ModelWithinHost

Models of natural history of malaria in humans

OpenMalaria supports multiple ways of modeling the behaviour of malaria in humans the absence of interventions. These model variants offer:

different algorithms for simulating how exposure (measured by the Entomological Inoculation Rate(EIR)) is translated into infections of humans. The way in which the EIR is specified is described here).
different ways of modeling parasite densities, and how they evolve over time in infected humans, described here.
different models for pathogenesis.

These different model variants are specified in the model element of the XML.

Within-host models

The within-host models calculate parasite densities and infectiousness of human hosts. Within the OpenMalaria code, these reside within a set of classes within the Host namespace.

The within-host models cover the following steps in the parasite life-cycle:

The incidence of blood stage infection as a function of the exposure to infective bites from mosquitoes (entomological inoculation rate). To allow for liver and pre-patent stages: blood-stage infection modelling is delayed by so many time-steps after infection to account for the pre-patent delay (see latentp parameter).
Asexual blood stage models: these are what we usually mean by "infection" models (see below).
Acquired natural immunity: in the Maire et al (2006) within-host models the effect of acquired immunity is to directly reduce the parasite density. Within the other within-host models, acquired immunity reduces the merozoite survival factor. In either case the effect depends on the history of previous infections. Most code is in the Infection and WithinHostModel base classes.
Sexual stage (gametocyte) model: Ross et al, 2006 This uses the results of a statistical model for the relationship between asexual parasite densities of Plasmodium falciparum and the infectivity of the host to mosquitoes, fitted to malariatherapy data. The model takes into account the delay between asexual parasitemia and infectivity resulting from the time course of gametocytemia, by making transmission depend on the asexual parasite density between 10 and 20 days earlier. It also allows for the need for the blood meal to contain gametocytes of both sexes if infection is to take place.

In some way the drug models could also be considered within-host models, though in the code they come under the PkPd directory and namespace.

Infection of humans

The InfectionIncidenceModel translates the EIR at each time step into simulated infections in specific Hosts. In all cases, the number of infections introduced at each time step is sampled from a Poisson Distribution. Different model variants, specified as described below may differ in how exposure (measured by the Entomological Inoculation Rate) is translated into the expected number of new infections.

All the model variables allow for variation between hosts in availability to mosquitoes. In the default variant the susceptibility of humans (proportion of bites from sporozoite positive mosquitoes resulting in infection) follows Equation (7) of Smith et al, 2006. In 'Mass Action' model variants the expected number of infections in any one host is proportional to that host's exposure. The corresponds to fixing the susceptibility of humans at a constant value, by default 0.702; this is computed as 0.19 (the value S from a negative binomial mass action model fitted to Saradidi data, divided by 0.302 (the ratio of body surface area in a 0.5-6 year old child (as per Saradidi) to adult).

Implemented variants of the models for human susceptibility are therefore:

The default model variant as described in Smith et al, (2006)
NEGATIVE_BINOMIAL_MASS_ACTION: Infections introduced by mass action. Baseline availability of humans is sampled from a gamma distribution (see below).
LOGNORMAL_MASS_ACTION: Infections introduced by mass action with log normal variation in human availability.

Asexual infection models

Currently implemented models are:

