Joint and single-stream models

The composers that assemble the submodels into a fittable model. bvd_joint is the headline fit over every included stream. The *_only_model composers fit one stream at a time for the stream-by-stream comparison.

Index

BVDOutbreakSize.bvd_joint
BVDOutbreakSize.cases_only_model
BVDOutbreakSize.confirmed_deaths_only_model
BVDOutbreakSize.confirmed_only_model
BVDOutbreakSize.deaths_only_model
BVDOutbreakSize.default_breakpoint
BVDOutbreakSize.exports_deaths_only_model
BVDOutbreakSize.exports_joint_only_model
BVDOutbreakSize.exports_only_model
BVDOutbreakSize.forecast_days
BVDOutbreakSize.joint_fit_args
BVDOutbreakSize.onsets_only_model
BVDOutbreakSize.patch_fit_args
BVDOutbreakSize.production_joint
BVDOutbreakSize.treatment_only_model

Reference

BVDOutbreakSize.bvd_joint Function

julia

bvd_joint(
    n::Integer,
    exported_cases::Union{Missing, Integer},
    total_deaths::Union{Missing, Integer};
    ...
) -> Any
bvd_joint(
    n::Integer,
    exported_cases::Union{Missing, Integer},
    total_deaths::Union{Missing, Integer},
    reported_cases::Union{Missing, Integer};
    ...
) -> Any
bvd_joint(
    n::Integer,
    exported_cases::Union{Missing, Integer},
    total_deaths::Union{Missing, Integer},
    reported_cases::Union{Missing, Integer},
    exports_deaths::Union{Missing, Integer};
    ...
) -> Any
bvd_joint(
    n::Integer,
    exported_cases::Union{Missing, Integer},
    total_deaths::Union{Missing, Integer},
    reported_cases::Union{Missing, Integer},
    exports_deaths::Union{Missing, Integer},
    confirmed_cases::Union{Missing, Integer};
    ...
) -> Any
bvd_joint(
    n::Integer,
    exported_cases::Union{Missing, Integer},
    total_deaths::Union{Missing, Integer},
    reported_cases::Union{Missing, Integer},
    exports_deaths::Union{Missing, Integer},
    confirmed_cases::Union{Missing, Integer},
    tests_analysed::Union{Missing, Integer};
    n_patches,
    importation_kernel,
    confirmed_deaths,
    recovered_cases,
    deaths_history,
    reported_history,
    confirmed_history,
    confirmed_deaths_history,
    lab_history,
    lab_daily_history,
    suspected_daily_history,
    suspected_daily_deaths_history,
    isolation_history,
    bed_capacity_history,
    recovered_history,
    treatment_admissions_history,
    treatment_deaths_history,
    treatment_ruleout_history,
    treatment_absconded_history,
    treatment_confirmed_incare_history,
    treatment_suspect_incare_history,
    occupancy_break_days,
    confirmed_break_days,
    confirmed_break_gross_cases,
    confirmed_break_gross_deaths,
    confirmed_break_sd,
    export_case_days,
    export_death_days,
    onset_curve_history,
    breakpoint,
    source_population,
    patch_infection,
    composition,
    province_increments,
    province_days,
    province_lab_increments,
    province_lab_days,
    province_lab_bins,
    lab_composition,
    background_split,
    province_isolation,
    province_capacity,
    province_admissions,
    province_death_increments,
    province_death_days,
    death_composition,
    death_ascertainment_sd_prior,
    province_cfr_sd_prior,
    export_pressure,
    exports,
    deaths,
    cases,
    confirmed,
    confirmed_deaths_stream,
    treatment,
    recovered,
    onset_report,
    dispersion,
    ascertainment,
    background_pooling,
    genetic,
    onset_to_sample,
    tmrca_days,
    tmrca_days_sd,
    renewal_start_lead,
    rt_walk_lead,
    background_onset_lead,
    treatment_defaults,
    forecast,
    simulated_data
) -> Any

Joint composer over all data streams. Runs the generating infection process once on a daily grid of length n (day n is the cut-off), stages it to daily onset incidence, then conditions on the DRC suspected cases, deaths and the laboratory pipeline, the confirmed deaths, the Uganda exports and deaths-among-exports, the digitised symptom-onset reporting triangle (onset_reporting_model, the only direct observation of the shared onset series, a no-op when onset_curve_history is empty), and the optional genetic seeding bound on the outbreak age. Each stream argument may be missing to drop it, so the model doubles as a prior- and posterior-predictive generator.

The confirmed-case stream shares its onset-to-report kernel with the suspected-case stream. A single analysed-specimen volume is fit through a report-to-analysed delay and the tested fraction. The confirmed positives are scored as a Binomial of the observed specimens-analysed denominator (lab_history) with a partially-pooled per-window positivity, so they do not pass through the multiplicative ascertainment ridge. After the national cumulative analysed series stops, the reporting format gives a 24h analysed count on some days (lab_daily_history). Those are fitted as per-day analysed volumes and also anchor that day's confirmed positives. Windows with no published denominator use the modelled analysed volume as the denominator, with the positivity (hence λ_bg) carried over from the windows that do have data (see confirmed_cases_model).

