Risk Factors and Attack Rates of Seasonal Influenza Infection: Results of the Southern Hemisphere Influenza and Vaccine Effectiveness Research and Surveillance (SHIVERS) Seroepidemiologic Cohort Study

Abstract

Background.

Understanding the attack rate of influenza infection and the proportion who become ill by risk group is key to implementing prevention measures. While population-based studies of antihemagglutinin antibody responses have been described previously, studies examining both antihemagglutinin and antineuraminidase antibodies are lacking.

Methods.

In 2015, we conducted a seroepidemiologic cohort study of individuals randomly selected from a population in New Zealand. We tested paired sera for hemagglutination inhibition (HAI) or neuraminidase inhibition (NAI) titers for seroconversion. We followed participants weekly and performed influenza polymerase chain reaction (PCR) for those reporting influenza-like illness (ILI).

Results.

Influenza infection (either HAI or NAI seroconversion) was found in 321 (35% [95% confidence interval, 32%–38%]) of 911 unvaccinated participants, of whom 100 (31%) seroconverted to NAI alone. Young children and Pacific peoples experienced the highest influenza infection attack rates, but overall only a quarter of all infected reported influenza PCR–confirmed ILI, and one-quarter of these sought medical attention. Seroconversion to NAI alone was higher among children aged <5 years vs those aged ≥5 years (14% vs 4%; P < .001) and among those with influenza B vs A(H3N2) virus infections (7% vs 0.3%; P < .001).

Conclusions.

Measurement of antineuraminidase antibodies in addition to antihemagglutinin antibodies may be important in capturing the true influenza infection rates.

Influenza infection results in various outcomes from asymptomatic infection to severe respiratory disease, cardiovascular complications, and death [1]. Few data exist on the seasonal influenza infection attack rates across age and other risk groups and the proportion of infections that progress to mild or severe illness. This lack of understanding hampers the ability to assess severity and burden in different target groups that guides our prevention and control measures such as vaccination, antiviral prophylaxis, and nonpharmaceutical interventions. It also poses challenges to assess the relative severity of novel influenza strains and predict their behaviors with modeling.

Seroepidemiologic studies can estimate the true age-specific incidence of influenza infection because asymptomatic infections are detected. Cross-sectional population serosurveys use differences in seropositivity proportions of antibodies before and after the influenza season to determine the attack rate [2, 3]. This approach may lead to misclassification of some infections due to inability to link pre- and postseason antibody titers to a specific individual, and low specificity resulting from cross-reactive antibodies, especially in seasonal influenza infections [4]. Cohort studies provide more reliable and precise estimates of incidence rates and risk profiles of infection by using paired pre- and postseason serum samples from the same individual. Household seroepidemiologic cohort studies for seasonal influenza [5-8] are useful for estimating the secondary attack rates and effects of intervention within households including children, but do not fully represent the entire community [9, 10].

The gold standard for serologic detection of recent influenza infections is to demonstrate seroconversion, a ≥4-fold increase in antibody titer relative to a baseline sample, to the circulating influenza suspected of causing the infection [4]. Antibody against hemagglutinin, which generally neutralizes viral infectivity, has been widely used in seroepidemiologic studies to define influenza infection through seroconversion. However, some influenza-infected individuals may not seroconvert to hemagglutinin, leading to underestimation of true infection rates [11-13]. Antibody against neuraminidase, while inducing “infection permissive” immunity by attenuating infection, limits viral spread within the host and ameliorates the clinical course of infection and disease [14, 15]. Although antineuraminidase seroconversion has been used in human challenge studies [13, 16], vaccine efficacy trials [17, 18], and investigations of naturally acquired infections [19, 20], it has not been applied to measure infection attack rates in population-based seroepidemiologic cohort studies. Estimation of influenza infection through antihemagglutinin and antineuraminidase seroconversion may capture additional infections missed when measuring antihemagglutinin seroconversion alone. Accurate estimate of the number of influenza infections is critically important in the calculation of case fatality or case hospitalization rates. This can result in more precise estimation of the burden and severity of seasonal influenza and help prediction of pandemic influenza transmission and optimize influenza countermeasures. Comparative patterns of antihemagglutinin and antineuraminidase responses are essential to improve understanding of host immune responses and correlates of protection and to optimize serologic diagnoses.

