Relative Effectiveness of Cell-Cultured and Egg-Based Influenza Vaccines Among Elderly Persons in the United States, 2017–2018

Abstract

Background

The low influenza vaccine effectiveness (VE) observed during the A(H3N2)-dominated 2017–2018 season may be due to vaccine virus adaptation to growth in eggs. We compared the effectiveness of cell-cultured and egg-based vaccines among Medicare beneficiaries.

Methods

Retrospective cohort study on Medicare beneficiaries aged ≥65 years who received an influenza vaccine (cell-cultured, egg-based quadrivalent; egg-based high-dose, adjuvanted, or standard-dose trivalent) during the 2017–2018 season. We used Poisson regression to evaluate relative VE (RVE) in preventing influenza-related hospital encounters.

Results

Of >13 million beneficiaries, RVE for cell-cultured vaccines relative to egg-based quadrivalent vaccines was 10% (95% confidence interval [CI], 7%–13%). In a midseason interim analysis, this estimate was 16.5% (95% CI, 10.3%–22.2%). In a 5-way comparison, cell-cultured (RVE, 11%; 95% CI, 8%–14%) and egg-based high-dose (RVE, 9%; 95% CI, 7%–11%) vaccines were more effective than egg-based quadrivalent vaccines.

Conclusions

The modest VE difference between cell-cultured and egg-based vaccines only partially explains the low overall VE reported by the Centers for Disease Control and Prevention, suggesting that egg adaptation was not the main contributor to the low VE found among individuals aged ≥65 years. The midseason interim analysis we performed demonstrates that our methods can be used to evaluate VE actively during the influenza season.

influenza vaccine, vaccine effectiveness, relative vaccine effectiveness, cell-cultured vaccine

(See the Editorial commentary by Flannery and Fry, on pages 1237–9.)

In the United States, approximately 140 000–710 000 influenza-related hospitalizations and 12 000–56 000 influenza-associated deaths occur annually, with most of the burden occurring among individuals aged ≥65 years [1–3]. Seasonal strain-specific vaccination is the main tool for influenza prevention [4]. A preliminary estimate of the effectiveness of influenza vaccines during the A(H3N2)-dominated 2017–2018 season by the US Influenza Vaccine Effectiveness Network, using a test-negative design, found 20% vaccine effectiveness (VE) (95% confidence interval [CI] −9% to 41%) against any influenza strain and 17% VE (−22% to 44%) against influenza A(H3N2) viruses among persons aged ≥65 years [5]. A prior interim analysis of the US data, and studies of the 2017 season in Australia and the 2017–2018 season in Canada had also found low VE against A(H3N2) strains [6–8].

Although there were differences in the dominant A(H3N2) influenza virus clades that circulated in these countries [7], the low VE found caused concern. One hypothesis for this low VE is the adaptation of influenza virus to growth in eggs [7]. Two vaccines not manufactured in eggs are licensed and distributed in the United States: a recombinant protein vaccine consisting of virus hemagglutinin (HA) produced in insect cells, not included in our study owing to the limited number of users, and a subunit vaccine prepared from influenza viruses propagated in mammalian cells (ie, cell cultured) [9, 10]. The A(H3N2) virus used for the manufacture of the cell-cultured vaccine for the 2017–2018 season did not undergo egg adaptation before culture in cells, whereas the 3 other components, the A(H1N1) virus and both influenza B vaccine seeds, originated from egg isolates [11]. All other influenza vaccines are produced in eggs, although some differences are potentially relevant to the evaluation of VE: one is high-dose (containing 60 μg of each HA antigen per dose) [12] and others are standard dose (15 μg per dose).

Among standard-dose vaccines, one is adjuvanted [13], some are trivalent, containing antigens from subtype A(H1N1) and A(H3N2) strains and a single type B (B/Victoria) lineage strain, and others are quadrivalent, containing antigens from both A subtypes and from 2 type B lineage strains. The relative VE (RVE) for the ensemble of these vaccines has never been evaluated. We used real-world data from Medicare claims to conduct both an interim evaluation, with data available as of 19 January 2018, and an end-of-season analysis, with updated data available as of 4 August 2018, of the RVE of cell-cultured versus egg-based influenza vaccines administered to US beneficiaries aged ≥65 years during the 2017–2018 season.

