malaria-modeling

name: malaria-modeling description: Disease-specific modeling guide for P. falciparum malaria. Covers transmission dynamics, intervention mechanisms, known modeling pitfalls, calibration targets, cost-effectiveness benchmarks, and published benchmark models. Use when the research question involves malaria incidence, prevalence, transmission, interventions, or resource allocation. Trigger phrases include "malaria", "PfPR", "ITN", "bednet", "IRS", "indoor residual spraying", "SMC", "seasonal malaria chemoprevention", "MDA", "mass drug administration", "Global Fund malaria", "malaria resource allocation", "malaria optimization".

Malaria (P. falciparum) Modeling Guide

1. Key Parameters

2. Critical Modeling Pitfalls

READ THIS SECTION BEFORE BUILDING ANY MALARIA MODEL. These are common errors that produce models with correct-looking calibration but fundamentally wrong intervention effects and optimization results.

Pitfall 1: PfPR-EIR Saturation

The PfPR-EIR relationship is nonlinear and saturating. At EIR > 100, PfPR plateaus around 70-80% due to superinfection and acquired immunity. An equilibrium formula without immunity saturation will force EIR to absurd values (1000+) to match observed PfPR in holoendemic areas, making the model insensitive to interventions in exactly the areas that need them most.

What goes wrong: The model calibrates to observed PfPR by inflating EIR far beyond published ranges. Then interventions that reduce EIR (ITN, IRS) have negligible effect on PfPR because the model is on the flat part of the saturation curve. The optimizer sees zero marginal benefit and allocates nothing to the highest-burden areas.

Solution: Use an empirical PfPR-EIR relationship that saturates correctly:

• Smith et al. 2005 (Malaria Journal): Hill function fit to field data
• Beier et al. 1999: logarithmic relationship from 31 African sites
• Griffin et al. 2010: mechanistic immunity acquisition model

Diagnostic: If any calibrated EIR exceeds 200 ib/p/yr, or if a calibrated parameter hits an optimizer bound, the model structure is inadequate for that setting. STOP and fix before proceeding.

Pitfall 2: Equilibrium Models Destroy Time-Limited Intervention Effects

SMC provides 4-month seasonal chemoprophylaxis to children 3-59 months. MDA provides a transient reduction in parasite prevalence. Both have effects concentrated in specific time windows.

What goes wrong: An equilibrium model averages the seasonal effect over 12 months, diluting a 73% reduction during 4 months into a ~20% annual average. This makes SMC look 3-4x less effective than it actually is. MDA's transient effect disappears entirely in equilibrium.

Rule: If ANY intervention has time-limited action (seasonal chemoprophylaxis, campaign-based MDA, seasonal IRS rounds), use a DYNAMIC model (ODE with seasonal time steps, difference equations, or agent-based), NOT algebraic equilibrium. This is a structural requirement, not a refinement.

Pitfall 3: Age Structure Is Required for SMC Evaluation

SMC targets children 3-59 months ONLY, which is ~15-17% of the population. Without age structure, the model cannot:

• Compute SMC's direct impact on the target population
• Correctly represent the age distribution of malaria burden
• Distinguish between interventions that protect everyone (ITN) vs those that protect a specific age group (SMC)

Rule: A model without age structure CANNOT evaluate SMC and should not include SMC as an optimization variable. Minimum age groups for SMC evaluation: 0-5y, 5-14y, 15+. Preferred: 0-5m, 6-59m, 5-14y, 15+.

Pitfall 4: ITN Effectiveness Is Context-Dependent

Pyrethroid resistance modifies ITN efficacy substantially. The relevant parameter depends on:

• Net type: standard LLIN (pyrethroid-only), PBO-pyrethroid, dual-AI (chlorfenapyr + pyrethroid)
• Vector species composition: An. gambiae s.s. vs An. arabiensis vs An. funestus (different insecticide susceptibility)
• Resistance intensity: metabolic vs target-site resistance

If the policy question involves net type selection or resistance scenarios: model at least 2 net types with different efficacy profiles.

Key reference: Protopopoff et al. 2018 (Lancet): PBO nets achieved 44% lower malaria prevalence than standard LLINs after 2 years in a pyrethroid-resistance setting.

Pitfall 5: IRS Has Threshold and Complementarity Effects

IRS efficacy is coverage-dependent due to community-level protection: the higher the coverage, the lower the overall vector density, benefiting even uncovered households.

What goes wrong: Setting IRS efficacy to zero below a coverage threshold (e.g., <40%) creates a cliff effect. A greedy optimizer sees zero marginal return from any IRS increment below the threshold, making it prohibitively expensive to reach the threshold. IRS then appears useless regardless of its true effectiveness.

Solution: Use a continuous dose-response for IRS efficacy, not a step function. Efficacy should increase smoothly with coverage.

Also critical: IRS complements ITN, especially in pyrethroid resistance settings (Pryce et al. 2022 Cochrane, CD012688.pub3). A model that reduces ITN efficacy for resistance but doesn't increase the relative value of non-pyrethroid IRS is missing the whole point of IRS in resistance contexts.

Pitfall 6: Published Benchmark Models MUST Be Compared Against

For Nigeria specifically:

• EMOD (IDM/Northwestern): Ozodiegwu et al. 2023 -- 774-LGA model with archetype calibration, directly informed GC7 request
• Optima Malaria (Burnet): Scott et al. 2017 -- allocative efficiency optimization across 6 geopolitical zones
• malariasimulation (Imperial College): Griffin et al. 2010+ individual-based with immunity dynamics
• OpenMalaria (Swiss TPH): Smith et al. -- detailed within-host dynamics, community-level effects

For any Global Fund or policy submission, the model must explain how and why its results differ from these established tools. If the model's allocation recommendation contradicts all published models, the model is almost certainly wrong -- not innovative.

