Nutrition Optimization Child Model Extension to include Inpatient SAM Treatment

Project overview

This model will be built off of the existing Nutrition Optimization simulation and extended to support the distinction between complicated and uncomplicated severe acute malnutrition in order to compare the relative priority of treatment for each condition. This model will use the nutrition optimization pregnancy simulation without changes from the previous version and will make several changes to the nutrition optimization child simulation (outlined below).

Simulation Design

Model components

Pregnancy model: unchanged from those listed in the nutrition optimization pregnancy simulation concept model document

Child model:

Components unchanged from the existing nutrition optimization child model:

Artifact-only updates (data updates to come):

Components that need to be altered to support new wasting state (more details to come):

  • Child underweight risk exposure: exposure lookup table needs to support additional wasting state (apply existing cat1 values to both cat1_complicated and cat1_uncomplicated)

  • CGF risk effects: EMR RRs for both SAM states set to zero, incidence rate RRs set to cat1_superstate value for both SAM substates

Components with substantial updates (links/updates to come)

Scenario descriptions

Baseline scenario:

The child baseline scenario will be run using fertility input data from the baseline scenario of the pregnancy model and with baseline coverage of all modeled child interventions.

Intervention scenarios:

We will run all child simulation scenarios using fertility input data from a new scenario of the pregnancy simulation: mms_at_anc1 where coverage of MMS is set to be equal to the proportion of ANC1 attendance according to the GBD covariate.

There will be a total of 54 child intervention scenarios.

Download this excel file for a list of intervention scenarios and their specified intervention coverage levels.

Default specifications

The below table is intended to outline the default specifications of your simulation. Included in the table is a column of parameter definitions. Please delete this column as you fill out the table.

Default simulation specifications

Parameter

Definition

Value

Note

Location(s)

Nigeria (modeled at the subnational level)

We will be modeling simulants at the subnational level in Nigeria and aggregating results to the national level. In practice, we should use the vivarium_gates_nutrition_optimization_child PopulationLineList component (as opposed to the EvenlyDistributedPopulation) and set subnational: All in the model spec file. This will ensure that simulants are distributed across all Nigerian subnational locations proportionately to subnational population size.

Number of draws

Desired number of draws that a given simulation is to be run for. (Generally, this should be a number between 1 and 1,000.)

20 to start

Population size per draw

Desired simulated population size per draw for a given simulation.

100,000 pregnancies

Note

All other specifications remain unchanged from the nutrition optimization pregnancy and nutrition optimization child simulation specifications

Simulation Observers

Default stratifications for all observers should include:

  • Input draw

  • Scenario

  • Sex

  • Age group

    • Default: GBD age groups

    • Alternative groupings for production runs:

      • 0-6 months

      • 6-18 months

      • 18-60 months

Simulation observers

Number

Observer

Observations

Default stratifications

Note

1

Mortality

  • deaths

  • ylls

Include: Wasting state

Cause-specific

2

Disability

  • ylds

Cause-specific

3

Child wasting

  • person_time_child_wasting

  • transition_count_child_wasting

Include: MAM treatment coverage, Uncomplicated SAM treatment coverage, Complicated SAM treatment coverage, SQ-LNS coverage

4

Child stunting

  • person_time_child_stunting

Include: SQ-LNS coverage

5

Child underweight

  • person_time_child_underweight

This observer is for V&V and can be excluded from production runs

6

Disease observers

For cause in [diarrheal_diseases, lower_respiratory_infections, measles, malaria]: * person_time_{cause} * transition_count_{cause}

Include: wasting state

These observers are for V&V and can be excluded from production runs

Verification & validation (V&V) tracking

This section is intended for tracking the progress of V&V of simulation results.

The below tables can be filled out iteratively as new model runs are requested and later V&V’d.

Note

More detail will be added to these runs as we become closer to launching them – details and V&V criteria may change depending on whether we are running with placeholder data or not

A note on fertility input data from the pregnancy model.

  • First pass fertility input data specs

    • Filepath: /mnt/team/simulation_science/pub/models/vivarium_gates_nutrition_optimization/results/vivarium_v4.0_vph_v5.0_update/v3/nigeria/2026_03_31_12_55_44

    • Input draws: [8, 13, 41]

    • Random seeds: [4344, 5616, 6810, 2787, 2284]

    • Maternal scenario: baseline

  • Updated fertility input data specs

    • Filepath: [path for TO-BE-GENERATED RESULTS]

    • Input draws: random sample of 20 draws between 1-99. Note we avoid draw 0 here as it has been overwritten with mean across all draws in N.O. artifacts

    • Random seeds: Unspecified, up to engineers

    • Maternal scenarios:

      • Baseline

      • NEW TO-BE-IMPLEMENTED SCENARIO of MMS coverage equal to ANC1 value

Todo

Fill in V&V summaries for all runs and run specs for 2.1 and 2.2

Model runs

Run number

Run description

Preg. Sim Specs

Scenario

Specification modifications

Stratification modifications

Observer modifications

1.0

Wasting model update (without updates to the PEM model, but with no mortality due to infectious diseases in the SAM states)

First pass fertility input data specs

Baseline

[8, 13, 41]