METHODS

Data Sources

We used Medicare administrative files providing data on enrollment, inpatient and outpatient care, physician office visits, and prescription drugs as the primary data source. We used respiratory sample data from the National Respiratory and Enteric Virus Surveillance System to define periods of high influenza circulation within seasons [14, 15], the National Plan and Provider Enumeration System and the National Council for Prescription Drug Programs databases to identify pharmacies, and the Minimum Data Set (www.cms.gov/Research-Statistics-Data-and-Systems/Computer-Data-and-Systems/Minimum-Data-Set-3-0-Public-Reports/index.html#; accessed 22 February 2018) to identify nursing facilities.

Observation Period

The study included the period from 6 August 2017 to 4 August 2018. The study period for the interim analysis extended through 19 January 2018, using data available as of that date. The study period for the end of season analysis extended through 4 August 2018, using data available as of that date.

Study Population

We included Medicare beneficiaries aged ≥65 years who received an egg-based or cell-cultured influenza vaccination from 6 August 2017 through 31 January 2018, with data available as of 4 August 2018. For the interim analysis, we included only beneficiaries vaccinated before 4 January 2018, with data available as of 19 January 2018. We required continuous enrollment in fee-for-service Medicare parts A and B in the 6 months before vaccination to allow for identification of chronic medical conditions. Because the case definition we used for the office visits outcome required a prescription for an influenza-specific antiviral (oseltamivir), we needed access to drug prescription information. Thus, for this outcome, we also required that the beneficiary was a participant in Medicare part D throughout the high influenza season. We excluded beneficiaries enrolled in Medicare part C and those residing in nursing homes on the vaccination date because their medical encounters may not be reliably captured. We also excluded beneficiaries who received an earlier influenza vaccination during the same season and those whose region of residence was not defined in the Department of Health and Human Services (DHHS) regions [14, 15].

Influenza Vaccine Exposure

We classified eligible beneficiaries into 5 cohorts based on the type of vaccine received. Vaccinations were identified using Healthcare Common Procedure Coding System and Current Procedural Terminology (CPT) codes for the cell-cultured quadrivalent vaccine (CPT 90674 and 90756), as well as the 4 egg-based vaccine types: quadrivalent (CPT 90630 and 90685–90688), high-dose trivalent (CPT 90662), adjuvanted trivalent (CPT 90653), and standard-dose trivalent (CPT 90656–90658; Healthcare Common Procedure Coding System Q2034-Q2038). Other influenza vaccines were not used in sufficient numbers for study inclusion. An unvaccinated cohort was not included because we were unable to reliably identify unvaccinated beneficiaries from Medicare claims alone, because they may have been vaccinated outside the system.

Study Outcomes

The primary outcome was influenza-related hospital encounters, defined as inpatient hospitalizations/emergency department visits listing an International Classification of Diseases, Tenth Revision, Clinical Modification, code for influenza (codes J09.xx, J10.xx, J11.xx, and J129). We also performed a post hoc analysis using only inpatient stays as outcomes, and a prespecified secondary analysis of influenza-related office visits, defined as community-based physician office visits or hospital outpatient visits with a rapid influenza diagnostic test performed (CPT 87804) followed by a therapeutic course of oseltamivir (75 mg twice daily for 5 days) prescribed within 2 days after the test [14, 15].

Person-Time Under Observation

Follow-up time began 14 days after vaccination (to allow for development of vaccine-specific immunity) and continued until one of the following occurred: outcome of interest, Medicare disenrollment, end of study period, death, admission into a nursing home, or administration of a subsequent influenza vaccine. To increase the positive predictive value of outcomes, we focused the analyses only on person-time observed in regions and weeks with high levels of influenza circulation [14–16]. Each week of the influenza season within each DHHS region was classified as a high influenza circulation period if ≥15% of respiratory samples tested positive for influenza [14–16].

Variables of Interest

For each beneficiary, we collected data on demographics, Medicaid eligibility, reason for entry into Medicare, DHHS region of residence, prior medical encounters, chronic medical conditions, and frailty indicators in the 6 months before vaccination. We adjusted for all study covariates as potential confounders. We assessed the covariate balance between cohorts using standardized mean differences (SMDs); SMDs <0.1 represented negligible imbalance [17].