3. Cost-Effectiveness Benchmarks

Published ranges for malaria interventions in Sub-Saharan Africa. These are MANDATORY checks for any optimization model.

Mandatory Check

IF THE MODEL PRODUCES COST-EFFECTIVENESS >5x OUTSIDE THESE RANGES FOR ANY INTERVENTION, THE MODEL STRUCTURE IS WRONG.

This is NOT a parameter calibration issue. It means the model's representation of that intervention's mechanism is inadequate. Do not proceed to optimization -- fix the intervention mechanism first.

Common causes of out-of-range cost-effectiveness:

• Equilibrium averaging of seasonal interventions (SMC, seasonal IRS)
• Missing age structure for age-targeted interventions (SMC)
• Step-function coverage thresholds creating optimizer cliff effects (IRS)
• PfPR-EIR saturation making interventions ineffective at high transmission

4. Model Structure for Malaria

Minimum Viable Model for Resource Allocation

Intervention Mechanisms

• ITN: Reduce vector-human contact rate (force of infection reduction). Effect proportional to coverage x efficacy. Waning: 2-3 year half-life for pyrethroid LLINs.

• IRS: Reduce vector density and survival. Continuous dose-response with coverage (NOT step function). Effect strongest in first 3-6 months post-spraying, requiring annual or biannual rounds.

• SMC: Seasonal chemoprophylaxis for children 3-59 months during peak transmission (typically 4 months). Reduce FOI for target age group during active window. Must be modeled with explicit seasonality and age structure -- cannot be represented in equilibrium.

• Treatment/case management: Reduce infectious duration (faster clearance with ACT) and prevent mortality. Coverage varies by treatment-seeking behavior and health system access.

What NOT to Do

• Do NOT use algebraic equilibrium for models with seasonal interventions
• Do NOT average SMC effect across 12 months
• Do NOT use step-function efficacy thresholds for IRS
• Do NOT use a single national R0 when sub-national variation is >5-fold
• Do NOT model ITN and IRS effects as independent when they target the same vector population (they are sub-additive at best)

Why Simple ODEs Fail for Malaria

Simple SIR/SEIR/SEDATU ODEs cannot sustain PfPR above ~30% because they deplete susceptibles without modeling superinfection and acquired immunity. This is a fundamental limitation, not a parameter tuning issue. The Hill function "solves" calibration by abandoning dynamics entirely, but then intervention effects are just static OR multipliers — not mechanistic.

The key missing mechanism is immunity: in malaria, people get reinfected repeatedly, developing partial immunity that limits disease severity without preventing infection. This is why PfPR can reach 70%+ in holoendemic areas — most infections are asymptomatic. Models need separate clinical immunity and anti-parasite immunity compartments, plus superinfection, to reproduce this.

Published Model Implementations

Several open-source malaria models solve the immunity problem. Consider adapting one rather than reimplementing these complex dynamics:

5. Intervention Effect Validation Ranges

BEFORE running optimization, verify the model produces plausible marginal intervention effects. Print a table:

Expected Ranges (from Cochrane/systematic reviews)

STOP Conditions

If any of these are true, STOP and report to the lead agent. Do NOT proceed to optimization with wrong intervention effects:

• ITN at 80% reduces PfPR by <1% in any zone (implausible)
• SMC at 80% averts <10% of cases in target age group during season
• IRS has zero effect at any coverage level (likely step-function artifact)
• Any intervention's cost-effectiveness is >5x outside published ranges
• A calibrated EIR exceeds 200 ib/p/yr or hits an optimizer bound

6. Nigeria-Specific Context

Epidemiology

• National PfPR (microscopy, 6-59m): 22.6% (NMIS 2021)
• Zone range: 14.2% (South West) to 33.5% (North West), 2.4-fold variation
• WHO 2023 estimates: 68M cases, 184,800 deaths, 299/1000 incidence
• Dominant vectors: An. gambiae s.s. (indoor), An. arabiensis (outdoor)
• Pyrethroid resistance: widespread, especially North West and North Central

Geographic Units

• 6 geopolitical zones: NW, NE, NC, SW, SE, SS
• 36 states + FCT (Federal Capital Territory)
• 774 Local Government Areas (LGAs)
• Zone-level is minimum for resource allocation; state-level preferred

Key Policy Context

• Global Fund GC7 (2024-2026): ~$993M total (HIV+TB+malaria); malaria component estimated ~$320M (not publicly disaggregated)
• Nigeria's GC7 request includes SMC expansion to 404 LGAs (from ~80)
• WHO 2024 guidelines recommend PBO/dual-AI nets for resistance areas
• NW zone: highest burden, ~25% of population. ANY credible resource allocation should invest in NW.

Calibration Targets

Published Allocation Results (for comparison)

If your model contradicts all three of these studies, investigate your model first, not the literature.

7. Key References

8. Handoff to Downstream Skills

When building malaria models, also use:

• epi-model-parametrization: For structured parameter spaces and calibration target design
• laser-spatial-disease-modeling: If using LASER framework for spatial ABM
• modelops-calabaria: For Bayesian calibration and cloud-scale optimization
• model-validation: For validation gates before accepting optimization results
• model-fitness: For evaluating whether model is fit for its stated purpose