Journal Design Clinical Emerald
African Food Systems Research (Interdisciplinary - incl Agri/Env) | 02 August 2000

A Bayesian Hierarchical Model for Assessing the Reliability of Public Health Surveillance Systems in Kenya

A Methodological Evaluation, 2000–2026
W, a, n, j, i, k, u, M, w, a, n, g, i, ,, K, a, m, a, u, O, c, h, i, e, n, g, ,, K, i, p, c, h, u, m, b, a, B, e, t, t, ,, A, m, i, n, a, H, a, s, s, a, n
Bayesian hierarchical modelsurveillance reliabilityhealth systemsspatial statistics
Substantial spatial variation in surveillance reliability identified across districts.
National annual reporting completeness estimated between 0.58 and 0.72.
Temporal random effects show declining reliability during health system strain.
Model provides framework to quantify uncertainty in surveillance data.

Abstract

{ "background": "Public health surveillance systems are critical for disease control, yet their reliability in resource-limited settings is often uncertain. Methodological evaluations of these systems are required to quantify their performance and guide improvements.", "purpose and objectives": "This study aimed to develop and apply a novel Bayesian hierarchical model to evaluate the reliability of national public health surveillance systems, using Kenya as a case study. The objective was to quantify system completeness and timeliness while accounting for spatial and temporal heterogeneity.", "methodology": "We developed a Bayesian hierarchical model integrating reported surveillance data with covariates including healthcare access and population density. The core model for the probability of a true event being reported in district $i$ and month $t$ is: $\\text{logit}(p{it}) = \\alpha + \\beta X{it} + ui + vt$, where $ui \\sim N(0, \\sigmau^2)$ and $vt \\sim N(0, \\sigmav^2)$ are random effects. Model parameters were estimated using Markov chain Monte Carlo simulation.", "findings": "The model identified substantial spatial variation in system reliability, with posterior probabilities of reporting completeness below 0.6 in over 30% of districts. The estimated national annual reporting completeness had a 95% credible interval of 0.58 to 0.72, indicating significant under-reporting. Temporal random effects showed a declining trend in reliability during periods of health system strain.", "conclusion": "The methodological evaluation reveals that surveillance reliability in the study context is moderate and highly variable, compromising data utility for real-time response. The Bayesian hierarchical model provides a robust framework for quantifying this uncertainty.", "recommendations": "Implement routine reliability assessments using this modelling framework to prioritise districts for surveillance strengthening. Integrate model outputs into data interpretation to adjust for under-reporting in epidemiological analyses.", "key words": "Bayesian hierarchical model, surveillance evaluation, health systems, reliability, spatial statistics, disease reporting", "contribution