Statistical Analyses

Our main analysis focused on the comparison between the cell-cultured and egg-based standard-dose quadrivalent vaccines to ensure comparability. For the adjusted analysis, we used inverse probability of treatment weighting (IPTW) to adjust for imbalances between cohorts on all covariates [18]. Stabilized IPTW weights were derived from propensity scores, representing each beneficiary’s probability of receiving the cell-cultured rather than egg-based vaccines, conditional on covariates [19]. We calculated propensity scores using a logistic regression model with vaccine cohort as the dependent variable and all covariates as independent variables. Weight values >5 were truncated to 5 to limit the bias that outlier beneficiaries could have on results [20]. We ran a univariate Poisson regression model on the weighted population to estimate the adjusted rate ratio between the 2 cohorts. The adjusted RVE was defined as (1 − rate ratio) × 100%. As sensitivity analyses, we ran a weighted univariate Poisson model for which weights were not truncated and a doubly-robust model that implemented covariate adjustment along with IPTW [21].

To provide further context, we conducted an analysis that also included the egg-based high-dose, adjuvanted, and standard-dose trivalent vaccines, using IPTW to adjust for covariate imbalances. We calculated 5 propensity scores for each participant using separate logistic regression models, with each score representing the probability of receiving a particular vaccine type relative to any other [22]. The outcome models and sensitivity analyses conducted were similar to those we used for the 2-way vaccine comparison.

We conducted the analyses using R 3.4.3 (R Foundation for Statistical Computing) and SAS 9.4 (SAS Institute). We performed all analyses with deidentified data collected for administrative purposes by the Centers for Medicare & Medicaid Services (CMS). This study was approved by the Food and Drug Administration’s Research Involving Human Subjects Committee. Medicare administrative data were used under a data use agreement with CMS, which was approved by the CMS privacy board.

RESULTS

More than 16 million fee-for-service Medicare beneficiaries received a cell-cultured or egg-based influenza vaccination in the 2017–2018 season through 31 January 2018. Approximately 13 million remained eligible after implementation of the study restrictions. Among all eligible vaccinees, 5% received the cell-cultured quadrivalent, 14% the egg-based quadrivalent, and 63%, 11%, and 7% received high-dose, adjuvanted, and standard-dose trivalent vaccines, respectively (Supplementary Figure 1). Before weighting, the cell-cultured and egg-based quadrivalent cohorts were imbalanced as to pharmacy vaccination status, region, and prior outpatient non–emergency room visits (Table 1). After weighting, the 2 cohorts were balanced, with SMDs ≤0.01 for all covariates (Table 1 and Supplementary Table 1). For the 5-way comparison analysis, all cohorts were balanced after weighting, with SMDs ≤0.06 (Table 1 and Supplementary Table 1). Total person-time, outcome rates, and unadjusted RVEs for all vaccine cohorts are provided in Table 2.

Table 1.

Distribution of Covariates Across Vaccine Cohorts for the 2017–2018 Season Before Implementation of IPTW

