The P. vivax claims ask the same questions as those for P. falciparum, for a
get_parameters(parasite = "vivax") list. They use vivax’s
own transmission grid, EIR 0.3 to 30. Radical cure, by primaquine and by
tafenoquine, replaces chemoprevention, which
malariasimulation cannot run under vivax, and vaccines, for
which it has no vivax calibration. malariasimulation’s
vivax model has no severe disease, so there are no severe claims. Two
claims are specific to vivax: relapse incidence and the share of people
carrying hypnozoites. The tier-3 site-file comparison is not yet part of
the vivax evidence.
malariasimulation 3.0.0 evaluates vivax relapses only on
days when at least one person is bitten, so on a day with no infectious
bite anywhere the IBM loses every relapse, where fleet
keeps them. In the scenario suite’s 10,000-person runs such days are
common under indoor residual spraying, and below an EIR of about 0.14
whatever is deployed. intervention-impact-pv sets out the
effect on the comparison.
This page is generated from claims.yml. A claim’s tier
is the cost of reproducing it; see Methods.
10 claims: 2 failing, 0 untested, 0 open, 8 pass.
| claim | tier | criterion | measured | verdict |
|---|---|---|---|---|
prevalence-eir-pv |
2 | inside the IBM replicate band at every EIR on the grid | inside at 5 of 5; largest departure 2.1% of the IBM median, at EIR 3 | pass |
clinical-allage-eir-pv |
2 | inside the IBM replicate band at every EIR on the grid | inside at 5 of 5; largest departure 2.0% of the IBM median, at EIR 1 | pass |
clinical-under5-eir-pv |
2 | inside the IBM replicate band at every EIR on the grid | inside at 5 of 5; largest departure 2.4% of the IBM median, at EIR 0.3 | pass |
relapse-allage-eir-pv |
2 | inside the IBM replicate band at every EIR on the grid | inside at 5 of 5; largest departure 2.7% of the IBM median, at EIR 0.3, and within 0.9% from EIR 1 up | pass |
hypnozoite-carriage-eir-pv |
2 | inside the IBM replicate band at every EIR on the grid | inside at 5 of 5; largest departure 2.6% of the IBM median, at EIR 0.3, and within 0.7% from EIR 1 up | pass |
age-profile-clinical-pv |
2 | inside the IBM replicate band in every age band carrying at least 5% of clinical episodes, at EIR 1, 3 and 10 | inside at 23 of 23 tested bands across the three EIRs, holding 90% of episodes; largest departure 1.24 replicate SD, in the 3-5 y band at EIR 10 | pass |
intervention-impact-pv |
2 | outside the IBM replicate band in no greater a share of cells than the best held-out IBM replicate, over every intervention and outcome at EIR 1, 3 and 10 | fleet outside in 26.7% of 60 cells, against 8.3% for the closest of the twenty IBM replicates and 24.2% for the median one | FAIL |
population-age-structure-pv |
2 | inside the IBM replicate band in every age band below 60 years, on shares renormalised to the 0-60 population | inside at 11 of 11 at EIR 3; largest departure 2.1% of the IBM median | pass |
speed-pv |
2 | at least 10x faster per simulated year than one IBM run of 10,000 people, each timed alone on one core, at EIR 1, 3 and 10, with nothing deployed and with a seasonal programme | 1.3x to 2.5x per simulated year at 10,000 people, 4.5x to 7.3x at 50,000; fleet 0.63 to 0.92 s per simulated year, the IBM 1.2 to 1.6 s at 10,000 people | FAIL |
seed-stability-pv |
1 | PvPR(2-10) flat to under 1% over the last five of 30 years, at EIR 0.3, 1, 3 and 10 | flat to 0.27% at EIR 0.3, 0.09% at 1, 0.007% at 3 and 0.002% at 10; settling 15%, 10%, 6% and 3% above the seeded PvPR | pass |
prevalence-eir-pv
pass LM prevalence in 2-10 year olds tracks the IBM across transmission intensity.
Criterion: inside the IBM replicate band at every
EIR on the grid
Measured: inside at 5 of 5; largest departure 2.1% of
the IBM median, at EIR 3
clinical-allage-eir-pv
pass All-age clinical incidence tracks the IBM across transmission intensity.
Criterion: inside the IBM replicate band at every
EIR on the grid
Measured: inside at 5 of 5; largest departure 2.0% of
the IBM median, at EIR 1
fleet sits 1.2 to 2.0% above the IBM median at every EIR from 1 up, a small offset of one sign, inside the band throughout. Part of it is the school-age excess of immunity sorting within cells, which fleet’s one immunity distribution per cell does not carry (vignette(“model”, package = “fleet”), V.7).
clinical-under5-eir-pv
pass Under-5 clinical incidence tracks the IBM across transmission intensity.
