Flu

Scenario Modeling Hub

Technical report: The uncertain burden of COVID-19 and influenza in the upcoming flu season (October 25, 2022)

The technical report can be downloaded in a pdf format with supplemental plots.

Summary

As the 2022-2023 flu season approaches, there is substantial concern as to the potential for a “twindemic” from the combined influenza season and a new wave of SARS-CoV-2. This heightened concern comes in the context of multiple emerging SARS-CoV-2 variants of concern and the potential for a large influenza season due to limited transmission during the past two seasons. Of particular concern is the potential burden to healthcare facilities, with excessive hospitalizations straining health systems. To better understand what this burden might look like, we combined the ensemble estimates of hospitalizations from the latest rounds of the COVID-19 Scenario Modeling Hub (Round 15) and the Flu Scenario Modeling Hub (Round 1). As of October 25, 2022, all scenarios of COVID-19 Round 15 and the pessimistic immunity scenarios of Flu Round 1 are tracking well with observed hospitalizations from each virus.

Results

Combining COVID-19 and influenza hospitalization projections, we find that while there is substantial variability among the combinations of the various scenarios, in each combination, we expect to see a substantial burden on the healthcare system. In particular, with the most pessimistic COVID-19 scenario (i.e., a new SARS-CoV-2 variant with late boosters) combined with all scenarios of influenza, hospitalizations are projected to exceed the highest levels of weekly incident hospitalizations observed since the first Omicron surge (n=46,000 in July 2022). In the most pessimistic combination of scenarios, we project 68,000 peak hospitalizations (median ensemble estimates).

In scenarios assuming pessimistic prior immunity to influenza driven by limited transmission during the COVID-19 pandemics, we project large and early influenza seasons. In the most pessimistic influenza scenario in particular, which assumes low 2022-23 vaccination protection, influenza hospitalizations are projected to peak early, during the week of December 17 (50% PI, November 26-January 7), and in the most optimistic scenario where immunity is the highest of all scenarios considered, the influenza ensemble peaks in the week of January 14 (50% PI, December 3-January 28). Given our assumptions about circulation of new COVID-19 variants, a COVID-19 surge is projected to precede an influenza surge. A combined hospital load is projected to peak in December-January.

Caveats and limitations

These projections were produced by combining separate multimodel ensemble projections of COVID-19 and influenza. We do not account for any interaction between COVID-19 and influenza, which could include behavioral or immunological interactions that might modify the impacts of one or both of these viruses. Additionally, these projections were produced without empirical data on either influenza for the 2022-2023 season or on the currently emerging SARS-CoV-2 variants. Despite this, they are tracking well with observed hospitalizations from each virus.

Methods

We combined the most recent rounds of COVID-19 (Round 15) and influenza (Round 1) projections. The projection period for COVID-19 Round 15 was July 31, 2022 to May 6, 2023, and the scenario axes considered were the timing of the updated bivalent boosters (available from September 11 in the optimistic scenario versus November 13 in the pessimistic) and the emergence of a new variant of concern (no new variant beyond BA.5 versus an immune escape variant with increased severity emerging in Sept 2022). The projection period for Flu Round 1 was August 14, 2022 to June 3, 2023, and the scenario axes addressed vaccination protection (high or low) and assumptions around prior flu immunity (optimistic or pessimistic, with optimistic representing a typical influenza season and pessimistic being driven by 2 years of limited influenza transmission). The overlapping projection period for the two rounds covered August 14, 2022 to May 6, 2023.

Seven teams contributed scenario projections for COVID-19 Round 15, and ten teams for flu Round 1 scenarios. Ensembles of these scenarios were obtained, and the medians of these ensembles were combined to obtain an aggregate number of incident hospitalizations. We assumed independence of COVID-19 and flu, with no interactions between the pathogens or diseases, or the behaviors toward them.

Ensemble median projections - incident hospitalizations

The dashed horizontal line is the prior peak incident hospitalizations and deaths for influenza, from seasons 2012-13 to 2019-20. These seasons are taken from FluSurv-NET (which is used as a proxy for national hospitalizations). This is from the 2017-18 season. The dotted horizontal line is the highest national COVID peak since the Omicron surge (~46,000).

Round Scenario Specifications

Table 1. Flu Scenario Modeling Hub round 1 scenarios. More detailed scenario definitions and model characteristics can be found at https://github.com/midas-network/flu-scenario-modeling-hub.

Table 2. COVID-19 Scenario Modeling Hub round 15 scenarios. More detailed scenario definitions and model characteristics can be found at https://github.com/midas-network/covid19-scenario-modeling-hub.