	Distribution of Covariate, %	Maximum SMDa
	Egg-Based Quadrivalent (n = 1 863 654)	Cell-Cultured Quadrivalent (n = 659 249)	Egg-Based HD Trivalent (n = 8 489 159)	Egg-Based Adjuvanted Trivalent (n = 1 473 536)	Egg-Based SD Trivalent (n = 1 018 494)	Before IPTW	After IPTWb
Vaccinated at pharmacy	9.2	19.3	44.3	67.4	11.6	1.50	0.03
Age bracket, y
65–74	52.0	49.6	51.1	50.6	48.4	0.07	0.02
75–84	33.0	34.3	34.6	34.1	34.4	0.03	0.02
≥85	15.0	16.1	14.4	15.2	17.2	0.08	0.02
Sex
Female	58.6	59.0	57.9	58.3	59.2	0.03	0.03
Male	41.4	41.0	42.1	41.7	40.8	0.03	0.03
Race
White	85.8	84.3	89.4	90.2	81.7	0.25	0.02
Black	6.2	7.3	4.6	4.3	7.7	0.14	0.01
Asian	1.2	1.5	0.8	0.7	2.3	0.13	0.02
Hispanic	2.4	3.2	1.7	1.4	3.7	0.15	0.01
Other	4.4	3.7	3.5	3.4	4.6	0.06	0.02
Regionc
Region 1	6.9	4.2	6.3	6.6	5.6	0.12	0.02
Region 2	9.2	10.8	8.3	9.4	13.4	0.17	0.02
Region 3	10.9	9.7	12.3	11.3	10.6	0.08	0.02
Region 4	20.6	30.3	19.2	30.7	20.9	0.27	0.02
Region 5	18.2	11.9	18.1	12.6	14.8	0.18	0.03
Region 6	10.2	13.5	10.6	8.1	14.3	0.20	0.01
Region 7	5.4	2.4	6.5	4.2	2.8	0.20	0.02
Region 8	3.4	1.2	4.0	2.4	2.0	0.18	0.01
Region 9	10.7	12.6	10.8	9.2	13.1	0.12	0.01
Region 10	4.5	3.3	4.1	5.4	2.5	0.15	0.02
Reason for entering Medicare
Aging	89.2	88.8	91.8	92.2	88.5	0.12	0.02
Disability	10.6	11.1	8.0	7.8	11.3	0.12	0.02
End-stage renal disease	0.2	0.1	0.2	0.1	0.2	0.04	0.02
Dual eligible	11.4	13.4	6.9	6.8	16.5	0.31	0.06
Month of vaccination
August or September	26.0	27.4	33.5	30.9	22.5	0.25	0.03
October	44.2	41.2	43.0	44.0	47.1	0.12	0.02
November	18.3	17.7	15.4	14.9	19.2	0.11	0.01
December or January	11.5	13.7	8.1	10.2	11.3	0.18	0.01
≥1 Total hospitalizations	11.7	10.1	8.6	7.7	11.1	0.13	0.02
≥1 Outpatient ER visits	16.9	15.7	14.2	13.6	16.4	0.09	0.02
Outpatient non-ER visits
0	32.2	43.5	36.9	40.4	37.5	0.23	0.04
1 or 2	30.8	30.0	30.9	31.0	29.9	0.02	0.01
≥3	37.0	26.5	32.2	28.7	32.6	0.23	0.05
All physician visits
0–5	34.0	30.4	34.4	34.6	31.1	0.09	0.01
6–10	28.1	28.5	28.7	28.8	28.1	0.02	0.00
11–15	15.7	16.8	15.7	15.8	16.5	0.03	0.00
16–25	13.8	15.0	13.5	13.4	14.8	0.04	0.01
≥26	8.5	9.3	7.8	7.5	9.5	0.07	0.01
Respiratory failure and pneumonia
Hospitalizations	2.6	2.1	1.6	1.3	2.4	0.09	0.01
ER outpatient visits	0.5	0.4	0.3	0.3	0.4	0.03	0.01
Non-ER outpatient visits	1.2	0.8	0.8	0.7	1.0	0.05	0.01
Physician office visits	4.9	4.4	3.6	3.3	4.8	0.08	0.01
Allergies
Anaphylaxis	0.1	0.1	0.1	0.1	0.1	0.01	0.00
Drug allergy	0.3	0.3	0.3	0.3	0.3	0.01	0.00
Food allergy	0.4	0.4	0.3	0.4	0.4	0.01	0.00
Concomitant pneumococcal vaccine	8.0	6.0	9.0	8.0	5.9	0.12	0.04
Health conditions
Asthma	6.7	6.8	6.1	5.7	6.9	0.05	0.01
Blood disorders	22.8	25.4	19.8	19.0	26.5	0.18	0.02
Chronic lung disease	20.4	21.4	17.4	16.6	21.8	0.13	0.01
Diabetes	30.3	32.9	26.4	24.3	34.6	0.23	0.02
Heart disease	42.9	45.3	39.9	38.8	45.7	0.14	0.01
Kidney disorders	16.5	16.8	14.2	12.4	17.0	0.13	0.02
Liver disorders	3.6	3.7	3.2	2.9	3.8	0.05	0.01
Neurological or neurodevelopmental conditions	17.2	19.0	14.9	15.2	19.4	0.12	0.02
Immunocompromising conditions	5.6	5.4	4.9	4.6	5.4	0.05	0.01
Other malignant neoplasms (not included elsewhere)	11.8	11.7	11.6	11.3	11.7	0.02	0.01
Frailty covariate
Home oxygen use	4.5	4.4	3.5	3.0	4.5	0.08	0.01
Wheelchair use	0.1	0.1	0.1	0.1	0.1	0.01	0.00
Walker use	1.4	1.4	1.2	1.1	1.5	0.04	0.01
Dementia	0.8	0.9	0.6	0.6	0.9	0.04	0.01
Urinary catheter use	5.6	6.6	4.2	4.8	6.7	0.11	0.03
Falls	6.6	6.3	5.6	5.4	6.6	0.05	0.01
Fractures	1.4	1.3	1.1	1.1	1.4	0.02	0.00

