of care at the present time for influenza immunization of patients treated by maintenance hemodialysis. The findings of our population-level study should not be interpreted to discourage influenza vaccination in the dialysis population. Rather, dialysis patients should continue to receive annual influenza immunization per CDC guidelines.⁸ In addition, future studies of alternative strategies (e.g., booster doses) and alternative vaccine production technologies (e.g., adjuvanted or cell-based vaccines) are warranted, as there remains a need for improved influenza prevention efforts in this population.

Figure 4.

The risk of hospitalization due to influenza / pneumonia among patients who received the high-dose versus standard-dose influenza vaccines, by subgroup. Risk differences were calculated as the difference between weighted cumulative risk functions, accounting for the competing risk of death, with bootstrap confidence intervals. Analyses were performed in adults ≥65 years, with the exception of the age stratified analyses. Abbreviations: CI, confidence interval; HDV, high-dose vaccine; SDV, standard-dose vaccine.