Abstract
Background
In the United States, it is recommended that healthcare providers offer influenza vaccination by October, if possible. However, if the vaccine’s effectiveness soon begins to wane, the optimal time for vaccination may be somewhat later. We examined whether the effectiveness of influenza vaccine wanes during the influenza season with increasing time since vaccination.
Methods
We identified persons who were vaccinated with inactivated influenza vaccine from 1 September 2010 to 31 March 2017 and who were subsequently tested for influenza and respiratory syncytial virus (RSV) by a polymerase chain reaction test. Test-confirmed influenza was the primary outcome and days-since-vaccination was the predictor of interest in conditional logistic regression. Models were adjusted for age and conditioned on calendar day and geographic area. RSV was used as a negative-control outcome.
Results
Compared with persons vaccinated 14 to 41 days prior to being tested, persons vaccinated 42 to 69 days prior to being tested had 1.32 (95% confidence interval [CI], 1.11 to 1.55) times the odds of testing positive for any influenza. The odds ratio (OR) increased linearly by approximately 16% for each additional 28 days since vaccination. The OR was 2.06 (95% CI, 1.69 to 2.51) for persons vaccinated 154 or more days prior to being tested. No evidence of waning was found for RSV.
Conclusions
Our results suggest that effectiveness of inactivated influenza vaccine wanes during the course of a single season. These results may lead to reconsideration of the optimal timing of seasonal influenza vaccination.
Keywords: influenza, vaccine effectiveness, waning, negative-control outcome
Vaccine effectiveness appears to decrease with time since vaccination. The odds of influenza in remote vaccinees vs recent vaccinees increased by about 16% per 28 days. Influenza vaccination may be more effective if received closer to the start of the influenza season.
(See the Editorial Commentary by Lipsitch on pages 1631–3.)
The Advisory Committee on Immunization Practices recommends annual influenza vaccination for all persons aged ≥6 months who do not have contraindications and recommends that healthcare providers in the United States offer vaccination by October, if possible [1]. The public health goal is to vaccinate everyone before influenza viruses begin to circulate widely, which in North American is late fall through early spring [2]. This recommendation is made whether or not the vaccine influenza strains are the same as in the prior year (as in the 2011–2012 season) [3–5] and is based on the assumption that vaccine effectiveness (VE) wanes from one year to the next [5]. This raises the question of whether the vaccine’s effectiveness also wanes during the season in which it is used.
A number of recent studies [6–13] have provided evidence consistent with intraseason waning of influenza vaccines [8, 9, 12]. All but 1 [11] of these studies used some form of a “case-positive, control test-negative” approach in which persons who tested positive for influenza were compared to persons who tested negative. The test-negative approach can reduce certain kinds of biases [14–19], although it also has some potential disadvantages [20]. None of the studies of intraseason waning used a negative control to detect possible residual confounding. A negative-control outcome is one that is expected to be subject to the same potential sources of bias as the outcome of interest but cannot plausibly be related to the exposure of interest [21, 22]. If a relationship between the exposure of interest and the negative-control outcome is detected, this may indicate residual confounding.
Our goals in this study were to use data from 7 influenza seasons to determine whether there is intraseason waning of inactivated influenza vaccine (IIV) effectiveness and to assess potential residual confounding using respiratory syncytial virus (RSV) as a negative-control outcome. Using a cohort of vaccinated persons who were subsequently tested for influenza virus, we determined if the risk of testing positive for influenza, compared to testing negative, increased with time since vaccination.
METHODS
Setting and Data Sources
Kaiser Permanente Northern California (KPNC) is a nonprofit, integrated healthcare delivery system that provides health services to 4 million members. KPNC’s electronic health record captures diagnostic and demographic information on KPNC members, as well as influenza immunizations and viral tests. Immunizations are provided at no additional cost to members, and most are received within the system [23]. KPNC performs real-time reverse-transcriptase polymerase chain reaction (RT-PCR) tests for the simultaneous detection of influenza A and influenza B viruses and RSV. KPNC does not perform influenza A and B subtyping. The KPNC Institutional Review Board approved this study.
Study Population
We defined the “influenza season” as running from 1 September to 31 March. We identified all vaccinations with IIV given in 7 seasons from 2010–2011 to 2016–2017 and extracted all viral RT-PCR tests ordered as part of routine clinical care for these vaccinated persons. We retained only vaccinated persons who, in the same season as their vaccination, had an RT-PCR test subsequent to vaccination. To exclude repeated tests for the same episode of illness, we retained the first RT-PCR test per person per season. Because vaccination with live-attenuated influenza vaccine was relatively uncommon, we did not include it in this study. We included only individuals with continuous KPNC membership during the 12 months prior to 1 September of each season and who were at least aged 2 years at vaccination. Persons who received more than 1 vaccination during a season or received an RT-PCR test prior to vaccination were excluded from that season. Consistent with prior studies [6–10, 12], we excluded persons who received an influenza test within 14 days of vaccination to allow time for full immunologic response. For each vaccination, we identified the recipient’s age, gender, and which of 7 geographic areas (eg, East Bay) the patient received their care within KPNC. As a proxy of health status and propensity to use services, we calculated the patient’s Diagnostic Cost Group (DxCG) predictive risk score based on data in the year prior to 1 September of the season in which they were vaccinated. The DxCG risk score is a commonly used risk adjustment system that is used to predict healthcare cost using an individual’s age, gender, and prior year’s procedure and diagnostic codes [24, 25].