Maire et al (2006) (implemented in the DescriptiveInfection class). This model takes, as a representation of a malaria infection in a naive human, a statistical description of the parasite densities over time during a malariatherapy infection, with stochastic noise added independently at each time point. Acquired immunity, due to previous exposure, is assumed to reduce these parasite densities. This model uses discrete 5-day time-steps.
(As-yet no publication; implemented in the EmpiricalInfection class). This is a model of Plasmodium falciparum asexual parasite densities based on an autoregressive time-series model to fitted to malariatherapy data using a Bayesian simulation-based algorithm. The fitting algorithm immediately suggests a corresponding prediction method that provides stochastic predictions of parasite density profiles, and a natural approach for incorporating effects on parasite multiplication of blood stage vaccines and of sub-therapeutic drug concentrations. This model uses discrete 1-day time-steps.
Molineaux et al (2001) (implemented in the MolineauxInfection class). This is a mass action model of Plasmodium falciparum asexual parasite densities fitted to malariatherapy data. In incorporates (i) intra-clonal antigenic variation, (ii) large variations of the variants' baseline growth rate, depending on both variant and case, (iii) innate autoregulation of the asexual parasite density, variable among cases, (iv) acquired variant-specific immunity and (v) acquired variant-transcending immunity, variable among cases. The published model uses a discrete time-step of 2 days. The OpenMalaria implementation adapts this to a 1-day time-step by interpolating the parasite multiplication rate. Drug-, vaccine-, and natural immune effects of previous exposure are represented by further modifying the parasite multiplication rates. Parameterisations in use up to and including Version 34 use 1-day timesteps for pharmacokinetic and pharmacodynamic models with intervention deployment and monitoring are specified by 5-day step (in these parameterisations the interval attribute is set to 5 days).

Specification of model variants

The <model> element of the XML contains the specification of which of the different model variants is required. In addition the <model> element specifies the specific parameter values required for the the chosen variant. The variant is specified by <ModelOption> sub-elements, as in the following example:

  <model>
    <ModelOptions>
      <option name="LOGNORMAL_MASS_ACTION" value="true"/>
      <option name="MAX_DENS_CORRECTION" value="true"/>
      <option name="INNATE_MAX_DENS" value="false"/>
      <option name="INDIRECT_MORTALITY_FIX" value="false"/>
    </ModelOptions> 
    ....

where each option is assigned a value of "true" or "false". Options that are not included are by default "false".

The main choices to make with regards to the model to use is the within-host / infection model:

"Descriptive" model: 5-day time-step; several parameterizations have been fit for this model
"Empirical" model: 1-day time-step; currently has not been parameterized
Molineaux model: 5-day or 1-day time-step; parameter fitting is currently in progress

This will also determine which clinical / health system model may be used:

5-day timestep model with simple cure/fail response to treatment and immediately determined clinical outcome
1-day timestep model: treatments use full PKPD modelling with a daily effect on parasite densities; clinical outcome determined by both the parasite density and clinical decisions.

Each model variant has a unique model structure and it is challenging to parameterise this with appropriate values, so after selecting the within-host model(s) to use, an appropriate pre-parameterised <model>...</model> element should be obtained from (https://github.com/SwissTPH/openmalaria.svn-archive/tree/master/download/experiments), or from the standard set of 14 parameterisations for the 5-day timestep in the snippet library. Arbitrary parameterisations are likely to reproduce extremely unrealistic behaviour.

For completeness the possible model options (which include bug fixes as well as model variant switches), are listed in the following table:

