Last updated: 23-12-2023 for Round 1 Scenarios.
Even the best models of infectious disease transmission struggle to give accurate forecasts at time scales greater than 3-4 weeks due to unpredictable drivers like changing policy environments, behavior change, development of new control measures, and stochastic events. However, policy decisions around the course of infectious diseases, particularly emerging and seasonal 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.
We have specified a set of scenarios and target outcomes to allow alignment of model projections for collective insights. Scenarios have been designed in consultation with academic modeling teams and government agencies (e.g., CDC).
This repository follows the guidelines and standards outlined by the hubverse, which provides a set of data formats and open source tools for modeling hubs.
The RSV Scenario Modeling Hub is open to any team willing to provide projections at the right temporal and spatial scales, with minimal gatekeeping. We only require that participating teams share point estimates and uncertainty bounds, along with a short model description and answers to a list of key questions about design. A major output of the projection hub is ensemble estimates of epidemic outcomes (e.g., infection, hospitalizations, and deaths), for different time points, intervention scenarios, and US jurisdictions.
Those interested to participate, please read the README file and email us at scenariohub@midasnetwork.us .
Model projections should be submitted via pull request to the model-output/ folder and associated metadata should be submitted at the same time to the model-metadata/ folder of this GitHub repository. Technical instructions for submission and required file formats can be found here, here, for the metadata file and in the Wiki.
The primary goal of SMH RSV Round 1 is to build capacity within the hub to model Respiratory Syncytial Virus (RSV) dynamics in the US and assess the feasibility of modeling this pathogen at the scale of the US given limited availability of epidemiological data. A secondary goal is to use nascent RSV modeling capabilities to generate ensemble projections for the impact of new interventions that are coming online during the 2023-24 season. We will consider 5 scenarios in total, following a 2*2 table describing the impact of long-acting monoclonals targeted at infants (first dimension, optimistic and pessimistic protection) and senior vaccination (second dimension, optimistic and pessimistic protection). A 5th counterfactual scenario will consider status quo RSV mitigation. Projections will be generated for a 29-week period, running Sun Nov 12, 2023 to Sat June 1, 2024. The scenario structure is as follows:
Weekly cumulative age-specific coverage for vaccines and monoclonals are provided.
We describe important details of the planned implementation of RSV interventions below as well as our rationale for vaccine coverage and effectiveness assumptions.
Long-acting monoclonal antibodies (nirsevimab) were recommended for prophylactic use in infants on August 3, 2023, and an update due to limited availability in the 2023-24 season was issued on October 23, 2023. Current recommendations are that all infants aged <6 months who are born during or entering their first RSV season should be prioritized to receive the new monoclonals. Older children up to 19 months who are at increased risk for severe RSV disease should not be prioritized for the new monoclonals. Older children at high risk will continue to receive the older monoclonals used in prior years. We require that teams implement new monoclonal interventions in infants <6 months, while explicit consideration of interventions in older high-risk babies (a small fraction of all US babies, see later) is at teams’ discretion. The new long-acting monoclonals will be administered in most of the continental United States from October through the end of March, the months of highest RSV activity.
For infants < 6 months the process is as follows:
- during the RSV season Oct 2023-Mar 2024, a fraction of all newborns will receive long-acting monoclonals at birth (fraction based on weekly monoclonal coverage)
- a fraction of infants who are born from Apr-Sep 2023 and are <6 months at the start of the season will receive a catch up dose of monoclonals during the RSV season, Oct 2023-Mar 2024 (also based on weekly coverage). No infant >= 6 months can receive monoclonals.
Senior RSV vaccination was recommended in July 2023 for individuals over 60 yrs. We recommend that teams consider vaccination in all individuals 60+ yrs, even though the hospitalisation target data is only available for individuals 65+ yrs. However the exact implementation of targeting 60+ yrs is left at teams’ discretion. Vaccination will proceed seasonally, similarly to the flu vaccine, between Sep 2023 (slightly slower start than flu vaccine) and June 2024. Because RSV is not a well known pathogen among seniors, and recommendations stipulate that vaccination should occur after consultation with a physician (unlike flu), we assume that the intervention will have limited uptake.
We will index the coverage of RSV interventions on flu vaccine coverage in a given state and relevant age group to reflect different propensities of different states to adopt new health interventions. The roll out of RSV interventions is likely to be slow since it is the first year interventions have been approved, and there are shortages and fee reimbursement issues for infant interventions.