Analyses
To describe the predictors of the timing of vaccination within the season, we calculated the number of days from 1 September to the date of vaccination for each person and used this as the dependent variable in an ordinary least squares regression. The independent variables were age at vaccination (categorical by age in years: 2 to <5, 5 to <12, 12 to <18, 18 to <30, 30 to <50, 50 to <65, 65 to <75, 75 to <85, and 85+), gender, DxCG, and a dichotomous indicator of influenza vaccination in the prior season.
To estimate VE waning, we used a case-positive, control test-negative design [14, 16, 18] to determine whether there was a trend in the association of days-since-vaccination with odds of testing positive for influenza among vaccinated persons who were tested. The cases were persons who tested positive for influenza virus and the comparison group was persons who tested negative. Conditional logistic regressions were run in which testing positive for any influenza (A + B), influenza A, and influenza B were the dependent variables and days-since-vaccination (at the time of the RT-PCR test) was the independent variable of interest. Days-since-vaccination was parameterized as a categorical variable in 28-day increments: 14–41 (the reference group), 42–69, 70–97, 98–125, 126–153, and 154+ days. The other independent variables were the same as in the model predicting the timing of vaccination within the season. The logistic regressions were conditioned (“stratified”) on calendar date (eg, “1/15/2015”) of the influenza test and the patient’s geographic region. Models were also run separately by influenza season and age group. To increase power, 3 age groups were used in the latter analyses: <18, 18–64, and 65+ years. To determine the effect on waning of adjusting for age and health status, we performed a post hoc analysis in which we excluded age and DxCG as covariables in the models. In another post hoc analysis, the days-since-vaccination variable was treated as a continuous variable, divided by 28, to estimate the average effect per 28-day period. The above analyses were repeated with positive RSV test as the outcome. Both the influenza and RSV analyses used the same comparison group—persons who test negative for influenza and RSV. Analyses were performed using SAS, version 9.3.
RESULTS
During the study period, 49272 vaccinations (representing 44959 unique patients) met inclusion criteria. Across all 7 seasons, testing peaked between 1 January and 23 February (Figure 1). Of the 49272 influenza tests during the study period, 6381 (13%) were positive for influenza A only, 1566 (3%) for influenza B only, 4088 (8%) for RSV only, and 61 for both influenza and RSV (Table 1). Persons who tested positive for influenza A were older than persons who tested positive for influenza B (aged 59 vs 36 years) or RSV (aged 59 vs 49 years).
Figure 1.
Number of reverse-transcriptase polymerase chain reaction influenza tests by week among persons vaccinated with inactivated influenza vaccine, influenza seasons 2010–2011 to 2016–2017 combined. Abbreviation: RT-PCR, reverse-transcriptase polymerase chain reaction.
Table 1.
Description of Persons Vaccinated With Inactivated Influenza Vaccine Who Were Tested for Influenza Virus, by Test Result, 2010–2011 to 2016–2017: Kaiser Permanente Northern California
| Characteristic | Tested Positive for Influenza A Only (n = 6381) | Tested Positive for Influenza B Only (n = 1566) | Tested Positive for RSV Only (n = 4088) | Tested Negative for Influenza A, B, and RSV (n = 37237) |
|---|---|---|---|---|
| Age (years), no. (%) | ||||
| 2–4 | 228 (4) | 117 (7) | 1139 (28) | 2250 (6) |
| 5–11 | 526 (8) | 433 (28) | 237 (6) | 2097 (6) |
| 12–17 | 331 (5) | 260 (17) | 60 (1) | 1333 (4) |
| 18–29 | 180 (3) | 36 (2) | 29 (1) | 908 (2) |
| 30–49 | 639 (10) | 125 (8) | 175 (4) | 3223 (9) |
| 50–64 | 947 (15) | 158 (10) | 433 (11) | 5919 (16) |
| 65–74 | 1034 (16) | 172 (11) | 602 (15) | 7146 (19) |
| 75–84 | 1401 (22) | 155 (10) | 808 (20) | 8255 (22) |
| 85+ | 1095 (17) | 110 (7) | 605 (15) | 6106 (16) |
| Age, mean (median) in years | 59 (68) | 36 (16) | 49 (64) | 60 (69) |
| Gender, no. (%) | ||||
| Female | 3400 (53) | 822 (52) | 2201 (54) | 19886 (53) |
| Male | 2981 (47) | 744 (48) | 1887 (46) | 17351 (47) |
| Diagnostic Cost Group risk score, mean (median) | 7 (4) | 4 (1) | 7 (4) | 9 (6) |
| Influenza vaccination in prior season, no. (%) | 5133 (80) | 1191 (76) | 3377 (83) | 29904 (80) |
| Year, no. (%) | ||||
| 2010–2011 | 390 (6) | 182 (12) | 441 (11) | 3118 (8) |
| 2011–2012 | 492 (8) | 175 (11) | 349 (9) | 3152 (8) |
| 2012–2013 | 980 (15) | 175 (11) | 618 (15) | 4103 (11) |
| 2013–2014 | 495 (8) | 73 (5) | 378 (9) | 5184 (14) |
| 2014–2015 | 1304 (20) | 110 (7) | 801 (20) | 6306 (17) |
| 2015–2016 | 327 (5) | 597 (38) | 486 (12) | 6558 (18) |
| 2016–2017 | 2393 (38) | 254 (16) | 1015 (25) | 8816 (24) |
