Bayesian population-based assessment of ascertainment bias in flea-borne typhus surveillance in California, 2011–2019

Abstract

In California, public health disease surveillance data for flea-borne typhus (FBT) are generated by healthcare providers and laboratories who are responsible for notifying local health jurisdictions when the disease is detected. When accounting for the associations between socioeconomic status (SES) factors (age, race and ethnicity, poverty, unemployment, housing status and land use), R. typhi reservoir host presence (e.g., rats, cats, opossums), and healthcare-seeking behaviours, it is reasonable to consider whether these factors also lead to under-reporting of FBT surveillance and distorted estimations of incidence. This study aimed to evaluate population-level associations between SES factors and FBT surveillance reporting using a Bayesian hierarchical model including a spatially autocorrelated random effect. Census tract-level covariates were sourced from the American Community Survey and Healthy Places Index (HPI). Specifying a zero-inflated Poisson distribution to FBT surveillance report counts, we estimated spatially smoothed, census tract-level estimates of FBT surveillance report rates and attributed variability in report rates to census tract characteristics. Socioeconomic advantage, as measured by the HPI, had the largest effect (IRR = 1.34 [1.07, 1.69]), corresponding to a 34% increase in FBT surveillance reporting for every point increase in HPI score for census tracts. The results herein suggest that FBT surveillance may be biased in its ascertainment of surveillance data, which may be helpful in contextualising and interpreting current trends in FBT epidemiology.