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

Endemic–epidemic time-series models partition surveillance counts into endemic, within-area and neighbor-associated components; for directly transmitted diseases a large neighbor-associated component is often read as evidence of geographic spread. Scrub typhus, Japanese spotted fever and severe fever with thrombocytopenia syndrome are acarine-borne zoonoses in which sustained person-to-person transmission does not occur, so they provide a biologically motivated negative-control setting for interpreting short-lag associations between human notifications as transmission between areas. We analyzed weekly notifications of the three infections for the 47 prefectures of Japan, 2013–2025, fitting negative-binomial endemic–epidemic models under five area-baseline specifications, from a single national intercept to prefecture-specific effects. Simulations crossed baseline structure and coupling strength, with identical generating and fitted spatial operators in the principal reference grid; disease-calibrated simulations used row-normalized adjacency for generation and binary adjacency for fitting, matching the empirical specification. The neighbor-associated fraction under a single national intercept was 0.235 (approximate 95% interval 0.211–0.259) for Japanese spotted fever, 0.151 (0.132–0.172) for scrub typhus and 0.149 (0.120–0.180) for severe fever with thrombocytopenia syndrome; under prefecture-specific baselines these fell to below 0.001, 0.063 and 0.002. In simulation with coupling set to zero, spatially smoothed baseline heterogeneity produced mean fractions of 0.083 under the common-intercept estimator in the reference grid and 0.027–0.060 in simulations calibrated to each disease—18–40% as large as the empirical estimates—and four additional zero-coupling mechanisms yielded mean fractions of 0.069–0.112 under the common-intercept estimator. With genuine coupling (true fraction 0.142) the common-intercept estimator returned 0.224 whereas prefecture-specific baselines returned 0.141 when the fitted spatial operator matched the generating one. The estimated neighbor-associated fraction changed substantially with the specification of background risk, shared temporal structure and spatial weights; appreciable fractions arose with no between-area coupling, and recovery of genuine coupling depended on alignment between generating and fitted spatial operators. Neighbor-associated fractions should be treated as specification-dependent summaries of statistical dependence, reported across alternative baselines and weights and corroborated with independent data before being interpreted as transmission.