Criterion: inside the IBM replicate band at every
EIR on the grid
Measured: inside at 5 of 5; largest departure 2.4% of
the IBM median, at EIR 0.3
relapse-allage-eir-pv
pass All-age relapse incidence tracks the IBM across transmission intensity.
Criterion: inside the IBM replicate band at every
EIR on the grid
Measured: inside at 5 of 5; largest departure 2.7% of
the IBM median, at EIR 0.3, and within 0.9% from EIR 1 up
Relapses are counted over everyone and divided by the 0-100 band’s population, which in the IBM leaves out its 0.7% over 100 and in fleet holds everyone, so fleet reads about 0.7% low by construction; most of its shortfall from EIR 1 up is that. Dividing both by the whole population needs the IBM rows re-summarised, at the vivax suite’s next IBM run.
hypnozoite-carriage-eir-pv
pass The share of people carrying hypnozoites tracks the IBM across transmission intensity.
Criterion: inside the IBM replicate band at every
EIR on the grid
Measured: inside at 5 of 5; largest departure 2.6% of
the IBM median, at EIR 0.3, and within 0.7% from EIR 1 up
At EIR 30 fleet is 0.66% below the IBM median, against a band 0.79% wide, and nearly all of that is the denominator of relapse-allage-eir-pv: carriage is counted over everyone and divided by the 0-100 band, which in the IBM leaves out the 0.7% over 100.
age-profile-clinical-pv
pass The age distribution of clinical incidence tracks the IBM.
Criterion: inside the IBM replicate band in every
age band carrying at least 5% of clinical episodes, at EIR 1, 3 and
10
Measured: inside at 23 of 23 tested bands across the
three EIRs, holding 90% of episodes; largest departure 1.24 replicate
SD, in the 3-5 y band at EIR 10
The band nearest the edge is 3-5 y at EIR 10, where fleet’s 2.225 episodes per child-year sit 4.1% above the IBM median of 2.137, 0.04 replicate SD inside the edge. Most of that is the offset of clinical-allage-eir-pv: fleet sits above the IBM median in most school-age bands, where immunity sorting within cells leaves the mean field a little high (vignette(“model”, package = “fleet”), V.7). The pass is that close: with a level for every hypnozoite batch count (ode_tuning(batch_bins = FALSE)) the band sits at 1.30 SD, 0.02 SD outside. No untested band is outside.
intervention-impact-pv
FAIL The modelled impact of each intervention, radical cure included, matches the IBM.
Criterion: outside the IBM replicate band in no
greater a share of cells than the best held-out IBM replicate, over
every intervention and outcome at EIR 1, 3 and 10
Measured: fleet outside in 26.7% of 60 cells, against
8.3% for the closest of the twenty IBM replicates and 24.2% for the
median one
Twelve of the sixteen cells outside are indoor residual spraying, and there most of the distance is a defect in malariasimulation 3.0.0. The IBM evaluates vivax infections, relapses included, only on days when at least one person receives an infectious bite: simulate_infection() returns early on an empty bitten set, which is right for falciparum and wrong for a relapse, which needs no bite. Spraying makes days with no infectious bite across the whole 10,000-person population common, every relapse is lost on them, and the IBM eliminates vivax (every replicate at EIR 1 and 3, 13 of 20 at EIR 10) where fleet, relapsing every day, keeps it. The other four are bed nets at EIR 1, where the IBM’s nets cut every outcome 6 to 9 points more than fleet’s: nets kill mosquitoes as well as deflecting them, and at EIR 1 that leaves days with no infectious bite too. A patched copy of malariasimulation that evaluates relapses every day brought the IBM’s reductions under spraying at EIR 3 and 10 and under nets at EIR 1 within 2.5 points of fleet’s, from up to 20 points away; that check was made against fleet’s earlier, continuous-time model and is not yet reproduced by a script here. Without the spraying scenarios fleet is outside in 8.3% of 48 cells, against 4.2% for the closest IBM replicate. Radical cure, by primaquine and by tafenoquine, agrees inside the band in every cell, as do treatment scale-up and bed nets at EIR 3 and 10.
population-age-structure-pv
pass The population age structure matches the IBM’s.
Criterion: inside the IBM replicate band in every
age band below 60 years, on shares renormalised to the 0-60
population
Measured: inside at 11 of 11 at EIR 3; largest
departure 2.1% of the IBM median
The vivax population ages and dies through compartments of its own, so its denominator is checked apart from falciparum’s, under the same 0-60 convention.
speed-pv
FAIL fleet is fast enough to be worth using in place of the IBM.
