Changelog • sircovid

Added keep_strains_Rt option to Rt calculation, allowing return of both the strain-specific Rts and the weighted Rt together

Severity now calculated based on daily incidence instead of step incidence

D_hosp is now disaggregated by age and vaccine class

New historic variants layer added to R compartment. These individuals can now get reinfected with either strain 1 or strain 2. Those recovered from strain 1 (including from a reinfection) can get reinfected with strain 2, while those recovered from strain 2 are immune to strain 1

Reinfections move to PCR pre-positive instead of PCR positive

Add severity outputs by vaccine class and age

Can now fit to ONS prevalence data

Add severity outputs by strain

Add logical flag to turn on/off exporting of severity outcomes and weight parameter in combine_rt function, to use more flexibly for Rt and severity aggregation

Multiple vaccine skip moves now allowed and lancelot_parameters now allows multiple boosters

Add logical flag to turn on/off exporting all death classes by age

Fit to REACT prevalence by age

Fix compatibility with mac M1

Fit to hospital admissions by age

Fit to deaths in the community by age

gamma_U is now time-varying

IFR post-processing calculation removed

Fit to deaths in hospital by age

Have strain_rel_p_death, strain_rel_p_icu, strain_rel_p_hosp_if_sympt and strain_rel_p_sympt as separate parameters in odin code to clarify effects of strains and vaccines on severity and so these parameters can be changed more easily without necessarily going through the lancelot_parameters function

Remove vestiges of “initial step” control, now we do this directly from the particle filter data

Add ability to change duration of latency period (E) for strain 2 in lancelot_parameters

Add new function vaccine_eligibility for computing lancelot-compatible vaccine eligibility vectors based on a minimum age.

Support for the multistage particle filter added

New function rotate_strains to allow cycling of strains

modify_severity now caps at 1 for all vaccine strata

Corrected implementation of cross immunity in effective Rt calculation

Fixed issue #365 and simulation to the UK level is now restored

Fixed initial seeding of the epidemic replace by a Poisson seeding

Add an extra vaccination move between vaccine strata vacc_skip_from and vacc_skip_to, for the purpose of allowing boosting before waning

Ensure booster_population_left is rounded in vaccine_schedule_future

vaccine_schedule_from_data function refactored to deal with flexible number of doses

carehomes model removed from the package

D_inc now returned as a state from odin for daily deaths (deaths_inc)

p_NC and phi_pillar2_cases now must be age-specific for lancelot model, including when fitting to aggregated Pillar 2 data.

Default value of I_A_transmission updated to actual value currently used

booster_groups renamed to boosters_proportion and now allows partial exclusion of age groups. For backward compatibility defaults have the same effect on the priority population.
Added sircovid_models and check_sircovid_model to list and check sircovid models respectively

Introduce lancelot model. Initially a clone of the carehomes model with the added flexibility to fit to positivity or cases (pillar 2) by age. Respectively, p_NC and phi_pillar2_cases parameters can be generic or age-specific for fitting to aggregate or age stratified data.

England NHS region populations now obtained directly from mid-2020 ONS CCG-level estimates

Introduced phi_pillar2_cases_weekend, which is used instead of phi_pillar2_cases on weekends

Can now choose between a random sample or thinned sample in carehomes_forecast via the random_sample logical input

Introduced p_NC_weekend, which is used instead of p_NC on weekends

Increase code coverage

NHS regions populations updated to ONS mid-2020 estimates

Added p_R, the probability a non-fatal infection leads to post-infection immunity, otherwise the individual instantly becomes susceptible again post-infection

Added beta_type option to carehomes_parameters, now allowing it to be "piecewise-linear" or "piecewise-constant"

Fix bug in modify_severity for single strain model

Add modify_severity function to modify severity and transmission params of a VOC from VE assumptions

Relax checks on relative vaccination parameters so only first vacc class of first strain needs to be 1 for all age groups

prob_strain now measures the proportion of total I_weighted (the number of infectives weighted by the infectivity of their compartment) for each strain

IFR calculation now works fully for a multistrain model

Can now account for vaccine efficacy against death

Can now also output ALOS (average length of stay) from carehomes_ifr_t

Fix coverage hole in Rt

Fixes bug in calculating Rt when only one time-point is non-NA in prob_strain

Fixes bug in calculating Rt for multiple strains when the prob_strain matrix is all NA

Can now fit to variant data for all ages (previously just for over 25s)

Booster doses can now be added to a vaccine schedule with parameters to determine which groups get boosters

Fix Rt calculations for multistrain models when including interpolation

For simplicity, it is now assumed that individuals that progress through model compartments at a given time step cannot also move vaccine classes in that same step

carehomes_Rt now only returns NA when prob_strain is NA and weight_Rt is TRUE

Can now input population and carehome_beds to carehomes_parameters to allow use for regions where these are not included in the package

rel vacc params can now be less than one in unvaccinated classes for second infections (strains 3 and 4)

Fix Rt calculations for multistrain models with cross-immunity

Add lag_groups and lag_days parameters to vaccine_schedule_future to manually add lags for given age groups to the vaccine schedule

New function sircovid::inflate_state_strains for adding empty strain compartments to a model state that was run without multiple strains

Added carehomes_check_severity to check all severity probabilities lie in [0, 1] after transformation and all relative parameters are mirrored as expected