CDC flu vaccine curves from 2021-22 will be used to estimate anticipated coverage of RSV interventions (similarly to recent flu and COVID-19 rounds). Specifically, for infants, weekly interpolated flu coverage will be shifted by 2 months (monoclonals will start on Oct 1, 2023) and administration will end at the end of March 2024 to reflect i) a late start of RSV interventions in the 2023-24 season and ii) CDC recommendations. For seniors, we will use the actual timing of flu vaccination shifted by only one month since the RSV vaccine was recommended in the summer. For both infants and seniors, weekly flu vaccine estimates will be adjusted for the intended level of RSV coverage saturation. We generally expect real coverage in 2023-24 to track with our pessimistic coverage assumptions, while we have chosen optimistic levels of coverage that would reflect potential benefits in a future season with no shortage and more awareness of these interventions. Senior and infant vaccination coverage curves are provided for all projection weeks and target locations, in the Github auxiliary data folder
We stipulate VE against hospitalizations for both senior vaccination and infant
monoclonals, since severe RSV disease is the primary endpoint of most clinical
trials (generally lower respiratory tract infection or RSV-associated
hospitalizations, see literature below).
Scenario assumption values are based on optimistic and pessimistic
interpretation of the randomized control trial data. For seniors, VE values
considered are 70% (scenarios B, D) and 90% (scenarios A, C).
For infants, VE values considered are 60% (scenarios C, D) and
80% (scenarios A, B).
RCTs – senior vaccines:
- GSK, Papi et al
- Pfizer, Walsh et al
RCTs in infants - Nirsevimab:
- Full term infants, Hammitt et al
- Pre-term infants, Griffin et al
It is at teams’ discretion to proportionate VE values into protection against infection, protection against severe disease given infection, and any effect on transmission. However note that the current thinking and available data suggests a very limited protection against infection, Wilkins et al, if any. There is no data on transmission and the possible impact on these interventions on RSV shedding, so a small to moderate transmission effect cannot be ruled out.
We do not prescribe a specific duration of protection for senior vaccination or infant monoclonals. Teams can refer to existing literature cited in the above section. As a general guideline, monoclonals are expected to provide protection for a mean of 6 months, while senior protection remains relatively stable within 2 years of vaccination.
None of the scenarios consider maternal vaccination (to protect infants after birth) since maternal interventions are not expected to have high enough coverage to make a significant impact on the 2023-24 season at the time of scenario specification. Teams should not model maternal vaccines. Consideration of non-specific RSV interventions such as a low level of residual masking is allowed.
In this scenario, we consider no change to the historic policy of RSV mitigation, which consists in a limited coverage of palivizumab monoclonals to high-risk premature infants (~2% of the US birth cohort receives a partial or full dose, Ambrose et al). The calibration data available from 2017-present takes into account the impact of this intervention. Teams have discretion to consider this policy explicitly or ignore it given the small fraction of infants covered. We note that high risk premature infants <6mo who previously would have received palivizumab (the older treatment) will now receive the new monoclonal nirsevimab, with comparable effectiveness. Due to shortages in 2023-24, older high risk babies will keep receiving palivizumab this season.
calibration process
It is important that all scenario simulations share the same values (or ranges of values) for intrinsic transmissibility and severity of RSV infection at the start of projections (that is, severity and transmissibility in the absence of interventions, where severity is risk of hospitalization given infection). To do so, teams should ensure that the calibration step uses similar transmissibility and severity parameters across scenarios. Only by using assumptions that lead to starting conditions for transmissibility and severity that effectively come from a shared distribution across scenarios, we can evaluate the impact of different interventions in the projection period.
We suggest that teams use one of the 3 following options for calibration (i) choose one of the 5 proposed scenarios for calibration (ideally, the scenario that seems the most plausible given what is known about interventions until the cut-off date on Nov-11), or (ii) use real-world coverage in infants and seniors that has been reported until the cut-off date in the calibration step. Once transmissibility and severity have been estimated, go back to the start of interventions prescribed in scenarios A-E (ie, early Oct) and apply assumed level of interventions in each scenario until Nov 11 and throughout the projection period. iii) use a shared distribution of parameters that is consistent with all scenarios.