| Number of days from 1 September to vaccination, mean (median) | 46 (40) | 55 (46) | 52 (44) | 50 (42) |
| Number of days from vaccination to influenza test (by 4-week periods), no. (%) | ||||
| 14–41 | 255 (4) | 52 (3) | 313 (8) | 4140 (11) |
| 42–69 | 697 (11) | 101 (6) | 596 (15) | 5428 (15) |
| 70–97 | 1812 (28) | 206 (13) | 948 (23) | 7634 (21) |
| 98–125 | 2022 (32) | 341 (22) | 1075 (26) | 7671 (21) |
| 126–153 | 1033 (16) | 392 (25) | 770 (19) | 5964 (16) |
| 154+ | 562 (9) | 474 (30) | 386 (9) | 6400 (17) |
| Number of days from vaccination to influenza test, mean (median) | 104 (102) | 130 (131) | 101 (102) | 105 (102) |
Study cohort consisted of all influenza vaccinations from 2010–2011 to 2016–2017. Persons contributed a record for each vaccination they received during the study period. Percents are column percents. In addition to those reported here, 61 tests were positive for both influenza virus and RSV.
Abbreviation: RSV, respiratory syncytial virus.
Older persons, females, persons with higher DxCG scores, and persons vaccinated in the previous season were more likely to be vaccinated earlier in the season (Supplementary Table 1). In models that assessed the waning of VE, odds of testing positive for any influenza increased with days-since-vaccination (Table 2). Compared to persons vaccinated 14 to 41 days prior to being tested, persons vaccinated 42 to 69 days prior had 1.32 (95% confidence interval [CI], 1.11 to 1.55) times the odds of testing positive, and the odds ratio (OR) increased approximately linearly to 2.06 (95% CI, 1.69 to 2.51) for persons vaccinated 154 or more days prior to being tested. When age and DxCG were excluded from the models, estimates of waning were less than when those variables were included. When days-since-vaccination was treated as continuous, every 28 days after vaccination was associated with 1.16 (CI, 1.13 to 1.20) increased odds of testing positive for any influenza. Results were similar for influenza A. For influenza B, the ORs for testing positive in all days-since-vaccination categories were above 1.0, but only days 126 to 153 were statistically significant. In contrast, there were no significant differences or any trend in the relationship between days-since-vaccination and testing positive for RSV.
Table 2.
Relationship Between Days Since Vaccination With Inactivated Influenza Vaccine and Risk of a Positive Viral Test, 2010–2011 to 2016–2017: Kaiser Permanente Northern California
| Test-positive Outcome, Odds Ratio (95% Confidence Interval) | |||||
|---|---|---|---|---|---|
| Parameterization of Days Since Vaccinationa | Days Since Vaccination | Any Influenza | Influenza A | Influenza B | Respiratory Syncytial Virus |
| Categorical in 28-day periods, including all covariables | 14–41 | Reference | Reference | Reference | Reference |
| 42–69 | 1.32 (1.11,1.55) | 1.34 (1.12,1.61) | 1.03 (0.67,1.59) | 0.96 (0.80,1.15) | |
| 70–97 | 1.56 (1.33,1.82) | 1.55 (1.31,1.83) | 1.41 (0.95,2.08) | 0.89 (0.74,1.05) | |
| 98–125 | 1.78 (1.52,2.09) | 1.82 (1.53,2.15) | 1.23 (0.84,1.80) | 0.95 (0.80,1.14) | |
| 126–153 | 1.91 (1.61,2.26) | 1.98 (1.64,2.39) | 1.50 (1.02,2.21) | 0.96 (0.79,1.17) | |
| 154+ | 2.06 (1.69,2.51) | 2.42 (1.92,3.05) | 1.49 (0.99,2.25) | 1.14 (0.89,1.46) | |
| Categorical in 28-day periods, without adjustment for age and Diagnostic Cost Group | 14–41 | Reference | Reference | Reference | Reference |
| 42–69 | 1.29 (1.10,1.52) | 1.34 (1.13,1.60) | 1.06 (0.71,1.57) | 0.93 (0.79,1.10) | |
| 70–97 | 1.44 (1.24,1.68) | 1.50 (1.27,1.77) | 1.23 (0.86,1.76) | 0.77 (0.66,0.91) | |
| 98–125 | 1.57 (1.35,1.83) | 1.71 (1.45,2.02) | 1.02 (0.72,1.44) | 0.75 (0.64,0.88) | |
| 126–153 | 1.56 (1.32,1.84) | 1.80 (1.50,2.16) | 0.98 (0.69,1.39) | 0.73 (0.61,0.87) | |
| 154+ | 1.59 (1.32,1.93) | 2.15 (1.72,2.70) | 0.86 (0.59,1.24) | 0.80 (0.63,1.00) | |
| Continuous; scaled as number of days/28 (model includes all covariables) | Per 28 days | 1.16 (1.13,1.20) | 1.20 (1.15,1.24) | 1.10 (1.03,1.17) | 1.01 (0.97,1.06) |
Case positive, control test-negative analyses included only those persons who received inactivated influenza vaccination and who subsequently received an reverse-transcriptase polymerase chain reaction (PCR) test for influenza and respiratory syncytial virus (RSV) during the season in which they were vaccinated. The statistical model was a conditional logistic regression where the dependent variable was a dichotomous indicator of whether the test was positive for influenza (or RSV). The comparison group was persons who tested negative for each of the other viruses. The logistic regression was conditioned on the calendar date of the PCR test and geographic area within Northern California, and models were adjusted for gender, whether patients received influenza vaccination in the prior year, and (unless otherwise noted) age and Diagnostic Cost Group. Separate models were run for each outcome.
