Skip to contents

Estimate Beta Prior for Rho (Care-Seeking Rate) and Save Parameter Data Frame

Source: R/get_rho_care_seeking_params.R
get_rho_care_seeking_params.Rd

Why pool the two Wiens strata?

Symptomatic cholera in MOSAIC's setting spans the full severity spectrum (mild watery diarrhea to severe rapid dehydration). The two Wiens 2025 meta-analytic strata bracket that spectrum:

  • General diarrhea (3+ loose/liquid stools): 29.9% (95% CI [25.3, 35.1]) from 122 observations. Represents care-seeking for the broader population of diarrheal episodes, which is appropriate for the mild-to-moderate tail of symptomatic cholera (the majority of cases by count).

  • Severe diarrhea + cholera: 58.6% (95% CI [39.9, 75.2]) from 22 observations. Represents care-seeking when symptoms are severe enough to be flagged as "severe" or specifically cholera, which is appropriate for the severe tail of symptomatic cholera.

Anchoring on the severe+cholera stratum ALONE biases rho upward (it represents only the severe tail and is dominated by outbreak-response settings with enhanced care-seeking, e.g. the Haiti 2010-19 outbreak which contributes 50% of underlying rows). Anchoring on general diarrhea ALONE biases rho downward (the broader category includes many mild, self-resolving episodes that don't reflect cholera-specific severity). Random-effects pooling of the two strata produces an estimate that honestly reflects the severity spectrum and its substantial heterogeneity.

Why not include GEMS Nasrin 2013 as a separate co-anchor?

Earlier versions (through v0.32.5) pooled GEMS pediatric moderate-to-severe diarrhea (4 SSA sites, 12 stratum-level estimates) WITH the Wiens severe+cholera pooled estimate. That pooling had three problems: (1) GEMS measures pediatric MSD care-seeking, a different population than MOSAIC's all-ages cholera; (2) 12 GEMS strata vs. 1 Wiens meta-analytic summary was upside-down dimensionally; (3) Wiens 2025 already includes GEMS-derived data (6 of its Study IDs are tagged GEMS/HUAS), so the direct GEMS entries partially double-counted the same source populations. Dropping GEMS and using Wiens's two strata directly resolves all three.

Methodology. Random-effects meta-analysis on the logit scale (DerSimonian-Laird tau^2) with the two Wiens strata as the unit of pooling. The 95% CI of the pooled mean is then fit to a Beta distribution via get_beta_params.

Geographic provenance (severe+cholera stratum). Of 23 underlying studies: 16 SSA, 3 Latin America/Caribbean (mostly Haiti 2010-19), 2 Bangladesh, 1 MENA, 1 South Asia (non-BGD). Predominantly SSA + outbreak settings - the regime MOSAIC calibrates.

Usage

get_rho_care_seeking_params(PATHS)

Arguments

PATHS

A list containing paths for saving outputs. Typically generated by get_paths() and must include:

  • MODEL_INPUT: Directory where the parameter CSV will be saved.

Value

A data frame with one row per parameter (shape1, shape2, mean) for the rho Beta prior, invisibly. Also writes param_rho_care_seeking.csv to PATHS$MODEL_INPUT.

Details

This function encodes the prior on rho (the care-seeking rate: probability that a true symptomatic cholera infection presents to the surveillance pipeline as a suspected case) by random-effects pooling of the TWO case-definition strata reported by Wiens et al. 2025 (PMC12013865): general diarrhea and severe-diarrhea + cholera.