We describe here a seroepidemiologic cohort study conducted in a Southern Hemisphere country, New Zealand, through the Southern Hemisphere Influenza and Vaccine Effectiveness Research and Surveillance (SHIVERS) project. It aimed to estimate attack rates (a measure of cumulative incidence of infection over the season) of both influenza infection (symptomatic or not) and influenza polymerase chain reaction (PCR)–confirmed influenza-like illness (ILI) across risk groups in our unvaccinated cohort and the proportion of those that sought medical care.

METHODS

Full methods are described in the Supplementary Materials. The study was approved by the Northern A Health and Disability Ethics Committee (NTX/11/11/102 AM13). We obtained written informed consent from all participants (or guardians of children).

Sampling Frame and Sample Size

We included all 88 011 persons enrolled in 14 selected Auckland general practices (GPs) as our source population [21]. We stratified the population by age (0–4, 5–19, 20–64, ≥65 years) and ethnicity (Maori, Pacific, Asian, European, and others). Within each stratum, we performed simple random sampling to select a sample to represent different ages and ethnicities but purposely oversampled young children and Maori and Pacific ethnicities to get more robust rates in these hard-to-reach groups. We estimated a minimum sample size of 800 for an assumed 26% risk of infection and 2-sided 95% confidence intervals (CIs) of ±12%, 30% risk of ILI (±11%), 7% risk of influenza PCR–confirmed ILI (±29%).

Data Collection

During February–April 2015, we contacted the selected individuals by letter with follow-up phone calls. If individuals declined or could not be reached after 6 attempts, they were replaced by another randomly selected person from the same stratum. We administered phone questionnaires to participants before and after the study period for information on influenza vaccination, occurrence of respiratory illness during study period, and risk factors. We collected paired sera from all participants: preseason during March to mid-June and postseason during mid-October to December.

During May–September 2015, we conducted surveillance for cases of ILI (defined as “an acute respiratory illness with a history of fever or measured temperature of ≥38°C and cough, and onset within the past 7 days”). We texted/emailed participants weekly whether or not they had ILI. Participants could also report ILI by phones. Subsequently, nurses contacted those with reported ILI to verify their ILI status and arranged to collect a nasopharyngeal or throat swab within 14 days of reported onset.

Study data were captured using REDCap electronic data capture tools [22].

Laboratory Methods

We tested respiratory samples for influenza RNA by real-time reverse-transcription PCR assay [23]. We used standard hemagglutination inhibition (HAI) and neuraminidase inhibition (NAI) assays [2, 24] to assess antihemagglutinin and antineuraminidase antibody titers (the Supplementary Materials detail the influenza strains used in the assays).

Outcomes

We defined seroconversion as a ≥4-fold rise in HAI or NAI antibody titers in paired sera with the second HAI titer ≥40. Titers of 10 were assigned a value of 5 for computational purposes. Influenza infection was defined as HAI or NAI seroconversion or influenza PCR–confirmed ILI.

Illnesses were classified as “ILI” when nurses confirmed participant responses of “yes” to the weekly question “Have you had cough and fever in the previous 7 days?” Illnesses among those who responded “yes” but were determined upon a nurse’s examination to not meet the above criteria were classified as “mild illness–not ILI.” Those who responded “no” to the weekly question were classified as “not ILI.” Mild illness–not ILI and not ILI were combined as “no ILI”

Statistical Analysis

Analyses were performed in Stata version 14.1 (StataCorp LLC) software. Participants who had received the 2015 Southern Hemisphere influenza vaccine were excluded from the analysis because serologic assays cannot distinguish whether antibody titer rises are due to vaccinations or naturally acquired infections.

The observed attack rates of ILI and influenza PCR–confirmed ILI were corrected each week to account for (1) missed weekly reports by applying the percentage of reports that were classified as ILI to those nonresponders (corrected number of ILI events = total possible number of responses × actual number of ILI / actual number of responses); (2) missed swabs from ILI cases by applying the influenza positivity rate of those tested to those nontested (corrected number of influenza PCR–confirmed ILI events = corrected number of ILI × actual number of influenza PCR–confirmed ILI / actual number of ILI swabs). The above adjustments for the missed weekly reports and missed ILI swabs were applied on a week-by-week basis for each stratum (age and ethnicity) to account for differences over the season. Then we aggregated the relevant events for the season (22 weeks). To estimate the number of persons who experienced at least 1 ILI or influenza PCR–confirmed ILI event during the season, we divided the number of the relevant events by the average number of the events per person: Corrected number of persons with ILI or influenza PCR–confirmed ILI = corrected ILI or influenza PCR–confirmed ILI events / (actual number of ILI or influenza PCR–confirmed ILI events / actual number of persons with ILI or influenza PCR–confirmed ILI events).