Fig 1: RSV vaccination coverage ramp up in seniors 60+ yrs
Fig 2: RSV monoclonals usage ramp up in infants <6 mo throughout the RSV season Oct-Mar
Age- and state-specific data on laboratory-confirmed RSV hospitalization rates are available for 12 states and the US from RSV-NET spanning 2017-18 to present (RSV-NET CDC Webpage). RSV-NET is an RSV hospitalization surveillance network that collects data on laboratory-confirmed RSV-associated hospitalizations through a network of acute care hospitals in a subset of states (12 states as of August 2023). Age-specific weekly rates per 100,000 population are reported in this system.
The data has been standardized and posted on the SMH RSV github target-data/ folder and is updated weekly. The target in this data is the weekly number of hospitalizations in each given state (inc_hosp variable), for all ages and by age group. To obtain counts, we have converted RSV-NET weekly rates based on state population sizes. This method assumes that RSV-NET hospitals are representative of the whole state. To obtain national US counts, we have used the rates provided for the “overall RSV-NET network”. The data covers 2017-present. Reported age groups include: [0-6 months[, [6-12 months[, [1-2 yr[, [2-4 yr[, [5-17 yr[, [18-49 yr[, [50-64 yr[, and 65+ years. The standardized dataset provided by SMH includes week- state- and age-specific RSV counts (the target), rates, and population sizes. Note that different states joined RSVnet in different years (between 2014 and 2018) while RSV surveillance throughout the network was initially limited to adults. Children RSV surveillance began in the 2018-19 season.
The source of age distribution used for calibration (RSV-NET vs other estimates) should be provided in the abstract metadata that is submitted with the projections.
A few auxiliary datasets have been posted in the GitHub repositority auxiliary-data/ folder including:
- state-specific CDC surveillance from NVERSS (only last year of data available)
- state-specific ED data (only last year of data available)
In this round, we will require submission of 100 individual trajectories for each target while submission of quantiles is optional. Targets will be based on the RSV-NET dataset. The required targets for trajectories will be weekly RSV incident hospital admissions (counts) in the 12 RSV-NET states, nationally for all ages, and for a set of minimal age groups; a more resolved set of age groups is strongly encouraged (see below). Estimates of cumulative counts can be obtained from weekly trajectories and hence we do not require trajectories for cumulative counts. Similarly, peak targets (peak hospital admission magnitude and peak timing) can be reconstructed from weekly trajectories. Teams who wish to submit quantiles along with trajectories should provide quantiles for weekly and cumulative counts, as well as for hospital admission peak size and peak timing.
- Weekly reported all-age and age-specific state-level incident hospital admissions, based on RSV-NET. This dataset is updated daily and covers 2017-2023. There should be no adjustment for reporting (=raw data from RSV-NET dataset to be projected). A current and standardized version of the weekly data has been posted here
- No infection target
- No case target
- No death target
- All targets should be numbers of individuals, rather than rates
- Hospital admissions should be provided for the following age groups: all ages, <1 yr, 1-4, 5-64, and 65+. (Most of the RSV burden on hospitalizations comes from the 0-1 and 65+ age groups)
- Weekly state-specific and national RSV hospitalizations among individuals <1 yr, 1-4, 5-17, 18-49, 50-64, 65+, and all ages
- Cumulative hospital admissions. Cumulative outcomes start at 0 at the start of projections, on Nov 12, 2023
- State-level peak hospital admissions
- State-level timing of peak hospital admission
- Scenarios set (no changes after): Friday, Oct 27, 2023
- Projections due: Tuesday, Nov 14, 2023
- Report finalized: No later than November 24, 2023
Several reference studies are worth considering to set (or guide) RSV model
parameters that cannot be estimated from the available hospitalization data.
These include work by Ginny Pitzer and colleagues in the US (see
Pitzer et al for state-specific
models driven by environmental drivers in the pre-intervention era, including
Table 2 for parameters; and
Zheng et al for an
updated model with interventions).
Risk of severity given infection was parametrized in these models based on
children cohort studies in the US and Kenya: see
Nokes et al,
Glezen et al, and
Breese Hall at al.