Default

Default

2.0

Updates to PEM model

Ideally Updated fertility input data specs, but first pass can be used if updated spec results are not ready

Baseline

10 draws with updated fertility input specs or [8, 13, 41] with first pass specs

Default

Default

3.0

  • Intervention model update implementation

  • Used Kano-only calibration data wasting_processed_rates_kano_20260428.csv found here. Note this data was missing the complicated_fraction parameter, which was filled using values from the prior run (2.2)

First pass fertility input data specs

  • Baseline maternal scenario

  • Baseline child scenario and: 0-7, 36 (uncomplicated_sam_tx__complicated_sam_stabilization__), 45 (uncomplicated_sam_tx__complicated_sam_recovery__)

10 draws

Default

Default

3.1

  • Update to complicated SAM treatment baseline coverage (was zero in 3.0)

  • Updated wasting calibration values in wasting_processed_rates_kano_20260429.csv found here (fixed issue with missing complicated_fraction parameter from 3.0)

  • Updated SQ-LNS effect values [ALI TO LINK SOON]

First pass fertility input data specs

Baseline maternal scenario and the following child scenarios: [baseline, zero_coverage__, targeted_mam_tx__, universal_mam_tx__, targeted_mam_tx__universal_sqlns__, universal_sqlns__, universal_mam_tx__universal_sqlns__, complicated_sam_stabilization__, complicated_sam_stabilization__targeted_mam_tx__, complicated_sam_stabilization__universal_mam_tx__, complicated_sam_stabilization__targeted_mam_tx__universal_sqlns__, complicated_sam_stabilization__universal_sqlns__, complicated_sam_stabilization__universal_mam_tx__universal_sqlns__, complicated_sam_recovery__, complicated_sam_recovery__targeted_mam_tx__, complicated_sam_recovery__universal_mam_tx__, complicated_sam_recovery__targeted_mam_tx__universal_sqlns__, complicated_sam_recovery__universal_sqlns__, complicated_sam_recovery__universal_mam_tx__universal_sqlns__, uncomplicated_sam_tx__, uncomplicated_sam_tx__targeted_mam_tx__, uncomplicated_sam_tx__universal_mam_tx__, uncomplicated_sam_tx__targeted_mam_tx__universal_sqlns__, uncomplicated_sam_tx__universal_sqlns__, uncomplicated_sam_tx__universal_mam_tx__universal_sqlns__, uncomplicated_sam_tx__complicated_sam_stabilization__, uncomplicated_sam_tx__complicated_sam_stabilization__targeted_mam_tx__, uncomplicated_sam_tx__complicated_sam_stabilization__universal_mam_tx__, uncomplicated_sam_tx__complicated_sam_stabilization__targeted_mam_tx__universal_sqlns__, uncomplicated_sam_tx__complicated_sam_stabilization__universal_sqlns__, uncomplicated_sam_tx__complicated_sam_stabilization__universal_mam_tx__universal_sqlns__] aka: numbers [baseline, 0, 1, 4, 6, 7, 8, 9, 10, 13, 15, 16, 17, 18, 19, 22, 24, 25, 26, 27, 28, 31, 33, 34, 35, 36, 37, 40, 42, 43, 44]

10 draws

Use the following age groups: * Early neonatal (GBD) * Late neonatal (GBD) * 1-5 months (GBD) * 6-12 months (GBD) * 12-18 months (Custom for SQLNS eligible age) * 18-24 months (Custom for SQLNS eligible age) * 2-4 years (GBD)

Default

4.0

Production run test

Updated fertility input data specs

Baseline and MMS at ANC1 maternal scenarios, baseline child scenario

10 draws

  • Same age groups as 3.1, or 0-6 months, 6-18 months, 18-60 months for run time

  • Can exclude wasting state stratification from mortality observer for run time

Can exclude child underweight and disease observers (observers #5 and #6) for run time

4.1

Production runs

Updated fertility input data specs

  • Baseline (maternal and child)

  • MMS at ANC maternal scenario and child scenarios listed for 3.1

20 draws

Same as 4.0

Same as 4.0
Model verification and validation tracking

Run number

V&V criteria

V&V summary

1.0

Note this model is not expected to validate to GBD with regard to mortality or wasting exposure given that PEM mortality has not been included in this run. The following checks can be performed:

  1. Confirm that underweight exposure in the SAM substates from the simulation does not vary by SAM substate and matches the expectation of underweight exposure in the SAM superstate from the artifact

  2. Confirm that no deaths due to diarrheal diseases, LRI, malaria, or measles occurred in either SAM substate

  3. Confirm that CGF RRs are functioning as expected (given asssumption that SAM substates both have the SAM superstate RRs for incidence rates while having EMR RRs equal to zero)

  4. Confirm that the wasting transition rates match input data expectation

  5. Confirm that the wasting state initialization (in neonatal age groups) in the sim matches input data expectation for the substate-specific input data from the artifact (relevant age group = 1-5 months)

2.0

  1. Verify population ACMR to GBD

  2. Verify wasting state-specific mortality rates are as expected

  3. Verify CGF exposures

  4. Confirm all 1.0 criteria as well

3.0

  • Confirm all intervention effects and coverage match expectations
Outstanding verification and validation issues

Issue

Explanation

Action plan

Timeline