EIR, HBR, prevalence, and cases

The four quantities

The entomological inoculation rate (EIR) is infectious bites per person per year, and it sets the force of infection, which is why stateMINT runs from it. estiMINT infers it from a prevalence or an HBR.

The human biting rate (HBR) is bites per person per year, infectious or not, which is the EIR divided by the sporozoite rate, so the numbers run to tens or hundreds of thousands per year. mosquito_delta scales mosquito density and reaches the transmission model through the HBR.

Prevalence is reported throughout MINTverse as the under-five value on a 0 to 1 scale.

Cases are clinical malaria episodes, all-age, per 1000 population, and an emulator output only. There is no cases-to-anything inversion to run. They are floored at zero.

Values and series

Quantity	What it is	Units	Value or series	Produced by
EIR	transmission intensity	infectious bites / person / year	single value	estiMINT
HBR	biting pressure	bites / person / year	single value	estiMINT
Prevalence	infected share, under-5	proportion, 0–1	input value or 157-step series	input or stateMINT
Cases	clinical episodes, all-age	per 1000 population	157-step series	stateMINT

estiMINT returns an equilibrium, the transmission intensity a setting sustains. It has no time axis, so it cannot say what the EIR will be two years after a PBO campaign, only what it would be in a setting already settled under one.

Only stateMINT returns a path.

EIR and equilibrium

estimate_eir_with_mosquito_delta returns two numbers, eir_baseline and eir_new. Both are equilibria. eir_new is the intensity a setting would settle at with a different mosquito density, not the EIR at some later date, so the two values do not bracket a trajectory.

The time grid

Every trajectory is laid on one grid of 157 fortnightly steps, built around the day the campaign lands. The simulations behind stateMINT run a six-year burn-in first, letting the transmission model settle into equilibrium. That burn-in is discarded and never reaches the emulator.

What remains starts at day 2190, the start of year 6, and the first three years are pre-campaign baseline. At day 3285, the start of year 9, the campaign switches on. dn0_future, itn_future, irs_future, lsm and routine take their value from that day onwards. They are zero before it.

The grid is written into the artefact.

import numpy as np
from stateMINT.model import Mamba2Regressor

artifact = Mamba2Regressor.from_pretrained(
    "dide-ic/stateMINT", predictor="prevalence", revision="v1.2.2",
)
cfg = artifact.preprocessing_config

abs_t = cfg["model_start_day"] + cfg["window_size"] * np.arange(cfg["n_steps"])
years = (abs_t - cfg["intervention_day"]) / 365
idx_y9 = int(np.argmin(np.abs(abs_t - cfg["intervention_day"])))

print(f"steps                {cfg['n_steps']}")
print(f"window               {cfg['window_size']} days")
print(f"grid starts          day {abs_t[0]}")
print(f"campaign             day {cfg['intervention_day']} -> step {idx_y9} (day {abs_t[idx_y9]})")
print(f"grid ends            day {abs_t[-1]}")
print(f"years from campaign  {years[0]:+.2f} to {years[-1]:+.2f}")

steps                157
window               14 days
grid starts          day 2190
campaign             day 3285 -> step 78 (day 3282)
grid ends            day 4374
years from campaign  -3.00 to +2.98

prev_y9 is the annual mean under-five prevalence in year 9, the year the campaign lands, so passing a measured prevalence to estiMINT asserts that it describes the setting on the day the nets are distributed.

The pre-campaign years show the setting under the nets it already has, so a baseline that starts somewhere other than the prevalence you supplied means the covariates do not describe your setting. Plotting in years puts the campaign at zero.

See also

The interventions that switch on at day 3285 are described in Interventions and nets, and the estimators and the emulator in The models. Please see Estimating EIR from prevalence for the inversion in practice.