aEach different parameterization of the days-since-vaccination variable was a separate model.
Although there was limited power to detect waning effectiveness by year or age group, there was a tendency toward increased risk of any influenza with days-since-vaccination in every season (especially the 2012–2013, 2014–2015, and 2016–2017 seasons; Table 3) and in each of the 3 age groups (Table 4).
Table 3.
Relationship Between Days Since Vaccination With Inactivated Influenza Vaccine and Risk of a Positive Viral Test by Season, 2010–2011 to 2016–2017: Kaiser Permanente Northern California
| Days Since Vaccination (in Periods 28 days in Length) | Season, Odds Ratio (95% Confidence Interval) | |||||||
|---|---|---|---|---|---|---|---|---|
| Outcome | 2010–2011 | 2011–2012 | 2012–2013 | 2013–2014 | 2014–2015 | 2015–2016 | 2016–2017 | |
| Any influenza | … | Cases = 578 | Cases = 672 | Cases = 1162 | Cases = 572 | Cases = 1419 | Cases = 937 | Cases = 2668 |
| 14–41 | Reference | Reference | Reference | Reference | Reference | Reference | Reference | |
| 42–69 | 1.66 (0.78,3.55) | 2.43 (0.99,5.94) | 1.68 (1.04,2.71) | 1.26 (0.78,2.05) | 1.08 (0.71,1.65) | 1.08 (0.61,1.94) | 1.25 (0.97,1.60) | |
| 70–97 | 1.40 (0.67,2.90) | 1.67 (0.73,3.82) | 1.96 (1.27,3.03) | 1.22 (0.78,1.91) | 1.66 (1.12,2.47) | 1.71 (0.99,2.95) | 1.41 (1.10,1.79) | |
| 98–125 | 1.78 (0.88,3.62) | 2.16 (1.01,4.66) | 2.42 (1.58,3.71) | 1.75 (1.11,2.76) | 1.82 (1.22,2.72) | 1.56 (0.92,2.66) | 1.51 (1.18,1.95) | |
| 126–153 | 1.87 (0.90,3.87) | 2.35 (1.10,5.01) | 2.69 (1.71,4.22) | 1.50 (0.83,2.69) | 1.98 (1.28,3.07) | 1.99 (1.17,3.41) | 1.60 (1.19,2.15) | |
| 154+ | 1.74 (0.78,3.87) | 2.73 (1.27,5.85) | 2.51 (1.44,4.40) | 1.23 (0.43,3.51) | 2.10 (1.20,3.68) | 2.28 (1.29,4.03) | 2.09 (1.43,3.05) | |
| Respiratory syncytial virus | … | Cases = 447 | Cases = 354 | Cases = 625 | Cases = 382 | Cases = 806 | Cases = 499 | Cases = 1036 |
| 14–41 | Reference | Reference | Reference | Reference | Reference | Reference | Reference | |
| 42–69 | 0.82 (0.47,1.44) | 0.92 (0.40,2.11) | 0.91 (0.58,1.42) | 1.19 (0.66,2.12) | 1.19 (0.75,1.88) | 0.95 (0.55,1.64) | 0.90 (0.63,1.27) | |
| 70–97 | 0.82 (0.47,1.40) | 0.79 (0.38,1.67) | 0.80 (0.52,1.21) | 1.27 (0.74,2.20) | 1.01 (0.64,1.58) | 0.76 (0.45,1.30) | 0.89 (0.64,1.25) | |
| 98–125 | 0.70 (0.39,1.26) | 0.73 (0.37,1.45) | 0.77 (0.50,1.18) | 1.69 (0.98,2.91) | 1.10 (0.70,1.72) | 1.00 (0.58,1.71) | 0.97 (0.68,1.38) | |
| 126–153 | 0.62 (0.32,1.21) | 0.76 (0.38,1.52) | 1.16 (0.72,1.86) | 1.41 (0.78,2.54) | 0.86 (0.53,1.40) | 1.10 (0.63,1.93) | 1.01 (0.68,1.49) | |
| 154+ | 1.07 (0.43,2.65) | 0.54 (0.25,1.18) | 1.37 (0.71,2.62) | 1.69 (0.78,3.65) | 1.30 (0.70,2.38) | 1.38 (0.71,2.67) | 1.24 (0.75,2.07) | |
Case positive, control test-negative analyses included only those persons who received inactivated influenza vaccination and who subsequently received a reverse-transcriptase polymerase chain reaction (PCR) test for influenza and respiratory syncytial virus (RSV) during the season in which they were vaccinated. The statistical model was a conditional logistic regression where the dependent variable was a dichotomous indicator of whether the test was positive for influenza (or RSV). The comparison group was persons who tested negative for each of the other viruses. The logistic regression was conditioned on the calendar date of the PCR test and geographic area within Northern California, and models were adjusted for age, gender, Diagnostic Cost Group, and whether patients received influenza vaccination in the prior year. Separate models were run for each outcome and each season.