ModelOption	Purpose
PENALISATION_EPISODES	No longer available. This was one of four hypotheses in the literature investigated in the IPTi study. Clinical episodes reduce the level of acquired immunity.Effective cumulative exposure to blood stage parasites is reduced by a clinical sickness event, so that clinical bouts have a negative effect on blood stage immunity.(ImmediateOutcomes model: per event; EventScheduler: once per bout.) Default: Clinical events have no effect on immune status except secondarily via effects of treatment. This originated in the now superseded IPTi code Ross et al 2008).
NEGATIVE_BINOMIAL_MASS_ACTION	Baseline availability of humans is sampled from a gamma distribution. Infections introduced by mass action with negative binomial variation in numbers of infection. Default: New infections are introduced via a Poisson process as described in AJTMH 75 (suppl 2) pp11-18.
ATTENUATION_ASEXUAL_DENSITY	No longer available. This was one of four hypotheses in the literature investigated in the IPTi study. No longer used. Does nothing if IPT is not present.
LOGNORMAL_MASS_ACTION	Baseline availability of humans is sampled from a log normal distribution. Infections introduced by mass action with log normal variation in infection rate. Default: New infections are introduced via a Poisson process as described in AJTMH 75 (suppl 2) pp11-18.
NO_PRE_ERYTHROCYTIC	Infections are introduced without using preerythrocytic immunity.
MAX_DENS_CORRECTION	Bug fixes in Descriptive & DescriptiveIPT within-host models. For new parameterisations, both MAX_DENS_CORRECTION and INNATE_MAX_DENS should be used. When using parameter sets from an old fitting run which didn't originally use these options, turn them off for consistency.MAX_DENS_RESET is not used since it is unneeded when MAX_DENS_CORRECTION is present and wouldn't make sense when not.
INNATE_MAX_DENS	(See MAX_DENS_CORRECTION)
MAX_DENS_RESET	No longer available. (See MAX_DENS_CORRECTION)
DUMMY_WITHIN_HOST_MODEL	Parasite densities are predicted from an autoregressive process. Default: Parasite densities are determined from the descriptive model given in AJTMH 75 (suppl 2) pp19-31.
PREDETERMINED_EPISODES	Clinical episodes occur if parasitaemia exceeds the pyrogenic threshold. Default: Clinical episodes are a stochastic function as described in AJTMH 75 (suppl 2) pp56-62.
NON_MALARIA_FEVERS	The presentation model includes simulation of non-malaria fevers. Default: Non-malaria fevers are not simulated.
INCLUDES_PK_PD	No longer available. PKPD code is now enabled in all compatible within-host models, making this option obsolete: Use a PK & PD model for drug effects
CLINICAL_EVENT_SCHEDULER	Use revised clinical and case management model, ClinicalEventScheduler. Default: use the Tediosi et al case management model (Case management as described in AJTMH 75 (suppl 2) pp90-103), ClinicalImmediateOutcomes.
MUELLER_PRESENTATION_MODEL	Clinical episodes occur in response to a simple parasite density trigger. Default: Use the Smith et al presentation model (Clinical episodes are a stochastic function as described in Smith et al 2006)
TRANS_HET	Simple transmission heterogeneity as described in Ross&Smith 2010. The default is no transmission heterogeneity. This simple transmission heterogeneity is incompatible with NEGATIVE_BINOMIAL_MASS_ACTION and LOGNORMAL_MASS_ACTION because both try to adjust _EIRFactor and it is not confirmed that the ways they do this are compatible.
COMORB_HET	Allow simple heterogeneity in comorbidity as described in Ross & Smith 2010
TREAT_HET	Allow simple heterogeneity in treatment seeking as described in Ross & Smith 2010
COMORB_TRANS_HET	Allow correlated heterogeneities in transmission and comorbidity as described in Ross & Smith 2010
TRANS_TREAT_HET	Allow correlated heterogeneities in transmission and treatment seeking as described in Ross & Smith 2010
COMORB_TREAT_HET	Allow correlated heterogeneities comorbidity and treatment seeking as described in Ross & Smith 2010
TRIPLE_HET	Allow correlated heterogeneities in transmission, comorbidity and treatment seeking as described in Ross & Smith 2010
EMPIRICAL_WITHIN_HOST_MODEL	Selection of within host models. Parasite densities are predicted from an empirical model
MOLINEAUX_WITHIN_HOST_MODEL	Use Molineaux within host model
PENNY_WITHIN_HOST_MODEL	Use Penny infection model
MEAN_DURATION_GAMMA	Selection of gamma distribution for Molineaux or Penny within host models. Use Gamma distribution for mean duration (Molineaux model)
FIRST_LOCAL_MAXIMUM_GAMMA	Use Gamma distribution for first local maximum (Molineaux model)
PARASITE_REPLICATION_GAMMA	Use Gamma distribution for first local maximum (Molineaux model)
IMMUNE_THRESHOLD_GAMMA	Use Gamma distribution for immune threshold (Penny model)
UPDATE_DENSITY_GAMMA	Use Gamma distribution for update density (Penny model)
GARKI_DENSITY_BIAS	Obsolete: it is preferred to specify diagnostics units instead. Use the Garki density bias instead of the default one in the detection limit. The default bias corresponds to counting parasites and white blood cells(assuming a white blood cell density of 8000 per µl), the Garki bias to estimations from a probability function.
IPTI_SP_MODEL	No longer available. Use the IPT(i) drug model (DescriptiveIPTWithinHost and DescriptiveIPTInfection classes) with its simple SP model. This has been removed; mass drug interventions can be used as a replacement.
REPORT_ONLY_AT_RISK	No longer available. Turn off reporting of several outputs for humans suffering a recent clinical episode and therefore not currently at risk of what would clinically be regarded as a separate episode. This is a compatibility option only. It only works with the 5-day model and the length of the not-at-risk period is hard-coded
VECTOR_LIFE_CYCLE_MODEL	Turn on vector life-cycle model. Allows better larviciding and vector population dynamics modelling. Requires vector model.
VECTOR_SIMPLE_MPD_MODEL	Turn on the simple mosquito population dynamics model (simpler version of life-cycle model). Requires vector model.
MOLINEAUX_PAIRWISE_SAMPLE	Sample case-specific densities Pc and Pm as a pair from one of the 35 patient records.
VIVAX_SIMPLE_MODEL	Use a simple vivax model instead of falciparum. See Vivax page.
PROPHYLACTIC_DRUG_ACTION_MODEL	No longer available.
INDIRECT_MORTALITY_FIX	Bug fixes: Without this, the 5-day case management leaves uncomplicated cases with indirect mortality untreated, and the 1-day case management forgets to apply indirect mortality if the sickness state doesn't change. This option fixes both bugs (though only one would have any effect,
CFR_PF_USE_HOSPITAL	Potentially a bug-fix: for in-hospital severe malaria patients where treatment fails to clear parasites, enabling this option selects the use of hospital Case-Fatality-Rate (as described in AJTMH supplement); otherwise the community CFR is used (old model behaviour, whether a bug or intended behaviour). This is disabled by default since model parameters have been fitted with this disabled. New models could be fit with the option enabled.