Two independent methods were employed to calculate the percentage of infections as measured by seroconversion that led to influenza disease: (1) dividing the corrected number of persons with influenza PCR–confirmed ILI by the total number of seroconverters; and (2) subtracting ILI rates in nonseroconverters from those in seroconverters to account for the fact that some ILIs would be attributable to other noninfluenza pathogens or factors unrelated to infections or that some influenza-associated ILI cases would be negative by PCR due to late swabbing, sample type, and sampling techniques.

The risks of influenza infection and of developing ILI once infected were examined for 5 potential risk factors (age, sex, ethnicity, socioeconomic deprivation, underlying medical conditions). Univariate and multivariate analyses were performed using rate ratios from Poisson regression generalized linear models.

Finally, we weighted the attack rates in our study population by age and ethnicity strata to arrive at a total infection rate for the whole Auckland region; 95% CIs for proportions were calculated using the binomial distribution, applying weighting where appropriate.

RESULTS

In 2015, New Zealand’s influenza surveillance data showed influenza A(H3N2) and B (mainly B/Victoria lineage) predominance [25].

Of 14 825 randomly selected persons, 1514 (10%) agreed to participate, supplied paired sera, and completed the study questionnaires (Supplementary Figure 1). Of these participants, 603 (40%) reported receipt of the 2015 influenza vaccine. This report focuses on the remaining 911 unvaccinated individuals with full information and paired sera. When the unvaccinated study population was compared to the central and southeastern Auckland population, children aged <5 years were overrepresented and older adults (≥65 years) were underrepresented (Supplementary Table 1).

Overall Attack Rate of Influenza Infection

Of all 911 unvaccinated participants, 321 (35%) seroconverted to either HAI or NAI (Figure 1). Of these, 175 (55%) seroconverted to both HAI and NAI, 46 (14%) to HAI only, and 100 (31%) to NAI only. Of 590 nonseroconverters, no influenza PCR–positive results were detected.

Figure 1.

Serology and polymerase chain reaction (PCR) testing flow and results among the cohort. Hemagglutination inhibition (HAI) and neuraminidase inhibition (NAI) assays were performed for the 911 unvaccinated participants of all ages. Of them, 175 had HAI and NAI seroconversion, 46 HAI alone seroconversion, and 100 NAI alone seroconversion. Among the 321 seroconverters, 156 had influenza-like illness (ILI) symptoms and 109 (70% [109/156]) had swabs taken for influenza PCR, with 50 being positive (46% [50/109]). The rate of ILI was 48.6% (156/321) among HAI or NAI seroconverters and 22.2% (131/590) among nonseroconverters.

The risk of influenza infection was significantly higher in children aged 0–19 years than other age groups (46% vs 26%; P < .001), among Pacific peoples compared to other ethnicities (46% vs 34%; P = .014), and among A(H3N2)-infected persons compared with influenza B–infected persons (21% vs 18%; P = .046) (Table 1).

Table 1.

Influenza Seroconversion, Influenza-like Illness (ILI), Influenza-Confirmed ILI, and Proportion of Seroconversions Leading to Influenza Polymerase Chain Reaction–Confirmed ILI Among the Cohort