Abbreviations: ER, emergency room; HD, high-dose; IPTW, inverse probability of treatment weighting; SMD, standardized mean difference; SD, standard-dose.

^aSMD values ≥0.1 indicate that the 2 groups are imbalanced as to the specified covariates.

^bThe covariate frequencies across vaccine cohorts after IPTW is presented in Supplementary Table 1.

^cRegion 1 includes Connecticut, Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont; region 2, New Jersey, New York, Puerto Rico, US Virgin Islands; region 3, Delaware, District of Columbia, Maryland, Pennsylvania, Virginia, and West Virginia; region 4, Alabama, Florida, Georgia, Kentucky, Mississippi, North Carolina, South Carolina, Tennessee; region 5, Illinois, Indiana, Michigan, Minnesota, Ohio, Wisconsin; region 6, Arkansas, Louisiana, New Mexico, Oklahoma, Texas; region 7, Iowa, Kansas, Missouri, Nebraska; region 8, Colorado, Montana, North Dakota, South Dakota, Utah, Wyoming; region 9, Arizona, California, Hawaii, Nevada, Guam, American Samoa, Commonwealth of the Northern Mariana Islands, Federated States of Micronesia, Republic of the Marshall Islands, Republic of Palau; region 10, Alaska, Idaho, Oregon, Washington.

Table 2.

Unadjusted Outcome Rates and RVEs Using the Egg-Based Quadrivalent Cohort as Reference for 2017–2018 Season

Outcome by Cohort	Outcomes, No.	Total Person-Time, ×10 000 Person-Weeks	Outcome Rate	RVE (95% CI), %
Influenza-related office visits
Egg-based quadrivalent	9607	1829	5.25	Reference
Cell-cultured quadrivalent	3299	599	5.51	−4.8 (−9.1 to −.8)
Egg-based HD trivalent	45 941	8782	5.23	0.4 (−1.8 to 2.6)
Egg-based adjuvanted trivalent	8202	1439	5.70	−8.6 (−11.8 to −5.4)
Egg-based SD trivalent	4868	982	4.96	5.6 (2.3–8.8)
Influenza-related hospital encounters
Egg-based quadrivalent	14 417	2472	5.83	Reference
Cell-cultured quadrivalent	4370	808	5.41	7.3 (4.1–10.4)
Egg-based HD trivalent	56 763	11 588	4.90	16.0 (14.5–17.5)
Egg-based adjuvanted trivalent	9393	1896	4.96	15.0 (12.8–17.2)
Egg-based SD trivalent	8239	1303	6.32	−8.4 (−11.4 to −5.5)

Abbreviation: CI, confidence interval; HD, high-dose; RVE, relative vaccine effectiveness; SD, standard-dose. RVEs for the egg-based quadrivalent cohort were not presented because the egg-based quadrivalent cohort was used as the reference for the RVE estimates calculation.