Since these studies have been published, there has been an increased recognition of the burden of RSV among seniors (see Jackson et al, Rha et al, and McLaughlin et al
- Prior immunity is at each team’s discretion. Immunity against infection is waning rapidly for RSV; based on prior modeling work, estimates of duration of immunity against infection range between 200-365 days. However, immunity against severe disease can be more long-lasting and generally increases with age and number of prior infections (eg see discussion in Pitzer et al). Overall, most individuals will get reinfected multiple times throughout life, but severe RSV infections that lead to hospitalizations tend to only occur among young children and seniors. Maternal immunity is expected to be brief
- Teams are allowed to vary prior immunity by age or other demographic characteristic, and state
- Unlike influenza virus, antigenic evolution is not a key feature of RSV
- No major interactions with future COVID-19 and flu surges (e.g., immunological, social, behavioural) should be considered in this round
- We note that many studies have reported that RSV circulation was perturbed during the COVID19 pandemic, as can be seen in the RSVnet data. Whether and how to fit the COVID19 pandemic period is left at teams discretion
- Sun Nov 12, 2023 to Sat June 1, 2024 (29 weeks)
- Variability in severity and reporting to RSV-NET between states is possible
Teams should include their best estimates of RSV seasonality in their model but we do not prescribe a specific level of seasonal forcing.
No reactive NPIs to COVID-19 or influenza should be modeled in this round; low level masking is allowed at groups’ discretion.
We leave seeding intensity, timing and geographic distribution at the discretion of the teams. In addition to the RSV-NET hospital admission dataset, CDC’s NVERSS viral surveillance dataset is a good resource for state-specific information on epidemic intensity (e.g., weekly % positive, or weekly ILI*%positive), and can be used to adjust seeding.
Prior immunity and amount of infections at the start of the projection period is at the discretion of the teams based on their interpretation of the scenarios. Variation in initial prevalence between states is left at teams’ discretion.
All of the teams’ specific assumptions should be documented in meta-data and abstract.
| Scenario | Scenario name | Scenario ID for submission file (scenario_id) |
|---|---|---|
| Scenario A. Optimistic protection from both senior and infant interventions | optInf_optSen | A-2023-10-27 |
| Scenario B. Optimistic protection from infant monoclonals, Pessimistic protection from senior vaccination | optInf_pessSen | B-2023-10-27 |
| Scenario C. Pessimistic protection from infant monoclonals, Optimistic protection from senior vaccination | pessInf_optSen | C-2023-10-27 |
| Scenario D. Pessimistic protection from both senior and infant interventions | pessInf_pessSen | D-2023-10-27 |
| Scenario E. Counterfactual | counter_fact. | E-2023-10-27 |
- Projection Due date: Tuesday, Nov 14, 2023
- End date for fitting data: Between Saturday Oct 28, 2023 and Saturday Nov 11, 2023
- Start date for scenarios: Sunday November 12, 2023 (first date of simulated transmission/outcomes)
- Simulation end date: June 1, 2024 (29-week horizon)
- Desire to release results by late-November 2023
-
Simulation trajectories: We ask that teams submit a sample of 100 simulation replicates. Simulations should be sampled in such a way that they will be most likely to produce the uncertainty of the simulated process. For some models, this may mean a random sample of simulations, for others with larger numbers of simulations, it may require weighted sampling. Trajectories will need to be paired across age groups (eg, for a given model, location, scenario and week, all age data for simulation 1 corresponds to the sum of age-specific estimates for simulation 1).
-
Geographic scope: state-level and national projections
- 12 states or a subset of 12 states, US overall recommended.
-
Results:
- Summary: Results must consist of a subset of weekly targets listed below; all are not required.
- Weeks follow epi-weeks (Sun-Sat) dated by the last day of the week.
- Weekly Targets: Weekly incident hospitalizations by location, all ages and age-specific
-
Metadata: We will require a brief meta-data form, from all teams.
-
Uncertainty:
- For trajectories (required submission): we require 100 trajectories.
- For quantiles (optional submission) We ask for 0.01, 0.025, 0.05, every 5% to 0.95, 0.975, and 0.99. Teams are also encouraged to submit 0 (min value) and 1 (max) quantiles if possible.
The target-data/ folder contains the RSV hospitalization data (also called "truth data") standardized from the Weekly Rates of Laboratory-Confirmed RSV Hospitalizations from the RSV-NET Surveillance System.
The weekly hospitalization number per location are going to be used as truth data in the hub.
The repository stores and updates additional data relevant to the RSV modeling efforts in the auxiliary-data/ folder:
-
Population and census data:
- National and State level name and fips code as used in the Hub and associated population size.
- State level population size per year and per age from the US Census Bureau.
-
Birth Rate:
- Birth Number and Rate per state and per year from 1995 to 2022 included.
- Data from the US Census Bureau and from the Centers for Disease Control and Prevention, National Center for Health Statistics. National Vital Statistics System, Natality on CDC WONDER Online Database.