		A	B	C	D	E	F	G	H	I	J	K	L
		Cohort		Seroconverted		Observed			Corrected
Characteristics		No.	col %	No.	Infection Attack Rate (Col C/A), % (95% CI)	No. of Persons With ILI (Col E/A %)	No. of ILIs With Swabs (Col F/E %)	No. of Swabs With PCR- Confirmed Influenza (Col G/F %)	No. of Persons With ILI	ILI Attack Rate (Col H/A), % (95% CI)	No. of ILIs With PCR-Confirmed Influenza	Influenza PCR– Confirmed ILI Attack Rate (Col J/A), % (95% Cl)	% of Influenza PCR–Confirmed ILIs Among Serocoonverted (Col J/C), % (95% CI)
Overall		911	100	321	35 (32–38)	287 (32)	209 (73)	50 (24)	329	36 (33–39)	76	8 (7–10)	24 (19–29)
Age, y	0–4	160	18	77	48 (40–56)	89 (56)	73 (82)	16 (22)	107	67 (59–74)	22	14 (9–20)	29 (19–40)
	5–19	264	29	117	44 (38–51)	70 (27)	34 (49)	15 (44)	81	31 (25–37)	27	10 (7–15)	23 (16–32)
	20–64	444	49	116	26 (22–31)	119 (27)	94 (79)	18 (19)	133	30 (26–35)	25	6 (4–8)	22 (15–30)
	≥65	43	5	11	26 (14–41)	9 (21)	8 (89)	1 (13)	10	23 (12–39)	1	2 (0–12)	9 (0–41)
Sex	Female	538	59	179	33 (29–37)	172 (32)	126 (73)	34 (27)	197	37 (33–41)	50	9 (7–12)	28 (22–35)
	Male	373	41	142	38 (33–43)	115 (31)	83 (72)	16 (19)	132	35 (31–41)	24	6 (4–9)	17 (11–24)
Ethnicity	Maori	123	14	43	35 (27–44)	39 (32)	27 (69)	7 (26)	44	36 (27–45)	11	9 (5–15)	26 (14–41)
	Pacific	97	11	45	46 (36–57)	38 (39)	26 (68)	5 (19)	51	53 (42–63)	9	9 (4–17)	20 (10–35)
	Asian	197	22	78	40 (33–47)	59 (30)	40 (68)	9 (23)	68	35 (28–42)	14	7 (4–12)	18 (10–28)
	Othersa	494	54	155	31 (27–36)	151 (31)	116 (77)	29 (25)	167	34 (30–38)	40	8 (6–11)	26 (19–33)
NZDepb	1 or 2	206	23	64	31 (25–38)	62 (30)	48 (77)	15 (31)	70	34 (28–41)	23	11 (7–16)	36 (24–49)
	3 or 4	195	21	60	31 (24–38)	57 (29)	40 (70)	6 (15)	64	33 (26–40)	10	5 (3–9)	17 (8–29)
	5 or 6	170	19	64	38 (30–15)	47 (28)	38 (81)	10 (26)	51	30 (23–38)	13	8 (4–13)	20 (11–32)
	7 or 8	164	18	63	38 (31–46)	63 (38)	45 (71)	13 (29)	73	45 (37–53)	19	12 (7–18)	30 (19–43)
	9 or 10	176	19	70	40 (33–47)	58 (33)	38 (66)	6 (16)	73	42 (34–49)	11	6 (3–11)	16 (8–26)
Underlying conditionc	No	756	83	275	36 (33–40)	229 (30)	170 (74)	41 (24)	262	35 (31–38)	64	9 (7–11)	23 (18–29)
	Yes	155	17	46	30 (23–38)	58 (37)	39 (67)	9 (23)	66	43 (35–51)	14	9 (5–15)	30 (18–46)
Virus	A(H3N2)	911	100	193	21 (19–24)	…	209 (73)	24 (12)	…	…	37	4 (3–6)	19 (14–25)
	B	911	100	159	18 (15–20)	…	209 (73)	26 (12)	…	…	39	4 (3–6)	25 (18–32)

ILI is defined as an acute respiratory illness with a history of fever or measured temperature of ≥38°C and cough, with onset within the past 7 days. Influenza-PCR–confirmed ILI is defined as PCR-confirmed influenza from those participants with ILI. Influenza infection is defined as hemagglutination inhibition (HAI) seroconversion (≥4-fold HAI titer rise in paired sera with the second titer at least 1:40) or neuraminidase inhibition (NAI) seroconversion (≥4-fold NAI titer rise) or influenza RNA detection by real-time reverse-transcription PCR.

Abbreviations: CI, confidence interval; ILI, influenza-like illness; PCR, polymerase chain reaction.

^aRefers to Europeans and others.

^bNZDep scale measures socioeconomic deprivation on an ordinal scale of 1–10 where 1 indicates the individual is living in a household that is in the least socioeconomic-deprived decile of all New Zealand households.

^CUnderlying condition includes 1 or more of the following conditions: asthma, diabetes, chronic respiratory illness, heart disease (angina, myocardial infarction/heart attack, chronic heart failure, atrial fibrillation, rheumatic heart disease, congenital heart condition), and mental illnesses.

The overall infection attack rate, when weighted to the age and ethnicity structure of the Auckland population, was 32% (95% CI, 29%–35%) when influenza infection was defined as HAI or NAI seroconversion or influenza RNA detection; 24% (95% CI, 21%–27%) when only influenza RNA detection or HAI seroconversion were included.