Parameter values

In addition to the specification of the model variants, the <model> element contains sub-elements specifying the health system memory, the age pattern of availability to mosquitoes, and the specific parameters required for the chosen model variant, as follows:

    <clinical healthSystemMemory="6"/>
    <human>
      <availabilityToMosquitoes>
        <group lowerbound="0.0" value="0.225940909648"/>
        ....
        .... 
        <group lowerbound="15.0" value="0.839587932303"/>
        <group lowerbound="20.0" value="1.0"/>
      </availabilityToMosquitoes>
    </human>
    <parameters interval="5" iseed="2359" latentp="3">
      .....
      .....
    </parameters> 
  <model>

The interval attribute; this specifies the length (in days) of the fundamental time-step of the simulator. The original model variant uses 5-day time steps. There is also a set of alternative models, also using 5-day time steps. Specifying a 1-day time-step results in more complex and slower runs for which parameterisations are less developed: in particular this invokes the 1-day time-step case management model, with 1-day discrete time stochastic processes determining access to care and treatment patterns. Several other parts of the configuration are entered using units of the length of a time-step.
The iseed="2359" attribute assigns the seed for the random number generator.
delta — unused and removed in schema version 19.
latentp — number of timesteps by which blood-stage infection is delayed from receipt of a mosquito bite
The '' elements make up a list of the specific parameter values required by the chosen model variant. The list of which parameters may be specified can be found here (in some versions numbers, rather than names, are used to specify each parameter in the scenario file).The parameters element describes a list of parameters which are fitted. Usually, this section should be copied from an appropriate scenario. Varying the length of the timestep or changing several of the model's options will affect the ideal values of these parameters.
The model element also includes data for some small human-related models.
A pharmacology sub-element is required when a 1-day time-step is used, and is essentially a library of parameters for the drug model. It should therefore not need to be edited (just copied from a source).