Skip to content

Latest commit

 

History

History
88 lines (48 loc) · 13 KB

intervention-description.md

File metadata and controls

88 lines (48 loc) · 13 KB

CovidSim: how interventions and policies are implemented

See model overview for further information.

CovidSim models infections through time in a spatially explicit representation of individuals in a population. Infected people are assumed to infect susceptible people at household, spatial and place level. Here, place is a generalisation of either primary schools, secondary schools, universities or workplaces (e.g. offices). The model includes disease progression for symptomatic individuals, and is parameterised to reflect current knowledge of Covid-19 disease progression and death.

Interventions are implemented in the code through the parameter and pre-parameter files (see input and outputs). These parameters are read in function ReadParams in CovidSim.cpp and collected in single structure PARAMS, of which there is a single, global instance P.

Intervention overview

The model considers various interventions that alter either the infectiousness of infected people or the susceptibility of susceptible people (or both). Fundamentally, interventions are implemented by changing the person-specific period during which a person is subject to a particular intervention (given in #define functions in ModelMacros.h). These values in turn toggle parameters in functions governing infectiousness and susceptibility in CalcInfSusc.cpp, and in function InfectSweep in Update.cpp. The interventions broadly fall into the following categories:

  • Case Isolation (CI),
  • Household Quarantine (HQ),
  • Place Closure (PC),
  • Social Distancing (SD).

Example parameter files have been provided for the following scenarios: i) no interventions (p_NoInt.txt); ii) case isolation; household quarantine; social distancing and place closure (p_PC7_CI_HQ_SD.txt).

Units of time t are in days, and day 1 is taken to be January 1st 2020 (e.g. day 76 is March 16th 2020) in output files. The model is calibrated at the start of runtime to match the user-specified inputs. It will approximately (through an iterative calibration process) match a user-specified threshold cumulative epidemic size (either [Number of deaths accummulated before alert] for deaths or [Number of detected cases needed before outbreak alert triggered] for detected cases, depending on value of P.PreControlClusterIdUseDeaths [Trigger alert on deaths]) occuring up to the day speecified by [Day of year trigger is reached] (P.PreControlClusterIdCalTime). If [Alert trigger starts after interventions] is set to 1 in the input pre-parameter file, this threshold can occur after the start of interventions, in which case the start day of interventions, [Day of year interventions start] (P.PreIntervIdCalTime) should be also specified.

Interention times in the intervention parameter files are relative to P.PreIntervIdCalTime. (e.g. if P.HQuarantineTimeStartBase / [Household quarantine start time] is set to 10, then in the start date of household quarantine policy will be ten days after P.PreIntervIdCalTime).

Importantly, the model distinguishes between the period during which a policy is implemented, and the period during which the actual intervention is implemented. If a policy is implemented from time t, the intervention will not be implemented until a user-specified threshold has been met by a particular “trigger” or count (see below). Further, CI and HQ policies have durations at the level of a person, which denote the length of time that a person would either self-isolate or be under home quarantine. These person-level durations are distinct from the duration of policy at the population level. For example, a policy could be implemented for a year, during which time people would home quarantine for 14 days if a member of their household becomes sick.

Intervention triggers

A trigger is defined flexibly, with reference to either the numbers (cumulative incidence) of confirmed cases, numbers of critical cases that require Intensive Care Units (ICU beds), or numbers of deaths. Numbers can be either absolute or per-capita, and the model may be refined so that the number is defined in relation to ICU bed capacity. The window over which cumulative incidence is calculated is specified by P.TriggersSamplingInterval [Number of sampling intervals over which cumulative incidence measured for global trigger] days. Further, triggers may be specified at either admin unit (regional) or at country (global) level. So too can the start times for when each intervention is considered. Finally, each intervention has its own threshold value.

Currently, all interventions must use the same epidemiological variable for their trigger (e.g. all using confirmed cases or all using ICU beds), and interventions do not differ in whether they consider absolute or per-capita triggers. Note that triggers can be set to zero, indicating that interventions are implemented throughout the time window specified by the start time and duration of that intervention.

The steps below summarise the above process:

  • Simulation starts
  • Policy implemented
  • Trigger (count) surpasses threshold for that intervention
  • Intervention implemented
  • infectiousness and/or susceptibility parameters turned on for individuals affected by intervention

The following interventions are described with reference to particular functions and parameters in code, and the parameter descriptions in parameter files (p_XX.txt), which are read in through ReadParams function in CovidSim.cpp. Functions TreatSweep in Update.cpp and RecordSample in CovidSim.cpp determine whether thresholds have been exceeded.

Notes on variable naming: Parameters with prefix us are unsigned short versions of their namesakes, that have been multiplied by parameter P.TimeStepsPerDay (e.g. P.usHQuarantineHouseDuration = P.HQuarantineHouseDuration * P.TimeStepsPerDay).

CASE ISOLATION (CI)

Policy implemented after time P.CaseIsolationTimeStartBase (“Case isolation start time”) for duration of P.CaseIsolationPolicyDuration days. A value of P.CaseIsolationTimeStartBase greater than simulation duration (P.SampleTime “Sampling time” in pre-parameter file) indicates that policy will not be implemented (e.g. if simulation runs for 720 days and start time set to 1000 days). If threshold P.CaseIsolation_CellIncThresh exceeded, symptomatic cases stay home for period of P.CaseIsolationDuration days, starting after P.CaseIsolationDelay days. Durations and delays can alternatively be specified at admin unit level.

After infection and upon case detection (DoDetectCase function in Update.cpp), a case isolates with probability P.CaseIsolationProp. Their isolation_start_time is set (Isolation stop time is isolation_start_time + P.usCaseIsolationDelay + P.usCaseIsolationDuration). These values are relevant to HOST_ISOLATED macro, which returns true or false. HOST_ISOLATED affects every infectiousness function (at person, house, place, spatial levels) but not susceptibility functions in CalcInfSusc.cpp.