-
RSV data:
- The National Respiratory and Enteric Virus Surveillance System (NREVSS) data at national and state level.
- The Weekly Rates of Laboratory-Confirmed RSV Hospitalizations from the RSV-NET Surveillance System
- The National Emergency Department Visits for COVID-19, Influenza, and Respiratory Syncytial Virus
For more information, please consult the associated README file.
All source code that is specific to the overall project is available under an open-source MIT license. We note that this license does NOT cover model code from the various teams, model scenario data (available under specified licenses as described above) and auxiliary data.
Those teams interested in accessing additional computational power should contact Katriona Shea at k-shea@psu.edu. Additional resources might be available from the MIDAS Coordination Center - please contact questions@midasnetwork.us for information.
-
CEPH Lab at Indiana University - MetaRSV
- Marco Ajelli (Indiana University Bloomington), Shreeya Mhade (Indiana University Bloomington), Paulo C. Ventura (Indiana University Bloomington), Maria Litvinova (Indiana University Bloomington), Snigdha Agrawal(Indiana University Bloomington)
-
Columbia University — RSV_SVIRS
- Teresa Yamana (CU), Sen Pei (CU)
-
Johns Hopkins University and University of North Carolina at Chapel Hill - flepiMoP
- Shaun Truelove (JHU), Alison Hill (JHU), Justin Lessler (UNC), Sara Loo (JHU), Joseph Lemaitre (UNC), Anjalika Nande (JHU), Anne Hebert (JHU), Madeleine Gastonguay (JHU), Sung-mok Jung (UNC), Vivek Murali (JHU)
-
MOBS Lab at Northeastern University - GLEAM RSV Model
- Alessandro Vespignani (Network Science Institute, NEU), Matteo Chinazzi (The Roux Institute, NEU, Portland (ME)), Jessica T. Davis (Network Science Institute, NEU), Clara Bay (Network Science Institute, NEU), Guillaume St-Onge (Network Science Institute, NEU)
-
National Institutes of Health - RSV_MSIRS
- Kaiyuan Sun (Fogarty International Center, NIH), Cécile Viboud (Fogarty International Center, NIH)
-
National Institutes of Health - RSV_Phenomenological
- Kaiyuan Sun (Fogarty International Center, NIH), Cécile Viboud (Fogarty International Center, NIH)
-
National Institutes of Health - RSV_WIN
- Samantha Bents (Fogarty International Center, NIH), Cécile Viboud (Fogarty International Center, NIH), Chelsea Hansen (Fogarty International Center, NIH)
-
Predictive Sciences - Package for Respiratory Disease Open-source Forecasting
- James Turtle (Predictive Science Inc), Michal Ben-Nun (Predictive Science Inc), Pete Riley (Predictive Science Inc)
-
University of Notre Dame - FRED
- Sean Moore (UND), Alex Perkins (UND, Guido Espana (CDC Center for Forecasting and Analysis)
-
University of Southern California - SIkJalpha
- Ajitesh Srivastava (USC), Majd Al Aawar (USC)
-
University of Texas at Austin - UT-ImmunoSEIRS
- Kaiming Bi (UTA), Shraddha Ramdas Bandekar (UTA), Anass Bouchnita (The University of Texas at El Paso), Spencer J. Fox (The University of Georgia), Lauren Ancel Meyers (UTA)
-
University of Virginia - EpiHiper Scenario Modeling for RSV
- Jiangzhuo Chen (UVA), Stefan Hoops (UVA), Bryan Lewis (UVA), Srini Venkatramanan (UVA), Lacey Critchley (UVA), Parantapa Bhattacharya (UVA), Dustin Machi (UVA), Madhav Marathe (UVA)
- Shaun Truelove, Johns Hopkins University
- Cécile Viboud, NIH Fogarty
- Justin Lessler, University of North Carolina
- Sara Loo, Johns Hopkins University
- Lucie Contamin, University of Pittsburgh
- Emily Howerton, Penn State University
- Claire Smith, Johns Hopkins University
- Harry Hochheiser, University of Pittsburgh
- Katriona Shea, Penn State University
- Michael Runge, USGS
- Erica Carcelen, John Hopkins University
- Sung-mok Jung, University of North Carolina
- Jessi Espino, University of Pittsburgh
- John Levander, University of Pittsburgh
- Samantha Bents, NIH Fogarty
- Katie Yan, Penn State University