A Note on Scenario Modeling Hub Round 1 (September 29, 2022)

A consortium of ten modeling groups convened to generate long-term scenario projections of hospitalizations and deaths that cover the period of 10 months from Aug 14, 2022 to June 3, 2023, across four scenarios. In this first round of influenza projections, we assessed the impact of reduced prior population immunity coming into the 2022-2023 influenza season as a result of decreased influenza circulation during the COVID-19 pandemic. We also assess the impact of low versus high vaccine-induced immunity (vaccine effectiveness combined with vaccination coverage)

Our main findings include:

  • There is large variability in the projected burden of the 2022-23 epidemic depending on vaccine and prior immunity assumptions; yet the 50% projection intervals of all scenarios support a larger cumulative burden this season compared to the 2021-22 winter and the lowest previous pre-pandemic season (2015-2016 season).
  • In the worst case scenario D, weekly hospitalizations are projected to peak at or above the highest pre-COVID-19 season (2017-2018), with an ensemble median of 35,800 nationally (50% PI, 17,300-53,400). In the best case scenario A, this is significantly reduced to 7,500 (50% PI, 4,900-17,800), which is still 2.1-fold higher than in the 2021-22 season.
  • With a large immunity gap due to COVID-19 (pessimistic immunity scenarios), hospitalizations are projected to be 27%-89% higher than with a typical level of immunity in the pre-COVID-19 period (optimistic immunity scenarios; range of ensemble medians across vaccine assumptions).
  • Increased vaccine effectiveness and coverage is expected to substantially decrease peak and cumulative hospitalizations regardless of existing population immunity, reducing hospitalizations by 45% in low prior immunity scenarios (representing around 165,000 hospitalizations averted) and 67% in high prior immunity scenarios (representing around 187,000 hospitalizations averted). Cumulative deaths would be reduced by around 67% and 76%, or 32,000 and 17,000 deaths averted, respectively.
  • In scenario D, where immunity from prior seasons and 2022-23 vaccination is at its lowest, ensemble hospitalizations are projected to peak in the week of December 17 (50% PI, November 26-January 7). In scenario A, where immunity is the highest of all scenarios considered, the ensemble peaks in the week of January 14 (50% PI, December 3-January 28).
  • A few caveats are worth noting:
    • There is substantial uncertainty in this first round of influenza projections due to lack of complete historical surveillance data from prior seasons. Further, the transmissibility of influenza strain(s) circulating in the 2022-23 season in the US is still unknown. As the season starts this uncertainty should reduce
    • There is also uncertainty in the amount of influenza reporting in the coming months as testing practices for respiratory viruses have changed during the COVID-19 pandemic.
    • The trajectories of individual models are asynchronous, likely due to differences in seasonality assumptions, among others. This flattens the ensemble median and 50% PI for incident hospitalizations and deaths. Ensemble estimates of the peak size and cumulative burden are less affected by differences in the timing of individual models and are considered a more reliable indicator of the potential impact of the 2022-23 season than the trajectory ensembles.
    • Scenarios did not consider immunological interactions with SARS-CoV-2 or reactive behavior changes and interventions in response to a new SARS-CoV-2 variant that may arise in the 2022-23 respiratory virus season, either of which could affect the transmission and disease burden of influenza.
    • These hospitalization projections represent the expected number of influenza hospitalizations reported to the HHS system. These are not meant to reflect the final CDC estimates from the pyramid approach, which takes into account underreporting and various delays.
    • The most pessimistic assumed vaccine effectiveness in these scenarios was VE=30% against hospitalization. It should be noted that recent seasons have had substantially lower estimated seasonal VE (e.g., 2014-15 with VE=19% overall), thus a substantial mis-match of the vaccine to circulating influenza strains could drive higher transmission than these scenarios.
    • These projections were made with a data cutoff of August 14, 2022; no data after that day was to be used to calibrate or otherwise inform the model.

Table 1. Flu Scenario Modeling Hub round 1 scenarios. More detailed scenario definitions and model characteristics can be found at https://github.com/midas-network/flu-scenario-modeling-hub.


Rationale

Even the best models of emerging infections struggle to give accurate forecasts at time scales greater than 3-4 weeks due to unpredictable drivers such as a changing policy environment, behavior change, the development of new control measures, and stochastic events. However, policy decisions around the course of emerging infections often require projections in the time frame of months. The goal of long-term projections is to compare outbreak trajectories under different scenarios, as opposed to offering a specific, unconditional estimate of what “will” happen.

As such, long-term projections can guide longer-term decision-making while short-term forecasts are more useful for situational awareness and guiding immediate response.