Estimating historical disease burden and the impact of vaccination by influenza type and subtype in the United States, 2016-2020

Surveillance data

We used surveillance data spanning the 2016/17 to 2019/20 influenza seasons. Unless otherwise specified, data were aggregated into six age groups: 0–6 months, 6 months–4 years, 5–17 years, 18–49 years, 50–64 years, and ≥65 years. Monthly numbers of laboratory-confirmed influenza-associated hospitalizations were obtained from the Influenza Hospitalization Surveillance Network (FluSurv-NET), a population-based hospitalization surveillance network that represents approximately 29 million people in the United States⁵. These data included the total number of hospitalizations, the number of hospitalizations attributed to influenza A and influenza B, and the proportions of influenza A hospitalizations that were attributed to A(H1N1) and A(H3N2) (Tables S1–S2, Figure S1). The latter proportions incorporated imputed influenza A subtype information (which was originally missing for approximately 55-60% of influenza A cases each season)^6,7,8. FluSurv-NET also provided information on the total catchment population for each age group⁹. To account for under-ascertainment of laboratory confirmed influenza hospitalizations, we calculated age-specific under-detection multipliers based on testing practice information from FluSurv-NET sites and diagnostic test sensitivity⁹.

In addition to the distributions of virus type and subtype among influenza-associated hospitalizations, we obtained corresponding distributions for laboratory-confirmed influenza-associated outpatient visits from the US Influenza Vaccine Effectiveness (Flu VE) Network¹⁰ (Table S2, Figure S2). Since this system recruits participants for vaccine effectiveness studies, there were no distributions for infants aged 0–6 months (who are not eligible for influenza vaccination). We therefore assumed distributions in that age group were equivalent to distributions among children aged 6 months–4 years. We verified that the distributions from the Flu VE Network and FluSurv-NET were similar to independent data obtained from the World Health Organization (WHO) and National Respiratory and Enteric Virus Surveillance System (NREVSS) Collaborating Labs and reported to CDC’s FluView (Figure S3)¹¹.

Finally, vaccine effectiveness estimates for A(H1N1), A(H3N2), influenza B, and all influenza viruses combined were obtained from the Flu VE Network, and vaccine coverage estimates were obtained from FluVaxView for all age groups ≥6 months¹² (Table S1, Figure S4). We approximated the population size for each age group and season using US Census data¹³.

Framework for estimating disease burden for influenza A(H1N1), A(H3N2), and B

The methods employed by CDC to estimate disease burden and burden prevented by vaccination for all influenza (i.e., for all viruses combined) have been described previously^3,4,14–17 (Figure 1A). Briefly, the monthly number of influenza-associated hospitalizations for each age group are first translated to rates by dividing by the FluSurv-NET catchment population. These rates are then corrected for under-detection of influenza-related hospitalizations and extrapolated to estimate total influenza hospitalizations in the US population using the age-specific US census estimates. Numbers of symptomatic influenza illnesses and deaths are obtained by multiplying the extrapolated number of hospitalizations by age-specific ratios for the number of symptomatic influenza cases per hospitalization (case-hospitalization ratio) and deaths per hospitalization (death-hospitalization ratio), respectively^3,18,16. The number of medically-attended illnesses are obtained by multiplying the number of symptomatic illnesses by age-specific estimates of the fraction of people with influenza-associated illness that seek medical care¹⁹. Values for the under-detection, case-hospitalization, death-hospitalization, and medically-attended multipliers are provided in Table S3.

• Download figure
• Open in new tab

Figure 1. Framework for estimating disease burden.

(A) Framework for estimating age-specific burden estimates for all influenza viruses. For each age group, reported hospitalizations rates are corrected for under-detection and extrapolated to the total population. These are then translated to estimates of deaths and symptomatic illnesses using age-specific death-hospitalization ratios and case-hospitalization ratios, respectively. Symptomatic illnesses are translated to medically-attended illnesses using age-specific estimates of the fraction that are medically-attended. (B) Burden estimates are translated to type or subtype specific estimates using virus distributions reported among influenza-positive hospitalizations (from FluSurv-NET) or outpatient cases (from the Flu VE Network).