Table 4.
Relationship Between Days Since Vaccination With Inactivated Influenza Vaccine and Risk of a Positive Viral Test by Age, 2010–2011 to 2016–2017: Kaiser Permanente Northern California
| Days Since Vaccination (in Periods 28 Days in Length) | Age in Years, Odds Ratio (95% Confidence Interval) | |||
|---|---|---|---|---|
| Virus Type | 2 to <18 | 18 to <65 | 65 and Older | |
| Any influenza | … | Cases = 1915 | Cases = 2099 | Cases = 3994 |
| 14–41 | Reference | Reference | Reference | |
| 42–69 | 1.19 (0.78,1.83) | 1.52 (1.06,2.18) | 1.13 (0.86,1.48) | |
| 70–97 | 1.42 (0.95,2.10) | 1.73 (1.23,2.42) | 1.37 (1.06,1.76) | |
| 98–125 | 1.77 (1.20,2.60) | 1.85 (1.31,2.60) | 1.56 (1.21,2.02) | |
| 126–153 | 2.10 (1.38,3.20) | 2.18 (1.49,3.18) | 1.70 (1.29,2.25) | |
| 154+ | 1.86 (1.11,3.10) | 2.62 (1.68,4.10) | 1.92 (1.37,2.68) | |
| Respiratory syncytial virus | … | Cases = 1456 | Cases = 651 | Cases = 2042 |
| 14–41 | Reference | Reference | Reference | |
| 42–69 | 0.85 (0.58,1.24) | 1.26 (0.68,2.33) | 1.08 (0.78,1.50) | |
| 70–97 | 0.92 (0.64,1.31) | 1.48 (0.81,2.69) | 0.99 (0.72,1.35) | |
| 98–125 | 0.97 (0.66,1.42) | 1.59 (0.88,2.89) | 1.04 (0.76,1.42) | |
| 126–153 | 0.63 (0.40,0.99) | 2.10 (1.13,3.91) | 1.03 (0.74,1.43) | |
| 154+ | 0.84 (0.37,1.87) | 1.62 (0.78,3.36) | 1.35 (0.91,2.00) | |
Case positive, control test-negative analyses included only those persons who received inactivated influenza vaccination and who subsequently received a reverse-transcriptase polymerase chain reaction (PCR) test for influenza and respiratory syncytial virus (RSV) during the season in which they were vaccinated. The statistical model was a conditional logistic regression where the dependent variable was a dichotomous indicator of whether the test was positive for influenza (or RSV). The comparison group was persons who tested negative for each of the other viruses. The logistic regression was conditioned on the calendar date of the PCR test and geographic area within Northern California, and models were adjusted for age, gender, Diagnostic Cost Group, and whether patients received influenza vaccination in the prior year. Separate models were run for each outcome and age group. Results of these models need not be a weighted average of the results in Table 2 due to changes in the number of tests that are in informative strata.
DISCUSSION
In this study encompassing 7 influenza seasons from 2010–2011 to 2016–2017, we found evidence that the effectiveness of IIV wanes during the course of the influenza season. Across all seasons, every additional 28 days between vaccination and influenza testing was associated with, approximately, a 16% increase in the odds of testing positive for any influenza. These results were mainly due to waning associated with influenza A, which represented 80% of positive influenza tests. We found no comparable significant relationship between day-since-vaccination and testing positive for RSV.
Our finding of intraseason waning effectiveness of IIV is consistent with a number of other recent studies. A US study of the 2007–2008 influenza season found an adjusted OR for influenza of 1.12 for every 14 day increase in time since vaccination [10]. A Spanish analysis of the 2011–2012 season found that VE was 61% against influenza A in the first 100 days after vaccination, 39% between 100 and 119 days, and zero thereafter [7]. A UK study of the 2011–2012 season estimated VE against influenza A of 53% for those vaccinated less than 3 months and 12% for those vaccinated 3 months or more before onset of symptoms [6]. A US study of the 2011–2012 through 2014–2015 seasons estimated a 7% absolute decline in VE against influenza A (H3N2) with every 30 days since vaccination [9]. None of these studies included as many seasons as the present study, nor did any include a negative-control outcome.
