A new surveillance modeling study reveals that spatial statistical models can generate substantial apparent neighbor-associated transmission signals even in the documented absence of person-to-person spread. Evaluating weekly notification data from Japan between 2013 and 2025, researchers tested negative-binomial endemic–epidemic time-series models against three tick- and mite-borne zoonoses: Japanese spotted fever, scrub typhus, and severe fever with thrombocytopenia syndrome. Because these acarine-borne infections lack sustained human-to-human transmission, they provide an empirical biological negative-control framework for examining whether short-lag associations between notification counts genuinely reflect transmission between administrative areas.
Endemic–epidemic surveillance models decompose disease counts into endemic components, within-area epidemic dynamics, and neighbor-associated spatial components. In public health research, elevated neighbor-associated components for directly transmissible pathogens are frequently interpreted as quantitative evidence of geographic pathogen spread across jurisdictional boundaries. However, the study demonstrated that model specifications regarding baseline background risk, shared temporal dynamics, and spatial weight matrices exert a profound influence on whether spurious spatial coupling emerges.
The empirical analysis evaluated weekly notifications across all 47 prefectures of Japan over the 13-year observation window from 2013 to 2025. Investigators fitted negative-binomial endemic–epidemic models across five distinct area-baseline specifications. These configurations ranged from a single, uniform national intercept to highly localized prefecture-specific baseline effects, allowing direct observation of how baseline risk assumptions alter the partitioning of observed case counts.
When models were fitted using a single national intercept, the estimated neighbor-associated fractions were substantial across all three non-transmissible infections. The neighbor-associated fraction reached 0.235, with an approximate 95% interval of 0.211 to 0.259, for Japanese spotted fever. For scrub typhus, the estimated fraction under the single national intercept was 0.151, with an approximate 95% interval of 0.132 to 0.172, while severe fever with thrombocytopenia syndrome yielded a fraction of 0.149, with an interval of 0.120 to 0.180.
Shifting the model specification to account for prefecture-specific baselines markedly diminished these spatial components. Under prefecture-specific baseline effects, the estimated neighbor-associated fraction plummeted to below 0.001 for Japanese spotted fever and to 0.002 for severe fever with thrombocytopenia syndrome. For scrub typhus, the neighbor-associated fraction decreased to 0.063. This sharp contraction highlights how unmodeled regional baseline heterogeneity can be incorrectly absorbed into spatial neighbor terms under simpler model structures.
Extensive simulation experiments further elucidated the mechanisms that drive these statistical artifacts. Simulations systematically crossed baseline structures and spatial coupling strengths, employing identical generating and fitted spatial operators in a principal reference grid. Additional disease-calibrated simulations generated counts using row-normalized adjacency matrices and fitted them with binary adjacency matrices to mirror the empirical modeling specifications.
In synthetic simulations where true coupling strength was explicitly set to zero, spatially smoothed baseline heterogeneity produced mean neighbor-associated fractions of 0.083 under the common-intercept estimator in the reference grid. In disease-calibrated zero-coupling simulations, baseline heterogeneity yielded mean fractions ranging from 0.027 to 0.060, representing 18% to 40% of the empirical estimates. Furthermore, four additional mechanisms operating under zero true coupling produced mean neighbor-associated fractions between 0.069 and 0.112 when evaluated with the common-intercept estimator.
The study also assessed the recovery of genuine coupling when true spatial spread was present at a known fraction of 0.142. Under this scenario, the common-intercept estimator markedly inflated the signal, returning a fraction of 0.224. In contrast, prefecture-specific baselines accurately recovered the true value, returning 0.141 when the fitted spatial operator properly matched the generating operator.
Based on these findings, the authors concluded that estimated neighbor-associated fractions should be treated as specification-dependent summaries of statistical dependence rather than direct evidence of geographic transmission. Because non-zero fractions readily emerge under zero-coupling conditions and genuine coupling recovery requires precise operator alignment, researchers emphasized that neighbor-associated fractions must be reported across alternative baseline specifications and spatial weights, and corroborated with independent external data before being interpreted as spatial transmission.
Subject of Research: Evaluation of neighbor-associated components in endemic–epidemic disease surveillance models
Article Title: Neighbor-associated components in endemic–epidemic surveillance models: scrub typhus, Japanese spotted fever and severe fever with thrombocytopenia syndrome in Japan as a biological negative-control setting, 2013–2025
Article References: Kutsuna, S. (2026). Neighbor-associated components in endemic–epidemic surveillance models: scrub typhus, Japanese spotted fever and severe fever with thrombocytopenia syndrome in Japan as a biological negative-control setting, 2013–2025. BMC Infectious Diseases. https://doi.org/10.1186/s12879-026-14800-7
Image Credits: AI Generated
DOI: 10.1186/s12879-026-14800-7
Keywords: endemic-epidemic models, surveillance modeling, spatial epidemiology, scrub typhus, Japanese spotted fever, severe fever with thrombocytopenia syndrome, zoonotic infections, negative control, statistical dependence, spatial coupling, Japan prefectures, time-series analysis, negative-binomial models, disease surveillance, epidemiological methods, transmission modeling, baseline heterogeneity
News Source: Phoebe Ingram. (November 12, 2026). Surveillance Model Artifacts Can Mimic Geographic Spread in Negative Control Zoonoses. Scienmag.