The IPTW-adjusted results indicate that the cell-cultured vaccine was significantly more effective than the egg-based quadrivalent vaccine in preventing influenza-related hospital encounters (RVE, 10.0%; 95% CI, 7.0% to 13.0%) (Figure 1). These estimates were consistent for the secondary outcomes and all sensitivity analyses (Supplementary Figure 2). The 19 January interim analyses results also showed that the cell-cultured vaccine was significantly more effective in preventing influenza-related hospital encounters than the egg-based quadrivalent vaccine (RVE, 16.5%; 95% CI, 10.3% to 22.2%). However, the interim analysis point estimates for all outcomes were higher than those from the end-of-season analysis (Figure 1). Results from the IPTW-adjusted 5-way comparisons indicate that the cell-cultured and high-dose vaccines were both significantly more effective in preventing influenza-related hospital encounters than the egg-based quadrivalent and trivalent vaccines (Tables 3 and 4 and Figure 1). The adjuvanted vaccine was also slightly more effective than the egg-based quadrivalent and trivalent vaccines in preventing influenza-related hospital encounters but was less effective than both the cell-cultured and high-dose vaccines (Tables 3 and 4 and Figure 1). The RVE estimates for the post hoc inpatient stays outcome were consistent with those of the influenza-related hospital encounter outcome, although the adjuvanted vaccine was no longer significantly more effective than the egg-based quadrivalent vaccine (Tables 3 and 4 and Figure 1).

Table 3.

IPTW-Adjusted Pairwise RVE Estimates for Influenza-Related Hospital Encounters in the 2017–2018 Season

Cohort	RVE by Reference Group (95% CI), %
	Egg-Based Quadrivalent	Egg-Based SD Trivalent	Egg-Based Adjuvanted Trivalent	Egg-Based HD Trivalent
Cell-cultured quadrivalent	11.0 (7.9–14.0)a	10.8 (7.4–14.1)a	7.5 (4.1–10.7)a	2.3 (−.8, to 5.3)
Egg-based HD trivalent	9.0 (7.2–10.6)a	8.7 (6.5–10.9)a	5.3 (3.3–7.3)a	…
Egg-based adjuvanted trivalent	3.9 (1.4–6.3)a	3.6 (.7–6.4)a	…	…
Egg-based SD trivalent	0.3 (−2.6 to 3.1)	…	…	…

Abbreviations: CI, confidence interval; HD, high-dose; IPTW, inverse probability of treatment weighting; RVE, relative vaccine effectiveness; SD, standard-dose.

^aPairwise comparison RVE estimates that are significant at the P ≤.05 level.

Table 4.

IPTW-Adjusted Pairwise RVE Estimates for Influenza-Related Office Visits and Inpatient Stays in the 2017–2018 Season

Outcome by Cohort	RVE by Reference Group (95% CI), %
	Egg-Based Quadrivalent	Egg-Based SD Trivalent	Egg-Based Adjuvanted Trivalent	Egg-Based HD Trivalent
Influenza-related office visits
Cell-cultured quadrivalent	5.7 (1.9–9.4)a	1.0 (−3.5 to 5.3)	11.5 (7.9–15.0)a	5.1 (1.6–8.4)a
Egg-based HD trivalent	0.7 (−1.5 to 2.9)	−4.3 (−7.4 to −1.3)a	6.8 (4.6–8.9)a	…
Egg-based adjuvanted trivalent	−6.6 (−9.7 to −3.5)a	−11.9 (−15.9 to −8.1)a	…	…
Egg-based SD trivalent	4.8 (1.5–8.0)a	…	…	…
Influenza-related inpatient stays
Cell-cultured quadrivalent	9.5 (5.3–13.4)a	11.4 (7.0–15.7)a	7.1 (2.7–11.3)a	−0.7 (−4.8 to 3.4)
Egg-based HD trivalent	10.0 (7.8–12.3)a	12.0 (9.2–14.8)a	7.7 (5.1–10.2)a	…
Egg-based adjuvanted trivalent	2.5 (−.8 to 5.8)	4.7 (.9–8.3)a	…	…
Egg-based SD trivalent	−2.2 (−6.1 to 1.5)	…	…	…

Abbreviations: CI, confidence interval; HD, high-dose; IPTW, inverse probability of treatment weighting; RVE, relative vaccine effectiveness; SD, standard-dose.

^aPairwise comparison RVE estimates that are significant at the P ≤.05 level.

Figure 1.