The finding of intraseason waning of VE raises the question of how many influenza cases might be prevented as a result of delaying vaccination. Specifically, we considered the effect on the number of influenza cases of an 8-week delay in vaccination from 1 October to 26 November in a hypothetical season in which influenza incidence in the unvaccinated is 60 per 1000 persons; all influenza cases occur between 7 January and 3 February; VE for the 1 October vaccinees is 40%; the mix of influenza A and B is average (80% influenza A, 20% influenza B); and the 8-week delay in vaccination results in no decrease in vaccine coverage. Our analysis (using ORs of 1.32 and 1.78 for 42 to 69 and 98 to 125 days since vaccination, respectively) implies that 9 influenza episodes would be prevented per 1000 vaccinations delayed from 1 October to 26 November.
Although our results suggest that some number of influenza cases may be averted by delaying vaccination, any changes in recommendations regarding the timing of vaccination should be approached with caution. The test-negative approach may lack generalizability [20] if the observed waning effect only applies (for example) to the type of persons who are tested.
In addition, attempts by persons or institutions to delay vaccination may inadvertently result in persons failing to get vaccinated and/or the influenza season may arrive before persons have a chance to get vaccinated. Nevertheless, our findings may help inform the decisions by individual patients (especially those determined to be vaccinated) and their healthcare providers regarding the optimal time for their annual vaccination. Our findings may also inform policy decisions regarding the timing of vaccination for healthcare workers and provide an additional rationale for the development of new vaccines.
Test-negative designs relating to influenza vaccination have often been used to assess overall VE [26] and typically have included both vaccinated and unvaccinated persons. As with some other studies of vaccine waning [10, 12], we included only vaccinated persons in our analyses, which allowed us to avoid confounding related to differences between vaccinated and unvaccinated persons. In addition, our implementation of the test-negative design rigorously controlled for calendar time and geographic area. Thus, we only compared influenza-positive cases with influenza-negative “controls” where the test was performed on the same calendar day in the same area. This reduces or eliminates biases relating to changes in the circulation of influenza and other viruses in the underlying population throughout the course of the year [16, 17] and confounding that might result, for example, if the vaccine was delivered earliest to those clinics in areas where influenza arrived late. Although we view waning as likely related to declining seroprotection in individuals, it could also be related to antigenic drift in circulating viruses and emergence of viruses not included in that year’s vaccine.
We adjusted for age and DxCG risk score in our primary models. We might expect older persons and persons with poorer health to be both more likely to get vaccinated early and to subsequently test positive for influenza. If this were so, failure to adjust for these characteristics could result in overestimating vaccine waning. However, because older age and poor health status may make persons more vulnerable to the consequences of getting influenza, providers may have a lower threshold for testing such persons. In this scenario, older age and poorer health may be associated with testing negative for influenza rather than testing positive, and failing to adjust for them would underestimate waning rather than overestimate it. Indeed, when age and DxCG were left out of the models, the estimates of waning decreased. This suggests that if there remain unmeasured confounders and if those confounders operate similarly to the ones we have measured, our estimates of waning may be underestimated rather than overestimated. In addition, if there is waning VE, it is possible that over time there could be differential depletion of susceptible individuals among the earlier vaccinees. Given our vaccine-only study design, this source of bias would cause us to underestimate risk in the early vaccinees and thus underestimate VE waning.
Some prior test-negative studies [9, 10, 12] of VE waning were prospective studies that recruited and consented patients who presented to the health system, met study criteria for influenza-like illness (ILI), and agreed to be tested for influenza. In our retrospective observational study, testing was performed as part of routine clinical care for a modest portion of patients who presented with ILI. Both approaches are subject to biases relating to care-seeking behavior; persons who present for care may be different from the underlying population in their timing of vaccination. In routine clinical care, the selection process by which the provider identifies the subset of patients to be tested may either exacerbate or mitigate this bias, and the same is true for the selection process used by a prospective study to recruit and consent a subset of patients. Thus, it remains unclear which of the 2 approaches yields a tested population that is less prone to selection bias and more representative of the overall population of vaccinees.
Although potential sources of bias and confounding remain with the test-negative approach, the use of the RSV negative-control outcome helps us assess whether such residual confounding actually exists. A negative-control outcome should be one that is subject to the same source of bias as the outcome of interest but is not affected by the exposure of interest [21, 27]. In our analysis, residual bias would exist if there are unmeasured differences between persons who get vaccinated earlier vs later, and these differences also are associated with testing positive for influenza. RSV is a good negative control to the extent that we expect those differences to be similar with respect to testing positive for RSV. The most important known difference in risk factors between RSV and influenza relate to age. However, our models adjusted for age (and did not include children aged <2 years). A study of adults found no differences between persons hospitalized with RSV and influenza A with respect to their demographics, chronic illnesses, residence in long-term care facilities, or smoking status [28].