We extended these methods to generate virus type- and subtype-specific estimates (Figure 1B). For each month and age group, hospitalizations were attributed to either influenza A or B using the relative proportions of each virus type in FluSurv-NET. Hospitalizations attributed to influenza A were further partitioned by subtype using the relative proportions of A(H1N1) and A(H3N2). We used the same type and subtype proportions to partition the total number of deaths into those attributable to A(H1N1), A(H3N2) and B. For symptomatic and medically-attended illnesses, we instead used the relative type and subtype proportions among Flu VE Network outpatient cases as these more accurately reflect virus distributions among non-hospitalized cases of influenza. When presenting results, we combined estimates for the two youngest age groups (0–6 months and 6 months–4 years) into all children aged 0–4 years.

Framework for estimating burden prevented by vaccination for influenza A(H1N1), A(H3N2) and B

For each virus type or subtype (A(H1N1), A(H3N2), B, or all influenza viruses combined), the monthly, age-specific estimates of disease burden were combined with estimates of vaccine coverage and effectiveness to estimate the burden prevented by vaccination following previously described methods⁴ (Figure 2). We excluded the 0–6 months age group since they are ineligible for vaccination. The population was divided into three categories, susceptible, effectively vaccinated, and (symptomatically) infected as follows. For each age group, a, and type or subtype, i, the number effectively vaccinated (EV) in each successive month m is calculated as where P_a is the starting population size (assuming everyone is susceptible at the beginning of the season), VC_a,m is the proportion vaccinated in that month, VE_a,i is the vaccine effectiveness against type or subtype i, and is the cumulative number of symptomatic illnesses estimated during the disease burden calculations and occurring up to the previous month (with C_{a,i, 0} = 0). This formulation assumes that vaccination provides all-or-nothing protection (i.e., individuals either gain complete immunity following vaccination or remain entirely susceptible) and that susceptible and infected individuals are equally likely to be vaccinated. The number susceptible (S) is then estimated as with the implicit assumption that individuals can be infected by more than one virus type or subtype within a season.

• Download figure
• Open in new tab

Figure 2. General framework for estimating disease burden prevented by vaccination within each age group and virus type or subtype for a given season.

Independent calculations are performed for each age group and virus type or subtype. The procedure is performed iteratively for each month (denoted by the subscript ‘m’). Abbreviations: VE = vaccine effectiveness.

The risk of symptomatic illness (R) is the number of symptomatic illnesses that occurred in the current month, m, divided by the susceptible (i.e., at-risk) population in the previous month, m–1, Similarly, the risks of medically-attended illness, hospitalization, and death are approximated by dividing the value in month m (estimated during the disease burden calculations) by the susceptible population in month m–1. We use these risks to estimate the number of symptomatic illnesses, medically-attended illnesses, hospitalizations, and deaths in the absence of vaccination. The estimated number of symptomatic illnesses without vaccination, Ĉ, is where the susceptible population without vaccination, Ŝ is now calculated iteratively as with Ŝ_a,i,0 = P_a and Ĉ _a,i,0 = 0. The number of symptomatic illnesses prevented by vaccination is then Ĉ_a,i,m − C_a,i,m. Analogous calculations were used to estimate the number of medically-attended illnesses, hospitalizations, and deaths prevented by vaccination. For each variable, age group, and type or subtype, we calculated the sum of burden prevented by vaccination across all months in the season. We also calculated disease burden prevented by vaccination as a percentage of the total estimated disease burden in the absence of vaccination, for example, . Finally, we note that estimates of burden prevented by vaccination for all influenza viruses combined may not always equal the sum of burden prevented for A(H1N1), A(H3N2), and B viruses because we use independent estimates of VE in each calculation.

Uncertainty distributions

We generated uncertainty distributions for each disease burden and vaccine-prevented burden estimate using Monte Carlo simulation methods described previously^20,21,22. Key input variables were each assigned a probability distribution (Table S4), from which we generated 10,000 independent samples. We then re-performed the disease burden and vaccine-prevented burden calculations for each sample and reported uncertainty as the 95^th percentiles of the resulting distribution. All analyses were performed in R version 4.2.0²³ and resulting estimates are reported to three significant figures.