If HOST_ISOLATED, place and spatial infectiousness scaled by P.CaseIsolationEffectiveness, whereas household infectiousness scaled by P.CaseIsolationHouseEffectiveness.

Additionally, case isolation amends absent start and stop time, in scenarios where isolated person is a child requiring adult supervision, and where host was already absent and requires their period of absence amended due to case isolation. Macro HOST_ABSENT relevant in InfectSweep in Sweep.cpp – all place infections are contingent on host not being absent from their place.

HOUSEHOLD/HOME QUARANTINE (HQ)

Policy implemented after time P.HQuarantineTimeStartBase (“Household quarantine start time”) for period of P.HQuarantinePolicyDuration days. Value of P.HQuarantineTimeStartBase greater than simulation duration indicates that policy will not be implemented (e.g. if simulation runs for 720 days and start time set to 1000 days). If threshold P.HHQuar_CellIncThresh exceeded, people who share household with a symptomatic case stay home for period of P. HQuarantineHouseDuration days, starting after P.HQuarantineHouseDelay days. Durations and delays can alternatively be specified at admin unit level.

Upon case detection (DoDetectCase function in Update.cpp), quar_start_time is set. A person’s quarantine stop time is quar_start_time + P.usHQuarantineHouseDuration. Compliance is set first at household and then at individual level, both of which are randomly determined with respective proportions P.HQuarantinePropHouseCompliant and P.HQuarantinePropIndivCompliant. Note that 50% compliance means that 50% of people are 100% compliant, and not that 100% of people are 50% compliant (although the latter probably more realistic).

Macro HOST_QUARANTINED asks whether host is a) compliant with HQ; b) currently between their quarantine start and stop times. HOST_QUARANTINED scales infectiousness: at household level by P.HQuarantineHouseEffect; at spatial level by P.HQuarantineSpatialEffect; and at place level by P.HQuarantinePlaceEffect[k] for place type k. HOST_QUARANTINED scales susceptibililty: at spatial level by P.HQuarantineSpatialEffect; and at place level by P. HQuarantinePlaceEffect[k] for place type k.

SOCIAL DISTANCING (SD)

Policy implemented after time P.SocDistTimeStartBase (“Social distancing start time”) for period of P.SocDistDuration days. Value of P.SocDistTimeStartBase greater than simulation duration indicates that policy will not be implemented (e.g. if simulation runs for 720 days and start time set to 1000 days). Durations and delays can alternatively be specified at admin unit level.

Social distancing implemented mainly in TreatSweep, which determines whether P.SocDistCellIncThresh incidence threshold exceed. If so, then the susceptibility of all people residing in microcell currently undergoing social distancing is scaled, either by enhanced or non-enhanced effects (see below). Infectiousness functions are unaffected by social distancing.

Distinction between social distancing and enhanced social distancing effect: at household level (P.SocDistHouseholdEffect vs. P.EnhancedSocDistHouseholdEffect); at place level (P.SocDistPlaceEffect[k] vs. P.EnhancedSocDistPlaceEffect[k], for place type k); and at spatial level (P.SocDistSpatialEffect vs. P.EnhancedSocDistSpatialEffect). These scale a person’s contact rate at household, place and spatial level.

Both sets of parameters can implemented at the same time, in which case a given person is compliant with enhanced social distancing (so person.esocdist_comply == 1) with age-specific probability P.EnhancedSocProportionCompliant[age] (decided for each person in SetupModel.cpp).

A social distancing policy can occur once only (if P.DoSocDistOnceOnly), or repeatedly.

SOCIAL DISTANCING FOR OVER 70s (SDO/SDOL70)

Differences between this and social distancing are in parameter values, rather than source code.

ReadParams in CovidSim.cpp populates P.ESocProportionCompliant array by scanning first for “Proportion compliant with enhanced social distancing by age group”, and failing that the non-age-specific “Proportion compliant with enhanced social distancing”, in which case all values of array same for all age groups. If this is not found, values default to 0, i.e. zero compliance with enhanced social distancing.

PLACE CLOSURE (PC)

Policy implemented after time P.PlaceCloseTimeStartBase (“Place closure start time”). Values of P.PlaceCloseTimeStartBase greater than simulation duration indicates that policy will not be implemented (e.g. if simulation runs for 720 days and start time set to 1000 days). Durations can alternatively be specified at admin unit level.

RecordSample in CovidSim.cpp determines whether trigger has exceeded P.PlaceCloseCellIncThresh, and if so sets P.PlaceCloseTimeStart. A subsequent call to TreatSweep function in Sweep.cpp, determines whether incidence threshold P.PlaceCloseCellIncThresh exceeded. If so, then all places in microcell marked as closed (and trigger reset), and DoPlaceClose in Update.cpp is called. DoPlaceClose changes quantities place.close_start_time and place.close_stop_time, which are used in macro function PLACE_CLOSED.

Unlike other measures, place closure does not affect infectiousness and susceptibility functions in CalcInfSusc.cpp, but it does affect infectiousness and susceptibility indirectly in InfectSweep function in Update.cpp with calls to PLACE_CLOSED. If person has their places closed, their household infectiousness is scaled by P.PlaceCloseHouseholdRelContact. Similarly, infected people’s spatial infectiousness also scaled by P.PlaceCloseSpatialRelContact, and potential infectee’s susceptibility to spatial infections is also scaled by P.PlaceCloseSpatialRelContact. Further, infected people cannot infect anyone in their place (and therefore have zero Place infectiousness) when their place is closed.

Like social distancing, place closure can occur intermittently if parameter P.DoPlaceCloseOnceOnly set to 0.