Because the formulation of influenza vaccine differs from season to season, as do the circulating viruses, we expect differences both in overall effectiveness and waning of effectiveness from year to year. Although the pooling of data from different years may, in certain circumstances, be misleading [29–31], our analyses by season indicated that waning tended to occur in most seasons, although to varying degrees. It may be perplexing that we found waning even during the 2014–2015 season when VE in the United States was reported being as low as 19% (CI, 10% to 27%) against any influenza and 6% (CI, −5% to 17%) against influenza A (H3N2) [32]. It is reasonable to assume that VE is never negative, and so during a season in which overall VE for the entire season is very low, we would expect to find less waning. However, our effect estimate, that is, that the odds of influenza in remote vaccinees vs recent vaccinees increases by about 16% per 28 days, could be consistent with a 12-week influenza season with overall VE of 14%, if VE is 36% during the first 4 weeks, 14% during the next 4 weeks month, and near zero during the last 4 weeks. (One study of waning VE estimated that overall VE against influenza A [H3N2] in the 2014–2015 season was 12%, but reached zero at 108 days post-vaccination [9].) With respect to seasonal VE estimates, it is worth noting that there may be considerable differences in circulating viruses in different regions, and VE may, in fact, vary between regions and different populations.
Our results contribute to a growing body of literature suggesting that the effectiveness of IIV wanes during the course of the season. The strengths of this study were the inclusion of 7 influenza seasons, the rigorous adjustment for calendar time and geographic area, and the use of RSV as a negative-control outcome. In combination with other findings, these results can contribute to the evaluation of the potential advantages and disadvantages associated with the timing of influenza vaccination.
Supplementary Data
Supplementary materials are available at Clinical 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
Disclaimer. The sponsor had no role in any of the following: design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, and approval of the manuscript; and decision to submit the manuscript for publication.
Financial support. This work was supported by the National Institute of Allergy and Infectious Diseases, National Institutes of Health (grant 1R01AI107721-01).
Potential conflicts of interest. G. T. R. has received research support on grants from Pfizer to Kaiser Permanente Division of Research in the past 3 years. N. P. K. reports research support from Sanofi Pasteur, GlaxoSmithKline, Protein Science, MedImmune, Pfizer, Merck, and Dynavax. All other authors report no potential 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.
References
- 1. Grohskopf LA, Sokolow LZ, Olsen SJ, Bresee JS, Broder KR, Karron RA. Prevention and control of influenza with vaccines: recommendations of the Advisory Committee on Immunization Practices, United States, 2015-16 influenza season. MMWR Morb Mortal Wkly Rep 2015; 64:818–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Grohskopf LA, Sokolow LZ, Broder KR, et al. Prevention and control of seasonal influenza with vaccines. MMWR Recomm Rep 2016; 65:1–54. [DOI] [PubMed] [Google Scholar]
- 3. World Health Organization. Recommended viruses for influenza vaccines for use in the 2010–2011 northern hemisphere influenza season. Available at: http://www.who.int/influenza/vaccines/virus/2010_11north/en/ Accessed 9 December 2017. [Google Scholar]
- 4. World Health Organization. Recommended composition of influenza virus vaccines for use in the 2011–2012 northern hemisphere influenza season. Available at: http://www.who.int/influenza/vaccines/virus/2011_12north/en/ Accessed 9 December 2017. [Google Scholar]
- 5. Grohskopf LA, Uyeki T, Bressee J, Cox N, Bridges C. Prevention and control of seasonal influenza with vaccines: recommendations of the Advisory Committee on Immunization Practices (ACIP), 2011. MMWR Morb Mortal Wkly Rep 2011; 60:1128–32. [PubMed] [Google Scholar]
- 6. Pebody R, Andrews N, McMenamin J, et al. Vaccine effectiveness of 2011/12 trivalent seasonal influenza vaccine in preventing laboratory-confirmed influenza in primary care in the United Kingdom: evidence of waning intra-seasonal protection. Euro Surveill 2013; 18:pii=20389. doi: 10.2807/ese.18.05.20389-en [DOI] [PubMed] [Google Scholar]
- 7. Castilla J, Martinez-Baz I, Martinez-Artola V, et al. Decline in influenza vaccine effectiveness with time after vaccination, Navarre, Spain, season 2011/12. Euro Surveill 2013; 18:pii=20388. doi: 10.2807/ese.18.05.20388-en [DOI] [PubMed] [Google Scholar]
- 8. Kissling E, Nunes B, Robertson C, et al. I-MOVE multicentre case-control study 2010/11 to 2014/15: is there within-season waning of influenza type/subtype vaccine effectiveness with increasing time since vaccination?Euro Surveill 2016; 21:pii=30201. doi: 10.2807/1560-7917.ES.2016.21.16.30201 [DOI] [PubMed] [Google Scholar]