ILI and Influenza PCR–Confirmed ILI

From 27 April to 27 September, participants responded to 17 121 of 20 042 (85%) potential person-weeks of ILI reminders. Responses included 347 (2% [347/17121]) ILI events reported from 287 persons for an observed ILI attack rate of 32% (287/911). The corrected ILI attack rate during the surveillance period was 36% (95% CI, 33%–39%). Children aged <5 years and Pacific peoples had significantly higher ILI attack rates than other age groups (67% vs 30%; P < .01) or other ethnicities (53% vs 34%; P < .01), respectively (Table 1).

Of the 287 persons with ILI, swabs were collected from 209 (73%) with a median of 8.2 days after onset (range, 2–15 days). Of these 209 persons, 50 (24%) were positive for influenza virus by PCR.

Influenza viruses were detected continuously from 27 June to 27 September with 2 distinct circulation patterns: A(H3N2) virus predominated during weeks 26–33, whereas influenza B virus (mainly B/Victoria lineage) predominated during weeks 34–39. No influenza A(H1N1)pdm09 virus was detected (Figure 2). The corrected number of influenza PCR–confirmed ILI (adjusting for nonswabbing and nonreporting) was 76 for a corrected influenza PCR–confirmed ILI attack rate of 8% (95% CI, 7%–10%) among the entire unvaccinated cohort. The highest attack rate of influenza PCR–confirmed ILI was in children aged 0–4 years (14%). Influenza PCR–confirmed ILI attack rates decreased with increasing age (Table 1).

Figure 2.

Temporal distribution of influenza-like illness (ILI) and influenza polymerase chain reaction–confirmed ILI and no ILI among the cohort during 27 April–27 September 2015. Influenza viruses were detected continuously during 27 June to 27 September with 2 distinct circulation patterns: A(H3N2) predominated during weeks 26–33, whereas influenza B (mainly B/Victoria lineage) predominated during weeks 34–39. No influenza A(H1N1)pdm09 was detected. ILI is defined as an acute respiratory illness with a history of fever or measured temperature of ≥38°C and cough, and onset within the past 7 days. No ILI refers those participants indicated not having ILI in the weekly text/email “Have you had cough and fever in the previous 7 days?” and also those indicated as having ILI but not verified by nurses.

Of all 321 who seroconverted to HAI or NAI, an estimated 24% (76/321) experienced influenza PCR–confirmed ILI. Using the alternative approach to estimate proportion of serological infections leading to ILI (subtracting rates of ILI in nonseroconverters from those in seroconverters), we found a corrected ILI proportion attributable to influenza of 30% (56% [180/321] in seroconverters minus 26% [151/590] in nonseroconverters). Infections of A(H3N2) and influenza B virus had similar proportions of ILI (19% [37/193] and 25% [39/159], respectively; Table 1).

We were concerned that we may have missed substantial numbers of symptomatic influenza illness from those who reported ILI but were subsequently not confirmed to meet the case definition (termed “mild illness–not ILI”) because the fever may have abated by the time the nurse examined the case of ILI. However, we found that the proportion of seroconverters with “mild illness–not ILI” was similar to those with “no ILI” reported (28% vs 23%, respectively; P = .125), but much lower than those with ILI (28% vs 48%, respectively; P < .001) (Supplementary Table 2).

Of the 50 influenza PCR–confirmed ILIs, 13 (26%) resulted in GP consultations. The corrected number of influenza PCR–confirmed ILIs resulting in consultation (adjusting for nonswabbing and nonreporting) was 15 for a rate of 1.6% (95% CI, .9–2.7) among the entire unvaccinated cohort. None of the influenza PCR–confirmed ILIs or ILI infections resulted in hospitalization (Supplementary Table 3).

HAI Versus NAI Seroconversion

Different patterns of HAI and NAI seroconversion rates were observed across risk groups. NAI-only seroconversion was significantly higher in young children aged <5 years than in other age groups (14% vs 4%; P < .001) and also in individuals infected with influenza B compared to those with A(H3N2) virus (7% vs 0.3%; P < .001; Figure 3 and Supplementary Table 4).

Figure 3.

Proportions of hemagglutination inhibition (HAI) or neuraminidase inhibition (NAI) seroconversion by age groups and by viruses. A, All influenza viruses. B, A(H3N2) virus. C, Influenza B virus. “%HAI” refers to proportion of individuals with HAI seroconversion (≥4-fold HAI titer rise in paired sera with the second titer at least 1:40). “%NAI” refers to proportion of individuals with NAI seroconversion (≥4-fold NAI titer rise in paired sera). “%NAI – %HAI” refers to the difference between NAI and HAI seroconversion rates by subtraction.