Inverse probability of treatment weighting–adjusted relative vaccine effectiveness (RVE) estimates for primary outcome analyses and 2-vaccine secondary outcome analysis. The RVE point estimates and 95% confidence intervals for the cell-cultured versus egg-based quadrivalent vaccine comparison are presented for both outcomes, as well as for the 5-way influenza-related hospital encounter comparison analysis for the 2017–2018 season. The egg-based quadrivalent vaccine cohort is used as the reference group for each RVE estimate presented.

In the 2-way influenza-related office visits outcome comparison, the cell-cultured vaccine was again significantly more effective than the egg-based quadrivalent vaccine (RVE, 10.5%; 95% CI, 6.8%–14.0%) (Figure 1). The 5-way analysis showed similar differences between the cell-cultured, high-dose, and adjuvanted vaccines for all outcomes (Table 4 and Supplementary Figure 3). However, office visit estimates for the comparisons with the standard-dose trivalent and quadrivalent vaccines were not consistent with findings from the hospital encounters and inpatient stays analyses (Table 4).

DISCUSSION

This analysis of the A(H3N2)-dominated 2017–2018 influenza season, which we conducted among approximately 13 million influenza vaccinees aged ≥65 years, showed that the effectiveness of the cell-cultured quadrivalent influenza vaccine was approximately 10%–11% higher than that of the comparable egg-based standard-dose quadrivalent products in preventing influenza-related hospital encounters, inpatient stays, and office visits (Figure 1). Among all 5 vaccine types investigated, RVE against influenza-related hospital encounters and inpatient stays was highest for the cell-cultured and high-dose vaccines (Tables 3 and 4). The higher effectiveness we observed for the egg-based high-dose relative to the egg-based standard-dose vaccine is consistent with findings from prior studies by our team and others [14, 15, 23].

To our knowledge, there is no prior real-world evidence of higher RVE for the cell-cultured vaccine versus the comparable egg-based vaccines, although there is virological evidence supporting this possibility [4, 6–8]. One hypothesis is that changes in the viral HA of influenza A(H3N2) virus during isolation, adaptation, and propagation in eggs of the influenza A(H3N2) vaccine strain might affect VE [4, 7, 24, 25]. However, our study does not rule out the possibility that other factors unrelated to the method of vaccine production could account for the observed differences in VE. Moreover, we did not separately analyze egg-based quadrivalent vaccines produced by different manufacturers.

The differential effectiveness we observed between the cell-cultured and egg-based quadrivalent vaccines only partially explains the 20% VE interim estimate reported by the Centers for Disease Control and Prevention (CDC) among individuals aged ≥65 years during the 2017–2018 season, low by historical standards [8]. Unlike in 2014–2015, when there was evidence of antigenic drift of the circulating A(H3N2) viruses compared with the vaccine reference viruses, there was no evidence of such HA drift in 2017–2018 [4, 26]. Other possibilities, including a neuraminidase drift, should be explored to help explain the low VE reported among persons aged ≥65 years [8, 27]. We also provide hypothetical VE estimates for each vaccine among beneficiaries aged ≥65 years, assuming a ±10 point difference from the CDC VE estimate (Supplementary Table 2) [8]. This method is described in the Supplementary Methods. The numbers can be considered only a very rough approximation, because our study did not estimate absolute VEs, the CDC study population could have differed from ours, the absolute VE estimate from the CDC study had wide CIs (−9% to 41%), and their estimates were preliminary.

Our interim 19 January 2018 analysis shows the potential value of CMS data to provide early estimates of RVE to help with strain selection for the following season. The 19 January interim analysis was our first attempt to conduct an RVE study during the high influenza circulation period, and the conclusions from that interim analysis were consistent with those of the end-of-season analysis. Our end-of-season RVE estimate for the cell-cultured versus the comparable egg-based standard-dose quadrivalent vaccines (RVE, 10.0%; 95% CI, 7.0%–13.0%) was somewhat lower than that of our 19 January interim analysis (16.5%; 10.3%–22.2%) (Figure 1).

This finding might be explained by an increase in the proportion of influenza B circulating viruses during the late part of the season [28–30]. Unlike the A(H3N2) component of the 2017–2018 cell-cultured vaccine, which was isolated in cells, the influenza B viruses in the cell-cultured vaccine were isolated in eggs and, therefore, are antigenically similar to the comparator influenza B viruses in egg-grown vaccines. This could explain why the RVE was lower in the end-of-season analysis, which included a late increase in the circulation of B/Yamagata-lineage viruses [4, 25, 28–30]. It is also possible that differences in the magnitude of waning effectiveness across vaccine types may have contributed to the differences in results between the interim and end-of-season analyses [31]. Nonetheless, we found no major differences by month of vaccination among the vaccines used in the study after weighting was implemented (Supplementary Table 1).