Estimated disease burden

For the 2016/17 to 2019/20 seasons, total estimated influenza disease burden ranged from 29.4–41.0 million symptomatic illnesses, 13.6–18.9 million medically-attended illnesses, 395,000–710,000 hospitalizations, and 22,100–51,800 deaths. The greatest number of hospitalizations and deaths occurred in 2017/18 (Figure 3A), and across all seasons the greatest proportion of hospitalizations and deaths occurred among adults ≥65 years (Figure S5).

• Download figure
• Open in new tab

Figure 3. Estimated disease burden.

(A) Point estimates of disease burden by type/subtype and season. (B) Distribution of type/subtypes causing burden in each age group, for all four seasons. Estimates by age group, season, and type/subtype, with corresponding 95^th percentile uncertainty intervals, are shown in Figure S6.

Among virus types and subtypes, A(H3N2) caused the greatest cumulative burden from 2016/17–2019/20, with 59.0 million total symptomatic illnesses, 27.3 million medically-attended illnesses, 1.12 million hospitalizations and 86,700 deaths (Table 1). However, the contribution of each type and subtype to overall disease burden varied by season and age group (Figure 3, Figure S6). Influenza A(H3N2) and B contributed most burden in 2016/17 and 2017/18, whereas A(H1N1) and A(H3N2) contributed most burden in 2018/19, and A(H1N1) and B contributed most burden in 2019/20 (Figure 3A). For all age groups, influenza A viruses comprised the greatest proportion of burden across all seasons, although the proportion attributed to influenza B viruses increased slightly among children <18 years (Figure 3B).

Estimated burden prevented by vaccination

Vaccine coverage was highest among children 6 months–4 years and adults ≥65 years, and during the 2018/19 and 2019/20 seasons (Table S1, Figure S4A). VE varied substantially by season, age group, and virus type or subtype. Broadly, VE was highest in 2017/18 and lowest in 2018/19, decreased with increasing age, and was lower against A(H3N2) than A(H1N1) and B (Table S1, Figure S4B). One notable exception occurred in 2019/20, when adults ≥65 years experienced relatively high VE against A(H1N1) and A(H3N2).

From 2016/17 to 2019/20, total estimated burden prevented by vaccination ranged from 2.85– 6.74 million symptomatic illnesses, 1.49–3.28 million medically-attended illnesses, 35,800–98,500 hospitalizations, and 2,520–6,370 deaths. The greatest burden prevented by vaccination occurred in 2019/20 (Figure 4A). For all seasons, the greatest number of hospitalizations were prevented among adults ≥65 years, whereas the greatest percent of hospitalizations were generally prevented among children 6 months–4 years (Figure S7).

• Download figure
• Open in new tab

Figure 4.

Estimated burden prevented by vaccination. (A) Point estimates for total averted burden by type/subtype and season. (B) Percent of hospitalizations averted by type/subtype and season (for all age groups). (C) Percent of hospitalizations averted by type/subtype and age group (across all four seasons). Estimates by age group, season, and type or subtype, with corresponding 95^th percentile uncertainty intervals, are shown in Figure S8.

Among type and subtypes, the largest cumulative number of hospitalizations and deaths were prevented by vaccination against influenza A(H3N2), whereas the largest number of symptomatic and medically-attended illnesses were prevented by vaccination against influenza B (Table 2). In line with our disease burden estimates, vaccination against influenza A(H3N2) and B prevented most burden in 2016/17 and 2017/18, whereas vaccination against A(H1N1) and A(H3N2) prevented most burden in 2018/19, and vaccination against A(H1N1) and B prevented most burden in 2019/20 (Figure 4A, Figure S8). Among seasons, the percent of hospitalizations prevented was generally highest for A(H1N1) and B, with substantially lower percentages for A(H3N2) (except in 2019/20 when the overall burden of A(H3N2) was low) (Figure 4B). This was also true among age groups, with vaccination against influenza B preventing the highest percentage of hospitalizations among children and adults <65 years, and vaccination against A(H1N1) preventing the highest percentage among adults ≥65 years (Figure 4C).