Risk Factors

Multivariate analysis showed that age was an independent risk factor for influenza infection. Children aged <5 and 5–19 years had significantly higher risk of infection compared to adults aged 20–64 years (Table 2). Although the risk of infection was similar among children <5 and 5–19 years, the risk of influenza PCR–confirmed ILI was significantly higher in children aged <5 years compared to adults aged 20–64 years. Pacific peoples also had significantly higher risk of influenza infections compared with Maori and European and other ethnic groups. The risk of influenza infection was 1.6 (95% CI, 1.1–2.3) times higher in persons with at least 1 child aged 5–19 years in the household compared to persons living in households without children (data not shown).

Table 2.

Multivariate Analysis for Risk of Influenza Infection and Disease Among the Cohort

		Influenza Infection			Influenza PCR–Confirmed ILI			% of Influenza PCR–Confirmed ILI Among Seroconverters
Risk Factor		Corrected No.	Adjusted Attack Rate Ratio (95% CI)	P Value	Corrected No.	Adjusted Attack Rate Ratio (95% CI)	P Value	Corrected %	Adjusted Attack Rate Ratio (95% CI)	P Value
Age, y	0–4	77	1.8 (1.4–2.5)	.000	22	2.5 (1.3–4.8)	.009	28.6	1.3 (.7–2.6)	.395
	5–19	117	1.7 (1.3–2.2)	.000	27	1.4 (.7–2.8)	.334	23.1	0.8 (.4–1.6)	.585
	20–64	116	ref		25	ref		21.6	ref
	≥65	11	1.0 (.5–1.8)	.947	1	0.6 (.1–4.3)	.589	9.1	0.6 (.1–4.4)	.603
Sex	Female	179	ref		50	ref		27.9	ref
	Male	142	1.1 (.9–1.4)	.231	24	0.7 (.4–1.2)	.201	16.9	0.6 (.3–1.1)	.085
Ethnicity	Maori	43	1.1 (.8–1.6)	.530	11	1.0 (.4–2.2)	.941	25.6	0.9 (.4–2.0)	.741
	Pacific	45	1.5 (1.1–2.1)	.021	9	0.9 (.3–2.3)	.788	20.0	0.6 (.2–1.5)	.282
	Asian	78	1.3 (1.0–1.7)	.094	14	0.8 (.4–1.6)	.511	17.9	0.6 (.3–1.3)	.205
	Othera	155	ref		40	ref		25.8	ref
NZDepb	1 or 2	64	ref		23	ref		35.9	ref
	3 or 4	60	1.0 (.7–1.4)	.957	10	0.4 (.2–1.1)	.075	16.7	0.4 (.2–1.1)	.078
	5 or 6	64	1.2 (.9–1.7)	.277	13	0.8 (.4–1.8)	.601	20.3	0.7 (.3–1.5)	.321
	7 or 8	63	1.2 (.9–1.8)	.232	19	1.1 (.5–2.3)	.823	30.2	0.9 (.4–1.9)	.737
	9 or 10	70	1.3 (.9–1.8)	.153	11	0.5 (.2–1.2)	.116	15.7	0.4 (.1–.9)	.037
Underlying conditionc	No	275	ref		64	ref		23.3	ref
	Yes	46	0.8 (.6–1.1)	.201	14	1.1 (.5–2.2)	.853	30.4	1.3 (.6–2.7)	.460

Influenza infection is defined as hemagglutination inhibition (HAI) seroconversion (≥4-fold HAI titer rise in paired sera with the second titer at least 1:40) or neuraminidase inhibition (NAI) seroconversion (≥4-fold NAI titer rise) or influenza RNA detection by PCR. Influenza PCR–confirmed ILI is defined as PCR-confirmed influenza among those symptomatic ILI participants. Percentage of infection leading to illness is defined as proportion of influenza PCR–confirmed ILI among influenza infection. The risks of influenza infection and of developing ILI once infected were examined for 5 potential risk factors (age, sex, ethnicity, socioeconomic deprivation, underlying medical conditions). Univariate and multivariate analyses were performed for these variables using rate ratios from Poisson regression generalized linear models.

Abbreviations: CI, confidence interval; ILI, influenza-like illness; PCR, polymerase chain reaction.

^aRefers to Europeans and others.

^bNZDep scale measures socioeconomic deprivation on an ordinal scale of 1–10 where 1 indicates the individual is living in a household that is in the least deprived decile of all New Zealand households.