Criterion: at least 10x faster per simulated year
than one IBM run of 10,000 people, each timed alone on one core, at EIR
1, 3 and 10, with nothing deployed and with a seasonal programme
Measured: 1.3x to 2.5x per simulated year at 10,000
people, 4.5x to 7.3x at 50,000; fleet 0.63 to 0.92 s per simulated year,
the IBM 1.2 to 1.6 s at 10,000 people
Seconds per simulated year; in brackets, the IBM’s multiple of fleet’s time.
| setting | EIR | fleet | IBM 10,000 | IBM 30,000 | IBM 50,000 |
|---|---|---|---|---|---|
| nothing deployed | 1 | 0.65 | 1.19 (1.8×) | 2.33 (3.6×) | 3.29 (5.1×) |
| 3 | 0.63 | 1.31 (2.1×) | 2.58 (4.1×) | 3.68 (5.8×) | |
| 10 | 0.63 | 1.56 (2.5×) | 3.19 (5.1×) | 4.54 (7.3×) | |
| seasonal programme | 1 | 0.92 | 1.22 (1.3×) | 2.76 (3.0×) | 4.17 (4.5×) |
| 3 | 0.91 | 1.23 (1.4×) | 2.72 (3.0×) | 4.48 (5.0×) | |
| 10 | 0.91 | 1.18 (1.3×) | 3.33 (3.7×) | 4.61 (5.1×) |
One run of each model, alone on one core (Snapdragon X 12-core
X1E80100 @ 3.40 GHz), over 10 years from set_equilibrium()’s seed,
start-up excluded: fleet’s takes 0.60 to 1.0 s, the IBM’s 3.4 to 3.7 s
at 10,000 people and 7.9 to 8.2 s at 50,000. fleet’s cost does not
depend on the population. The seasonal programme is case management with
chloroquine and primaquine radical cure, bed nets and indoor spraying,
running from the first day.
CMP_PARASITE=pv validations/02-scenarios/speed.R makes this
table.
Each run in the table starts from set_equilibrium()’s seed and is timed alone, start-up apart. Each fleet figure is one complete run_simulation_ode() call, the fastest of two. With nothing deployed a vivax fleet run costs about eight times a falciparum one, for the hypnozoite dimension (eight levels for the eleven batch counts; vignette(“model”, package = “fleet”), V.1) and the within-cell immunity spread, and the programme adds 42 to 45%, most of it primaquine radical cure’s daily transfers. The IBM costs a little less for vivax than for falciparum. So at 10,000 people fleet is 1.8 to 2.5 times faster with nothing deployed and 1.3 to 1.4 times faster under the programme. Its cost does not grow with the population, so at 50,000 people it is 4.5 to 7.3 times faster. One fleet run also stands in for the twenty replicates an IBM comparison needs: across the vivax scenario suite, whose IBM runs share a worker pool, the IBM takes 9.5 CPU-hours and fleet 20 minutes.
seed-stability-pv
pass An undisturbed run settles to a steady state, off the equilibrium it was seeded at, as the IBM’s does.
Criterion: PvPR(2-10) flat to under 1% over the last
five of 30 years, at EIR 0.3, 1, 3 and 10
Measured: flat to 0.27% at EIR 0.3, 0.09% at 1, 0.007%
at 3 and 0.002% at 10; settling 15%, 10%, 6% and 3% above the seeded
PvPR
No figure: the numbers above decide this claim.
The criterion is settling rather than holding the seed, because the vivax seed is neither model’s fixed point. malariaEquilibriumVivax solves each age, heterogeneity and batch cell with one immunity value, where both models build up a spread of immunity within a cell from the first day, and the vivax immunity curves are steep enough for that to move everything: over the four EIRs fleet’s clinical incidence settles 16 to 22% above the seed and its realised EIR 5 to 18% above init_EIR, PvPR overshooting first at the lowest EIRs (21.5% at EIR 0.3). The IBM moves as far: in the tier-2 runs its realised EIR ends 18%, 13% and 8% above init_EIR at EIR 0.3, 1 and 3, against fleet’s 18%, 14% and 9%. The control says why: with immunity_spread = FALSE fleet holds its seed at EIR 3 to within 1.3%. So every vivax comparison is made after a 30-year burn-in in both models (long enough at EIR 0.3, where settling takes over two decades), and a vivax run started from set_equilibrium() needs one. validations/01-seed-stability/run.R reproduces the figures.