- 9. Ferdinands JM, Fry AM, Reynolds S, et al. Intraseason waning of influenza vaccine protection: evidence from the US influenza vaccine effectiveness network, 2011-12 through 2014-15. Clin Infect Dis 2017; 64:544–50. [DOI] [PubMed] [Google Scholar]
- 10. Belongia EA, Sundaram ME, McClure DL, Meece JK, Ferdinands J, VanWormer JJ. Waning vaccine protection against influenza A (H3N2) illness in children and older adults during a single season. Vaccine 2015; 33:246–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Petrie JG, Ohmit SE, Truscon R, et al. Modest waning of influenza vaccine efficacy and antibody titers during the 2007–2008 influenza season. J Infect Dis 2016; 214:1142–9. [DOI] [PubMed] [Google Scholar]
- 12. Puig-Barbera J, Mira-Iglesias A, Tortajada-Girbes M, et al. Waning protection of influenza vaccination during four influenza seasons, 2011/2012 to 2014/2015. Vaccine 2017; 35:5799–807. [DOI] [PubMed] [Google Scholar]
- 13. Young B, Sadarangani S, Jiang L, Wilder-Smith A, Chen MI. Duration of influenza vaccine effectiveness: a systematic review, meta-analysis, and meta-regression of test-negative design case-control studies. J Infect Dis 2018; 217:731–41. [DOI] [PubMed] [Google Scholar]
- 14. Sullivan SG, Tchetgen Tchetgen EJ, Cowling BJ. Theoretical basis of the test-negative study design for assessment of influenza vaccine effectiveness. Am J Epidemiol 2016; 184:345–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. De Serres G, Skowronski DM, Wu XW, Ambrose CS. The test-negative design: validity, accuracy and precision of vaccine efficacy estimates compared to the gold standard of randomised placebo-controlled clinical trials. Euro Surveill 2013; 18:pii=20585. doi: 10.2807/1560-7917.ES2013.18.37.20585 [DOI] [PubMed] [Google Scholar]
- 16. Jackson ML, Nelson JC. The test-negative design for estimating influenza vaccine effectiveness. Vaccine 2013; 31:2165–8. [DOI] [PubMed] [Google Scholar]
- 17. Bond HS, Sullivan SG, Cowling BJ. Regression approaches in the test-negative study design for assessment of influenza vaccine effectiveness. Epidemiol Infect 2016; 144:1601–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Foppa IM, Haber M, Ferdinands JM, Shay DK. The case test-negative design for studies of the effectiveness of influenza vaccine. Vaccine 2013; 31:3104–9. [DOI] [PubMed] [Google Scholar]
- 19. World Health Organization. Evaluation of influenza vaccine effectiveness: a guide to the design and interpretation of observational studies. Geneva: World Health Organization, 2017. [Google Scholar]
- 20. Westreich D, Hudgens MG. Invited commentary: beware the test-negative design. Am J Epidemiol 2016; 184:354–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Lipsitch M, Tchetgen Tchetgen E, Cohen T. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology 2010; 21:383–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Arnold BF, Ercumen A, Benjamin-Chung J, Colford JM Jr. Brief report: negative controls to detect selection bias and measurement bias in epidemiologic studies. Epidemiology 2016; 27:637–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Greene SK, Shi P, Dutta-Linn MM, et al. Accuracy of data on influenza vaccination status at four vaccine safety datalink sites. Am J Prev Med 2009; 37:552–5. [DOI] [PubMed] [Google Scholar]
- 24. Wagner TH, Upadhyay A, Cowgill E, et al. Risk adjustment tools for learning health systems: a comparison of DxCG and CMS-HCC V21. Health Serv Res 2016; 51:2002–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Hamad R, Modrek S, Kubo J, Goldstein BA, Cullen MR. Using “big data” to capture overall health status: properties and predictive value of a claims-based health risk score. PLoS One 2015; 10:e0126054. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Sullivan SG, Feng S, Cowling BJ. Potential of the test-negative design for measuring influenza vaccine effectiveness: a systematic review. Expert Rev Vaccines 2014; 13:1571–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Arnold BF, Ercumen A. Negative control outcomes: a tool to detect bias in randomized trials. JAMA 2016; 316:2597–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Falsey AR, Hennessey PA, Formica MA, Cox C, Walsh EE. Respiratory syncytial virus infection in elderly and high-risk adults. N Engl J Med 2005; 352:1749–59. [DOI] [PubMed] [Google Scholar]
- 29. Baker SG, Kramer BS. Good for women, good for men, bad for people: Simpson’s paradox and the importance of sex-specific analysis in observational studies. J Womens Health Gend Based Med 2001; 10:867–72. [DOI] [PubMed] [Google Scholar]
- 30. Skowronski DM, Chambers C. Pooling and the potential dilution of repeat influenza vaccination effects. Clin Infect Dis 2017; 65:353–5. [DOI] [PubMed] [Google Scholar]
- 31. Cheng AC, Macartney KK, Waterer GW, Kotsimbos T, Kelly PM, Blyth CC. Reply to Skowronski and chambers. Clin Infect Dis 2017; 65:355. [DOI] [PubMed] [Google Scholar]
- 32. Zimmerman RK, Nowalk MP, Chung J, et al. 2014–2015 influenza vaccine effectiveness in the United States by vaccine type. Clin Infect Dis 2016; 63:1564–73. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.