^cUnderlying condition includes 1 or more of the following conditions: asthma, diabetes, chronic respiratory illness, heart disease (angina, myocardial infarction/heart attack, chronic heart failure, atrial fibrillation, rheumatic heart disease, congenital heart condition), and mental illnesses.

Other variables, sex, socioeconomic deprivation, underlying conditions, smoking, poor housing conditions, and household size did not show any independent effect on infection and influenza PCR–confirmed ILI.

DISCUSSION

The SHIVERS seroepidemiologic cohort study, to our knowledge, is the first report to quantify attack rates of influenza infection and disease using assays to measure seroconversion against both hemagglutinin and neuraminidase antigens in a seroepidemiologic cohort for all age groups. A third of unvaccinated individuals in our cohort were infected with influenza of which one-quarter developed influenza PCR–confirmed ILI and only a quarter of these sought medical attention. Seroconversion with antineuraminidase antibody alone constituted one-third of all seroconverted individuals, and was particularly frequent among children aged <5 years and influenza B virus–infected individuals.

When influenza infection was weighted to the Auckland population assuming unvaccinated, we estimate that 32% of the Auckland population were infected with influenza virus during the study. When NAI antibody titers were measured, we identified an additional 31% of influenza infections associated solely with antineuraminidase antibody. Our study highlights the importance of measuring serologically defined infections against not just hemagglutinin but also neuraminidase antigens to understand the true epidemiology and immunology of influenza and better guide prevention and control measures. Antineuraminidase antibody’s broad cross-protection and long duration of immunity against the 1968 Hong Kong A(H3N2) pandemic was demonstrated in seroepidemiologic cohort study in Tecumseh, Michigan. Infections with the newly emerged influenza A/Hong Kong/68 (H3N2) virus (against which the population had no specific H3 antihemagglutinin antibody) were significantly more frequent in subjects without antineuraminidase antibody than in those with antibody acquired by prior infection with the preceding influenza A(H2N2) viruses [20]. The importance of preexisting NAI titers as predictors of immunity against the 2009 pandemic virus has also been demonstrated [19, 26], and such data are valuable for better modeling influenza transmission.

Using the criteria of HAI seroconversion or influenza RNA detection, we found an attack rate of 24%, similar to studies in the United Kingdom (19%) [9], Vietnam (17%–26%) [10], and Tecumseh, Michigan (25%–34%) [27]. This consistency, despite varying study designs, laboratory methods, data availability, and circulating viruses, suggests that annually 15%–35% of the unvaccinated population is infected with seasonal influenza. Additionally, we found that seasonal influenza infections resulted in the highest infection attack rates in children and the risk decreased with increasing age. This pattern, also observed elsewhere from historical North America [7], contemporary Europe [9], and Southeast Asia [10] studies, may represent a generalizable finding globally. This probably reflects relative immunologic naivety of children and long-lived and cross-protective immunity against seasonal influenza virus [28]. Furthermore, we found that Pacific peoples had a significantly higher infection attack rate than other ethnicities, similar to our 2009 pandemic serosurvey [2]. This may be due to higher transmission of respiratory viruses [29] (possibly among communities with household crowding), higher levels of comorbidities, delays in seeking treatment, lack of knowledge of preventing spread, or genetic susceptibility. The proportion of Maori/Pacific populations may have lower herd immunity because of lower vaccination rates [30, 31].

Most influenza infections do not lead to illness. We found that 24% of HAI seroconverters developed PCR-confirmed influenza disease, similar to the UK study (25%) [9]. Additionally, we found that 19% of A(H3N2) and 25% of influenza B virus infections led to influenza disease, different from studies in Tecumseh, Michigan [27] with 15%–25% of A(H3N2) and 19%–34% of influenza B infections, and Vietnam [10] with 11% of A(H3N2) and 15% of influenza B infections leading to clinical illness. These varying proportions between studies may reflect different case definitions, study designs, climate settings, degrees of antigenic drift, circulating strains, and propensity to report illness.

Most individuals with symptomatic influenza did not consult healthcare providers. Our finding that 26% of the influenza PCR–confirmed ILI cases sought medical care was lower than studies in the United States (32%) [32] and Tecumseh, Michigan (38%) [8], but higher than the UK study (17%) [9]. These differences may reflect primary healthcare accessibility, health-seeking behavior, and cultural practices within these countries. We found a higher rate (1.6%) of GP consultations in participants with influenza PCR–confirmed ILI compared with that (0.6%) of the GPs registered population with influenza PCR–confirmed ILI [25]. This difference may be due to propensity to consult GPs if ill in our study cohort.