The optional suspected_daily_history adds the post-26 May daily new-suspect inflow ("nouveaux cas suspects du jour"), scored against the modelled daily suspected series on the report days where the frozen cumulative suspected stream stops, disjoint from it. The optional suspected_daily_deaths_history adds the deaths analogue, the daily new suspected deaths ("cas suspects du jour N (M deces)").

The confirmed deaths mirror the confirmed-case laboratory pipeline. A death "analysed" volume (the suspected deaths carried to laboratory receipt and scaled by the death testing intensity tau_death, specimens per suspected death, which may exceed one) is scored through a death-pool composition positivity p = s·q_death + (1−spec)(1−q_death), with q_death the BVD share of the suspected deaths (see confirmed_deaths_model). The suspected deaths carry a death ascertainment p_death and a non-BVD background tied to the case background by a background CFR cfr_bg (see deaths_model).

The optional isolation_history adds the daily isolation/treatment-bed occupancy ("Patients en isolement"), a prevalence stream fitted as the suspect inflow (BVD treatment stay plus non-BVD rule-out stay) carried through a length-of-stay survival into a daily stock (see treatment_flow_model). The optional recovered_history adds the recovered-among-confirmed stream ("cumul guéris"), survivors among the modelled daily confirmed cases scaled by the recovery probability and lagged by a confirmation-to-recovery delay (see recovered_model).

breakpoint is the intervention day passed to the reproduction-number walk (e.g. the first WHO situation report). genetic injects the genetic seeding submodel when tmrca_days is given. The analysed volume is scaled by specimens analysed per suspect (specimen_intensity_model), which τ_test alone, being a probability, cannot exceed one of.

Tracked deterministics: C_T (cumulative infections by the cut-off), the established reproduction number R0 (= the first R_t), r and doubling_time (current growth), r0 (the R0-implied cryptic growth rate), T (outbreak age), R_T (current reproduction number), the per-stream expected counts, the testing fraction tau_test, the specimens analysed per suspect (specimens_per_suspect), the background rate lambda_bg, the death ascertainment death_ascertainment, the background CFR background_cfr, the death testing intensity tau_death, the implied per-suspected (suspected_positivity) and per-test (test_positivity) positivities, and the death-pool BVD composition (death_composition) and death-confirmation positivity (death_confirmation).

With n_patches > 1 the latent process is the meta-population renewal of patch_infection_model, one renewal equation per province coupled by importation, with every national stream above fitted against the summed provinces. The default n_patches = 1 collapses it onto the single-population model, since the sum-to-zero deviations vanish, there is nothing to import between, and no per-province likelihood is scored.

The spatial information enters through two composition terms. The per-province confirmed cases and confirmed deaths in the situation reports' spatial tables are exact partitions of the national totals, so scoring them with their own count likelihoods would put the same data into the joint density twice. province_composition_model scores each split conditional on the national total instead, adding only the spatial signal the national series does not carry. Pass the reshaped data as province_increments with province_days, and province_death_increments with province_death_days, both built by province_increment_matrix. Omit them and the composition terms are skipped. The case composition identifies only the product of a province's incidence and its case-finding. The death composition identifies the incidence split, since the case-fatality ratio and the death-confirmation probability are national, and the case composition then identifies the relative case ascertainment as the residual.

A third composition scores the per-province analysed-specimen volume conditional on the national total in each laboratory bin. The modelled split is each patch's BVD suspects (its onsets through the onset-to-confirmation kernel, thinned by p_drc) plus its share of the non-BVD background, the share a partially pooled simplex (background_split_model) carries and this term identifies. The BVD suspects carry the case composition's relative ascertainment, so the two compositions agree on how many of a patch's cases reach the laboratory. Pass province_lab_increments with province_lab_days and province_lab_bins, built by province_lab_increment_matrix, which sums the printed days into calendar weeks for the production fit. The testing fraction stays national and the composition samples no ascertainment contrast of its own. The per-province positives are not fitted, being the differencing of the confirmed counts already scored.

Uganda exports are driven by the provinces in proportion to sampled relative export weights, with Ituri the reference at weight one (see province_export_pressure_model).

Per-patch quantities are surfaced as vector deterministics, one entry per patch, for patch_summary_table and patch_overview_table: C_T_patch, R_T_patch, infections_T_patch, delta_patch, log_rt_contrast and the deviation knots delta_knots from which reconstruct_patch_rt rebuilds the provincial trajectories. R_T is the force-of-infection-weighted reproduction number implied by the summed patch infections.

With forecast a ForecastHorizon (see with_horizon) the model runs past the cut-off n and draws each stream's future counts through its own likelihood, as missing observations predict generates: forecast_reports, forecast_deaths, forecast_confirmed, forecast_confirmed_deaths, forecast_recovered, the treatment flows (treatment_forecast_model), forecast_exports and the onset triangle's future vintages (onset_forecast_model). With province data, each future week's national confirmed cases and deaths are split across the provinces by the fitted compositions (composition_split_model), as forecast_province_confirmed and forecast_province_deaths. With province care data, the occupancy and admissions are drawn daily by province, as forecast_isolation.province.obs and forecast_admissions.province.obs, with the beds as forecast_province_beds, and each sums to the national forecast (treatment_forecast_model). Every quantity up to the cut-off, and the density there, is the fitted model's.

Joint and single-stream models ​

Index ​

Reference ​

Joint and single-stream models

Index

Reference