One limitation of the use of Medicare data in real-world evidence studies is the lack of access to virological case confirmation data. However, because the 2017–2018 season was dominated by influenza A(H3N2), we can infer that most influenza events were probably caused by influenza A(H3N2). Our definition of influenza-related office visits included a rapid test followed by a prescription for a therapeutic course of oseltamivir. Although the sensitivity and specificity of rapid influenza tests are lower than those from culture or polymerase chain reaction, rapid influenza tests can provide results to a healthcare provider within 15 minutes after sample collection [32].

We expect that these results guided the decision to prescribe influenza-specific antivirals [33, 34]. However, our influenza-related office visits 5-way comparison produced estimates inconsistent with the hospital encounters and inpatient hospitalizations outcomes, specifically in the comparisons with the standard-dose quadrivalent and trivalent vaccines (Table 4 and Supplementary Figure 3). These inconsistencies might be explained by potential vaccine cohort differences in health seeking behavior and in testing and treatment preferences by physicians, and they highlight the need for caution when considering results from observational studies that include outcomes particularly sensitive to differences in health-seeking and treatment preference behaviors not characterized in claims databases [35]. To resolve these issues, we have initiated projects linking surveys of Medicare beneficiaries, which provide information on frailty and on health-seeking and other behaviors, with Medicare claims databases [36]. Moreover, we a priori considered hospital encounters as our primary analysis outcome, and we included a post hoc inpatient stay–only analysis, which produced similar RVE estimates.

A strength of our study is that the population includes all eligible beneficiaries identified through Medicare fee-for-service claims. Our large sample size allowed us to perform analyses for serious outcomes including inpatient stays, difficult to obtain in any randomized trial. However, our study has limitations. Because the analysis is observational, residual confounding could be a concern. In particular, residual confounding associated with chronic conditions not sufficiently characterized by the dichotomous indicators derived from Medicare claims can be problematic [37, 38]. However, our use of IPTW yielded well-balanced vaccine cohorts with respect to potential confounders available in the Medicare database, and our past estimates using this database have been credible [14, 15, 36].

Our consistent finding that the cell-cultured and high-dose vaccines were more effective than the comparator egg-based vaccines among persons aged ≥65 years during the 2017–2018 season, along with our findings from prior studies [14, 15, 23, 39], show that the methods we developed for using real-world data to analyze the relative effectiveness of vaccines in the Medicare population could help optimize vaccine strategies for this vulnerable population during epidemics and pandemics. However, given the potential for residual confounding in all observational studies, and the small magnitude of the differences we obtained, our findings would benefit from replication in studies in other settings and healthcare systems. We plan to conduct a rapid-response VE comparison analysis for each influenza season going forward. The critical analysis of the ensemble of real-world evidence studies performed each season should provide valuable information for epidemic influenza control and pandemic preparedness and may help guide the future use of real-world evidence by the Food and Drug Administration and others [40–42].

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. Thanks to Maryna Eichelberger, Center for Biologics Evaluation and Research (CBER), Food and Drug Administration (FDA), for advice on egg adaptation of influenza viruses and critical review of the manuscript; Zhiping Ye, CBER/FDA, for advice on egg adaptation of influenza viruses; and Alicia Fry, Influenza Division, Centers for Disease Control and Prevention, and Steven Anderson, Philip Krause, and Peter Marks, CBER/FDA, for critical review of the manuscript.

Financial support. This work was supported by the Food and Drug Administration (FDA) as part of the SafeRx Project, a joint initiative of the Centers for Medicare & Medicaid Services and the FDA.

Disclaimer. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the FDA or the Centers for Medicare & Medicaid Services.

Potential conflicts of interest. All authors: No reported conflicts. 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.

Presented in part: Advisory Committee on Immunization Practices meeting, Atlanta, Georgia, 20–21 June 2018.