The serologically confirmed symptomatic and asymptomatic influenza infections provide an accurate denominator of the total burden of influenza for assessing severity by calculating case fatality and case hospitalization ratios. Previous studies showed that asymptomatic fractions among total influenza infections ranged from 65% to 85% in seroepidemiologic studies adjusted for background illnesses, a pooled mean of 16% (95% CI, 13%–19%) in outbreak investigations, and 33% in volunteer challenge studies, respectively [33-35]. For seroepidemiologic studies, varying case definitions of asymptomatic infection have been used, including completely asymptomatic [27], absence of acute respiratory illness [9], or absence of ILI (fever and cough) [10]. We showed that 70%–76% of seroconverted individuals did not develop ILI; however, some of these individuals would have developed milder afebrile illness.

Our study had several limitations: First, we only conducted the cohort study in 2015 rather than multiple influenza seasons. We were not able to study A(H1N1)pdm09 attack rates due to its low circulation. Second, we did not estimate true proportions of clinical disease among those infected as we did not elicit clinical information or test those not meeting ILI case definition. Some influenza cases with ILI may have been missed because of delays in reaching the case, but proportions of ILI due to influenza by subtracting rates of ILI in nonseroconverters from those in seroconverters produced similar results. Third, we had high weekly ILI responding rates (~85%). We assumed that ILI rates in not-always responders are likely to be similar to that in every-week responders. There did not appear to be any systematic bias by age or ethnicity between nonresponders and responders. Even if there was any bias, it would not contribute significantly due to small proportions (15%) of nonresponders. Fourth, ≥4-fold increase of HAI or NAI titers was used to define seroconversion. While this is relevant for diagnosis of an individual case to account for inherent measurement errors, we may miss those who were truly infected but did not show 4-fold rises [36]. Fifth, our sample had small numbers of participants aged <1 and ≥65 years, resulting in less precise estimates. Last, we only studied unvaccinated persons who may be different from vaccinated persons. Therefore, our extrapolation to all of Auckland only represents rates among unvaccinated people. The vaccination rate in Auckland is not recorded in registries, so calculating attack rates among the total population is not possible.

CONCLUSIONS

Our study provides more precise measurement for influenza infection and illness attack rates across risk groups. A substantial fraction of seroconverted individuals only to neuraminidase antigen in an age- and virus-specific manner highlights the importance of measuring serologically defined infections against not just hemagglutinin but also neuraminidase antigens in future seroepidemiologic cohort studies. Accurate estimates of true infection rates are critically important for modeling transmission and impact of influenza and guiding countermeasure strategies such as antivirals, vaccines, hospital use, and behavioral interventions. The comparative patterns of the antibody responses to the 2 most abundant and immunogenic antigens of influenza virus will improve understanding of immune correlates of protection and optimize pandemic and seasonal vaccine design and vaccination policies.

SHIVERS investigation team.

Research nurses at Auckland District Health Board (ADHB): Kathryn Haven, Bhamita Chand, Pamela Muponisi, Debbie Aley, Claire Sherring, Miriam Rea, Judith Barry, Tracey Bushell, Julianne Brewer, and Catherine McClymont. Research nurses at Counties Manukau District Health Board (CMDHB): Shona Chamberlin, Reniza Ongcoy, Kirstin Davey, Emilina Jasmat, Maree Dickson, Annette Western, Olive Lai, Sheila Fowlie, Faasoa Aupa’au, and Louise Robertson. World Health Organization National Influenza Centre, Institute of Environmental Science and Research: Pam Kawakami, Susan Walker, Robyn Madge, and Amanda des Barres. ADHB Laboratory: Fahimeh Rahnama. CMDHB Laboratory: Helen Qiao, Fifi Tse, Mahtab Zibaei, Tirzah Korrapadu, Louise Optland, Cecilia Dela Cruz. Labtests Laboratory in Auckland. Sentinel general practices: Avondale Family Doctor, Three Kings Family Medical Centre, East Tamaki Healthcare Bairds Road, East Tamaki Healthcare Dannemora, Dr D. M. de Lacey Surgery, Balmoral Doctors, Mount Eden Medical Centre, Mount Wellington Family Health Centre, Ranfurly Medical Centre, Blockhouse Bay Medical Centre, Westmere Medical Centre, Epsom Medical Centre, EastMed Doctors, and Pacifika Horizon Healthcare.