Hotspot or blind spot? Fine-scale spatial heterogeneity and methodological bias in Chlamydia psittaci prevalence rate reports from urban feral pigeons (Columba livia f. urbana).
Description
Urban wildlife populations often carry zoonotic pathogens that pose considerable public health risks through direct contact with humans. However, disease prevalence in urban settings is generally reported at the city level, which disregards spatial heterogeneity due to local differences in habitat structure or resource availability in the urban landscape. This study therefore examines the fine-scale spatial heterogeneity of Chlamydia psittaci, a generalist bacterium that frequently infects feral pigeons, in Antwerp, Belgium. We collected samples from 377 pigeons at 23 locations and performed qPCR screening for C. psittaci. At the same time, we performed an extensive literature review that includes 20 studies from 29 unique cities globally. Our findings indicate that the prevalence variance within Antwerp (CV = 0.70) is comparable to the variation reported among cities worldwide (CV = 0.88), suggesting that distinct locations within a single city can exhibit differences comparable to those between entirely different cities. Our data suggest that citywide or even countrywide are likely confounded by small scale spatial infection heterogeneity. A combination simulation showed that at least 12 unique sampling sites are necessary to accurately assess the true prevalence at the city level. Finally, we could show that also the screening method influences reported prevalence, with blood samples and non-PCR screening inflating reported prevalence rates. Taken together, we recommend that urban surveillance reports include at least 12 sampling sites, use standardized screening protocols and provide site-level data so that fine-scale heterogeneity can be taken into account.
Files
Steps to reproduce
All statistical analyses were performed in R 4.2.1. Variance tests used car (v3.1.2), mixed-effects modeling used lme4 (v1.1-35.5), and model diagnostics used DHARMa (v0.4.7). To assess sampling effort for stable city-level prevalence, we applied a combination simulation on Antwerp site-level prevalences (23 locations): for each 𝑘 = 1…23, we drew random subsets of 𝑘 distinct locations (without replacement), computed the mean prevalence, and repeated this 10,000 times per 𝑘; for each 𝑘, we derived the 95% interval (2.5th–97.5th percentiles) of simulated means, defined the “true” mean as the mean across all 23 locations, and identified the smallest 𝑘 at which the 95% interval remained within ±5 and, separately, ±10 percentage points of that mean. To compare fine-scale versus large-scale variability, we calculated the coefficient of variation (CV = sd/mean) for Antwerp site-level prevalences and for literature-based city–study prevalences comprising 31 estimates (29 unique cities from the literature, Madrid counted twice from two studies, plus the pooled Antwerp estimate); equality of variances between these two distributions was tested with Levene’s test using the median-centered Brown–Forsythe variant (car v3.1.2). To examine confounding by sample origin, we fitted a generalized linear mixed model with binomial error to literature-derived prevalence data assembled as follows: if a single city-wide prevalence was reported (text or tables), we used that value even when multiple sampling events were described; if no city-wide prevalence was available, we included all reported prevalence estimates from individual sampling events; when multiple sample types were reported, all such outcomes were included. The resulting dataset contained 39 prevalence estimates from 29 cities across 20 studies, with sample origins: blood (15), cloacal swab (11), combined cloacal + pharyngeal swabs (7), feces (4), and pharyngeal swab (2). In the GLMM, prevalences were the response, sample origin was included as a fixed effect, study ID as a random intercept to account for non-independence, and observations were weighted by the total number tested; detection method (e.g., PCR, RT-PCR, ELISA) was not included because ELISA was used exclusively with blood whereas PCR-based methods were used for the other sample types. Model fitting used lme4 (v1.1-35.5), and assumptions were evaluated with DHARMa (v0.4.7).
Institutions
- Universiteit Antwerpen Departement BiologieAntwerpen