Deaths and admissions by age and vaccination class compartments exported as predicted samples

Improve efficiency of Rt calculation for single strains when supplying multistrain parameters

Fix multistrain Rt calculation for single particle case

Fix IFR_t calculation for multi-strain models

New function sircovid::inflate_state_strains for adding empty strain compartments to a model state that was run without multiple strains

R compartment now exported as part of predicted samples

Added ability to fit 2nd serology assay with different serology parameters

Added ability to directly set time-varying severity and progression parameters in carehomes_parameters_progression and carehomes_parameters_severity.
sircovid_parameters_beta renamed to sircovid_parameters_piecewise_linear and sircovid_parameters_beta_expand renamed to sircovid_parameters_expand_step

Fixed Rt calculation for multistrain so that Rt is NA only in the steps where prob_strain is NA (previously all Rt set to NA if any prob_strain was NA)

Fixed carehomes_rt function to have flexibility for multiple dimensions, given strains, age categories and vaccinations classes.
test-support.R updated accordingly.

Can now input a list of observation parameters into the carehomes_parameters function instead of taking the default observation parameters.

Add cross_immunity to control the amount of immunity conferred by one strain that decreases the probability of immediately being exposed to the other strain (ignored in single strain model).
Removed model_super_infection as model_super_infection = 0 is equivalent to cross_immunity = 1 (default)

Parameters describing vaccine efficacy (rel_susceptibility, rel_p_sympt, rel_p_hosp_if_sympt and rel_infectivity) can now take array values for varying across age groups, vaccination classes, and pathogen strains.

Multi-strain model now allows for ‘super-infections’, after recovering from strain 1 or 2, one can now be immediately infected with the ‘other’ strain. This can be turned on with model_super_infection.

Relaxed restrictions on waning immunity - the number of recovered individuals losing immunity is no longer capped by the number of individuals in the T_sero_neg or T_PCR_neg compartments.

Add parameter strain_rel_severity for varying the probabilities of p_G_D, p_H_D, p_W_D and p_ICU_D by strain.

Add additional disaggregated outputs for deaths and diagnosed admissions

New set of parameter strain_rel_gamma_ for varying gamma_A, gamma_P, gamma_C_1, and gamma_C_2 by strain.

Support to allow catch up of vaccine doses that could not be delivered (e.g. when people were infected at the time they should have been vaccinated)

Added support for time-varying hospital durations

Output cumulative numbers vaccinated

Changed fixed strain_seed_value to stochastic strain_seed_rate vector
Changed strain_seed_date from vector of start and end seeding dates to a vector of dates corresponding to rates set by strain_seed_rate

EpiEstim now allows age-varying p_C

New function sircovid::compile_gpu which compiles sircovid to run on a GPU (#237)

New functions to support reordering samples in mcstate_pmcmc objects

Fix corner case in vaccination exposed by recent dust updates

New functions to estimate effective Rt using EpiEstim

Added utils for nested pmcmc

Can control the number of initial infected individuals used in seeding

Predefined vaccination schedule implemented

Rt corrected for restructured model

Carehomes model restructured
Symptomatic pathway now includes presymptomatic compartment and two symptomatic compartments
Cases for pillar 2 testing arise in the movement from presymptomatic to the first symptomatic compartment
NHS regions populations updated to ONS mid-2019 estimates

Output deaths by age

Can now fit non-hospital deaths as two separate streams: community deaths and care home deaths

Allow vector of absolute values in future_Rt()

Faster Rt calculation (#209)

IFR calculation added for carehomes model

Allow a subset of Rt types to be calculated, perhaps in parallel

Enables daily vaccine doses to be time-varying

Fixed Rt calculation for multistrain so that Rt is NA when prob_strain is NA

Allow specification of Rt calculation in add_future_betas()

Add over 25 strain data stream to likelihood

Corrected calculation of NGM for multistrain model

Added different infectivity levels depending on vaccination stage
Corrected calculation of NGM with vaccination (but not for multistrain model)

Rename compartments and parameters to match paper notation

Multistrain model: ability to model 2 strains (and flexibility for more in the future) with strain-specific transmissibility. 2nd strain can be seeded over a number of pre-specified days at a pre-specified seeding level.

Added functions add_future_betas and future_Rt for constructing future transmission scenarios
Added new helper functions add_trajectory_incidence, drop_trajectory_incidence and drop_trajectory_predicted

Move vaccination process inside odin code

Now drawing equal numbers of care home workers across worker age groups.

Correct Rt calculation accounting for vaccination

I_mild compartment removed from both models and I_ILI renamed to I_sympt

Extended model structure for vaccination modelling

Deaths in stepdown now incorporated in carehomes model
Triage compartments merged in carehomes model

Implement trajectory and Rt aggregation (e.g., across regions)

Support for vaccination modelling, with vaccination affecting only levels of susceptibility (age-specific)

Parameter N_age was split into n_age_groups (both models) and n_groups (carehomes)
Implement preliminary support for waning immunity

Implement preliminary support for seroreversion

Implement preliminary support for vaccination modelling

Support for probability of positive calculation directly in the package
Pillar 2 data can now be fitted to over 25s only, as well as all ages
Sensitivity is explicitly considered in the sero probability calculation

Add REACT to log-likelihood